text
stringlengths
27.9k
483k
the model of quantum lattice bose gas has a long history and has been suggested initially for conventional superconductors @xcite , quantum crystals such as @xmath4he where superfluidity coexists with a crystalline order.@xcite afterwards , the bose - hubbard ( bh ) model has been studied as a model of the superconductor - insulator transition in materials with local bosons , bipolarons , or preformed cooper pairs.@xcite two - dimensional bh models have been addressed as relevant to describe the superconducting films and josephson junction arrays . the most recent interest to the system of hard - core bosons comes from the delightful results on bose - einstein ( be ) condensed atomic systems produced by trapping bosonic neutral atoms in an optical lattice . @xcite the progress in boson physics generates an especial interest around nonlinear topological excitations ( skyrmions , vortices ) which play an increasingly significant role in physics . one of the fundamental hot debated problems in bosonic physics concerns the evolution of the charge ordered ( co ) ground state of 2d hard - core bh model ( hc - bh ) with a doping away from half - filling . numerous model studies steadily confirmed the emergence of `` supersolid '' phases with simultaneous diagonal and off - diagonal long range order while penrose and onsager @xcite were the first showing as early as 1956 that supersolid phases do not occur . the most recent quantum monte - carlo ( qmc ) simulations @xcite found two significant features of the 2d bose - hubbard model with a screened coulomb repulsion : the absence of supersolid phase at half - filling , and a growing tendency to phase separation ( co+bs ) upon doping away from half - filling . moreover , batrouni and scalettar @xcite studied quantum phase transitions in the ground state of the 2d hard - core boson hubbard hamiltonian and have shown numerically that , contrary to the generally held belief , the most commonly discussed `` checkerboard '' supersolid is thermodynamically unstable and phase separates into solid and superfluid phases . the physics of the co+bs phase separation in bose - hubbard model is associated with a rapid increase of the energy of a homogeneous co state with doping away from half - filling due to a large `` pseudo - spin - flip '' energy cost . hence , it appears to be energetically more favorable to `` extract '' extra bosons ( holes ) from the co state and arrange them into finite clusters with a relatively small number of particles . such a droplet scenario is believed to minimize the long - range coulomb repulsion . however , immediately there arise several questions . whether a simple mean - field approximation ( mfa ) and classical continuum model can predict such a behavior ? what is the detailed structure of the co+bs phase separated state ? what are the main factors governing the essential low - energy and long - wavelength physics ? is it possible to make use of simple toy models ? are there the analogies with fermion hubbard model ? the behavior of the latter under doping away from half - filling is extensively studied in last decade in frames of stripe scenario which implies that doping may proceed through the formation of stripes , or charged domain walls being specific topological solitons in nel antiferromagnets . in particular , emery and kivelson @xcite argued that the phase separation reflects a universal tendency of the correlated antiferromagnet to expel the doped holes . white and scalapino @xcite showed that the pure 2d t - j model , in the small - j / t regime and with dopings near @xmath5 , has a striped ground state . in general , the topological phase separation is believed to be a generic property of 2d fermion hubbard model . in the paper we present a topological scenario of co+bs phase separation in 2d hc - bh model with inter - site repulsion . the extra bosons / holes doped to a checkerboard co phase of 2d boson system are believed to be confined in the ring - shaped domain wall of the skyrmion - like topological defect which looks like a bubble domain in an easy - axis antiferromagnet . this allows us to propose the mechanism of 2d _ topological _ co+bs _ phase separation _ when the doping of the bare checkerboard co phase results in the formation of a multi - center topological defect , which simplified pseudo - spin pattern looks like a system of bubble co domains with bose superfluid confined in charged ring - shaped domain walls . the rest of the paper is organized as follows . in sec.ii we give a short overview of the conventional hc - bh model in frames of a pseudo - spin formalism and mfa description . in sec.iii we show that the doping in 2d hc - bh system can be accompanied by the formation of a topological defect like a single or multi - center skyrmion . we present the scenario of the essential low - energy physics for the topologically doped hc - bh system . in sec.iv we address a simple model of nanoscopic bubble - like domain in a checkerboard co phase for a discrete square lattice that allows us to demonstrate a subtle microscopic structure of such a center with a multi - component order parameter . the hamiltonian of hard - core bose gas on a lattice can be written in a standard form as follows : @xmath6 where @xmath7 is the projection operator which removes double occupancy of any site , @xmath8 , @xmath9 the chemical potential determined from the condition of fixed full number of bosons @xmath10 or concentration @xmath11 $ ] . the @xmath12 denotes an effective transfer integral , @xmath13 is an intersite interaction between the bosons . here @xmath14 are the pauli creation ( annihilation ) operators which are bose - like commuting for different sites @xmath15=0,$ ] if @xmath16 @xmath17=1 - 2n_i$ ] , @xmath18 ; @xmath19 is a full number of sites . it is worth noting that near half - filling ( @xmath20 ) one might introduce the renormalization @xmath21 , or neutralizing background , that immediately provides the particle - hole symmetry . the model of hard - core bose - gas with an intersite repulsion is equivalent to a system of spins @xmath22 exposed to an external magnetic field in the @xmath23-direction . for the system with neutralizing background we arrive at an effective pseudo - spin hamiltonian @xmath24 where @xmath25 , @xmath26 , @xmath27 , @xmath28 . in a linear approximation the hamiltonian for the coupling with an electromagnetic field reads as follows @xmath29_{z}-\frac{1}{2}(\phi _ { j}-\phi _ { i})^{2}({\hat { \bf s}_{i}}\cdot { \hat { \bf s}_{j}}));\ ] ] @xmath30 where @xmath31 is the vector - potential , and integration runs over line binding the @xmath32 and @xmath33 sites ; @xmath34_{z}=\frac{i}{2}(\hat b_{i}^{\dagger}\hat b_{j}-\hat b_{i}\hat b_{j}^{\dagger}),\ ] ] however , the pseudo - spins are assumed to lie in @xmath35 plane . then the boson current density operator one may represent to be a sum of the paramagnetic @xmath36_{z},\end{aligned}\ ] ] and diamagnetic @xmath37 terms , respectively . below we make use of a conventional two - sublattice mfa approach . for the description of the pseudospin ordering to be more physically clear one may introduce two vectors , the ferromagnetic and antiferromagnetic ones : @xmath38 where @xmath39 . for the plane where these vectors lie one may introduce two - parametric angular description : @xmath40 . the hard - core boson system in a two - sublattice approximation is described by two diagonal order parameters @xmath41 and two complex off - diagonal order parameters @xmath42 and @xmath43 . the complex superfluid order parameter @xmath44 is determined by the in - plane components of ferromagnetic vector : @xmath45 , @xmath46 being the length of the in - plane component of ferromagnetic vector . so , for a local condensate density we get @xmath47 . it is of interest to note that in fact all the conventional uniform @xmath48 states with nonzero @xmath49 imply simultaneous long - range order both for modulus @xmath50 and phase @xmath51 . the in - plane components of antiferromagnetic vector @xmath52 describe a staggered off - diagonal order . it is worth noting that by default one usually considers the negative sign of the transfer integral @xmath12 , that implies the ferromagnetic in - plane pseudospin ordering . the model exhibits many fascinating quantum phases and phase transitions . early investigations predict at @xmath48 charge order ( co ) , bose superfluid ( bs ) and mixed ( co+bs ) supersolid uniform phases with an ising - type melting transition ( co - no ) and kosterlitz - thouless - type ( bs - no ) phase transitions to a non - ordered normal fluid ( no).@xcite the detailed mean - field and spin - wave analysis of the uniform phases of 2d hc - bh model is given by pich and frey.@xcite mfa yields for the conventional uniform supersolid phase @xcite @xmath53 with a constant chemical potential @xmath54 . it should be noted that the supersolid phase appears to be unstable with regard to small perturbations in the hamiltonian . the mean - field energy per site of the uniform supersolid phase is written as follows : @xmath55 where @xmath56 . the cost of doping both for co and co+bs phase appears to be rather high as compared with an easy - plane bs phase at half - filling where the chemical potential turns into zero.@xcite magnetic analogy allows us to make unambiguous predictions as regards the doping of bh system away from half - filling . indeed , the boson / hole doping of checkerboard co phase corresponds to the magnetization of an antiferromagnet in @xmath23-direction . in the uniform easy - axis @xmath57-phase of anisotropic antiferromagnet the local spin - flip energy cost is rather big . in other words , the energy cost for boson / hole doping into checkerboard co phase appears to be big due to a large contribution of boson - boson repulsion . however , the magnetization of the anisotropic antiferromagnet in an easy axis direction may proceed as a first order phase transition with a `` topological phase separation '' due to the existence of antiphase domains.@xcite the antiphase domain walls provide the natural nucleation centers for a spin - flop phase having enhanced magnetic susceptibility as compared with small if any longitudinal susceptibility thus providing the advantage of the field energy . namely domain walls would specify the inhomogeneous magnetization pattern for such an anisotropic easy - axis antiferromagnet in relatively weak external magnetic field . as concerns the domain type in quasi - two - dimensional antiferromagnet one should emphasize the specific role played by the cylindrical , or bubble domains which have finite energy and size . these topological solitons have the vortex - like in - plane spin structure and often named `` skyrmions '' . the classical , or belavin - polyakov ( bp ) skyrmion @xcite describes the solutions of a non - linear @xmath58-model with a classical isotropic 2d heisenberg hamiltonian and represents one of the generic toy model spin textures . it is of primary importance to note that skyrmion - like bubble domains in easy - axis layered antiferromagnets were actually observed in the experiments performed by waldner,@xcite that were supported later by different authors ( see , e.g. refs . ) . although some questions were not completely clarified and remain open until now , @xcite the classical and quantum @xcite skyrmion - like topological defects are believed to be a genuine element of essential physics both of ferro- and antiferromagnetic 2d easy - axis systems . the magnetic analogy seems to be a little bit naive , however , it catches the essential physics of doping the hc - bh system . as regards the checkerboard co phase of such a system , namely a finite energy skyrmion - like bubble domain @xcite seems to be the most preferable candidate for the domain with antiphase domain wall providing the natural reservoir for extra bosons . the classical description of nonlinear excitations in hc - bh model includes the skyrmionic solution given @xmath59 . @xcite the skyrmion spin texture consists of a vortex - like arrangement of the in - plane components of ferromagnetic @xmath60 vector with the @xmath57-component reversed in the centre of the skyrmion and gradually increasing to match the homogeneous background at infinity . the simplest spin distribution within classical skyrmion is given as follows @xmath61 @xmath62 where @xmath63 is a global phase ( @xmath64 order parameter ) , @xmath65 skyrmion radius . the skyrmion spin texture describes the coexistence and competition of the staggered charge order parameter @xmath57 and bs order parameter @xmath46 ( see fig.[fig1 ] ) that reflects a complex spatial interplay of potential and kinetic energies . ; b ) the radial distribution of the modulus of the superfluid order parameter @xmath66 ; c ) the radial distribution of the charge density @xmath67 for the charged skyrmion ; d ) the orientation of the superfluid order parameter @xmath66 given two values of the phase order parameter @xmath68 : @xmath69 , respectively . rings in all the pictures correspond to the skyrmion ( bubble domain ) wall , the dashed circle being its center.,width=321 ] the skyrmion looks like a bubble domain in an easy - axis magnet . it should be noted that the domain wall in such a bubble domain somehow created in the checkerboard co phase of 2d hc - bh system represents an effective ring - shaped reservoir both for bose - superfluid and extra boson / hole . indeed , the soliton energy depends quadratically on the number @xmath70 of bosons bound in domain wall,@xcite similarly to that of homogeneous bs phase.@xcite in other words , one might say about a zero value of the effective boson / hole chemical potential for the co bubble domain configuration , provided it were a ground state . the numerical calculations performed in frames of a classical continuous model @xcite show that the doped bosons appear to be trapped inside the bubble domain wall . the spatial distribution of the doped boson / hole density ( @xmath71 ) in the charged skyrmion is shown schematically in fig . skyrmions are characterized by the magnitude and sign of its topological charge , by its size ( radius ) , and by the global orientation of the spin . the scale invariance of classical bp skyrmionic solution reflects in that its energy does not depend on radius , and global phase . an interesting example of topological inhomogeneity is provided by a multi - center bp skyrmion @xcite which energy does not depend on the position of the centers . the latter are believed to be addressed as an additional degree of freedom , or positional order parameter . in the continuous model the classical bp skyrmion is a topological excitation and can not dissipate . however , the classical static skyrmion is unstable with regard to an external field or anisotropic interactions both of easy - plane and easy - axis type . small easy - axis anisotropy or external field are sufficient to shrink skyrmion to a nanoscopic size when magnetic length @xmath72 : @xmath73 is of the order of several lattice parameters , and the continuous approximation fails to correctly describe excitations . nonetheless , abanov and pokrovsky @xcite have shown that the easy - axis anisotropy together with fourth - order exchange term can stabilize skyrmion with radius @xmath74 a skyrmionic scenario in hc - bh model allows us to make several important predictions . away from half - filling one may anticipate the nucleation of a topological defect in the unconventional form of the multi - center skyrmion - like object with ring - shaped bose superfluid regions positioned in an antiphase domain wall separating the co core and co outside of the single skyrmion . the specific spatial separation of bs and co order parameters that avoid each other reflects the competition of kinetic and potential energy . such a _ topological _ ( co+bs ) _ phase separation _ is believed to provide a minimization of the total energy as compared with its uniform supersolid counterpart . thus , the parent checkerboard co phase may gradually lose its stability under boson / hole doping , while a novel topological self - organized texture is believed to become stable . the most probable possibility is that every domain wall accumulates single boson , or boson hole . then , the number of centers in a multi - center skyrmion nucleated with doping has to be equal to the number of bosons / holes . in such a case , we anticipate the near - linear dependence of the total bs volume fraction on the doping . generally speaking , one may assume scenario when the nucleation of a multi - center skyrmion occurs spontaneously with no doping . in such a case we should anticipate the existence of neutral multi - center skyrmion - like object with equal number of positively and negatively charged single skyrmions . however , in practice , namely the boson / hole doping is likely to be a physically clear driving force for a nucleation of a single , or multi - center skyrmion - like self - organized collective mode in the form of multi - center charged topological defect which may be ( not strictly correctly ) referred to as multi - skyrmion system akin in a quantum hall ferromagnetic state of a two - dimensional electron gas.@xcite in such a case , we may characterize an individual skyrmion by its position ( i.e. , the center of skyrmionic texture ) , its size ( i.e. , the radius of domain wall ) , and the orientation of the in - plane components of pseudo - spin ( u(1 ) degree of freedom ) . an isolated skyrmion is described by the inhomogeneous distribution of the co parameter , or staggered boson density @xmath57 , order parameter @xmath75 characterizing the deviation from the half - filling , and @xmath46 that corresponds only to the modulus of the superfluid order parameter . it seems likely that for a light doping any doped particle ( boson / holes ) results in a nucleation of a new single - skyrmion state , hence its density changes gradually with particle doping . therefore , as long as the separation between skyrmionic centers is sufficiently large so that the inter - skyrmion interaction is negligible , the energy of the system per particle remains almost constant . this means that the chemical potential of a boson or hole remains unchanged with doping and hence apparently remains fixed . the multi - skyrmionic system in contrast with uniform ones can form the structures with inhomogeneous long - range ordering of the modulus of the superfluid order parameter accompanied by the non - ordered global phases of single skyrmions . such a situation resembles in part that of granular superconductivity . in the long - wavelength limit the off - diagonal ordering can be described by an effective hamiltonian in terms of u(1 ) ( phase ) degree of freedom associated with each skyrmion . such a hamiltonian contains a repulsive , long - range coulomb part and a short - range contribution related to the phase degree of freedom . the latter term can be written out in the standard for the @xmath76 model form of a so - called josephson coupling @xmath77 where @xmath78 are global phases for skyrmions centered at points @xmath79 , respectively , @xmath80 josephson coupling parameter . namely the josephson coupling gives rise to the long - range ordering of the phase of the superfluid order parameter in a multi - center skyrmion . such a hamiltonian represents a starting point for the analysis of disordered superconductors , granular superconductivity , insulator - superconductor transition with @xmath81 array of superconducting islands with phases @xmath82 . calculating the phase - dependent part of skyrmion - skyrmion interaction timm _ et al._@xcite arrived at @xmath83 sign of @xmath80 that favors antiparallel alignment of the u(1 ) pseudospins . in other words , two skyrmions are believed to form a peculiar josephson @xmath84 micro - junction . there are a number of interesting implications that follow directly from this result:@xcite the spontaneous breaking of time - reversal symmetry with non - zero supercurrents and magnetic fluxes in the ground state , long - time tails in the dynamics of the system , unconventional aharonov - bohm period @xmath85 , negative magnetoresistence . to account for coulomb interaction and allow for quantum corrections we should introduce into effective hamiltonian the charging energy @xcite @xmath86 where @xmath87 is a boson number operator for bosons bound in @xmath32-th skyrmion ; it is quantum - mechanically conjugated to @xmath51 : @xmath88 , @xmath89 stands for the capacitance matrix , @xmath90 for bosonic charge . the classical ground state energy of skyrmion lattice for all reasonable two - dimensional lattice structures was minimized by timm _ _ taking the u(1 ) order into account . @xcite besides the expected triangular lattice with frustrated antiferromagnetic u(1 ) order and square lattice with nel u(1 ) order the authors have also obtained ground - state energies for all 2d bravais lattices . such a system appears to reveal a tremendously rich quantum - critical structure . in the absence of disorder , the @xmath48 phase diagram of the multi - skyrmion system implies either triangular , or square crystalline arrangements ( skyrmion , or bubble crystal ) with possible melting transition to a skyrmion ( bubble ) liquid . the melting of skyrmion lattice is successfully described applying the berezinsky - kosterlitz - thouless ( bkt ) theory to dislocations and disclinations of the lattice and proceeds in two steps . the first implies the transition to a liquid - crystal phase with short - range translational order , the second does the transition to isotropic liquid . disorder pins the skyrmion lattice and also causes the crystalline order to have a finite correlation length . for such a system provided the skyrmion positions fixed at all temperatures , the long - wave - length physics would be described by an @xmath76 model with expectable bkt transition and gapless @xmath76 spin - wave mode . the classical phase diagram of skyrmion lattice is quite rich . depending on the relation between coulomb and josephson coupling , and density one may arrive at different lattice structures with continuous or first order phase transitions . as regards the superfluid properties the skyrmionic system reveals unconventional behavior with two critical temperatures @xmath91 and @xmath92 , @xmath93 being the temperature of the ordering of the modulus and @xmath94 that of the phase of order parameter @xmath95 . the low temperature physics in skyrmion crystals is governed by an interplay of two bkt transitions , for the u(1 ) phase and positional degrees of freedom , respectively . @xcite dislocations in most skyrmion lattice types lead to a mismatch in the u(1 ) degree of freedom , which makes the dislocations bind fractional vortices and lead to a coupling of translational and phase excitations . both bkt temperatures either coincide ( square lattice ) or the melting one is higher ( triangular lattice).@xcite quantum fluctuations can substantially affect these results . quantum melting can destroy u(1 ) order at sufficiently low densities where the josephson coupling becomes exponentially small . similar situation is expected to take place in the vicinity of structural transitions in skyrmion crystal . with increasing the skyrmion density the quantum effects result in a significant lowering of the melting temperature as compared with classical square - root dependence . the resulting melting temperature can reveal an oscilating behavior as a function of particle density with zeros at the critical ( magic ) densities associated with structural phase transitions . in terms of our model , the positional order corresponds to an incommensurate charge density wave , while the u(1 ) order does to a superconductivity . in other words , we arrive at a subtle interplay between two orders . the superconducting state evolves from a charge order with @xmath96 , where @xmath97 is the temperature of a melting transition which could be termed as a temperature of the opening of the insulating gap ( pseudo - gap ! ? ) . the normal modes of a dilute skyrmion system ( multi - center skyrmion ) include the pseudo - spin waves propagating in - between the skyrmions , the positional fluctuations , or phonon modes of the skyrmions which are gapless in a pure system , but gapped when the lattice is pinned , and , finally , fluctuations in the skyrmionic in - plane orientation and size . the latter two types of fluctuation are intimitely connected , since the @xmath23-component of spin and orientation are conjugate coordinates because of commutation relations of quantum angular momentum operators . so , rotating a skyrmion changes its size . the orientational fluctuations of the multi - skyrmion system are governed by the gapless @xmath76 model.@xcite the relevant model description is most familiar as an effective theory of the josephson junction array . an important feature of the model is that it displays a quantum - critical point . the low - energy collective excitations of skyrmion liquid includes an usual longitudinal acoustic phonon branch . the liquid crystal phases differ from the isotropic liquid in that they have massive topological excitations , _ i.e. _ , the disclinations . one should note that the liquids do not support transverse modes , these could survive in a liquid state only as overdamped modes . so that it is reasonable to assume that solidification of the skyrmion lattice would be accompanied by a stabilization of transverse modes with its sharpening below melting transition . in other words an instability of transverse phonon modes signals the onset of melting . a generic property of the positionally ordered skyrmion configuration is the sliding mode which is usually pinned by the disorder . the depinning of sliding mode(s ) can be detected in a low - frequency and low - temperature optical response . our speculations as regards the topological phase separation and long wave - length physics of 2d hc - bh model are summarized in a qualitative @xmath98-t ( @xmath99 ) phase diagram in fig.[fig3 ] . first of all , the phase diagram implies a scenario in which the topological phase separation state evolves with a minimal doping , though it is worth noting that the ideal checkerboard co phase can not be a nominal @xmath100 limit of any topologically phase separated phase . despite the qualitative character of phase diagram we took into account some quantitative results of quantum monte - carlo calculations for the 2d hc - bh model with nearest neighbor coulomb repulsion.@xcite first , it concerns the doping range of additive , or near - linear concentration behavior of bose - condensate density @xmath101 and co structure factor @xmath102 . the superfluid density increases approximately linearly with doping except for the most overdoped point @xmath103 , where it turns down again . diagonal long - range charge order is characterized by the equal time structure factor at the ordering vector @xmath104 , @xmath105 in the presence of long - range order @xmath102 will diverge with the system size @xmath106 for a given ordering momentum @xmath107 , which characterizes the ordered phase . for the checkerboard order @xmath108 . in the concentration range @xmath109 , where both quantities vary linearly with @xmath110,@xcite we may approximate the topological defect to be a system of @xmath111 interacting single - charged skyrmions . it should be noted that the both co and bs order parameters coexist in a rather narrow doping concentration interval : @xmath112 . beyond this `` overdoping '' region we deal with an inhomogeneous boson liquid which pseudo - spin picture implies a strongly frustrated singlet - triplet system that resembles the spin glass , and maybe termed as a dynamical `` singlet - triplet '' pseudo - spin glass . such a system can be characterized by a dynamical short - range diagonal and off - diagonal ordering with a wide distribution of respective correlation lengths and relaxation rates . interestingly , that in frames of a continuous model this phase is still described to be a strongly correlated multi - center topological defect . however , such a model fails obviously to describe the real system where the inter - center spacing is comparable with the lattice parameter . the temperature evolution of hc - bh system with large inter - site boson repulsion implies the highest critical temperature @xmath113 separating the high - temperature non - ordered no phase ( boson liquid ) and a low - temperature quasi - co phase with a disordered system of skyrmions . the next critical temperature @xmath114 points to the first order phase transition with a formation of inhomogeneous bose condensate in a single skyrmion with the vortex - like texture of the quasi - local order parameter @xmath115 . in other words , it is a temperature of the topological co+bs phase separation ( tps ) . in frames of our scenario such tps state emerges with a minimal doping , and @xmath116 is likely to be nearly constant in a linear doping regime ( @xmath117 ) . the low - temperature part of the phase diagram , which is schematically shown in fig . [ fig3 ] , describes the gradual skyrmionic solidification and may include the isotropic skyrmion(bubble ) liquid phase , the skyrmion liquid - crystal phase , the incommensurate skyrmion crystal phase , and u(1 ) ordering , or superfluid phase , restricted by the temperatures of the proper bkt transitions . for a small deviation from half - filling ( `` under - doping '' ) the temperatures of bubble crystallization / melting are seemingly to obey the square root concentration law @xmath118 with a strongly developed quantum melting effect when approaching an `` overdoped '' regime or concentrations limiting the linear regime . the superfluid phase in fig.[fig3 ] is arbitrarily chosen to lie inside the skyrmion solid phase . one should emphasize the specific role played by quantum fluctuations : these lead to the melting of the bubble crystal at high densities and orientational disordering @xcite at low densities . both effects are of primary importance in the overdoped and heavily underdoped regions of the phase diagram , respectively . moreover , quantum melting effect may strongly affect the phase diagram near the `` magic '' doping level , where the skyrmion lattice undergoes the structural phase transition . for illustration , in the right hand side panel in fig . [ fig3 ] we present the possible phase diagram for a hc - bh system with a quantum melting effect near the `` magic '' doping level @xmath119 . in all the cases , the critical doping level for the superfluid formation is determined by the magnitude of the josephson coupling constant . it is worth noting that the bubble crystallization is accompanied by different ( pseudo)gap effects . our interpretation of the phase transition at @xmath120 differs from that in ref .. this temperature is governed by the magnitude of transfer integral , and believed to be a temperature of the emergence of the nonzero magnitude of the modulus of the superconducting order parameter rather than a critical temperature for an insulator to superconductor transition as it is stated in ref .. this conclusion seems to be a result of finite size effects and boundary conditions in qmc calculations despite the authors make use of the most efficient qmc strategy . such problems seem to be typical for a finite - size simulation of many phase transitions . in addition , we should note that mc calculations need substantial increase in lattice size to reproduce quantitatively the long - wavelength physics because the size of skyrmion and skyrmion lattice parameter appear to be new characteristic lengths . the close inspection of the phase diagram in fig.[fig3 ] , where @xmath121 , does not provide the optimistic expectations as regards the high - temperature superconductivity in 2d hc - bh systems , even if the magnitude of the local boson transfer integral were as large as @xmath122 k. nevertheless , the attractively large temperatures @xmath120 of the topological phase separation with the emergence of nonzero local condensate density engender different reasonable speculations as regards its practical realization . the making use of the mean - field approximation together with simplified classical continuous models can hardly provide the quantitative description of a quantum lattice boson system . both quantum effects and the discreteness of skyrmion texture can result in substantial deviations from the predictions of classical model . the continuous model is relevant for discrete lattices only if we deal with long - wave length inhomogeneities when their size is much bigger than the lattice spacing . in the discrete lattice the very notion of topological excitation seems to be inconsistent . at the same time , the discreteness of the lattice itself does not prohibit from considering the nanoscale ( pseudo)spin textures whose topology and spin arrangement is that of a skyrmion.@xcite naturally , the typical continuous models fail to describe properly such short - wave length nanoscopic inhomogeneities . hereafter , we discuss a simple model which seems to catch the main effects of discreteness and quantization . what is the lattice counterpart of the small skyrmion - like bubble domain ? fig . [ fig4 ] presents the schematic view of such a smallest skyrmion - like bubble domain in a checkerboard co phase for 2d square lattice with an effective size of 4 lattice spacings . the domain wall is believed to include as a minimum eight sites forming a ring - shaped system of four dimers each composed of two sites . there are two types of such domains which differ by a rotation by @xmath123 . the formation of such a center seems not to require a big energy . indeed , the simple estimate of a change in potential energy : @xmath124 points to a near cancelation of @xmath0 and @xmath1 contributions . the remarkable feature of the domain is in a rather small magnitude both of scalar potential and electric field inside the 8-site domain wall ( see fig . [ fig4 ] ) . the flip - energy for the dimer dipole moment is estimated to be @xmath125 , compared with that for bare checkerboard co phase ( @xmath126 ) that implies rather subtle competition between @xmath0 and @xmath1 couplings . however , the difference @xmath127 is believed to be always positive and large . in other words , the flip - energy for the dimer dipole may be relatively small . respective dipole fluctuations would result in an effective screening of electrostatic repulsion energy thus providing a stabilization of a bubble - like defect . an important additional mechanism of the domain stabilization in the hc - bh systems with local bosons composed from electron pairs may arise from the electron and lattice polarization effects.@xcite as is well known , the respective energies are comparable in value with the intersite coulomb interaction . probably , namely both these effects might strongly contribute to the domain stabilization energy . on the other hand , the bubble geometry implies the formation of the electrostatic potential well inside the domain wall both for positive and negative charges . it means that the doping into a domain wall stabilizes the domain configuration . such a doping maybe energy costless , while the energy cost of the pseudo - spin - flip in a checkerboard co phase is rather high : @xmath128 , if the neutralizing background is taken into account . with taking into account the kinetic energy ( quantum tunneling ) only inside the 8-site domain wall we shall consider it as a quantum system in an external electrostatic potential field assuming a rigid checkerboard co ordering overall beyond the domain wall . further , taking into account the reasonable relation : @xmath129 , we shall consider the domain wall to be a system of four dimers , or pairs of nearest neighboring sites forming a quantum cluster . transfer integral . the arrows mark the dipole - allowed transitions ( see text for details).,width=321 ] we anticipate the model though being simplified would be very instructive analyzing the role played both by quantum and discreteness effects . let first address a simple two - site , or a dimer system . the effective pseudo - spin - hamiltonian for such a cluster , or a single dimer hamiltonian can be written as follows @xmath130\,+d { \hat s}_{z}^2\ , -\,h_s\,{\hat s}_{z}-\,h_t\,{\hat t}_{z } , \label{h}\ ] ] where @xmath131 is the total pseudo - spin momentum , @xmath132 is an operator that changes pseudo - spin multiplicity , @xmath133 , @xmath134 , @xmath135 . the first two terms in this effective pseudo - spin hamiltonian describe the intra - dimer interactions , while the last two ones describe its coupling with the off - domain - wall surroundings . it is worth noting that the condition @xmath134 means that the effective magnetic field produced by this surroundings turns into zero that provides the particle - hole symmetry of the dimer physics , in particular , for the domain wall doping . actually , we arrive at an effective singlet - triplet model.@xcite both @xmath136 and @xmath137 operators have a rather simple physical sense : the former corresponds to the total `` quantum '' charge of the dimer , while the latter does to the total `` quantum '' dipole moment . strictly speaking , the diagonal order parameters @xmath138 and @xmath139 describe the charge and dipole density , while the off - diagonal order parameters @xmath140 and @xmath141 describe the corresponding phase ordering . it is of primary importance to note that these order parameters are not independent because the respective operators obey the kinematic constraint : @xmath142 stemmed from a simple spin algebra . it should be noted that there are two operators : @xmath143 and @xmath144 $ ] that change the pseudo - spin multiplicity with their matrices being symmetric and antisymmetric , respectively : @xmath145 @xmath146 the hamiltonian ( [ h ] ) points to the competition of s- and t - orders in the ring - shaped domain wall . fig.[fig5 ] shows a step - by - step formation of the energy spectrum of such a two - site cluster , or dimer . the arrows mark the dipole - allowed transitions that could be revealed in optical spectra . the tunnel states @xmath147 and @xmath148 describing purely `` half - filled '' dimer states are mixed due to the electric field with the mixing level governed by the ratio @xmath149 . if @xmath150 , we arrive at classical nel - like `` up - and - down '' dimer state . hereafter , to describe our singlet - triplet quantum pseudo - spin system we start with trial functions @xmath151 where the spin functions @xmath152 in cartesian basis are used : @xmath153 and @xmath154 . the conventional spin operator is represented on this basis by a simple matrix : @xmath155 and for the order parameters one easily obtains : @xmath156 , \langle\{\hat{s}_i\hat{s}_j\}\rangle=2(\delta_{ij}-a_ia_j - b_ib_j ) \label{med}\ ] ] given the normalization constraint @xmath157 . thus , for the case of spin-1 system the order parameters are determined by two classical vectors ( two real components of one complex vector @xmath158 from ( [ wf1 ] ) ) . the two vectors are coupled , so the minimal number of dynamic variables describing the @xmath159 spin system appears to be equal to four . for the dimer as the singlet - triplet center we have additional unconventional t , j - order parameters . for the respective averages we can easily obtain @xmath160 given the real value of @xmath161 parameter we arrive at a simple form:@xmath162 , and @xmath163 . now we may proceed with the inter - dimer coupling in the domain wall . introducing the symmetrized superpositions of s and t vectors , we can write the pseudo - spin hamiltonian of the inter - dimer interaction as follows @xmath164 @xmath165 + \frac{1}{2}(v_{nnn}-t_{nnn})[s_{z}^{2}(a_{1g})- s_{z}^{2}(b_{1 g } ) + \ ] ] @xmath166,\ ] ] where @xmath167 @xmath168 @xmath169 @xmath170 @xmath171 @xmath172 @xmath173 @xmath174 ( see fig.[fig4 ] ) . the chemical potential @xmath175 for the system of four dimers in a domain wall is determined now by the condition : @xmath176 for undoped and singly - doped ( boson / hole ) domain wall , respectively . the respective mean values we can address to be order parameters that describe the subtle structure of a domain wall : as regards the diagonal order @xmath177 specifies the full charge ; @xmath178 and @xmath179 do the electric dipole moments , @xmath180 the component of the quadrupole momentum , and one might introduce the higher order multipole moments . as regards the off - diagonal order we should , in general , proceed with the three types ( @xmath181 , or @xmath182 ) of the s - order , and three types ( @xmath183 , or @xmath184 ) of the t - order , which can be defined as follows : @xmath185 noteworthy to mention the kinematic constraint that couples different order parameters . in the absense of an external magnetic field the energy does not depend on @xmath186 vector , it is restricted only to lie in @xmath35-plane . in other words , we deal with the uncertainty of j - order parameter . the mean - field domain - wall ground state corresponds to the coexistent @xmath187 modes given the negative sign of the transfer integral @xmath188 or @xmath189 modes given the positive sign of the transfer integral @xmath188 . the relation between s- and t - mode weight is specified by the relationship between the @xmath0 transfer integral @xmath190 and electric field @xmath191 . if we assume @xmath192 , then given @xmath193 the dimers have purely @xmath159 ground state , and we arrive at s - type off - diagonal order . in contrast , given @xmath194 we arrive at t - type off - diagonal order . in general , neglecting the st - mixing term leads to two independent phase order parameters . generally speaking , the interference dipole - dipole st - term would result in st - mixing accompanied by the constraint on the phase order parameters with an appearance of a noncollinearity effect . thus we arrive at the conclusion that the symmetry of the order parameter distribution in the domain wall would be specified only by the sign of the transfer integral . in addition , we see that the problem of the order parameter associated with our bubble domain is much more complicated than in conventional bcs - like approach @xcite due to its multicomponent nature . moreover , we deal with a system with different symmetry of low - lying excited states and competing order parameters that implies their possible ambiguous manifestation in either properties . the low - symmetry crystalline electric fields or crystal distortions would result in a mixing of the order parameters with different symmetry . the bubble domain ( fig.[fig4 ] ) yields a simple and instructive toy model to describe such effects . it is worth noting that the flux quantization effects for the bubble domain , in particular , the localizing effect of the magnetic field on the moving bosons , are expected to be observed only for rather large fields . indeed , nanoscopic atomic systems such as a square plaquette with a size of around 10 @xmath195 require a field of @xmath196 tesla for a half flux quantum per plaquette . above we have concerned a simple one - center topological defect . however , in general one has to make use of more complicated topological excitation like a multi - center bp skyrmion.@xcite the question arises , whether such an entity may be described as a system of weakly coupled individual one - center defects ? in practice , it seems to be a rather reasonable approach only for slight deviation from half - filling , when the mean separation between doped particles is much larger than the effective domain size . the bubble domain may be addressed to be well isolated only if its environment does not contain another domain(s ) which could be involved to a `` dangerous '' nearest - neighbor inter - dimer coupling . each domain in fig.[fig4 ] has @xmath197 of such neighbors . hence the concentration of well isolated domains can be written as follows : @xmath198 where @xmath23 is a number of `` dangerous '' neighbors , @xmath199 the boson concentration . the @xmath200 maximum is reached at @xmath201 . in our case @xmath202 , or @xmath203 . with increasing doping the deviation of @xmath200 from the linear law rises . on the other hand , knowing the effective domain area @xmath204 we can roughly estimate the limiting concentration of the single - domain model description to be @xmath205 , @xmath206 . the inter - domain coupling includes both long - range ss- and short - range tt - terms . we should emphasize a strong anisotropy of this coupling . for nearest neighbors along [ 1,0 ] and [ 0,1 ] directions namely the short - range ( dipole - dipole ) tt - coupling would result in an effective suppression of the main s ordering ( see fig.[fig4 ] ) , in contrast with [ 11 ] direction . thus we may conclude that the above mentioned simple continuous model for josephson junction arrays should be strongly modified to describe the multi - domain configurations in 2d quantum hc - bh lattices , both as regards the subtle internal domain structure , the competition of different order parameter , and the anisotropy of josephson coupling . in conclusion , the boson / hole doping of the hard - core boson system away from half - filling is assumed to be a driving force for a nucleation of a multi - center skyrmion - like self - organized collective mode that resembles a system of co bubble domains with a bose superfluid and extra bosons both confined in domain walls . such a _ topological _ co+bs _ phase separation _ , rather than an uniform mixed co+bs supersolid phase , is believed to describe the evolution of hc - bh model away from half - filling . starting from the classical model we predict the properties of the respective quantum system . in frames of our scenario we may anticipate for the hc - bh model the emergence of an inhomogeneous bs condensate for superhigh temperatures @xmath207 , and 3d superconductivity for rather high temperatures @xmath208 . the system is believed to reveal many properties typical for granular superconductors , cdw materials , wigner crystals , and multi - skyrmion system akin in a quantum hall ferromagnetic state of a 2d electron gas . topological inhomogeneity is believed to be a generic property of 2d hard - core boson systems away from half - filling . such a behavior represents a boson counterpart of the so - called _ topological doping _ being a general feature of mott - insulator or 2d fermion hubbard model.@xcite despite all shortcomings , mfa and continuous approximation are expected to provide a physically clear semiquantitative picture of rather complex transformations taking place in bare co system with doping , and can be instructive as a starting point to analyze possible scenarios . first of all , the mfa analysis allow us to consider the antiphase domain wall in co phase to be a very efficient ring - shaped potential well for the localization of a single extra boson ( hole ) thus forming a novel type of a topological defect with a single - charged domain wall . such a defect can be addressed as a charged skyrmion - like quasiparticle which energy can be approximated by its classical value for co bubble domain . it is of great importance to note that domain wall simultaneously represents a ring - shaped reservoir for bose superfluid . unfortunately , we have no experience to deal with multi - center skyrmions as regards its structure , energetics , and stability . it should be noted that such a texture with strongly polarizable centers is believed to provide an effective screening of long - range boson - boson repulsion thus resulting in an additional self - stabilization . nucleation of topological phase is likely to proceed in the way typical for the first order phase transitions . the role played by quantum effects and lattice discreteness has been illustrated in frames of the simplest nanoscopic counterpart of the bubble domain in a checkerboard co phase of 2d hc - bh square lattice . it is shown that the relative magnitude and symmetry of multi - component order parameter are mainly determined by the sign of the @xmath0 and @xmath1 transfer integrals . the topologically inhomogeneous phase of the hc - bh system away from the half - filling can exhibit the signatures both of @xmath2 , and @xmath3 symmetry of the off - diagonal order . the model allows us to study the subtle microscopic details of the order parameter distribution including its symmetry in a real rather than momentum space , though the problem of the structure and stability of nanoscale domain configurations remains to be solved . the present paper establishes only the framework for analyzing the subtleties of the phase separation in a lattice hc - bh model away from half - filling . much work remains to be done both in a macroscopic and microscopic approaches . i acknowledge stimulating discussions with c. timm , s .- l . drechsler , t. mishonov , and the support by smwk grant , intas grant no . 01 - 0654 , crdf grant no . rec-005 , rme grant no . e 02 - 3.4 - 392 and no . ur.01.01.062 , rfbr grant no . 04 - 02 - 96077 . i would like to thank leibniz - institut fr festkrper- und werkstoffforschung dresden where part of this work was made for hospitality . special thanks are to referees for their remarks improving the manuscript . it is worth noting that in frames of bcs - like scenario the symmetry of the order parameter is strictly defined in a momentum space albeit the discussion of different experimental data has usually been performed with a real - space distributed order parameter .
ever since landau@xcite suggested that the electron can be trapped by the deformable lattice , strongly coupled electron - phonon systems have been the subject of intensive examination . besides the investigations of those systems in which the lattice is coupled to the whole electronic band , there has been significant interest in the physics of a single polaron , in which the electron and the associated lattice deformation form a quasi - particle , spatially and spectrally decoupled from the rest of the system . lattice degrees of freedom make even a single polaron problem a many - body one . the analytical and numerical examinations of most electron - phonon models are thus difficult . for this reason , even the simple holstein model@xcite ( suggested in 1959 ) is still being investigated in recent works . various methods have been proposed in order to calculate the polaron ground state of the holstein model . almost exact results ( except for the adiabatic limit ) have been obtained with the quantum monte carlo calculations,@xcite the global - local method,@xcite the density matrix renormalization group method,@xcite and some exact diagonalization methods.@xcite the main goal of this paper is to determine the elements important for the qualitative description of the polarons in the whole range of electron - phonon coupling . some of them , although already known , are found to be better understood when supplemented with additional details . the ground state of the holstein system changes from the delocalized polaron state , in which the electron is nearly free , to the small and self - trapped polaron state , as the electron - phonon coupling @xmath0 increases . these two opposite limits are usually identified as the weak and strong coupling regime , respectively . in the weak coupling regime the influence of the small lattice deformation on polaron dynamics is very small , which makes the energies of polaron and electron hopping to neighboring sites similar , @xmath1 . the exact ground state is an eigenstate of the system momentum , regardless of the coupling.@xcite therefore the polaron ground state is delocalized for all parameters . however , in the strong coupling regime @xmath2 becomes negligible , leading to _ self - trapped _ polaron states . according to ref . , the dynamics of the small self - trapped polaron can be separated into two time scales . on the short time scale , the lattice deformation is centered at some lattice site , and the electron can virtually hop among the neighboring lattice sites . only after a certain number of such events ( of the order @xmath3 ) , the polaron as a whole tunnels to a new central site . the localized polaron states thus may lead to a very accurate estimation of the polaron ground state energy in the strong coupling regime . accordingly , for the self - trapped polarons , an exact diagonalization method of calculating the localized polaron states , rather than the translationally invariant ones , can be used . by comparison to the results of other methods , it is shown that this approach introduces only minor errors in the ground state energy of the self - trapped polaron . moreover , the localized polaron functions permit , unlike the translationally invariant ones , a separate analysis of the electron and phonon properties of the polaron . the local electron density , the mean lattice deformation , and the on - site zero point motions of the self - trapped polarons can be calculated in this way . in the crossover ( intermediate ) regime , which is between the weak and the strong coupling regime , no known perturbation calculation converges , which complicates the discussion of the polaron nature . although it has been proved that the change of the polaron ground state with @xmath0 is smooth,@xcite the physics of the rapid crossover between two opposite limits of the electron - phonon coupling , in the small interval of @xmath0 s , is not completely clear . in ref . it has been claimed that the phonon excitation associated with the first excited state of the system is uncorrelated to the electron in the weak coupling regime , while it is confined to the electron in the strong coupling regime . the transition occurs in the crossover regime in which , in addition , the energy difference between the ground and excited states becomes small . in the present paper , the excited polaron states are treated by the new method which uses variational approach in order to define and solve the generalized eigenvalue problem . even if this method does not converge systematically , it does provide new results concerning the nature of the polaron ground and first excited state . the holstein hamiltonian reads @xmath4 it describes the tight - binding electrons in the nearest - neighbor approximation , coupled to one branch of dispersionless optical phonons . @xmath5 is the creation operator for the electron of spin @xmath6 at lattice site @xmath7 , and @xmath8 is the creation operator for the phonon . @xmath9 is the transfer ( hopping ) integral of the electron . @xmath10 and @xmath0 are the phonon and the electron - phonon coupling energies , respectively . the holstein hamiltonian depends only on two ratios of relevant energy parameters : @xmath11 and @xmath12 , i.e. , the results will use @xmath10 as the energy unit . it is often convenient to express the lattice vibrations in terms of the nuclei space and momentum coordinates , @xmath13 where @xmath14 and @xmath15 are space and momentum uncertainties of the harmonic oscillator ground state . @xmath16 is the mass of a nucleus and @xmath17 denotes the spring constant , @xmath18 . by noting that the electron - lattice displacement coupling constant @xmath19 , in @xmath20 , is independent of @xmath16 , an alternative set of holstein hamiltonian parameters can be introduced , @xmath21 which is convenient for the discussion of the adiabatic regime @xmath22 . here , @xmath23 is the electron effective mass and @xmath24 is the lattice constant . it is worth noting that @xmath9 and @xmath25 ( the binding energy of the polaron for @xmath26 ) are independent of @xmath16 and thus they are the only parameters relevant in the adiabatic limit . by using standard conventions for @xmath27 and @xmath28 , @xmath29 eq . ( [ h0 ] ) can be rewritten in momentum space and separated in two mutually commuting parts , @xmath30 , @xmath31 where @xmath32 , @xmath33 , and @xmath34 . the part which involves only the @xmath35 phonon mode , @xmath36 , can easily be transformed into the diagonal form , by using the unitary operator @xmath37 , @xmath38 where @xmath39 . any eigenstate of @xmath40 , @xmath41 , has the same mean total lattice deformation , @xmath42 , @xmath43 which is independent of @xmath9 . since the @xmath35 part of the hamiltoniancommutes with @xmath44 , it can be concluded that eq . ( [ xtot ] ) is also valid for all eigenstates of the total holstein hamiltonian . the phonon part of these eigenstates can be represented in the form of a direct product of two groups of states , the first one includes @xmath45 phonon modes , while the second includes only the @xmath35 phonon mode . this is useful because one can always check approximate computations by calculating @xmath46 , or include this property in the computation itself . this specific property of the @xmath35 mode is not restricted to the particular dimension of the system , nor to the number of electrons . moreover , it can also be found in some other models in which the electron - phonon coupling consists of the lattice deformation linearly coupled to the local electron density . a hint in this direction was reported for the first time in ref . . the total momentum of the system , @xmath47 , is the sum of the electron and phonon momenta , @xmath48 and it commutes with the hamiltonian . in the present treatment , only the low energy polaron states ( the low lying states of the system for which the electron and lattice part of the wave function are spatially bound ) , are explicitly calculated . in this case , the total momentum @xmath49 of the system is also the polaron momentum . the eigenstate computations reported here are based on the variational approach . still , from the physical and mathematical point of view , there are significant differences among them . each method is therefore described separately in the present section . it is important to notice that two qualitatively different kinds of polaron functions are employed , the _ localized _ ones and the _ translationally invariant _ ones . the primary objective of the methods with localized functions are self - trapped polaron states , i.e. , the strong coupling regime . on the other hand , translationally invariant states are devised in the first place with the weak coupling and crossover regime in mind . the methods with a small number of variational parameters are meant to help in understanding the basic properties of the polarons in different regimes of the holstein model . again , the methods with a very large number of variational parameters are necessary to obtain accurate results for the polaron states . let us start with a simple localized polaron wave function formed as a product of the electron and the lattice part , centered at the lattice site @xmath51 , @xmath52 here , @xmath53 is the normalized electron function at site @xmath54 , @xmath55 , while @xmath56 denotes the coherent state operator acting on site @xmath57 , with a complex amplitude , @xmath58 , @xmath59 it is easy to see that operator @xmath56 shifts the space and momentum coordinates of lattice vibration at a site @xmath57 by @xmath60 and @xmath61 , respectively , @xmath62 } e^{[-i2\re(\xi_m)x_0\hat p_{j+m}/\hbar]}\\ & \times&e^{[i\re(\xi_m)\im(\xi_m)]}\;.\end{aligned}\ ] ] the variational energy of the state ( [ iafunction ] ) , @xmath63 , is independent of @xmath51 , and is given by @xmath64 the minimization of the energy with respect to @xmath65 establishes a simple relationship between the lattice mean deformation and electron density @xmath66 , @xmath67 so that only the equation for @xmath53 has to be solved . the well known approximate solution to this problem is the large holsteinpolaron@xcite valid in the long wave limit , @xmath68 the numerical scheme suggested in ref . , and denoted here by @xmath50 ( @xmath50 for localized ) , is not restricted to long waves , and is used here in order to obtain the exact minimum of eq . ( [ iaenergy ] ) . the energy , henceforth referred to as @xmath69 , depends only on two relevant hamiltonianparameters , @xmath9 and @xmath70 , and therefore @xmath69 and @xmath71 are both independent of @xmath16 . for this reason eq.([locking ] ) is sometimes referred to as the adiabatic locking of electron and lattice coordinates . it is noteworthy that the lattice part of the function @xmath72 is a simple product of coherent states with real amplitudes . consequently , the state of the lattice at some site is defined by the ground state of the displaced harmonic oscillator . the mean lattice deformation @xmath71 corresponds to the equilibrium position of that oscillator , while the lattice zero point motion is approximated in eq . ( [ iafunction ] ) by that of the free lattice . next , we shall study a translationally invariant solution composed of a linear superposition of the localized states ( [ iafunction ] ) , @xmath75 @xmath76 describes the polaron state with the momentum @xmath49 . a similar type of function was first proposed by toyozawa.@xcite in the present work , @xmath77 is used so that the mean total deformation of function ( [ tivfunction ] ) satisfies eq . ( [ xtot ] ) , @xmath78 the expectation value of the polaron energy , @xmath79 , may be written in terms of @xmath80 , @xmath81 @xmath82 and @xmath83 are given in the appendix . a simple method for calculating the minimum of the energy @xmath79 has not yet been proposed , but accurate results have been obtained in ref . by using the toyozawa method , which includes a very large number of variational parameters . some additional approximations may be found in refs . and . the approximation used here simplifies the general expression in eq.([tivfunction ] ) by introducing the exponential form for functions @xmath53 and @xmath84,@xcite @xmath85 equation ( [ ve ] ) defines a polaron function @xmath86 which is completely determined by two parameters , @xmath87 and @xmath88 . in what follows , two different approaches are presented . the first , denoted by the index @xmath73 ( @xmath73 for translational ) , treats @xmath87 and @xmath88 as the variational parameters for which the energy minimum @xmath89 has to be found , and its corresponding polaron function is @xmath90 . the @xmath73 method gives good results in the weak and the strong coupling regime . namely , in both of these limits , the function @xmath90 becomes similar to the polaron function obtained by the appropriate perturbative calculations . in the second approach , presented here for the first time , the variational method is used to define a generalized eigenvalue problem as follows . the polaron wave function , denoted by the index @xmath74 ( @xmath74 for combination of translational functions ) , is rewritten as a linear combination of functions @xmath91 , @xmath92 it is understood here that the functions @xmath93 form a set of @xmath94 generally nonorthogonal functions , defined by @xmath94 different pairs of parameters @xmath95 . again , the coefficients @xmath96 should be determined from the requirement that the expectation value of the energy , @xmath97 is minimal , @xmath98 or , @xmath99 the solution of this generalized eigenvalue problem is a set of @xmath94 orthogonal polaron functions , @xmath100 , with corresponding energies @xmath101 . the ground state energy is @xmath102 . one may always include the function @xmath103 in the sum ( [ lvfunction ] ) in order to ensure that the energy @xmath102 is the same or better than @xmath89 , the energy computed by the @xmath73 method . moreover , by paying some further attention to the starting set of functions @xmath91 in eq.([lvfunction ] ) at the outset , one is able to investigate the first excited state @xmath104 of the system , when this state corresponds to an excited polaron . the best results for the @xmath74 method are obtained when the number @xmath94 of @xmath91 functions in eq . ( [ tivfunction ] ) changes with the hamiltonian parameters . the special case where the @xmath74 method is used with constant @xmath105 is denoted by the index @xmath106 . finally , this paper presents the results of two numerical exact diagonalization methods.@xcite in order to compute the low energy polaron states , one approximates the infinite dimensional hamiltonian matrix with a finite one , and proceeds with the exact diagonalization of this matrix . the lowest eigenvalue and eigenvector in such a reduced hilbert space correspond to the polaron energy and wave function , respectively . for a large sparse matrix , the energy and the wave function can be calculated very accurately , by using an appropriate numerical scheme , in the present case the lanczos algorithm . the two exact diagonalization methods used here differ in the choice of the basis of the hilbert space . in the first method,@xcite denoted by @xmath107 ( @xmath109 stands for exact diagonalization and @xmath73 for translational ) , the general orthonormal state is given by @xmath110 which describes an eigenstate of the system with momentum @xmath49 . @xmath111 is the number of phonons at the @xmath112th lattice site away from the electron . for example , at the site @xmath51 , and at the nearest neighbor lattice sites left and right from it , there are @xmath113 , @xmath114 , and @xmath115 phonons , respectively . the hamiltonian ( [ h0 ] ) does not mix states ( [ tbstates ] ) with different momenta . therefore , the polaron function obtained by the @xmath107 method has the same @xmath49 momentum as the basis states . the current implementation of the @xmath107 method is highly accurate , and from a practical point of view may be treated as exact.@xcite for this reason , the @xmath107 results can be used to determine the numerical errors present in the other methods . the minimal number of states of the reduced basis necessary to obtain accurate results depends on the hamiltonian parameters . this number for the @xmath107 method increases very rapidly for @xmath116 , @xmath9 , which prevents its use for both large @xmath0 and @xmath9 . in the second exact diagonalization method , denoted by @xmath108 ( @xmath109 stands for exact diagonalization and @xmath50 for localized ) , the general orthonormal state of the chosen basis is more complicated than for the @xmath107 method , @xmath117\ ; \mid n_0 , n_{-1 } , n_1, ... ,n_m, ... \rangle_j\;.\label{lbstates}\end{aligned}\ ] ] here the @xmath118 and @xmath112 indices are given with respect to the center of polaron , which is placed at the site @xmath51 . thus @xmath119 creates an electron at the @xmath118th site from the polaron center at @xmath51 , @xmath111 is the number of extra phonons at the @xmath112th site away of the polaron center , when the lattice is already distorted by the coherent state operators @xmath56 , i.e. , @xmath120 . the @xmath108 method calculates localized polaron wave functions . namely , eq . ( [ lbstates ] ) describes a localized state , with the polaron center at site @xmath51 kept constant . the @xmath108 method is therefore accurate only in the strong coupling regime , in which the effects of polaron delocalization are negligible ( self - trapped polarons ) . for a given set of hamiltonian parameters , @xmath84 in eq . ( [ lbstates ] ) are determined by the use of the @xmath50 method , i.e. , by minimizing the energy ( [ iaenergy ] ) , @xmath121 . if only those states ( [ lbstates ] ) with all @xmath122 are used in calculations , the @xmath108 method gives the same polaron wave function as the @xmath50 method . the additional states ( [ lbstates ] ) , with @xmath111 phonon excitations , are necessary to obtain the actual equilibrium positions of the lattice in the exact localized polaron state and the zero point motion of the renormalized lattice vibrations . it is worth noting that the electron and phonon parts of the @xmath108 polaron wave function can not be completely separated , as in the case of the @xmath50 function . in the case of @xmath108 method , the mean lattice deformation of the localized polaron is approximately taken care of by the product of the coherent states operators , @xmath123 , which keeps the necessary number of states ( [ lbstates ] ) , in the @xmath108 method , relatively small . in order to reduce the basis of the hilbert space , the maximal allowed distance of the electron and phonons from the polaron center has been limited here by the choice @xmath124 . the distance @xmath125 has been determined from the condition that @xmath126 if @xmath127 . the maximal total number of phonons has been kept limited , retaining the states with @xmath128 . as the sum @xmath129 does not include the phonons associated with the coherent state operators @xmath56 , the small value of @xmath129 is not a restriction on the overall amplitude of the lattice deformation . the accuracy of the results obtained by the @xmath108 method , supplemented by the two abovementioned criteria , depends of course on the values of parameters . for the purpose of clarity , it seems appropriate at this point to review briefly the notation @xmath50 , @xmath73 , @xmath74 , @xmath107 , and @xmath108 of all five presented methods . all methods based on the localized polaron function are denoted by the letter @xmath50 ( @xmath50 and @xmath108 methods ) , whereas all methods based on the translational function are denoted by the letter @xmath73 ( @xmath73 , @xmath74 , and @xmath107 methods ) . the letter @xmath109 denotes an exact diagonalization method ( @xmath107 and @xmath108 methods ) , while a single letter notation ( @xmath50 and @xmath73 ) suggests the simplest form of the method . as the variational methods of the preceding section themselves , the results of the corresponding calculations may be best understood in terms of the weak and strong coupling limits and the crossover between them . it has been shown previously,@xcite on the basis of the global - local method for @xmath130 , that the empirical relation @xmath131 describes well the values of parameters for which the variation of the effective polaron mass with @xmath0 is the fastest . it will be also argued here that @xmath132 of eq . ( [ romero ] ) describes accurately the crossover from the weak to the strong coupling limit with respect to the nature of the @xmath133 ground state . the latter changes continuously from the light delocalized state in the weak coupling limit to the heavy self - trapped state in the strong coupling limit , with the anticrossing of the two states at @xmath134 . the physical content of eq . ( [ romero ] ) is best understood by considering the limits of small and large @xmath9 with respect to @xmath10 , when , respectively , @xmath135 and @xmath136 . both these conditions were qualitatively explained in ref . . in the present section we first discuss the strong and the weak coupling limit , and then devote most of our attention to the crossover between the two . the nature of the self - trapped polaron may be discussed by analyzing the properties of the two polaron wave functions given by eqs . ( [ iafunction ] ) and ( [ tivfunction ] ) . when the localized polaron functions @xmath80 at different lattice sites are not orthogonal , @xmath137 the local properties and delocalization effects of @xmath76 in eq . ( [ tivfunction ] ) are a complex mixture . however , in both the weak and strong coupling regimes , the translational polaron function can be approximately written in terms of orthogonal localized polaron functions . in the weak coupling regime , the orthogonality follows from the electron part of the wave function , @xmath138 while in the strong coupling regime , it follows from the lattice part , @xmath139 } \approx\delta_{0,\delta}\;.\label{phorth}\ ] ] in eq . ( [ phorth ] ) , @xmath140 is the debye - waller factor . the condition ( [ phorth ] ) corresponds to the regime of self - trapped polarons . the negligible contribution of the lattice part to the overlap of any two localized polaron functions at different lattice sites results then in a negligible polaron hopping energy @xmath2 . one may notice that in the limit @xmath141 the translational form of the polaron function ( [ tivfunction ] ) has no consequences on polaron energy , and the minimal values of the variational energies ( [ iaenergy ] ) and ( [ tivenergy ] ) coincide , i.e. , @xmath82 , @xmath142 in eq . ( [ tivenergy ] ) . the hopping of the self - trapped polaron occurs only at the time scale which is much larger than the scale relevant for the local interplay between the electron and the lattice deformation . therefore an accurate description of the local polaron properties may be obtained even if the hopping of the polaron is completely omitted by using only localized polaron functions . the translational invariance of the polaron may be , however , always restored in the same way as the function ( [ iafunction ] ) is used to obtain ( [ tivfunction ] ) . the @xmath108 method provides very accurate results for the ground state energy of the self - trapped polaron . except that it neglects the polaron hopping , the local polaron function is calculated exactly . the error of the @xmath108 method may be estimated by using the @xmath107 method , that is , by subtracting the exact energy of zero momentum polaron state @xmath143 from @xmath144 . for the electron - phonon coupling which is greater than the critical electron - phonon coupling in eq.([romero ] ) just by @xmath10 , i.e. , @xmath145 , the maximal error of the @xmath108 method is @xmath146 . this is shown in fig . [ fig1 ] , in which two qualitative estimates of @xmath2 in the strong coupling regime are plotted as well . the first estimate gives @xmath2 as one fourth of the polaron bandwidth computed exactly by the @xmath107 method , @xmath147\;,\label{bandw}\ ] ] while the second estimate , based on the @xmath50 method , multiplies the electron hopping energy with the small debye - waller factor , @xmath148}\;.\label{overlap}\ ] ] the estimation of the debye - waller factor in eq . ( [ overlap ] ) is based on the evaluation of the lattice part of the overlap of the two neighboring @xmath149 functions ( the @xmath50 method gives good results for the mean lattice deformation of the self - trapped polarons ) . all three curves in fig . [ fig1 ] are similar functions of @xmath0 , which is not surprising since all plotted quantities are related to the polaron hopping energy @xmath2 in the strong coupling regime . moreover , it may be seen from fig . [ fig1 ] that the error of the @xmath108 method is almost equal to one half of the polaron bandwidth , @xmath150 one of the advantages of the @xmath108 method is that it permits separate calculations of the electron and the lattice properties , in spite of the fact that the electron and the phonon part of the @xmath108 wave function are not separable . for instance , for the polaron centered at the origin , the associated mean lattice deformation @xmath151 is given simply by the expectation value of @xmath152 of eq . ( [ zpm ] ) . besides @xmath151 , in the present paper the mean real space uncertainty of the on - site lattice vibration @xmath153 and the product @xmath154 , where @xmath155 is the mean momentum uncertainty of the on - site lattice vibration , are calculated . figure [ fig2 ] shows the data for two sets of parameter , corresponding to a small self - trapped polaron and a self - trapped polaron extended over few lattice sites , respectively . for the second set of parameter ( large @xmath0 and @xmath9 ) , the @xmath107 method is not accurate enough , and the question arises of whether the @xmath108 polaron state is really a good approximation of the system ground state . it is difficult to prove that a polaron state , with non - negligible hopping energy @xmath2 and an energy close or lower than @xmath108 energy , does not exist . this question is currently under investigation . differences between the results of the @xmath50 and the @xmath108 methods are also analyzed . it is found that they are more pronounced for small @xmath0 . we may see from the results in fig . [ fig2 ] , that the mean lattice deformation differs between these two methods . in the case of the @xmath108 method , the mean lattice deformation is more extended and the width of the polaron is slightly larger . additionally , the electron density of the @xmath108 method remains approximately proportional to the mean lattice deformation ( as in eq . ( [ locking ] ) , valid for the @xmath50 method ) , since @xmath156/x_n^{\;el}<1\%\;.\ ] ] as has already been mentioned , the lattice part of the @xmath50 function describes a set of displaced , but unrenormalized , harmonic oscillators , so @xmath157 , @xmath158 , where @xmath159 and @xmath160 are defined in eq . ( [ zpm ] ) . one may notice from fig . [ fig2 ] that real space uncertainties of the lattice vibrations on the sites occupied by the polaron are larger in the @xmath108 case than in the @xmath50 case , @xmath161 , but uncertainties of on - site momentum lattice vibrations are smaller , @xmath162 . it may be concluded , from the first plot in fig . [ fig2 ] that the electron affects mostly the central site of the small polaron . the product of uncertainties for this site stays close to the free lattice value , @xmath163 . therefore the phonon mode at the central site of the small polaron may be treated , in a good approximation , as harmonic . in particular , the renormalized frequency of this mode , @xmath164 , can be roughly estimated from the relation @xmath165 since @xmath166 , it follows that @xmath167 . therefore in the strong coupling regime the first excited state should correspond to the excitation of the renormalized phonon mode , rather than to the excitation of the phonon of energy @xmath10 , which is uncorrelated to the polaron . one may also notice that the energy of the mean lattice deformation is larger for the @xmath108 method than the @xmath50 method ( for the @xmath50 method this energy is minimal ) . this is compensated for , however , with the lower energy associated with the zero point motion of the @xmath164 phonon mode , which makes the total polaron energy of the @xmath108 method lower . for the more extended self - trapped polarons , the renormalized normal phonon modes are expected to be spread over a number of lattice sites . consequently , a number of different phonon modes contribute to the lattice displacement at the lattice sites occupied by the polaron . thus the analysis of the on - site vibrations can not give direct information on the renormalized lattice modes . nevertheless , from the second plot in fig . [ fig2 ] one may notice that the product of uncertainties @xmath168 shows a minor deviation from that of the harmonic vibration . this suggests that the renormalized lattice modes of the extended self - trapped polaron are harmonic as well in a good approximation . in the weak coupling regime , the @xmath73 method gives results close to the @xmath107 results . since the form of the function @xmath73 is quite simple , it will be used in this section as a basis for further discussion . in the weak coupling regime , the minimum of the variational energy @xmath89 corresponds to small values of the variational parameters @xmath87 and @xmath169 . the standard perturbative ground state of the system with momentum @xmath49 in terms of the @xmath73 method polaron function may be written as follows : @xmath170 the wave function ( [ wc1function ] ) has two parts . the main part corresponds to the free electron of momentum @xmath49 , and the smaller part , proportional to @xmath169 , corresponds to the electron dressed by one spatially correlated virtual phonon . at the threshold @xmath171 , the energy of such a polaron state intersects with the energy of the system consisting of the zero momentum polaron and one extra phonon with momentum @xmath171 , eq . ( [ wc2function ] ) . so , for @xmath172 the ground state is achieved with one real phonon in the system which carries the system momentum and which is spatially uncorrelated with the polaron.@xcite for @xmath173 this state becomes the first excited state of the system . the difference between the energies of the ground state and the first excited state is the largest for @xmath133 , and is equal to @xmath10 . the validity of the perturbative treatment requires that the weight of the second term in eqs . ( [ wc1function ] ) and ( [ wc2function ] ) is small , i.e. , the mean number of phonons associated with the lattice deformation has to satisfy @xmath174 here , @xmath169 has been eliminated by using eq . ( [ xtot ] ) . there are two ways to satisfy condition ( [ nphpol ] ) . either the electron - phonon coupling is small , @xmath175 , or the lattice deformation is spread to a large number of lattice sites , @xmath176 . in the latter case , the total mean polaron deformation does not have to be small , @xmath177 , since @xmath0 can be larger than @xmath10 . the translationally invariant form of the wave function for the @xmath73 method , given by eq . ( [ tivfunction ] ) , provides an energy gain due to the polaron delocalization . at the same time , the spatial correlation between the electron and the lattice deformation has a finite length . for instance , in eq.([tivfunction ] ) , this length is of the same order for @xmath76 and @xmath80 . the perturbative calculation for @xmath88 in eq.([wc1function ] ) gives @xmath178 ^ 2 - 1}\;.\ ] ] @xmath88 measures the electron - lattice correlation length . @xmath88 is independent of @xmath0 , which makes the correlation length finite , even in the limit @xmath179 for which the lattice deformation vanishes , @xmath180 . on the other hand , the electron - lattice deformation correlation length and the polaron delocalization range are of the same order for the localized functions . this may be easily seen from eq . ( [ iafunction ] ) . in the limit @xmath179 they both become infinite . thus in the weak coupling regime one obtains a localized polaron state @xmath181 of very large width , but associated with a tiny lattice deformation . this is specific for one dimensional systems in which an attractive symmetric potential has always a bound electron state . therefore the corresponding polaron energy is less than the free electron energy @xmath182 . in higher dimensions , an arbitrary attractive symmetric potential does not have a bound electron state for sufficiently small g ( or the electron binding energy is too small to balance the lattice deformation energy ) , and the total polaron energy is larger than @xmath183 , where @xmath184 is the dimension of the system . this explains why , in the weak coupling regime , the adiabatic ( localized ) polaron functions@xcite fail to have an energy lower than the free electron energy , for the dimension of the system greater than one . a detailed parallel perturbative analysis of polarons in one , two , and three dimensions is given in ref . . in the crossover regime the polaron hopping energy @xmath2 is not negligible , and the translationally invariant polaron functions should be used in order to obtain the full physical picture of the polaron . in order to calculate the numerical errors of different methods accurately , the present discussion of the crossover regime is restricted to the values of @xmath9 smaller than @xmath185 , for which results from the @xmath107 method are available . in order to examine the crossover regime it is instructive to calculate the energy difference between the ground and first excited state , @xmath186 . let @xmath187 denote the value of @xmath0 for which this difference is minimal , @xmath188 our results indicate that @xmath187 is very close to the value of @xmath132 given by eq . ( [ romero ] ) . for example , for @xmath189 , @xmath190 , while @xmath191 . for smaller @xmath9 , @xmath187 and @xmath132 coincide even better . the analysis of the effective mass,@xcite variational energy of the polaron ground state,@xcite polaron size,@xcite as well as the behavior of the first excitation energy , suggest that a dramatic change in the nature of the polaron ground state and the first excited state are intimately related in the crossover regime . this can be well understood by considering the properties of the wave function in the @xmath73 method . near @xmath187 , this polaron wave function has two separate energy minima in the @xmath87-@xmath88 parameter space defined by eq . ( [ ve ] ) , which become degenerated for @xmath192 . let the symbol @xmath193 denotes the lower minimum at @xmath194 , and the symbol @xmath195 the lower minimum at @xmath196 . @xmath197 and @xmath198 are the corresponding polaron wave functions . even if they are not mutually orthogonal , they are still physically quite different . the numerical data show that the translational invariance of @xmath197 contributes strongly to the polaron energy . on the other hand , the translational invariance of @xmath198 has almost negligible energy contributions , i.e. , @xmath198 describes an almost self - trapped polaron . it is worth noting that the degenerate nature of the variational energy which has been reported for the toyozawa method@xcite is of the same kind as the one of the @xmath73 method discussed here . namely , both @xmath73 and toyozawa method are based on the same polaron function ( [ tivfunction ] ) . @xmath197 and @xmath198 can be combined to form new polaron functions , @xmath199 it should be mentioned that a similar combination of two states has been already used in ref . in order to calculate the polaron ground state . @xmath200 corresponds to eq.([lvfunction ] ) with @xmath105 , which means that the @xmath106 method is implied . from this treatment the improved ground state @xmath201 and the approximate first excited polaron state @xmath202 are obtained . it may be seen from fig . [ fig3 ] that @xmath106 method describes an anticrossing of @xmath198 and @xmath197 states , which yields two orthogonal states , @xmath201 and @xmath203 . in order to have a better illustration of that anticrossing in fig . [ fig3 ] , the exact ground state energy @xmath143 is subtracted from all plotted energy curves . for @xmath194 the _ light _ state @xmath197 is lower in energy , and participates in the ground state more than the _ heavy _ state @xmath198 . however , as @xmath0 increases this balance changes continuously in favor of @xmath198 . the opposite trend is observed for the first excited state , which is heavier than the ground state for smaller @xmath0 and lighter for larger @xmath0 . the @xmath74 method gives better results for the ground and first excited state when a large number of functions ( large @xmath94 ) in eq . ( [ lvfunction ] ) is used . from fig . [ fig3 ] , one may estimate that @xmath204 at @xmath192 . moreover , the energy @xmath205 , unlike @xmath206 , satisfies @xmath207 for all @xmath196 , which is an important improvement over the @xmath105 result . figure [ fig4 ] shows the results for the ground state energy and the energy of the first excited state with total system momentum @xmath133 , as functions of @xmath0 , obtained by several different methods . two plots correspond to @xmath208 and @xmath209 , respectively . the ground state energy obtained by the @xmath107 method is subtracted from all the other results . @xmath210 and @xmath211 are used to mark three different polaron regimes with respect to the strength of the electron - phonon coupling . for @xmath212 the mean number of phonons of the lattice deformation is less than one . thus for @xmath213 we recognize the weak coupling regime . for @xmath214 the @xmath108 polaron energy has a negligible error ( see fig . [ fig1 ] ) , which means that for @xmath215 polarons are self - trapped , and we recognize the strong coupling regime . the crossover regime is found in the interval @xmath216 , with @xmath217 . it may be noted from fig . [ fig4 ] that in the weak coupling regime , the energy obtained by the @xmath50 method ( @xmath69 ) is close to the free electron energy @xmath182 ( for @xmath218 , the absolute error of the @xmath50 method becomes greater than @xmath10 in some intervals of @xmath0 ) . in the strong coupling regime , for large @xmath0 , @xmath69 approaches the exact polaron energy . the error of the @xmath73 method , as may be seen from fig . [ fig4 ] , is the largest in the crossover regime in which the results can be improved by a better choice of @xmath53 and @xmath84 in eq . ( [ ve ] ) . in the strong coupling regime , the translationally invariant form of the function @xmath73 has no effect on the polaron energy , i.e. , both the @xmath73 and @xmath50 methods give very similar results . the energy of the first excited polaron state , @xmath219 , intersects the energy of the ground state plus one phonon , @xmath220 , for @xmath221 . after the minimum of @xmath186 is reached in the crossover regime at @xmath222 , @xmath219 approaches @xmath220 asymptotically for @xmath223 . as has already been pointed out in sec . [ f06 ] , the excitation of the renormalized phonon mode explains the nature of the first excited state in the strong coupling limit . in ref . perturbation theory was used to calculate the frequency of this local phonon mode , @xmath224 , to the lowest order in @xmath225 , @xmath226 the obtained perturbative correction to the phonon frequency is adiabatic , i.e. , the square root in eq . ( [ kabanov ] ) is independent of mass @xmath16 . it is worth noting that for large @xmath0 , @xmath219 calculated here shows approximately the same behavior as the energy of the ground state with one additional phonon of frequency ( [ kabanov ] ) , @xmath227 the lattice part of the @xmath74 function is spatially symmetric with respect to the electron . therefore in the strong coupling regime the first excited state , which has been identified here as a local renormalized phonon mode of the self - trapped polaron , should be basically a symmetric oscillation of the lattice deformation around the central polaron site . in the vicinity of @xmath132 , on the other hand , the nature of the first excited state is currently explained by the anticrossing of the self - trapped and the delocalized polaron state . the question of how the excited self - trapped polaron state in the strong coupling regime and the first excited state near @xmath132 may be linked together is a matter of further considerations . one may speculate that the anticrossing of the excited self - trapped polaron state and the delocalized state will give the answer . the present paper discusses the ground and first excited states of the polaron for three different regimes of the electron - phonon coupling parameter @xmath0 . the results can be briefly summarized as follows . in the strong coupling regime the polaron hopping energy to neighboring sites is negligible , and the self - trapped polaron states are obtained . the results of the @xmath108 method suggest that the adiabatic picture of the localized polaron state is valid , in which some of the local lattice vibrations are renormalized by the presence of the electron . the numerical data show no significant deviation from the adiabatic locking relation ( [ locking ] ) of the electron site density and the mean lattice deformation . in the small polaron case , the predominant effect of the electron is the lowering of the frequency of the vibration at the central polaron site . the excitation of the renormalized phonon mode corresponds to the first excited state of the small self - trapped polaron . the nature of the polaron ground state in the crossover regime has been discussed in a number of papers , and its rapid change with @xmath0 has been well established numerically . the difference between the energies of the first excited state and the ground state , as a function of @xmath0 , has a minimum for @xmath192 . it is shown , by using the @xmath106 method , that the anticrossing of the self - trapped and the delocalized polaron state can link the behavior of the ground and first excited polaron state . according to the @xmath74 method , for @xmath196 the effective mass of the ground state is larger than the effective mass of the first excited state , while for @xmath194 the opposite is true . in addition , it is found that @xmath187 , which characterizes the first excited state , and @xmath132 , obtained from ground state analysis ( eq.([romero ] ) ) almost coincide for @xmath228 . upon further reduction of @xmath0 the total mean number of phonons bound by the polaron becomes smaller than one , and the weak coupling regime is reached . the nearly free electron is dressed by a cloud of virtual phonons , and its mass is slightly renormalized . the first excited state of the system , with momentum @xmath229 , can be viewed as the ground state of the polaron plus one additional uncorrelated phonon , rather than as an excited polaron . finally , it is worth noting that there is a simple sum rule for the mean total lattice deformation , eq . ( [ xtot ] ) , which is valid for any number of electrons and is independent of the system dimension . this important sum rule may be extended to some other models in which the lattice deformation is linearly coupled to the local electron density . @xmath230}\;,\\ s_\delta&=&y_\delta\;\sum_n\eta^*_n\eta_{n+\delta}\;,\\ e_\delta&=&-t\;y_\delta\;\sum_n\eta^*_n(\eta_{n+\delta+1}+\eta_{n+\delta-1 } ) \\&+&\hbar\omega\;s_\delta\;\sum_m\xi_m^*\xi_{m+\delta}\\ & -&g\;y_\delta\;\sum_n\eta^*_n\eta_{n+\delta}(\xi_n^*+\xi_{n+\delta})\;.\end{aligned}\ ] ] l. d. landau , z. phys . * 3 * , 644 ( 1933 ) . t. holstein , ann . ( n.y . ) * 8 * , 325 ( 1959 ) . h. de raedt and a. lagendijk , phys . * 49 * , 1522 ( 1982 ) ; phys . rev . b * 27 * , 6097 ( 1983 ) ; * 30 * , 1671 ( 1984 ) ; p. e. kornilovitch and e. r. pike , phys . b * 55 * , r8634 ( 1997 ) ; d. w. brown , k. lindenberg , and y. zhao , j. chem . phys . * 107 * , 3179 ( 1997 ) . c. zhang , e. jeckelmann , and s. r. white , phys . lett . * 80 * , 2661 ( 1998 ) ; phys . b * 60 * , 14 092 ( 1999 ) ; e. jeckelmann and s. r. white , _ ibid . _ * 57 * , 6376 ( 1998 ) . p. gosar , j. phys . c * 8 * , 3584 ( 1975 ) . a. s. alexandrov , phys . b * 61 * , 12315 ( 2000 ) . b. gerlach and h. lwen , phys . b * 35 * , 4291 ( 1987 ) ; rev . mod . phys . * 63 * , 63 ( 1991 ) ; h. lwen , phys . rev . b * 37 * , 8661 ( 1988 ) . d. feinberg , s. ciuchi , and f. de pasquale , int . b * 4 * , 1395 ( 1990 ) . g. kalosakas , s. aubry , and g. p. tsironis , phys . b * 58 * , 3094 ( 1998 ) . y. toyozawa , prog . * 26 * , 29 ( 1961 ) . y. zhao , d. w. brown , and k. lindenberg , j. chem . phys . * 107 * , 3159 ( 1997 ) . h. b. shore and l. m. sander , phys . b * 7 * , 4573 ( 1973 ) . k. h. hock , h. nickisch , and h. thomas , helv . acta * 56 * , 237 ( 1983 ) . g. venzl and s. f. fischer , phys . b * 32 * , 6437 ( 1985 ) . a. klamt , j. phys . c * 21 * , 1953 ( 1988 ) . a. la magna and r. pucci , phys . b * 53 * , 8449 ( 1996 ) . j. ranninger and u. thibblin , phys . b * 45 * , 7730 ( 1992 ) . f. marsiglio , phys . a * 180 * , 280 ( 1993 ) ; physica c * 244 * , 21 ( 1995 ) . a. s. alexandrov , v. v. kabanov , and d. k. ray , phys . b * 49 * , 9915 ( 1994 ) . g. wellein and h. fehske , phys . rev . b * 56 * , 4513 ( 1997 ) ; * 58 * , 6208 ( 1998 ) . j. bona , s. a. trugman , and i. batisti ' c , phys . b * 60 * , 1633 ( 1999 ) . a. h. romero , d. w. brown , and k. lindenberg , phys . b * 59 * , 13 728 ( 1999 ) . m. capone , w. stephan , and m. grilli , phys . b * 56 * , 4484 ( 1997 ) . j. m. robin , phys . b * 58 * , 14 335 ( 1998 ) . d. emin and t. holstein , phys . lett . * 36 * , 323 ( 1976 ) . v. v. kabanov and o. yu . mashtakov , phys . b * 47 * , 6060 ( 1993 ) . a. h. romero , d. w. brown , and k. lindenberg , phys . a * 254 * , 287 ( 1999 ) . a. h. romero , d. w. brown , and k. lindenberg , phys . b * 60 * , 4618 ( 1999 ) . a. h. romero , d. w. brown , and k. lindenberg , phys . a * 266 * , 414 ( 2000 ) . v. cataudella , g. de filippis , and g. iadonisi , phys . b * 60 * , 15 163 ( 1999 ) ; * 62 * , 1496 ( 2000 ) .
the recent development in the autonomy and the capabilities of mobile robots greatly increases the number of applications suitable for a team of autonomous agents . particular interest has been received by those tasks requiring continual execution , such as the monitoring of oil spills @xcite , the detection of forest fires @xcite , the track of border changes @xcite , and the patrol ( surveillance ) of an environment @xcite . the surveillance of an area of interest requires the robots to continuously and repeatedly travel the environment , and the challenging problem consists in scheduling the robots trajectories so as to optimize a certain performance criteria . the reader familiar with network location , multiple traveling salesman , or graph exploration problems may observe a close connection with the patrolling problem we address , e.g. , see @xcite . it is worth noting , however , that these classical optimization problems do not capture the repetitive , and hence dynamic , aspect of the patrolling problem , nor the synchronization issues that arise when a timing among the visits of certain zones is required . a precise formulation of the patrolling problem requires the characterization of the robots capabilities , of the environment to be patrolled , and of the performance criteria . in this work , we assume the robots to be identical and capable of sensing and communicating within a certain spatial range , and of moving according to a first order integrator dynamics with bounded speed . we represent the environment as a graph , in which the vertices correspond to physical and strategically important locations , and in which the edges denote the possibility of moving and communicating between locations . we assume that , when a robot is placed at each of the graph vertices , the union of the sensor footprints provides complete sensor coverage of the environment . regarding the performance criteria of a patrolling trajectory , we consider ( i ) the time gap between any two visits of the same region , called _ refresh time _ , and ( ii ) the time needed to inform the team of robots about an event occurred in the environment , called _ latency_. loosely speaking , refresh time and latency reflect the effectiveness of a patrolling team in detecting events in the environment and in organizing remedial actions . for both the refresh time and latency optimization problem , we focus on the worst case analysis , even though the average refresh time and the average latency cases are also of interest . notice that for the latency to be finite , the motion of the robots needs to be synchronized . for instance , if two robots are allowed to communicate only when they simultaneously occupy two adjacent vertices of the graph , then they need to visit those vertices at the same time in a finite latency trajectory . the patrolling problem is receiving increasing attention because of its fundamental importance in many security applications , e.g. , see @xcite . although many solutions have been proposed , the problem of designing minimum refresh time and latency team trajectories for a general environment is , to date , an open problem . almost all traditional approaches rely on space decomposition , and traveling salesperson tour computation @xcite . in @xcite an empirical evaluation of existing patrolling heuristics is performed . in @xcite two classes of strategies are presented , namely the cyclic- and the partition - based strategy . in the cyclic - based strategy , the robots compute a closed route through the viewpoints , and travel repeatedly such route at maximum speed . clearly , in the case of a single robot , if the tour is the shortest possible , then the cyclic - based strategy performs optimally with respect to the refresh time and latency criteria . in the partition - based strategy , the viewpoints are partitioned into @xmath0 subsets , being @xmath0 cardinality of the team , and each robot is responsible for a different set of viewpoints . to be more precise , each robot computes a closed tour visiting the viewpoints it is responsible for , and it repeatedly moves along such tour at maximum speed . still in @xcite , the two classes of strategies are compared , and it is qualitatively shown that cyclic - based strategies are to be preferred whenever the ratio of the longest to the shortest edge of the graph describing the environment is small , while , otherwise , partition - based policies exhibit better performance . in @xcite and @xcite , an efficient and distributed solution to the perimeter patrolling problem for robots with zero communication range is proposed . by means of some graph partitioning and graph routing techniques , we extend the results along these directions , e.g. , by considering the case of a nonzero communication range for the perimeter patrolling , and by characterizing optimal strategies for different environment topologies . an important variant of the patrolling problem is known as persistent surveillance , e.g. , see @xcite . differently to our setup , a dynamically changing environment is considered for the persistent surveillance problem , and performance guarantees are offered only under a certain assumption on the rate of change of the regions to be visited . it is worth mentioning that a different approach to the design of patrolling trajectories relies on the use of pebbles or chemical traces to mark visited regions , e.g. , see @xcite . these techniques , although effective even without a global representation of the environment , and with severe communication constraints , do not explicitly deal with the optimality of the patrolling trajectories , and they represent therefore a complementary area of research with respect to this work . the main contributions of this work are as follows . we introduce and mathematically formalize the concept of refresh time and latency of a team trajectory , and we formally state the patrolling optimization problem . we propose a procedure to build a graph ( roadmap ) to represent the topological structure of the area to be patrolled , and we study separately the case of a chain , tree , and cyclic ( not acyclic ) graph . we exhaustively discuss the case of a chain roadmap . first , we characterize a family of minimum refresh time and latency team trajectories , which can be computed by optimally partitioning the chain graph among the robots . second , we derive a centralized polynomial time algorithm to compute an optimal partition , and , ultimately , to design an optimal team trajectory . our partitioning procedure is based upon a bisection method , and it is also amenable to distributed implementation . third , we develop a distributed procedure for the robots to converge and synchronize along an optimal trajectory , so as to minimize the refresh time and latency criteria . fourth and finally , we test the robustness of our methods through a simulation study . when the roadmap has a tree or cyclic structure , we focus on the refresh time optimization problem , and we do not consider the latency optimization nor the design of distributed algorithms . for the case of a tree roadmap , we reduce the minimum refresh time patrolling problem to a known graph optimization problem . we show that the computational complexity of the minimum refresh time patrolling problem is polynomial in the number of vertices of the roadmap , and , under the assumption of a fixed and finite number of robots , we identify a polynomial time centralized algorithm to compute a minimum refresh time team trajectory . for the case of a cyclic roadmap , we show that the patrolling problem is an _ np - hard _ optimization problem . we propose two approximation algorithms , and we characterize their performance . the first approximate solution is extremely easy to compute , but its performance depends upon the ratio between the longest and the shortest edge in the graph representing the environment . the second approximation algorithm is based on a polynomial time path - covering procedure , and it allows us to compute a team trajectory whose refresh time is within a factor of @xmath1 from the minimum refresh time for the given environment ( cf . [ fig : path_cover ] for an example ) . to the best of our knowledge , this algorithm is the first constant factor approximation algorithm for the _ np - hard _ minimum refresh time patrolling problem . a preliminary version of this work appeared in @xcite . with respect to the latter manuscripts , in this current work we introduce and solve the latency optimization problem , we perform a numerical study to analyze the robustness of our algorithmic procedures , and we improve the presentation of the results on the refresh time optimization problem . the rest of the paper is organized as follows . the notation and the problem under consideration are in section [ sec : prob_set ] , where we also show that the patrolling problem is , generally , computationally hard . section [ sec : refr_time ] , [ sec : latency ] , and [ sec : distr_alg ] contain our results for the patrolling of a chain environment . we characterize a minimum refresh time and latency team trajectory , and we derive a centralized and a decentralized algorithm for its computation . in section [ sec : simulations ] we perform a simulation study to show some robustness and reconfigurability properties of our distributed procedure . section [ sec : tree_patrol ] contains our results for the patrolling of a tree environment . we describe a minimum refresh time team trajectory on a tree roadmap , and we characterize the complexity of computing an optimal solution . section [ sec : heurist ] deals with the general case of cyclic environment , and it contains our approximation procedures . our conclusion and final remarks are in section [ sec : future_work ] . we will be using the standard motion planning notation , and we refer the reader to @xcite for a comprehensive treatment of the subject . we are given a team of @xmath2 identical robots , capable of sensing , communicating , and moving in a connected environment . we make the following combined assumptions on the robot capabilities and on the environment to be patrolled . regarding sensing , we assume that the environment can be completely covered by simultaneously placing a robot at each of a set of @xmath3 _ viewpoints _ in the configuration space . in other words , if @xmath4 robots were available and placed at the @xmath5 viewpoints , then the union of the sensors footprint of the robots would provide complete sensor coverage of the environment . we assume that each viewpoint is required for complete sensor coverage . finally , we assume @xmath3 so that at least one robot needs to visit more viewpoints for the entire environment to be monitored over time . regarding communication , we associate an undirected graph @xmath6 with the environment , whose vertices are the @xmath5 viewpoints , and in which there is an edge between two vertices if two robots placed at those viewpoints are able to communicate to each other . we assume that @xmath6 is connected . in what follows we design cooperative patrolling algorithms with sporadic communication , in the sense that two robots are required to communicate only when they occupy adjacent vertices . the occurrence of additional communication links can be easily incorporated into the algorithms and can not decrease their performance . regarding motion , we assume that the robots are holonomic , i.e. , they are modeled as first order integrators and move at most at unit speed . additionally , we turn the graph @xmath6 into a robotic roadmap @xcite and a metric weighted graph as follows : to each pair of viewpoints that are neighbors in @xmath6 , we associate a unique path connecting them . we select these paths so that the set of path lengths , adopted as edge weights , verify the triangle inequality . ( for example , the shortest paths between viewpoints constitute a suitable choice . ) we assume that each robot remains always on the roadmap . in summary , the combined assumptions on the robot capabilities and on the environment are that : the vertices of @xmath6 provide complete sensor coverage of the environment and each edge of @xmath6 corresponds to both a communication edge and a motion path . hence , we refer to @xmath6 as a roadmap with sensor coverage and communication connectivity . the problem of constructing a roadmap from an environment is here discussed . assume that the robots are holonomic vehicles moving at bounded speed , and equipped with an omnidirectional sensing device , and a line - of - sight communication device . if a map of the environment is available , then a valid roadmap is obtained by solving an art gallery problem with connectivity @xcite . a solution to the art gallery problem with connectivity is a set of locations , called guards , with the following two properties : each point in the environment is visible by at least one guard and the visibility graph of the guards is connected . an example roadmap is given in fig . [ fig : commgraph ] . ( a distributed sensor - based algorithm for the art gallery problem with connectivity is given in @xcite . ) the edges of the roadmap denote the possibility for the @xmath7 robots of both moving and communicating between pair of connected viewpoints . the weight of each edge corresponds to its length.,title="fig : " ] + if the environment is unknown , then an exploration algorithm should be used to obtain a representation of the environment , see for example @xcite . while or after exploring the environment , a robot should select viewpoints to provide full sensor coverage . by construction , the existence of a path between two viewpoints is automatically guaranteed by the exploration algorithm . moreover , if communication between two path - connected viewpoints @xmath8 is not guaranteed , then the graph may be augmented with additional viewpoints along the path connecting @xmath9 and @xmath10 . the roadmap resulting from these steps features sensor coverage and communication connectivity . the following remarks are in order . first , a roadmap representing the environment is in general not unique , and the patrolling performance depends upon the particular choice . in this work , we do not address the problem of choosing the roadmap that yields optimal performance . instead , we present efficient algorithms for the patrolling problem , which can also be used to compare different roadmaps on the basis of the corresponding patrolling performance . second , for the implementation of our algorithms , a robot does not need to travel exactly along the roadmap . indeed , a robot only needs to arrive sufficiently close to a viewpoint , or to be able to circle around it to provide sensor coverage . a related example is in section [ sec : simulations ] , where the arrival delay of a robot can be interpreted as the uncertainty in the motion of the robots . third , the global knowledge of the roadmap may not be required for the execution of a patrolling trajectory . indeed , in general , each robot only visits a subpart of the roadmap . fourth and finally , collisions are prevented by letting two robots exchange their roles every time they are about to collide . indeed , since no robot is assigned to a specific viewpoint , if robot @xmath11 is about to collide with robot @xmath12 at time @xmath13 , then , without affecting the performance of the team trajectory , collision is avoided by redefining the @xmath11-th and @xmath12-th trajectory as @xmath14 and @xmath15 for @xmath16 ; see the notion of order invariant trajectory below . communication or sensing devices can be used to detect and prevent possible collisions . for a team of @xmath0 robots with specific capabilities , let @xmath17 be a robotic roadmap with sensor coverage and communication connectivity . a _ team trajectory _ @xmath18 is an array of @xmath0 continuous and piecewise - differentiable trajectories @xmath19 defined by the motion of the robots on the roadmap @xmath6 , i.e. , @xmath20 \mapsto g$ ] , for @xmath21 , where @xmath22 is a time horizon of interest , much larger than the time required by a single robot to visit all viewpoints . we say that a viewpoint @xmath23 is visited at time @xmath24 by robot @xmath11 if @xmath25 . we define the _ refresh time _ of a team trajectory @xmath18 , in short @xmath26 , as the longest time interval between any two consecutive visits of any viewpoint , i.e. , @xmath27 where @xmath28 ^ 2 , t_1 \le t_2 , \ ; | \ ; x_i(t)\neq v , \forall i\in{\{1,\dots , m\ } } , t_1 < t < t_2\}}$ ] . we claim that there exists a team trajectory with minimum refresh time and prove it as follows . without loss of generality , we restrict our attention to team trajectories in which each robot moves at maximum speed along the edges and stops for certain durations , possibly equal to zero , at the viewpoints . thus , a team trajectory can be described as a tuple of the form @xmath29 , where @xmath30 contains the sequence of viewpoints visited by robot @xmath11 , and @xmath31 contains the waiting times at the visited vertices . notice that the time horizon @xmath32 is finite , the length of each edge is positive , the number of vertices is finite , and the speed of the robots is bounded . it follows that the length of each sequence @xmath30 is finite , and , therefore , each @xmath30 takes value in a finite set . now , for each possible sequence of visited vertices , the refresh time is a continuous function of only the waiting times , and each waiting time lies in the compact interval @xmath33 $ ] . because any continuous function defined over a compact region admits a point of minimum value , we conclude that there exists a team trajectory with minimum refresh time . [ prob : minrt ] given a roadmap and a team of robots , find a minimum refresh time team trajectory . in section [ sec : latency ] we present a different optimization problem , which deals with the possibility for a robot to communicate , possibly with multiple hops , with every other robot in the team . we now conclude this section with our first result on the computational complexity of the team refresh time problem . for a detailed discussion of the main aspects of the computational complexity theory , we refer the interested reader to @xcite . recall that an optimization problem is said to be _ np - hard _ if it is , informally , as hard as the hardest problem in the class _ np _ , for which no polynomial time algorithm is known to compute an optimal solution . [ thm : complexity ] the _ team refresh time _ problem is _ np - hard_. this statement can be shown by reduction from the traveling salesman problem @xcite . in fact , if @xmath34 , since the speed of the robots is bounded , then a minimum refresh time trajectory consists of moving the robot at maximum speed along a shortest closed tour visiting the viewpoints . the problem of finding a shortest tour through a set of points in the plane , also known as traveling salesman problem , is an _ np - hard _ problem @xcite . hence , by restriction , the team refresh time problem is also _ np - hard_. following theorem [ thm : complexity ] , the minimum refresh time optimization problem is generally computationally hard . in this work , we first identify two roadmap structures for which there exists an efficient solution to the team refresh time problem , and then we describe two approximation algorithms to deal with the general case . we characterize in this section an optimal refresh time team trajectory when the roadmap associated with the environment has a chain structure . let @xmath35 denote the neighbor set of the vertex @xmath11 , and let @xmath36 denote the degree of @xmath11 , i.e. , the cardinality of the set @xmath35 . a chain roadmap is an undirected , connected , and acyclic roadmap , in which every vertex has degree two , except for two vertices which have degree one . without losing generality , we assume that the @xmath5 vertices are ordered in a way that @xmath37 , @xmath38 , and @xmath39 for each @xmath40 . we define a relative order of the robots according to their position on the roadmap . a team trajectory is _ order invariant _ if the order of the robots does not change with time , i.e. , if @xmath41 for each @xmath42 and for every instant @xmath43 $ ] , where @xmath44 denotes the distance at time @xmath24 on the roadmap from the first vertex of the chain to the position of the @xmath11-th robot . [ prop : eqv_traj ] let @xmath18 be a team trajectory . there exists an order invariant team trajectory @xmath45 such that @xmath46 . let @xmath18 be a team trajectory , and consider the permutation matrix @xmath47 , that keeps track of the order of the robots at time @xmath24 , i.e. , such that the @xmath48-th entry of @xmath47 is @xmath49 if , at time @xmath24 , the @xmath11-th robot occupies the @xmath12-th position in the chain of robots , and it is @xmath50 otherwise . since @xmath18 is continuous , anytime the function @xmath47 is discontinuous , the positions of the robots directly involved in the permutation overlap . therefore , the order invariant team trajectory @xmath51 is a feasible team trajectory , and it holds @xmath52 . let @xmath53 be the set of viewpoints visited over time by the agent @xmath11 with the trajectory @xmath54 , and let the _ image _ of the team trajectory @xmath18 be the set @xmath55 . notice that different team trajectories may have the same image . throughout the paper , let @xmath56 , @xmath57 , and @xmath58 . finally , let @xmath59 . a team trajectory is _ non - overlapping _ if @xmath60 for all @xmath61 . [ prop : disjoint_sets ] given a chain roadmap , there exists an order invariant and non - overlapping team trajectory with refresh time @xmath62 . let @xmath63 be a minimum refresh time team trajectory , and let @xmath18 be the order invariant team trajectory obtained from @xmath63 as in proposition [ prop : eqv_traj ] . clearly @xmath64 . let @xmath55 be the image of @xmath18 , and note that @xmath65 . consider the partition of @xmath66 defined as @xmath67 for every nonempty @xmath68 , let @xmath69 , @xmath70 , and @xmath71 . note that , by construction , the viewpoint @xmath72 is visited by the robot @xmath11 and , possibly , by the robots @xmath73 . also , because @xmath18 is order invariant , we have @xmath74 . it follows that @xmath75 . consider now the team trajectory @xmath45 with image @xmath76 , and assume that the robots sweep periodically at maximum speed their segment . then @xmath77 , so that @xmath78 is an order invariant and non - overlapping team trajectory with minimum refresh time . given a chain graph on the viewpoints @xmath66 , let @xmath79 be an _ @xmath0-partition _ of @xmath66 , i.e. , @xmath80 is a collection of subsets of @xmath66 such that @xmath81 whenever @xmath61 , and @xmath82 . additionally , let the dimension of the partition @xmath83 equal the longest distance between any two viewpoints in the same cluster , i.e. , @xmath84 , where @xmath85 if @xmath86 . following proposition [ prop : disjoint_sets ] , there exists a minimum refresh time team trajectory whose image coincide with an @xmath0-partition of @xmath66 . we now show that the minimum refresh time equals twice the dimension of an optimal @xmath0-partition . [ thm : min_refre_time ] let @xmath6 be a chain roadmap , and let @xmath0 be the number of robots . then @xmath87 . as a consequence of propositions [ prop : eqv_traj ] and [ prop : disjoint_sets ] , there exists a minimum refresh time team trajectory whose image coincides with an @xmath0-partition @xmath83 . since each robot is assigned a different cluster , and the speed of the robots is bounded by @xmath49 , we have @xmath88 . consider a team trajectory @xmath18 in which each robot continuously sweeps at maximum speed the cluster it is assigned to . clearly , @xmath89 . @xmath90 for @xmath91 @xmath92 for @xmath93 we have shown that a minimum refresh time trajectory consists of letting the robot sweep at maximum speed a part of the chain graph . such a trajectory is more formally described for the @xmath11-th robot in trajectory [ algo : max_sweep ] , where we only characterize the instants of time at which robot @xmath11 changes its velocity vector , and we assume that it moves at maximum speed otherwise . by removing the longest edges in the chain the average length of the clusters is minimized . in general , such partition does not minimize the dimension of the @xmath0-partition , and hence it is not optimal in our sense . an example is in fig . [ fig : average ] . + + in the remaining part of the section we describe an algorithm to compute an optimal @xmath0-partition . for a set of viewpoints @xmath66 , we call _ left - induced _ partition of length @xmath94 the partition @xmath95 defined recursively as ( cf . fig . [ fig : leftinduc ] ) @xmath96 where @xmath97 the cardinality @xmath98 corresponds to the integer @xmath12 such that . observe that the function @xmath99 is monotone , non - increasing , and right - continuous ( cf . [ fig : leftinduccard ] ) . let @xmath100 be the discontinuity points of the function @xmath99 , then , for @xmath101 , @xmath102 note that two or more discontinuity points of @xmath103 may coincide , so that the function @xmath103 may not assume all the values in the set @xmath104 , e.g. , in fig . [ fig : leftinduccard ] the value @xmath103 is never equal to @xmath105 . [ thm : optimal_partition ] let @xmath6 be a chain roadmap . let @xmath83 be an @xmath0-partition of @xmath6 , and let @xmath106 be the left - induced partition of length @xmath107 of @xmath6 . then @xmath108 let @xmath83 be an @xmath0-partition , and let @xmath109 be the left induced partition of length @xmath107 of a chain roadmap @xmath6 . let @xmath110 . we want to show that @xmath111 is one of the discontinuity points of the function @xmath103 , i.e. , that @xmath111 verifies the conditions . by contradiction , if @xmath112 and @xmath113 , then an @xmath0-partition with dimension smaller than the optimal would exists . therefore we have @xmath114 if @xmath112 . suppose now that @xmath115 , and let @xmath116 be an @xmath0-partition with minimum dimension . notice that @xmath117 , because the cluster @xmath118 contains all the viewpoints within distance @xmath107 from @xmath119 , and hence also within distance @xmath111 . it follows that @xmath120 , and also that @xmath121 . by repeating the same procedure to the remaining clusters , we obtain that @xmath122 , so that , if @xmath123 and @xmath124 , then @xmath125 . following theorem [ thm : optimal_partition ] , an optimal left - induced partition of cardinality ( at most ) @xmath0 is also an optimal @xmath0-partition . notice that for the computation of an optimal left - induced partition only the lengths @xmath107 corresponding to the discontinuity points of @xmath106 need to be considered . since each discontinuity point coincides with the distance between a pair of vertices , only @xmath126 values need to be tested . therefore , an optimal left - induced partition can be computed with complexity @xmath127 . in what follows we describe an @xmath128-approximation algorithm with linear complexity for any @xmath129 . notice that @xmath128-approximation algorithms with linear complexity are often more convenient for a practical implementation than exact algorithms with higher complexity @xcite . we now present our algorithm for the computation of an optimal @xmath0-partition . since the function @xmath99 is monotone and continuous , a bisection method is effective for finding its discontinuity points , and , therefore , for determining the shortest length of a left - induced partition of cardinality @xmath0 . a bisection based procedure to compute an optimal left - induced partition is in algorithm [ algo : partition ] , where the function _ left - induced(@xmath130,@xmath107 ) _ returns the left - induced partition defined in equation . we next characterize the convergence properties of algorithm [ algo : partition ] . [ thm : conv_alg_part ] let @xmath6 be a chain roadmap , and let @xmath83 denote an @xmath0-partition of @xmath6 . let @xmath131 . algorithm [ algo : partition ] with tolerance @xmath128 returns a left - induced partition of dimension at most @xmath132 and cardinality at most @xmath0 . moreover , the time complexity of algorithm [ algo : partition ] is @xmath133 . algorithm [ algo : partition ] searches for the minimum length @xmath111 that generates a left - induced partition of cardinality at most @xmath0 . because of theorem [ thm : optimal_partition ] , the length @xmath111 coincides with one of the discontinuity points of the function @xmath103 , and it holds @xmath134 . indeed , @xmath135 because @xmath136 , and @xmath137 , because @xmath138 . recall from that @xmath114 for every @xmath112 , and that the function @xmath139 is monotone . note that the interval @xmath140 $ ] , as updated in algorithm [ algo : partition ] , contains the value @xmath111 at every iteration . the length of the interval @xmath140 $ ] is divided by 2 at each iteration , so that , after @xmath141 , the value @xmath111 is computed with precision @xmath128 . since the computation of @xmath103 can be performed in @xmath142 operations , the time complexity of algorithm [ algo : partition ] is @xmath133 . as a consequence of proposition [ prop : disjoint_sets ] and theorem [ thm : min_refre_time ] , in what follows we only consider team trajectories whose image coincide with an @xmath0-partition . therefore , for ease of notation , we use the set @xmath55 to denote both the image set of a team trajectory and an @xmath0-partition of the chain graph . we conclude this section with a summary of the presented results . [ thm : summary_chain_rt ] let @xmath6 be a chain graph with @xmath5 viewpoints and let @xmath143 be the number of robots . let @xmath144 be an optimal @xmath0-partition of @xmath6 computed by means of algorithm [ algo : partition ] with tolerance @xmath128 . let @xmath145 be the dimension of @xmath144 . a team trajectory with image @xmath144 , and minimum refresh time @xmath146 is computed as in trajectory [ algo : max_sweep ] . moreover , the time complexity of designing such trajectory is @xmath133 . the previous section considers the problem of designing team trajectories with optimal refresh time on a chain graph . in a patrolling mission it may be important for the robots to communicate with each other in order to gather information about the status of the entire environment . for instance , messages could be sent by a unit to ask for reinforcement , or to spread an alarm . informally , we call _ latency _ of a team trajectory @xmath18 , in short @xmath147 , the shortest time interval necessary for a message generated by any robot to reach all the other robots . in other words , given our communication model , the latency of a team trajectory is a measure of how fast a message spreads to all robots . in this section we describe team trajectories with minimum refresh time and latency . we now give a more formal definition of @xmath147 . recall that , by assumption , two robots are allowed to communicate when they lie on two adjacent viewpoints . in a chain roadmap , for a message to reach every robot in the chain , every pair of adjacent robots needs to communicate . for @xmath148 , let @xmath149 denote the union of the set of times at which the robots @xmath150 and @xmath11 communicate and @xmath151 . the _ up - latency _ of @xmath18 , in short @xmath152 , is the longest time interval between any two consecutive communications between the robots @xmath153 and @xmath154 . precisely , @xmath155 where @xmath156 . analogously , we call _ down - latency _ the quantity @xmath157 where @xmath158 . finally , we define the latency of a team trajectory as @xmath159 notice that our definitions of latency hold for @xmath160 , and that , if @xmath161 , then we have @xmath162 for every team trajectory @xmath18 . we envision that the up- and down - latency performance criteria should be adopted when it is of interest to report the status of the monitored area to a base station located at one end of the chain environment . the latency minimization problem is more appropriate for fully distributed scenarios . in this section we design synchronized team trajectories with the following two features . first , since a minimum refresh time trajectory is determined by an optimal partition of the chain graph , we aim at finding team trajectories associated with the same optimal partition . second , we design synchronized team trajectories with minimum up - latency ( resp . down - latency ) or latency . we start by showing a lower bound for @xmath163 and @xmath164 . recall that , for a partition @xmath55 , we have @xmath56 , @xmath165 , @xmath58 , and @xmath166 . [ lower_bound ] let @xmath6 be a chain roadmap , and let @xmath55 be an @xmath0-partition of @xmath6 . the latency of a team trajectory with image @xmath55 is lower bounded by @xmath167 . the proposition follows from the fact that the robots speed is bounded by @xmath49 , and that the robots need to travel their segment to communicate with the neighboring robots . for the up - latency of a team trajectory to equal the lower bound in lemma [ lower_bound ] , each robot @xmath11 needs to transfer a message from robot @xmath150 to robot @xmath168 in time @xmath169 . in order to do so , each robot @xmath11 needs to communicate with its neighbor @xmath168 as soon as @xmath170 . let @xmath6 be a chain graph with @xmath5 viewpoints and let @xmath143 be the number of robots . let @xmath144 be an optimal @xmath0-partition of @xmath6 computed by means of algorithm [ algo : partition ] with tolerance @xmath128 . let @xmath145 be the dimension of @xmath144 , and let @xmath169 be the length of the @xmath11-th cluster . a team trajectory with image @xmath144 , minimum refresh time @xmath171 , and minimum up - latency @xmath172 is computed as in trajectory [ algo : min_base ] . moreover , the time complexity of designing such trajectory is @xmath133 . the theorem follows by observing that the trajectory is @xmath173-periodic , and that no robot @xmath11 waits at @xmath174 to communicate with the neighboring robot @xmath168 . the up - latency equals the lower bound in lemma [ lower_bound ] , and it is therefore minimum . regarding the computational complexity , notice that it is determined by the computation of the optimal @xmath0-partition @xmath144 , and hence , by lemma [ thm : conv_alg_part ] , it equals @xmath133 . an example of a team trajectory with minimum refresh time and minimum up - latency is in fig . [ fig : traj3 ] . finally , observe that the minimization of the down - latency can be achieved in an analogous way . @xmath175 for @xmath176 and for @xmath177 @xmath178 for @xmath179 . notice that each robot @xmath11 communicates with the neighboring robot @xmath168 as soon as @xmath180 . a team trajectory with minimum down - latency is obtained by shifting in time the trajectory of robot @xmath49 , in a way that robot @xmath181 communicates with robot @xmath49 as soon as @xmath182 . ] let @xmath55 be an optimal @xmath0-partition , and suppose that @xmath183 for all @xmath184 . then the up - latency and the down - latency can be made minimum and equal to @xmath185 by arranging the robots trajectories to be in opposite phase . specifically , for @xmath186 , we set @xmath187 if @xmath11 is odd , and @xmath188 if @xmath11 is even . because of lemma [ lower_bound ] , the above trajectory has minimum latency . this particular case was studied in @xcite . we now consider the minimization of the latency criterion , and we restrict our attention to periodic team trajectories . to be more precise , let @xmath55 be an optimal @xmath0-partition of the environment , and let @xmath189 denote the longest length of the clusters . we aim at finding a @xmath190-periodic team trajectory with image @xmath55 and minimum latency . notice that , by imposing a periodicity of @xmath190 , the refresh time of the trajectory , if finite , is also minimized . we start by considering the pedagogical situation in which @xmath191 for all @xmath192 . in the next lemma , we show that the frequency of message exchange among the robots is limited by the periodicity of the trajectory . [ lemma : exchange_rate ] consider a @xmath190-periodic team trajectory , and let @xmath193 for all @xmath192 . for any @xmath194 $ ] and for any @xmath195 , there exist no two distinct sequences @xmath196 and @xmath197 , with @xmath198 , @xmath199 , such that @xmath200 moreover , for any @xmath201 , there exist no two sequences @xmath202 and @xmath203 with @xmath198 , @xmath204 , such that @xmath205 since @xmath206 , it follows @xmath207 . let @xmath208 . by contradiction , if two distinct sequences @xmath196 and @xmath197 exist , with @xmath209 , @xmath210 , @xmath211 and @xmath212 , then the @xmath213-th robot travels its cluster four times . since the speed of the robots is bounded by one , robot @xmath168 can not travel its cluster four times in a period of @xmath190 ( cf . [ fig : frequency_exchange ] ) . the second part of the theorem follows from an analogous reasoning . , at most one communication sequence is possible within each period of length @xmath214 . here @xmath215 . ] notice that in the above lemma the index @xmath11 belongs to the set @xmath216 ( resp . @xmath217 ) because we consider @xmath7 consecutive communication instants , and because @xmath149 denotes the sequence of times at which robots @xmath150 and @xmath11 communicate . because of lemma [ lemma : exchange_rate ] , in a @xmath190-periodic team trajectory with @xmath218 only one message can be passed from robot @xmath11 to robot @xmath219 in a period of time of @xmath190 . this limitation determines a lower bound on the latency of a periodic trajectory . notice that eventual communication instants @xmath220 , with @xmath221 , do not influence the latency , since all information can be transmitted at time @xmath222 without affecting the latency performance . [ lemma : lower_bound_latency ] let @xmath18 be a @xmath190-periodic team trajectory with @xmath223 for all @xmath192 . then @xmath224 because of lemma [ lemma : exchange_rate ] , a message can be transfered from robot @xmath11 to robot @xmath219 in at most @xmath190 instants of time , by traveling the clusters @xmath225 and @xmath226 . without losing generality , we let @xmath189 be the time to pass a message form @xmath168 to @xmath227 , and from @xmath227 to @xmath219 . notice that the same reasoning holds also for the time to transfer a message from @xmath219 to @xmath11 . therefore , the latency is lower bounded by @xmath185 . we now consider the situation in which an optimal @xmath0-partition does not verify the constraint @xmath218 for all @xmath228 . intuitively , for what concerns the latency , two consecutive clusters with length @xmath169 and @xmath229 may be regarded as one single cluster of length @xmath230 . therefore , in order to use lemma [ lemma : exchange_rate ] and lemma [ lemma : lower_bound_latency ] , we partition the clusters @xmath55 into groups such that the sum of the length of the clusters of two consecutive groups is greater than @xmath189 . in other words , let @xmath231 be the set of _ right - extreme _ viewpoints of the partition @xmath55 defined recursively as @xmath232 where @xmath233 such that @xmath234 . let @xmath235 be the set of _ left - extreme _ viewpoints defined recursively as @xmath236 additionally , define the set of aggregated clusters @xmath237 , where @xmath68 contains all the clusters within @xmath72 and @xmath238 , and let @xmath239 be the sum of the length of the clusters in @xmath240 . [ thm : lat_general ] let @xmath18 be a @xmath190-periodic team trajectory with image @xmath55 , and let @xmath241 be the number of aggregated clusters . then , @xmath242 consider the clusters defined by the right extreme viewpoints , and notice that they verify @xmath218 . then , the theorem follows from lemma [ lemma : lower_bound_latency ] , and from the fact that the minimum latency on the image @xmath243 equals the minimum latency on the image @xmath55 . the terms @xmath244 and @xmath245 are due to the fact that we are interested in delivering a message from robot @xmath49 to robot @xmath0 in the original configuration , and not on the aggregated chain . a team trajectory with minimum refresh time and minimum latency is formally presented in trajectory [ algo : opt_traj ] , where we specify the instants of time at which a robot changes its velocity , and we assume that it moves at maximum speed otherwise . an example is reported in fig . [ example_latency_opt ] , and here we give an informal description . let @xmath55 be an optimal @xmath0-partition of a chain graph , and let @xmath246 be the set of aggregated clusters . recall that @xmath247 is a subset of @xmath55 , and that the sum of the length of two consecutive elements is larger than @xmath189 . the procedure in trajectory [ algo : opt_traj ] ( lines @xmath248 and @xmath249 ) is such that the robots in the same group behave as a single robot assigned to the whole set of viewpoints . in other words , the motion of the robots in the same group is determined by a token passing mechanism , in which robot @xmath168 moves towards @xmath250 only when @xmath180 and @xmath251 , and , analogously , robot @xmath11 moves towards @xmath252 only when @xmath253 and @xmath254 . instead , lines @xmath255 and @xmath256 in trajectory [ algo : opt_traj ] guarantee the transfer of one message in a period of @xmath190 between three consecutive groups , and , consequently , the minimization of the latency . indeed , since the sum of the length of two consecutive groups is larger than @xmath189 , because of lemma [ lemma : exchange_rate ] , no more than one message can be transferred between three consecutive groups in a period of @xmath190 . let @xmath6 be a chain graph with @xmath5 viewpoints and let @xmath143 be the number of robots . let @xmath144 be an optimal @xmath0-partition of @xmath6 computed by means of algorithm [ algo : partition ] with tolerance @xmath128 . let @xmath145 be the dimension of @xmath144 , and let @xmath257 ( resp . @xmath258 ) be the length of the first ( resp . last ) cluster in @xmath144 . let @xmath241 be the number of aggregated clusters , and let @xmath259 ( resp . @xmath260 ) be the length of the first ( resp . last ) aggregated cluster . a team trajectory with image @xmath144 , minimum refresh time @xmath146 , and minimum latency @xmath261 is computed as in trajectory [ algo : opt_traj ] . moreover , the time complexity of designing such trajectory is @xmath133 . by inspection , the team trajectory described in trajectory [ algo : opt_traj ] is @xmath190-periodic , and therefore it has minimum refresh time . moreover , by construction , the communications at the extreme viewpoints happen every @xmath190 instants of time , so that the latency is equal to @xmath262 , and hence , by lemma [ thm : lat_general ] , minimum . regarding the computational complexity , notice that it is determined by the computation of the optimal @xmath0-partition @xmath144 , and hence , by lemma [ thm : conv_alg_part ] , it equals @xmath133 . and minimum latency @xmath263 is here reported . a message with delivery time equal to @xmath264 is reported in red . the right - extreme viewpoints are @xmath265 , the left - extreme viewpoints are @xmath266 , and @xmath267 . the represented team trajectory has minimum refresh time @xmath268 and minimum latency @xmath269 . a message with delivery time equal to @xmath264 is reported in red . ] in the previous sections we have shown that , for the computation of a minimum refresh time and latency team trajectory , first an optimal @xmath0-partition of the roadmap needs to be found , and , then , a synchronization of the motion of the robots needs to be achieved to ensure communication between neighboring robots . the distributed computation of an optimal @xmath0-partition follows directly from algorithm [ algo : partition ] , by letting the robots compute the left - induced partition of length @xmath270 in a distributed way . a simple solution consists of the following three operation : 1 . the robots gather at the leftmost viewpoint , and 2 . determine the cardinality of the team and elect a leader ; 3 . the leader computes an optimal left - induced partition , and assigns a different cluster to each robot . notice that , by only assuming the capability of detecting the viewpoints in the roadmap ( in particular the leftmost and rightmost viewpoint ) the robots can distributively compute an optimal partition . indeed , the leader can simply travel the roadmap , and evaluate if , for a given length @xmath107 , the cardinality of the corresponding left - induced partition is greater , equal , or smaller than the cardinality of the team . we believe that , by simple programming operations , the above procedure can be improved , so as to handle reconfigurations due to addition or removal of robots in the team . in this section we focus instead on the design of a distributed feedback law to synchronize the motion of the robots so as to guarantee the minimization of the latency of the trajectory . recall that @xmath271 denotes the position on the chain of the robot @xmath11 at time @xmath24 . moreover , let the @xmath11-th cluster of an optimal @xmath0-partition be delimited by @xmath272 and @xmath174 . let @xmath273 denote the direction of motion of robot @xmath11 . c - time _ denote the current time of simulation , let _ a - time _ be the time at which a robot arrives at his right extreme . n - meet(i , j ) _ be a function that returns the number of times that the robots @xmath11 and @xmath12 have communicated . timer(@xmath274 ) _ be a function that returns @xmath49 after a time interval of length @xmath274 . an algorithm for the robots to distributively converge to a minimum refresh time and latency team trajectory is in algorithm [ algo : synchro ] . it should be noticed that the algorithm assumes the knowledge of an optimal partitioning of the chain graph , and of the left- and right - extreme sets . algorithm [ algo : synchro ] is informally described as follows . first , the velocity of a robot changes only when its position coincides with the leftmost or the rightmost of the assigned viewpoints . when a robot reaches an extreme viewpoint , it waits until a communication with the neighboring robot happens ( lines @xmath275 ) . this mechanism determines the feedback behavior of our procedure . the behavior of a robot after a communication is determined by the lines @xmath276 , which reproduce the optimal behavior described in trajectory [ algo : opt_traj ] . to be more precise , the function _ token(@xmath277 ) _ coordinates the motion of the robots in the same group ( see section [ sec : realistic_case ] ) , so that they move as if there was a single robot sweeping the viewpoints in the same group . the function _ timer(@xmath278 ) _ , instead , ensures the maximum frequency of messages exchange between consecutive groups , so as to minimize the latency of the resulting trajectory . [ thm : latency ] let @xmath18 be the team trajectory generated by algorithm [ algo : synchro ] . there exists a finite time after which @xmath18 has minimum refresh time and latency . let @xmath55 be an optimal @xmath0-partition , let @xmath279 be the set of right - extreme viewpoints , and let @xmath280 be the set of left - extreme viewpoints . let @xmath281 denote the set of robots patrolling a viewpoint between @xmath72 and @xmath238 , where @xmath72 and @xmath238 are as previously defined . first , notice that robot @xmath49 ( resp . @xmath0 ) sets @xmath282 ( resp . @xmath283 ) as soon as @xmath284 ( resp . @xmath285 ) . second , the function @xmath286 guarantees that , when @xmath287 communicate , exactly one robot among @xmath288 maintains a zero velocity and in an alternate way . therefore , after a finite time @xmath289 , independent of the initial robots motion direction , the velocities of the robots in @xmath281 are such that , upon communication , @xmath290 and @xmath291 . in other words , after @xmath289 , the robots in @xmath281 behave as a single robot sweeping the segments between @xmath72 and @xmath238 . finally , the function @xmath292 and the parameter @xmath293 guarantee that the communications at the extreme viewpoints happen every @xmath294 instants of time . we conclude that the trajectory generated by algorithm [ algo : synchro ] converges in finite time to a team trajectory with minimum refresh time and latency . it should be observed that , differently from the team trajectories presented in the previous sections , algorithm [ algo : synchro ] contains a feedback procedure to synchronize the motion of the robots . our algorithm is independent of the initial configuration , and , as it is shown in the next section , it is robust to a certain class of robot failures . + given the theoretical nature of this work , in this section we propose a simulation study to show the effectiveness of our procedures . for our simulations , we use the matlabsimulation environment , and we model the robots as holonomic vehicles of zero dimension . the communication edges and the motion paths are described by a given roadmap . consider the chain roadmap with @xmath295 viewpoints in fig . [ fig : simulation ] . suppose that @xmath296 robots are assigned to the patrolling task . in order to obtain a team trajectory with minimum refresh time , an optimal @xmath296-partition of the roadmap is computed ( cf . [ fig : simulation ] ) . additionally , to obtain a minimum latency team trajectory , each robot is assigned to a different cluster , its initial position is chosen randomly inside its cluster , and its velocity is initialized randomly . the motion of each robot is then determined by algorithm [ algo : synchro ] . the resulting team trajectory is in fig . [ fig : trajectories ] , where the team of robots synchronize on a minimum refresh time and latency team trajectory after a finite transient . , the team trajectory obtained with algorithm [ algo : synchro ] is reported here . the dashed lines denote a communication among two neighboring robots . at time @xmath297 the robots have synchronized their motion , and from that moment up to time @xmath298 the team trajectory has minimum refresh time and latency . from time @xmath298 up to time @xmath299 , robot @xmath300 undergoes a temporary failure , causing all the other robots to lose synchronization . the team of robots synchronizes again when robot @xmath300 resume its motion . at time @xmath301 , the failure of robot @xmath300 is detected by the remaining robots , which compute and synchronize on a new partition.,title="fig : " ] + we now test the robustness of our synchronization procedure . as first case study , we consider a temporary stopping failure , in which a robot stops for a certain time interval . for instance , suppose that robot @xmath300 stops from time @xmath302 up to time @xmath303 ( cf . [ fig : trajectories ] ) . notice that , after the failure , each robot @xmath12 , with @xmath304 , gathers at @xmath305 , and each robot @xmath306 , with @xmath307 , gathers at @xmath308 waiting for a communication with the corresponding neighboring robot . as soon as robot @xmath300 resumes its functionalities , the team of robots recover the desired synchronization . notice that the transient failure of robot @xmath300 can be easily detected by its neighbors by means of a timer mechanism with a predefined threshold . as a second case study , we let the robots actuation be affected by noise , so that the speed of the robots becomes a random variable with a certain distribution . precisely , let @xmath309 be the equation describing the dynamics of robot @xmath11 , where @xmath310 is a zero mean gaussian variable with variance @xmath311 @xmath312 $ ] . we let @xmath313 and we run @xmath314 simulations for each possible value of @xmath311 on the roadmap of fig . [ fig : simulation ] . the refresh time and the latency of the team trajectories obtained with algorithm [ algo : synchro ] are plotted in fig . [ refresh_error ] and in fig . [ latency_error ] , respectively , as a function of @xmath311 . note that the performance degrade gracefully with the noise magnitude . as third and final case study , we consider the situation in which a robot definitively stops . the remaining robots need to compute a new optimal partition and to synchronize in order to guarantee an optimal patrolling of the environment . notice that for the computation of such a partition by algorithm [ algo : partition ] the chain graph and the number of the robots is required . suppose that the failure of the robot @xmath300 is detected at time @xmath315 by the well - behaving robots , and assume that each robot knows the chain roadmap and the number of operative robots . algorithm [ algo : partition ] and algorithm [ algo : synchro ] allow the team to synchronize on a new team trajectory with minimum refresh time and latency . notice that the initial and the final partitions do not coincide . the problem of designing minimum refresh time and latency team trajectories on a chain roadmap has been discussed . in this section we consider the more general cases of tree and cyclic roadmap , we characterize the computational complexity of determining optimal trajectories , and we describe two approximation methods with performance guarantees . the results we are going to present are intended for a team of more than one robot . indeed , if only one robot is assigned to the patrolling task , then a minimum refresh time trajectory follows from the computation of the shortest tour through the viewpoints , for which efficient approximation algorithms already exist @xcite . let @xmath316 denote an undirected , connected , and acyclic roadmap ( tree ) . recall that a vertex path is a sequence of vertices such that any pair of consecutive vertices in the sequence are adjacent . a tour is a vertex path in which the start and end vertices coincide , and in which every vertex of @xmath13 appears at least once in the sequence . a depth - first tour of @xmath13 is a tour that visits the vertices @xmath66 in a depth - first order @xcite . let dft@xmath317 denote the length of a depth first tour of @xmath13 . notice that the length of a depth - first tour of a connected tree equals twice the sum of the length of the edges of the tree , and that any depth - first tour is a shortest tour visiting all the vertices . we now show that , for the case of tree roadmap , the set of cyclic - based and partition - based trajectories described in @xcite does not contain , in general , a minimum refresh time trajectory . recall that in a cyclic - based strategy the robots travel at maximum speed and equally spaced along a minimum length tour visiting all the viewpoints . consider the tree roadmap of fig . [ tree_1 ] , and suppose that two robots are assigned to the patrolling task . clearly , the minimum refresh time is @xmath318 , while the refresh time of a cyclic strategy equals @xmath319 . consider now the tree roadmap in fig . [ tree_equal ] , where the edges have unit length , and assume that two robots are in charge of the patrolling task . observe that any partition of cardinality @xmath181 contains a chain of length @xmath181 , so that , since only one robot is assigned to each cluster , the minimum refresh time that can be obtained is @xmath320 . suppose , instead , that the robots visit the vertices of the roadmap as specified in table [ table : visiting_order ] , where @xmath321 denotes the position of a robot at time @xmath24 . since the refresh time of the proposed trajectory is @xmath7 , we conclude that neither the cyclic - based nor the partition - based strategy may lead to a minimum refresh time team trajectory on a tree roadmap . [ cols="^,^,^,^,^,^,^,^,^",options="header " , ] [ table : visiting_order ] we now introduce some definitions . let @xmath18 be a team trajectory on the tree roadmap @xmath13 . we say that the edge @xmath322 is _ used _ by @xmath18 if there exists @xmath184 and @xmath323 ^ 2 $ ] such that @xmath324 and @xmath325 , and it is _ unused _ otherwise . note that , because in a tree there exists only one path connecting two vertices , the above condition ensures that the edge @xmath326 is traveled by the robot @xmath11 . let @xmath327 denote the set of unused edges , and let @xmath328 be the forest obtained from @xmath13 by removing the edges @xmath327 from @xmath329 , i.e. , the collection of vertex - disjoint subtrees @xmath330 , with @xmath331 , such that @xmath332 and @xmath333 , for each @xmath334 . let @xmath335 be the number of robots that visit at least one vertex of @xmath289 in the interval @xmath336 $ ] , and note that @xmath337 in a finite refresh time trajectory . let @xmath338 . notice that the same subtree collection can be associated with different team trajectories . we say that a team trajectory is _ efficient _ if its refresh time is the smallest among all the team trajectories associated with the same subtree collection . [ eff_team ] let @xmath339 be the subtree collection associated with the team trajectory @xmath18 on the tree roadmap @xmath13 , where @xmath340 , and @xmath338 . then , @xmath18 is efficient if @xmath341 let @xmath342 , and let @xmath335 be the number of robots assigned to @xmath289 . notice that the robots in @xmath289 travel , in total , at least @xmath343 to visit all the vertices . since the speed of the robots is bounded by @xmath49 , the smallest refresh time for the vertices of @xmath289 is @xmath344 . an efficient team trajectory , can be computed with the following procedure . see table [ table : visiting_order ] for an example . [ lemma_team ] let @xmath339 be a subtree collection of a tree roadmap , where @xmath340 , and @xmath338 . an efficient team trajectory is as follows : for each @xmath334 , 1 . compute a depth - first tour @xmath345 of @xmath289 , 2 . equally space @xmath335 robots along @xmath345 , and 3 . move the robots clockwise at maximum speed on @xmath345 . since every vertex of @xmath346 appears at least once in a depth first tour @xmath345 of @xmath289 , and the robots move with maximum speed and equally spaced along @xmath345 , every vertex is visited at most every @xmath347 . the statement follows . let @xmath348 be the partition set of @xmath0 , i.e. , the set of all the sequences of integers whose sum is @xmath0 . the following problem is useful to characterize the complexity of designing minimum refresh time trajectories on a tree roadmap . [ prob : optimal_forest ] let @xmath13 be a tree roadmap and let @xmath0 be the number of robots . find a subtree collection @xmath339 that minimizes @xmath349 subject to @xmath350 and @xmath351 . [ equiv_prob ] for the case of a tree roadmap , the team refresh time problem and the optimal subtree collection problem are equivalent . as a consequence of theorem [ eff_team ] , the minimum refresh time on a tree roadmap @xmath13 can be written as @xmath352 , where @xmath353 is a subtree collection of @xmath13 , and @xmath354 . it follows that a solution to problem [ prob : minrt ] can be derived in polynomial time from a solution to problem [ prob : optimal_forest ] by using the procedure described in lemma [ lemma_team ] . suppose now we have a solution to problem [ prob : minrt ] . then an optimal subtree collection follows from the identification of the unused edges . we conclude that the two optimization problems are equivalent . we now state our main result on the design of minimum refresh time team trajectory on a tree roadmap . [ complexity ] let @xmath13 be a tree roadmap with @xmath5 vertices , and let @xmath0 be the number of robots . a minimum refresh time team trajectory on @xmath13 can be computed in @xmath355 time . recall from @xcite that an optimal subtree collection can be computed in @xmath355 . then , by using lemma [ equiv_prob ] and lemma [ lemma_team ] , the claimed statement follows . as a consequence of theorem [ complexity ] , the problem of designing minimum refresh time team trajectories on a tree roadmap is computationally _ easy _ for any finite number of robots . in our design procedure , we first compute an optimal subtree collection of the given tree , and then we schedule the robots trajectory according to lemma [ lemma_team ] . in this section we propose two approximation methods for the team refresh time problem in the case of a cyclic , i.e. , not acyclic , roadmap . these solutions are obtained from a transformation of the cyclic roadmap into an acyclic roadmap . let @xmath17 , with @xmath356 , be an undirected and connected roadmap . note that there exists an open tour @xmath293 with at most @xmath357 edges that visits all the vertices . edges that visits all the vertices can be constructed starting from a leaf of a spanning tree of @xmath6 . ] we construct a chain roadmap @xmath358 from @xmath293 by doubling its repeated vertices and edges , so that @xmath358 has at most @xmath359 vertices and at most @xmath357 edges , and such that the length of the @xmath11-th edge of @xmath358 equals the length of the @xmath11-th edge of @xmath293 ( cf . [ fig : treechain ] ) . our first approximation method consists of applying algorithm [ algo : synchro ] to an optimal @xmath0-partition of @xmath358 . [ thm : perf_ratio ] let @xmath6 be a connected roadmap , let @xmath5 be the number of vertices of @xmath6 , and let @xmath360 be ratio of the longest to the shortest length of the edges of @xmath6 . let @xmath62 be the minimum refresh time on @xmath6 . let @xmath293 be an open tour with @xmath357 edges that visits all the @xmath5 vertices , and let @xmath358 be the chain roadmap associated with @xmath293 . let @xmath361 be the minimum refresh time on @xmath358 . then @xmath362 let @xmath363 be the shortest length of the edges of @xmath6 , and note that the length of @xmath358 is upper bounded by @xmath364 . it follows that @xmath365 . since @xmath366 by assumption , some robots need to move along @xmath6 for all the viewpoints to be visited . because each robot can visit only a vertex at a time , at least @xmath367 steps are needed to visit all the vertices of @xmath6 , and therefore @xmath368 . by taking the ratio of the two quantities the statement follows . it should be noticed that , when @xmath360 grows , the performance of our procedure might degrade . for instance , suppose that the roadmap is as in fig . [ fig : ratio_inf ] , and suppose that @xmath320 robots are assigned to the patrolling task . as long as @xmath369 , a minimum refresh time strategy requires one robot to patrol the viewpoints @xmath370 , while the second , third , and fourth robot stay on the viewpoints @xmath371 , @xmath372 , and @xmath373 respectively . it follows that @xmath374 . on the other hand , an optimal @xmath0-partition of any chain graph associated with a tour that visits all the viewpoints has dimension at least @xmath49 . consequently , the refresh time of the team trajectory obtained with algorithm [ algo : synchro ] equals @xmath181 , and the ratio @xmath375 grows proportionally to @xmath376 . , then the performance ratio @xmath377 grows with @xmath376.,title="fig : " ] + we next describe a polynomial time constant factor approximation algorithm for the minimum refresh time problem . given a roadmap @xmath17 and a positive integer @xmath378 , we define a path cover of cardinality @xmath306 as the collection of paths @xmath379 such that @xmath380 . let the cost of a path equal the sum of the lengths of its edges . the min - max path cover problem asks for a minimum cost path cover for the input graph , where the cost of a cover equals the maximum cost of a path in the cover . the following result is known . [ result_path_cover ] there exists a @xmath320-approximation polynomial algorithm for the _ np - hard _ min - max path cover problem . following theorem [ result_path_cover ] , given a graph @xmath6 , there exists a polynomial time algorithm that computes a path cover of @xmath6 with cost at most @xmath320 times greater than the cost of any path cover of @xmath6 . we now state our approximation result for the _ np - hard _ team refresh time problem . [ 4_approx ] there exists an @xmath1-approximation polynomial algorithm for the _ np - hard _ team refresh time problem . let @xmath381 be a @xmath320-approximation path cover of the graph @xmath6 . note that the length of each path is within @xmath382 . indeed , in a minimum refresh time team trajectory starting at time @xmath50 and with unitary velocity , every vertex is visited within time @xmath62 . let @xmath18 be a team trajectory obtained by letting robot @xmath11 sweep at maximum speed the path @xmath383 . clearly , @xmath384 . because of theorem [ result_path_cover ] , the team trajectory @xmath18 can be computed in polynomial time . robots for the patrolling of a part of the ucsb campus . the viewpoints ( red circles ) are chosen so as to provide sensor coverage of the whole area . for the design of the patrolling trajectory , a roadmap is first constructed as described in section [ sec : prob_set ] . then , a path covering of cardinality @xmath300 is computed with the procedure in @xcite , and each robot is assigned a different path . finally , the trajectory of each robot consists in sweeping at maximum speed the tour obtained by shortcutting the assigned path . the refresh time of the proposed team trajectory is proven to be within a factor of @xmath1 of the minimum refresh time for the given roadmap . ] following lemma [ 4_approx ] , for any given roadmap and any number of robots , a team trajectory with refresh time within a factor of @xmath1 of the optimal refresh time can be constructed by computing a path covering of the roadmap , and by assigning a different path to each robot . an example is in fig . [ fig : path_cover ] , a movie of which is included in the multimedia material . several existing heuristics can be used to improve upon the trajectories in fig . [ fig : path_cover ] . for instance , since the robots move in a metric space , shortcutting techniques may be applied @xcite . because these heuristics do not guarantee an improvement of the optimality gap of our trajectories , they are not considered in this work , and they are left as the subject of future investigation . the design of team trajectories to cooperatively patrol an environment has been discussed . after defining the problem and the performance criteria , we analyze the computational complexity of the design problem as a function of the shape of the environment to be patrolled . for the case of a chain environment , we describe a polynomial algorithm to compute minimum refresh time and latency team trajectories . for the case of a tree environment , under the technical assumption of a constant number of robots , we identify a polynomial time algorithm to compute minimum refresh time team trajectories . finally , the general case of cyclic environment is shown to be _ np - hard _ , and two approximation algorithms with performance guarantees have been proposed . interesting aspects requiring further investigation include a throughout study of the latency optimization problem for cyclic roadmaps , the development of more efficient approximation algorithms , and the introduction of a more general communication framework , in which the robots are allowed to communicate while traveling the edges of the roadmap . the study of average performance criteria and the extension to dynamically changing environments are also of interest . finally , an hardware implementation of our algorithms would further strengthen our theoretical findings . y. elmaliach , a. shiloni , and g. a. kaminka , `` a realistic model of frequency - based multi - robot polyline patrolling , '' in _ international conference on autonomous agents _ , estoril , portugal , may 2008 , pp . 6370 . m. batalin and g. s. sukhatme , `` the design and analysis of an efficient local algorithm for coverage and exploration based on sensor network deployment , '' _ ieee transactions on robotics _ , vol . 23 , no . 4 , pp . 661675 , 2007 . f. amigoni , n. basilico , and n. gatti , `` finding the optimal strategies for robotic patrolling with adversaries in topologically - represented environments , '' in _ ieee int . conf . on robotics and automation _ , kobe , japan , may 2009 , pp . 20052010 . a. marino , l. parker , g. antonelli , and f. caccavale , `` behavioral control for multi - robot perimeter patrol : a finite state automata approach , '' in _ ieee int . conf . on robotics and automation _ , kobe , japan , may 2009 , pp . 33503355 . a. almeida , g. ramalho , h. santana , p. tedesco , t. menezes , v. corruble , and y. chevaleyre , `` recent advances on multi - agent patrolling , '' in _ advances in artificial intelligence - sbia 2004_.1em plus 0.5em minus 0.4emspringer , 2004 , vol . 3171 , pp . 474483 . a. machado , g. ramalho , j. d. zucker , and a. drogoul , `` multi - agent patrolling : an empirical analysis of alternative architectures , '' in _ multi - agent - based simulation ii _ lecture notes in computer science.1em plus 0.5em minus 0.4emspringer , 2003 , pp . 155170 . i. a. wagner , m. lindenbaum , and a. m. bruckstein , `` efficiently searching a graph by a smell - oriented vertex process , '' _ annals of mathematics and artificial intelligence _ 24 , no . 1 , 211223 , 1998 . k. j. obermeyer , a. ganguli , and f. bullo , `` multi - agent deployment for visibility coverage in polygonal environments with holes , '' _ international journal on robust and nonlinear control _ , sep . 2010 , to appear . i. a. wagner , m. lindenbaum , and a. m. bruckstein , `` mac vs. pc : determinism and randomness as complementary approaches to robotic exploration of continuous unknown domains , '' _ international journal of robotics research _ , vol . 19 , no . 1 , 1231 , 2000 . a. franchi , l. freda , g. oriolo , and m. vendittelli , `` the sensor - based random graph method for cooperative robot exploration , '' _ ieee / asme transactions on mechatronics _ , 14 , no . 2 , pp . 163175 , 2009 .
in the late 70 s , montonen and olive discovered that a given four dimensional gauge theory might be described using different sets of elementary fields @xcite . the fields in one of this set create the standard perturbative spectrum of the theory ( photon , w bosons , quarks , ) , while the fields in the `` dual '' set correspond to the solitonic states ( dyons ) . an example of such a phenomenon was observed even before in some two dimensional theories @xcite . since then , a natural and exciting question arose : can we find theories where the description in terms of the solitonic states is the same ( same field content and symmetries in the lagrangian ) as the description in terms of the perturbative states ? answering this question requires to understand the strong coupling behaviour of the gauge theories under study , since exchanging electrically charged perturbative states with magnetically charged solitonic states amounts to inverting the gauge coupling constant due to the dirac quantization condition . this was out of reach in the 70 s , though some important progress was made . first , it was realized that the natural arena for electric - magnetic duality was yang - mills theories with extended supersymmetries @xcite . second , it was pointed out that in order to have dyons of spin 1 , which could be the dual of the w bosons , @xmath1 supersymmetry seemed to be required @xcite . actually , an exact electric - magnetic duality , combined with the dirac quantization condition , implies that the @xmath8 function must vanish . this is indeed the only case where the electric coupling @xmath9 and the magnetic coupling @xmath10 , being constant , have the same behaviour under the renormalization group flow . the @xmath1 theory is known to be conformally invariant at the quantum level , perturbatively @xcite as well as non - perturbatively @xcite . at the perturbative level , this property is shared with some @xmath2 theories @xcite , which are believed to be finite even non - perturbatively at least when the rank of the gauge group is one . the latter theories are strongly believed to have , together with the @xmath1 theory , an exact electric - magnetic duality symmetry ; we will study them below . there are various ways to test these duality conjectures . one of them , advocated in @xcite , consists of looking at the ( hyper)elliptic curve from which the low energy physics can be deduced , and check whether some sensible duality transformations can be defined in order to insure exact electric - magnetic duality symmetry . another approach has been to look at the free energy @xcite . but the most popular test certainly is to determine the content of a particular sector of the hilbert space , the bps sector , and to check whether it is compatible with duality @xcite . to study solely the bps states is not too restrictive . actually , all the perturbative and known stable solitonic states are bps states . moreover , this is in general a very accurate test . for instance , it is well known that the bps spectra of the asymptotically free @xmath2 theories certainly do not have the same symmetries as the associated ( hyper)elliptic curve . there are also some technical reasons to study the bps spectra . an exact quantum mass formula is known for these states , stemming from the fact that they lie in small representations of the supersymmetry algebra and thus saturate the bogomolnyi bound @xcite . the semiclassical quantization can give reliable results when one can go continuously from weak coupling to strong coupling without altering the stability of the states , which is the case for instance in the @xmath1 theory . and , what is maybe their most intriguing property , it seems that the bps spectra can be computed completely once one knows the low energy structure of the theory , that is all the information about these spectra seems to be contained , in a very hidden way , in the ( hyper)elliptic curve . in this paper , i will study in detail the bps spectra of two @xmath2 theories having zero @xmath8 function . the first one will be the so(3 ) gauge theory with one flavour of bare mass @xmath11 , which reduces to the @xmath1 theory when @xmath12 , and the second one will be the su(2 ) gauge theory with four hypermultiplets whose bare masses will be taken to be @xmath13 , @xmath14 . the fact that these seemingly very different theories can be treated in parallel stems from the fact that their low energy effective action are formally identical . i will use a method whose spirit originated in @xcite and which is completely understandable in the framework of non perturbative field theory la seiberg - witten @xcite . this will lead to a natural and rigorous proof that the spectra of both the @xmath1 and the @xmath2 theories are self - dual when the bare masses are zero . we will also see how the duality can still work in the massive theories though , as it will be shown , the spectra are no longer in general self - dual _ at a given point _ in the moduli space . the plan of the paper is as follows . in section 2 some generalities on the theories under study are recalled . particular emphasis is put on the quantum numbers carried by the solitonic states , as suggested by a semiclassical analysis , and it is explained why they are compatible with an exact duality symmetry . in section 3 , after a short presentation of the seiberg - witten curves whose exactitude will be our unique , very mild hypothesis , some of the physical ideas which are at the basis of the present work are presented . this leads to a limpid understanding of _ why _ states having any magnetic charges must exist in the theory , and also gives a flavour of what the spectra in the massive cases look like . it appears that in some regime , the magnetic charge is quantized much as if it were a periodic variable , while only one value of the electric charge is allowed . though the spectrum does not appear in general as being self - dual at a given point in moduli space , it is argued that it is nevertheless perfectly compatible with an exact s duality of the massive theories . it is also pointed out that semiclassical reasonings might be able to account for the curious disappearance of all but very few states in some regions of moduli space , a phenomenon first discovered in @xcite at strong coupling . in order to prove the complete sl@xmath15 invariance of the spectra , the consideration of theories with non zero bare @xmath16 angle is needed . this is done in section 4 . we also discuss the appearance of superconformal points in some particular theories . in section 5 , the uniqueness of the states is proven , as required by duality . in section 6 , a general presentation of the curves of marginal stability , across which the bps spectra may be discontinuous , these curves are then used in section 7 to rigorously establish the existence of all the @xmath0 states , for @xmath3 and @xmath4 relatively prime integers , in the massless theories . finally , in section 8 , a general physical argument is presented which shows that the theories under study must have at least an exact @xmath5 duality symmetry . in particular , the bps spectra of the _ massless _ theories must be invariant under the monodromy group @xmath5 of the _ massive _ theories . we then deduce that the vector @xmath6 states exist in the massless @xmath7 theory . in appendix a , the computation of the seiberg - witten periods is presented , and in appendix b the curves of marginal stability are used to prove the existence of some particular states required by duality in the massive theories . as already noted in the introduction , one easy way to check whether a given theory may be , or can not be , self - dual , is to look at the quantum numbers carried by the states which are supposed to be transformed into each other by an electric - magnetic rotation . for example , the w bosons must be the dual of a charged particle having spin one in the one monopole sector . the very existence of such a spin one solitonic state is already a non trivial test of duality and was proven in @xcite . the main goal of this paper will be to extend this kind of result to all the electric and magnetic quantum numbers . in addition to the electric and magnetic charges , there is another abelian quantum number , which we will call the @xmath17 charge , which plays an important r^ ole , since it also appears in the central charge of the supersymmetry algebra and thus in the bps mass formula . the aim of this section is to explain the relations that may exist between these quantum numbers , and also to discuss the way one should compute the @xmath17 charges . the theory whose matter content is one @xmath2 adjoint `` quark '' hypermultiplet will be called hereafter the so(3 ) theory . this matter multiplet consists in two chiral @xmath18 superfields @xmath19 , which correspond in terms of ordinary fields to one dirac spinor @xmath20 and two complex scalars @xmath4 and @xmath21 in the adjoint representation of the gauge group . the microscopic lagrangian also contains one yang - mills @xmath2 multiplet ( one vector field @xmath22 , two majorana spinors @xmath23 and @xmath24 and one complex scalar @xmath25 ) . when the bare mass @xmath11 of the matter multiplet is zero , we recover the @xmath1 theory . when @xmath26 , the moduli space of vacua is a coulomb branch where the gauge symmetry is spontaneously broken down to u(1 ) by a non zero higgs vacuum expectation value , @xmath27 . when @xmath12 and @xmath1 , we have other equivalent coulomb branches , permutted by the su(4)@xmath28 symmetry , corresponding to the other scalar fields having a vev . in addition to the u(1)s associated to the electric @xmath29 and magnetic @xmath30 charges , the theory has an abelian u(1 ) global symmetry @xmath31 , @xmath32 , with corresponding charge @xmath17 . classically , the central charge of the supersymmetry algebra is @xmath33 where @xmath9 is the gauge coupling constant . semiclassically , the electric charge @xmath29 is @xmath34 , @xmath35 integer , up to terms coming from cp violation ( like a standard witten term proportional to the bare @xmath16 angle ) . the magnetic charge is given by the dirac quantization condition , @xmath36 for an integer @xmath37 . we will loosely call in the following the integers @xmath35 and @xmath37 the electric and magnetic charges . the mass of a bps state @xmath38 will then be @xmath39 for instance , an elementary quark @xmath40 has a tree level mass @xmath41 . the exact quantum formula for @xmath42 can be found . it was shown in @xcite that it can be cast in the form @xmath43 here @xmath44 is the dual variable of @xmath45 and can be , at least in principle , computed very explicitly as a function of a gauge invariant coordinate @xmath46 on the moduli space ( @xmath46 is @xmath47 up to a constant ) @xcite . at the tree level , we have @xmath48 , where @xmath49 is a very convenient combination of the gauge coupling @xmath9 and the theta angle @xmath16 which has a simple transformation law under s duality . note that @xmath50 is not renormalized since the @xmath8 function is zero . together with the mass @xmath11 and the coordinate on the moduli space @xmath46 , it is the only parameter in the theory . the constants @xmath51 in ( [ cc ] ) are only indirectly related to the physical @xmath17 charges @xcite , but can be computed very explicitly , see below . this theory differs from the preceding one by its matter content . now we have four @xmath2 hypermultiplets @xmath52 , @xmath53 , transforming in the fundamental representation of the gauge group su(2 ) . we will limit ourselves , for reasons that will become clear in section 3 , to the case where two of these hypermultiplets are massless , say @xmath14 , and the other two have identical masses @xmath13 . the global non abelian symmetry of the theory is then @xmath54 ( concerning the occurence of the universal cover spin(4 ) of so(4 ) instead of so(4 ) itself for the symmetry associated with the two massless hypermultiplet , see the next subsection ) . this shows that some states are likely to come in doublets in this theory , as we will see later . the moduli space of vacua has again a coulomb branch where dyons can exist , in addition to higgs branches which we will not study . the formula ( [ cccl ] ) is still valid along the coulomb branch , but now we will have in general @xmath55 , @xmath35 integer , instead of @xmath56 , because the gauge group is really su(2 ) and no longer so(3 ) in this theory . the elementary quarks have @xmath57 for instance . moreover , the @xmath17 charge will correspond to the transformations @xmath58 , @xmath59 for the quarks @xmath60 and @xmath61 simultaneously . at a given point on the moduli space of vacua , the parameters of the theory will be @xmath11 and a generalized unrenormalized gauge coupling @xmath62 . note the difference between this definition and the corresponding one in the so(3 ) theory . the advantage of this choices will become apparent in section 3 . moreover , in order to keep the formula ( [ cc ] ) in the same form in the two theories , we will define here the parameter @xmath45 by @xmath63 . these different sets of conventions were already used in @xcite and allow to make the formal similarity between the two theories under study very explicit , see section 3 . from the semiclassical point of view , the solitonic states appear as bound states of a supersymmetric quantum mechanics describing the low energy dynamics of interacting dyons @xcite . the dynamical variables of this supersymmetric quantum mechanics correspond to zero modes of the elementary fields of the quantum field theory . in a @xmath37 monopole configuration , there are @xmath64 real bosonic zero modes . the four zero modes that are present for all @xmath65 correspond to the center of mass motion ( three modes ) and to the global electric charge ( one periodic mode ) . the other @xmath66 zero modes describe the relative motions and electric charges . to these bosonic zero modes are associated some fermionic one . from callias index theorem @xcite we know that the two adjoint majorana spinors of the @xmath2 vector multiplet will give @xmath67 complex zero modes @xmath68 , @xmath69 , carrying spin @xmath70 . the other fermionic zero modes come from the matter fermions and differ in the so(3 ) and @xmath7 theory . there , the adjoint fermions in the matter multiplet will yield @xmath67 additional complex fermionic zero modes @xmath71 . these zero modes carry spin @xmath70 and @xmath17 charge @xmath72 . the quantization of the four @xmath73 and @xmath20 fermionic zero modes corresponding to @xmath74 yields to a spectrum of @xmath75 states . when @xmath12 , this correspond to a full short @xmath1 multiplet . this shows that monopoles @xmath76 may be the dual of the w bosons . when @xmath77 , there are some additional zero modes . however it will become clear in the following sections that the @xmath77 stable bound states that exist in the theories under study all have the same quantum numbers as the @xmath76 monopoles ( in some sense , they can be generated by a continuous deformation of the @xmath76 monopoles or of the elementary excitations ) . this means that for the stable states , the additional zero modes play no rle , and thus we will discard them in the following . when the mass @xmath11 is set to a non zero value , the @xmath1 multiplet splits into three parts corresponding to states having the same @xmath17 charge and thus the same mass . more precisely , from the sixteen states of the @xmath1 multiplet we get : eight states corresponding to one @xmath2 vector multiplet , which may be the dual to the @xmath2 w bosons @xmath78 ; four states corresponding to one - half of a cpt self conjugate @xmath2 hypermultiplet ( the other half is obtained by quantization in the @xmath79 monopole sector ) , which may be the dual of one component of the adjoint elementary quark , say @xmath80 ; and four states which may be the dual to @xmath81 . this is summarized in table 1 . [ states1 ] .the sixteen states of a @xmath1 multiplet , with their spin and @xmath17 charge . @xmath82 is the a priori unknown @xmath17 charge of the state @xmath83 . the @xmath2 content of the @xmath1 multiplet is also indicated . [ cols="<,<,<,<,<,<,<",options="header " , ] in the bps mass formula ( [ bpsmass],[cc ] ) appears , in addition to the electric and magnetic quantum numbers @xmath35 and @xmath37 , a term @xmath84 proportional to the mass . the computation of @xmath51 for each bps state is crucial in order to have quantitative predictions for the physical masses , as we will need in the following . however , the meaning of the constant @xmath51 is not clear at first sight . it can not be identified with the physical @xmath17 charge , because the latter has a non trivial dependence on the higgs expectation value due to the spontaneous breaking of cp invariance . this is , for the @xmath17 charge , a similar effect as the witten phenomenon for the electric charge @xcite , and was discussed in @xcite . there it was shown that the non trivial part of the physical @xmath17 charge is actually automatically included in the periods @xmath45 and @xmath44 , and is responsible for the constant shifts these variables can undergo under duality transformations . this possibility is related to the fact that @xmath45 and @xmath44 are period integrals of a meromorphic one form having poles with non zero residus @xcite , see appendix a. the constants @xmath51 are thus simply a constant part of @xmath17 not already included in @xmath44 and @xmath45 . they can always be determined by consistency considerations , see section 4.4 and the discussion in section 5 . in this section , some arguments are presented which lead to a clear physical understanding of _ why _ states with any magnetic charges must exist in the theories we are studying , in the context of quantum field theory alone . we will heavily rely on the seiberg - witten low energy effective action , encoded in the elliptic curve presented in @xcite . thus this section begins with a brief review of this result , emphasizing that it is by now established at a high level of rigour , and above all _ does not _ rely on any duality assumption . this short survey will also serve to set our notations . because of @xmath2 supersymmetry , the variables @xmath86 and @xmath45 , which not only completely determine the form of the low energy effective action ( up to two derivatives and four fermion terms ) , but also enter the bps mass formula ( [ bpsmass ] ) , must be analytic functions of the gauge invariant coordinate @xmath46 on the coulomb branch @xcite . these analytic functions have branch cuts and sl@xmath15 monodromies due to singularities caused by charged particles becoming massless for some values of @xmath46 . the number and type of singularities occuring on the coulomb branch can most easily be found by analyzing the theory in a limit where the mass @xmath11 is very large compared to some scale @xmath87 which represents the dynamically generated scale of the asymptotically free theory we obtain after integrating out the supermassive hyper(s ) . quantitatively , this limit corresponds to @xmath88 with @xmath89 let us focus for concreteness on the so(3 ) theory . in this regime , and at scales of order @xmath87 , the theory must look like the pure gauge theory , since the ultra - massive quarks must decouple . the singularity structure of the pure gauge theory was already greatly deduced in @xcite , and it was recently studied at an even higher degree of rigour @xcite . the result is that , without making any duality assumption , one can prove that we have two singularities at strong coupling whose monodromy matrices are conjugate to @xmath90 . moreover , we must have at scales of order @xmath11 , where a tree level analysis must be valid when @xmath91 , an additional singularity coming from an elementary quark becoming massless . the monodromy matrix is here again conjugate to @xmath90 , as can be deduced with the help of the @xmath8 function of the low energy theory @xcite , or even directly in the microscopic theory @xcite . actually , the monodromy transformation for @xmath92 around a singularity at @xmath93 due to a state @xmath94 becoming massless is @xmath95 where @xmath96 searching for an elliptic curve of the form @xmath97 , where @xmath98 is a polynomial of degree three , whose moduli space will reproduce the singularity structure described above , is then a fairly simple mathematical exercice . the solution is unique and given by @xcite @xmath99 where @xmath100 and @xmath101 the discussion above was for the so(3 ) theory , but as noted in @xcite , the singularity structure and thus the curve is exactly the same in the @xmath102 theory , provided one chooses bare masses @xmath103 and @xmath104 for the matter hypermultiplets , uses the different sets of convention introduced in section 2.1 , and expresses the curve in terms of a dimensionless constant @xmath50 which is @xmath105 only at the tree level but receives one loop as well as non perturbative corrections @xcite . though this useful relation will allow us to study the so(3 ) and the @xmath102 theories in parallel , be careful that it is only a formal one . for instance , only even , @xmath106-instantons corrections exist in the @xmath102 theory , whereas all @xmath107-instantons contribute in the @xmath108 theory . with our conventions , these corrections will both be proportional to @xmath109 . there are also important differences at the level of the spectra . one of them is that in the so(3 ) theory , a state @xmath110 may correspond to a vector multiplet ( w bosons ) , whereas in the @xmath102 theory the ws are labeled as @xmath111 . moreover , due to the flavour symmetry , the singularities in the @xmath7 theory are produced by su(2 ) doublets of hypermultiplets becoming massless . the general form of the monodromy matrices ( [ monomat ] ) remains valid since if the solution for the periods is noted @xmath112 for the so(3 ) theory , it will be @xmath113 for the @xmath7 theory . the parameter @xmath46 appearing in ( [ curve ] ) is a good gauge invariant coordinate on the coulomb branch , which is related to the physical expectation value @xmath114 by a formula of the form @xmath115 the @xmath116 correspond to instanton corrections @xcite . we conventionally extracted the combination @xmath117 because it tends towards the physical parameter @xmath118 under the renormalization group flow towards the pure gauge theory or the @xmath119 massless theory ( [ rg ] ) . the transformation properties of the parameters @xmath120 of the theories under s duality can easily be read from the curve ( [ curve ] ) . under a general sl@xmath15 transformation given by a matrix @xmath121 we have @xmath122 the fact that these strong - weak coupling transformations can be coherently implemented on the curve ( [ curve ] ) is already a non - trivial evidence that the theories may be self - dual @xcite . below , we will investigate the transformation properties of the variables @xmath86 and @xmath45 ( or equivalently of the electric @xmath123 and magnetic @xmath124 charges ) . this will give rise to predictions on the bps spectra . checking that these predictions are indeed true will be the main goal of the forecoming sections . in the remainder of section 3 , we will set the bare @xmath16 angle to zero . the very interesting physics associated with the possibility of having a varying real part in @xmath50 will be considered in section 4 . to guess what the analytic structure ( position of the singularities and of the branch cuts ) looks like , it is simplest to choose the parameters so that we are not far away from the pure gauge theory ( or massless @xmath125 ) , where we easily control what is going on . at strong coupling , we have two singularities on the real axis , at points @xmath126 and @xmath127 such that @xmath128 , with an analytic structure already worked out in @xcite . moreover , at weak coupling , we have a quark becoming massless at @xmath129 , with the following asymptotics for @xmath86 and @xmath45 : @xmath130 the exact location of the singularities @xmath131 can be found by setting the discriminant of the curve ( [ curve ] ) to zero , @xmath132 [ figure1 ] = 14.7 cm we have @xmath133 . we choose the position of the branch cut for the logarithm appearing in the asymptotics ( [ asympq ] ) in such a way that it does not disturb the analytic structure of the pure gauge theory . this is the most natural choice , since the existence of a singularity at a scale @xmath134 can not influence the physics at scale @xmath87 . in figure 1 are also indicated the quantum numbers of the particles becoming massless . note that due to a `` democracy '' between the dyons @xcite , the most general choices for the electric and magnetic charges of these states are @xmath135 where @xmath136 corresponds to the singularity @xmath137 and @xmath35 is any integer . our choice is @xmath138 . note also that due to the cuts , the particle becoming massless at @xmath126 is described by two different sets of integers , @xmath139 or @xmath140 , depending on whether one looks from the @xmath141 or @xmath142 half - plane . this shows that one must introduce two different monodromy matrices at @xmath126 , @xmath143 the other monodromy matrices are @xmath144 it is now possible to study the transformation properties of @xmath112 ( or equivalently of @xmath145 ) under duality . the naive guess would be that @xmath92 transforms exactly according to the corresponding sl@xmath15 matrix @xmath146 acting on the parameters of the microscopic theory as in ( [ dualt ] ) . however , we will see that this is not true in general . we must introduce an sl@xmath15 matrix @xmath147 which acts on the low energy effective action ( i.e. @xmath44 and @xmath45 ) while @xmath146 acts on the microscopic parameters . this subtlelty stems from the fact that the monodromy group associated with the curve ( [ curve ] ) is @xmath5 , the subgroup of sl@xmath15 consisting in the matrices congruent to the identity modulo 2 . the only straightforward and correct statement that can be done is thus that @xmath148 mod 2 . moreover , in the massive theories , @xmath44 and @xmath45 can also pick up some constants under a duality transformation . this is allowed by the form of the bps mass formula ( [ bpsmass],[cc ] ) , since a shift in @xmath44 or ( and ) @xmath45 can be reabsorbed in a shift of @xmath51 . a very similar and related phenomenon occurs when one is studying the duality transformations of the low energy theory @xcite , but one must keep in mind that by now we are studying duality transformations of the whole , microscopic , theory . let us be more concrete . what we want to deduce is a formula relating the dual theory solution characterised by the parameters @xmath149 given by ( [ dualt ] ) and whose periods are @xmath150 to the original solution described by @xmath151 . let us focuss in this section on the transformation @xmath152 . the analytic structure of @xmath153 can be readily deduced from the one of @xmath112 depicted in figure 1 . indeed , as @xmath154 , @xmath155 and @xmath156 , the effect of the duality transformation will simply be to exchange the singularities at @xmath129 and @xmath126 while the singularity at @xmath127 remains fixed . the dual theory thus has singularities at exactly the same points as the original theory ( a direct consequence of the self - duality of the curve ( [ curve ] ) ) , but the quantum numbers of the states becoming massless at a given point , as well as the position of the branch cuts , are changed , see figure 2 . [ figure2 ] = 14.7 cm this is enough to deduce that @xmath157 the shifts by @xmath158 come from the fact that a dyon is exchanged with a quark , and these two states have different @xmath51 . moreover , the s duality transformation does not simply exchange @xmath45 and @xmath44 ( up to a sign , and up to the shifts ) . this is not surprising since the state becoming massless at @xmath159 is @xmath160 in both formulations , and this state is not self - dual ( is not an eigenvector of s ) . second , to eliminate the cut between @xmath159 and @xmath161 in the dual formulation ( and thus recover the same structure for the cuts as in the original formulation ) , it is necessary to perform an sl@xmath15 transformation , given on one side of the cut by the monodromy @xmath162 . this also explains why the transformation law depends on the sign of @xmath163 . we are by now in a position to state precisely what the duality @xmath85 implies for the dyon spectrum . these predictions will be tested in the following . at a very general level , suppose you postulate that the theories described by the variables @xmath112 and @xmath164 are equivalent . if @xmath165 where @xmath166 is an sl@xmath15 matrix and @xmath167 and @xmath168 are @xmath169-valued constant sections ( this is the most general transformation law we will encounter ) , then the existence of a state @xmath170 of charge @xmath51 in the theory @xmath112 will imply the existence of a state @xmath171 of charge @xmath172 in the theory @xmath164 with @xmath173 the @xmath174 product denotes the standard symplectic product , @xmath175 the formula ( [ galtrans2 ] ) is a straighforward consequence of the bps mass formula ( [ bpsmass ] ) . let us apply ( [ galtrans2 ] ) to the s transformation @xmath176 described by ( [ periodstrans ] ) . suppose that you have a state @xmath145 in the theory at given @xmath46 , @xmath50 and @xmath11 . this state may be a vector or a hyper multiplet of @xmath2 supersymmetry , and lie in a given representation of the flavour group . then s duality gives a prediction for the spectrum of the theory whose parameters are @xmath177 . if @xmath142 , we must have there a state @xmath178 , in a similar @xmath2 multiplet as the original state . the flavour quantum numbers may be changed as discussed in section 2.2 . if @xmath141 , the state predicted by duality will be @xmath179 . it is very important to realize that these predictions of duality relate theories having different parameters , even at the self - dual point @xmath180 . thus a priori they say nothing about the dyon spectrum of a given theory ( i.e. , at fixed @xmath46 , @xmath50 and @xmath11 ) . one may be tempted to argue that , due to the stability of the bps states , the parameters can be varied continuously without changing the spectrum . this means that a state which exists in the theory @xmath177 will also exist in the theory @xmath120 . this reasoning is indeed correct when @xmath12 , but is wrong in the massive case , due to the presence there of curves of marginal stability , accross which otherwise stable bps states become degenerate in mass with multiparticle states . on these curves , the decay of bps states is possible . we will investigate these decays in the following . their mere existence implies that _ the self - duality of a general @xmath2 theory does not entail that the spectrum of bps states is self - dual at a given point on the moduli space . _ let us now penetrate at the heart of our problem : is there an easy way to understand why the spectra of the massless theories should be self - dual ? can we readily imagine what the spectrum of the massive theories look like ? there is a regime where the second question can be easily answered , at least partially . by choosing the parameters as indicated in ( [ rg ] ) , we can integrate out the quarks having a bare mass , and we are thus left with the pure gauge theory ( for so(3 ) ) or with the su(2 ) theory with two massless flavours . the spectra of these theories are well - known @xcite . discarding for the moment the states having @xmath181 , which can have arbitrarily high physical masses when @xmath182 , two regions in the coulomb branch must be distinguished . indeed , to the possible decays of @xmath183 states into other @xmath183 states corresponds a unique curve of marginal stability is associated a whole family of curves . ] @xmath184 which looks like an ellipse and contains the singularities @xmath185 and @xmath159 @xcite , see figure 3 . [ figure3 ] = 14.7 cm it can be shown that inside the curve ( in the strong coupling region @xmath186 ) only the two bps states which become massless at @xmath126 or at @xmath127 can be present @xcite . these are @xmath187 and @xmath188 ( for this latter state , two different descriptions in terms of electric and magnetic quantum numbers are necessary , due to the non trivial analytic structure @xcite ) . outside the curve , in what we call the weak coupling region @xmath189 , the bps spectrum can be understood in different ways . the most natural certainly is to perform a semiclassical analysis , which is valid here because of asymptotic freedom ( in the context of the full , finite , theory , this means that we are considering a region of the coulomb branch where @xmath190 ) . this analysis first tells you that the perturbative states , created by the fundamental fields , must be present . in the so(3 ) theory , we have the w bosons @xmath191 and the photon @xmath192 which lie in vector multiplets . in the @xmath119 theory we again have the w bosons , now represented as @xmath193 , the photon , but also the elementary quarks @xmath194 and @xmath195 . beyond this perturbative spectrum , we have a solitonic spectrum . in the one monopole sector , we can construct all the states @xmath196 . the fact that all the integers @xmath123 are realized simply comes from the quantization of the periodic zero mode associated with electric charge rotations . statements about higher magnetic charges would be much harder to do . it was shown in @xcite that no state with @xmath197 exists , but a complete analysis for @xmath198 seems by now impossible , for the explicit form of the multimonopole metric is not known in general in these cases . however , there is another method , initiated in @xcite , which consists in remarking first that the spectum in @xmath199 must be invariant under the transformation @xmath200 this can be shown by transporting the state @xmath201 along a closed path encircling the two singularities and which does not cross @xmath202 , so that the state remains stable ( see figure 3 ) . in this operation , we cross the cut @xmath203 and thus should perform an analytic continuation of the periods @xmath86 and @xmath45 in order to insure that the physical mass of the state varies continuously . an analytic continuation @xmath204 across a cut will in general be related to the main solution @xmath92 by a relation of the form ( [ galtrans1 ] ) ( @xmath168 is indeed @xmath169-valued because the residues of the seiberg - witten differential form are @xmath205 , see appendix a ) . this shows that a state represented by @xmath94 on the one side of the cut will be represented by @xmath206 on the other side , see ( [ galtrans2 ] ) . for the cut @xmath203 , we will have @xmath207 and @xmath208 ( or the inverse matrix depending on whether we turn clockwise or counterclockwise ) , which indeed generates the transformation ( [ scinv ] ) . since the states @xmath160 and @xmath209 must be present ( they are responsible for the singularities ) , we immediately deduce that all the states @xmath196 are also present , for any integer @xmath123 . the reader might think that this is really a roundabout mean to show that electric charge is quantized . however , this reasoning also provides a statement about the @xmath210 sectors , powerful enough to show that no state with @xmath210 can exist . indeed , if such a state @xmath201 would exist , it would be associated with the whole tower of states @xmath211 , for any integer @xmath212 , angle shifts by mean of which you can not show that all these states exist at the same time . ] and _ one of these states will inevitably becomes massless _ somewhere on the curve @xmath202 since @xmath213 takes all real values between 0 and 2 on this curve @xcite . this would produce an additional singularity on the coulomb branch which does not exist . the greatest virtue of this kind of reasoning is that one can generalize it to other cases . we have just studied the rg flow ( [ rg ] ) towards the pure gauge ( or @xmath119 ) theory . it would be as natural to study the s dual of this rg flow , which would correspond to @xmath214 where @xmath215 this is inherently a strong coupling limit where the semiclassical method can not be applied . the solution for the dual pure gauge ( or massless @xmath119 ) theory is given by @xmath216 , and its analytic structure has already been worked out , see figure 2 . it is of course very tempting to try to repeat the analysis already made for the original theory . however , we can not generalize the reasoning at once because the position of the branch cuts are now completely different . the fact that we will nevertheless obtain a spectrum in complete accordance with duality is a nice evidence that s duality is indeed correct . the aim of this subsection is to work out the bps spectra implied by the analytic structure depicted in figure 2 . it is convenient to use , instead of @xmath216 , the periods @xmath217 this simply amounts to setting @xmath51 to zero for the two states @xmath110 and @xmath160 becoming massless . we will note with an uppercase @xmath146 the @xmath51 charge computed with the new variables , @xmath218 we have again a curve of marginal stability @xmath219 , which coincides with @xmath202 because of ( [ periodstrans ] ) . [ figure4 ] = 14.7 cm the coulomb branch is thus separated into two regions . the one inside the curve will be called in this context the weak coupling region @xmath220 and is the dual of the strong coupling region @xmath221 one had previously . outside the curve , we will have the strong coupling region @xmath222 . these names deserve some comments . in any case , one should keep in mind that the bare gauge coupling of the whole ( finite ) theory is very large in the limit ( [ rgdual ] ) . thus , by strong or weak coupling , we only refer to the effective coupling which governs the low energy physics . moreover , even in @xmath220 for instance , the effective theory may be strongly coupled ( it is near @xmath159 since there a magnetically charged particle is massless ) . the point is that the regions in @xmath220 where the coupling is strong can always be joined to other regions in @xmath220 where the coupling is weak , the dyon spectrum remaining unchanged . indeed , the _ electric _ effective coupling is zero at @xmath161 where the quark is massless . this suggests that the bps spectrum might be probed in the vicinity of @xmath161 by using standard , `` semiclassical , '' approximation schemes , since we do have in this region a small parameter in the original , `` electric , '' theory . actually , if duality is correct ( and we will prove that its predictions indeed are concerning the dyon spectrum ) , the physics should be the same in the weakly coupled region surrounding @xmath161 in figure 4 as in the strongly coupled region surrounding @xmath185 in figure 3 , and we should be able to account for the `` strong coupling '' jumps of the bps spectrum seen in @xcite by using such methods . however , we will not try to do that here . let us rather apply our method to the weak coupling spectrum @xmath223 . it contains of course the states @xmath160 and @xmath110 , which are responsible for the singularities and thus exist on both sides of the curve . we will now show that no other state , having @xmath224 , can exist . this can be understood by studying the variations of @xmath225 along the curve @xmath226 . @xmath225 is related to @xmath213 by the formulas ( [ newsol ] ) and ( [ periodstrans ] ) , and @xmath213 varies monotonically from 0 to 2 when one follows @xmath226 clockwise starting from and ending at the singularity where @xmath110 is massless . this immediately implies that @xmath225 takes all values between @xmath227 and 1 in the upper half plane , and all values between 1 and @xmath228 in the lower half plane . thus any state @xmath201 existing in @xmath229 will become massless somewhere on @xmath226 , at the point where @xmath230 , and then must be either @xmath110 or @xmath160 . this is in perfect agreement with s duality . when looking at the strong coupling spectrum , the complications come from the fact that one must consider analytic continuations across the cuts @xmath231 and @xmath232 . we then need to introduce two different descriptions @xmath233 and @xmath234 in terms of electric and magnetic quantum numbers of the _ same _ physical spectrum @xmath235 , depending on whether we are in the upper or lower half plane . this is satisfactory from the point of view of duality , see section 3.3 . quantitatively , a state @xmath201 in @xmath236 will be @xmath237 after crossing the cut @xmath232 and @xmath238 after crossing the cut @xmath231 . by looping around the curve of marginal stability , taking care of remaining in @xmath239 , we thus immediately deduce that @xmath234 must be invariant under the subgroup of sl@xmath15 generated by @xmath240 : @xmath241 applying this transformation to the states @xmath110 and @xmath160 , which we know must exist , will then generate states having _ any _ magnetic charge : @xmath242 actually , this shows that all the states dual to the @xmath196 states of the original theory will be present . moreover , to each state @xmath196 corresponds one and only one state in the dual theory , since there the states are in one to one correspondence with states becoming massless . are there any other states with @xmath224 ? if so , let @xmath201 be such a state in @xmath243 . if @xmath244 , it is the state @xmath110 which indeed exists for it is responsible for one of the singularity . , @xmath245 can exist since they would modify the structure of the singularity at @xmath161 . ] otherwise , let us consider the ratio @xmath246 . this ratio necessarily is in the interval @xmath247-\infty , 1[$ ] , since otherwise the corresponding state would become massless somewhere on the curve @xmath226 ( note that we are working in the lower half plane and thus we see only half of the curve ) . as explained above , in the upper half plane the same state is described by other electric and magnetic quantum numbers . crossing the cut @xmath248 , we will obtain another ratio @xmath249 , which must be in the interval @xmath2471,+\infty [ $ ] unless the particle becomes massless somewhere on the coulomb branch . one can return in the lower half plane by crossing the cut @xmath250 , and obtain again a new ratio @xmath251 . by repeating this operation , we obtain two homographic sequences @xmath252 and @xmath253 , @xmath254 with the following important property : if , for some @xmath255 , are obtained by looping counterclockwise around @xmath226 . ] @xmath256 or @xmath257-\infty , 1]$ ] , then the state @xmath201 we started from must be one of the states already obtained above ( see ( [ provspec ] ) ) . it is elementary to check that this is always the case , except if @xmath258 . this case corresponds , in the so(3 ) theory , to the dual of the w bosons , and is represented by @xmath259 in @xmath236 and by @xmath260 in @xmath261 . in the @xmath102 theory , we would again have , in @xmath236 , the state @xmath259 , now interpreted as being the dual of the elementary quarks , and also @xmath262 , interpreted as being the dual of the w. up to now , we did not take into account eventual decays of @xmath245 states into @xmath263 states along the path depicted in figure 4 , although this path can cross curves of marginal stability corresponding to such decays . however , the states @xmath264 have a mass of order @xmath11 , whereas the states @xmath224 are likely to have a mass of order @xmath87 in the region surrounding the singularities @xmath127 and @xmath129 we are considering . since in the rg flow , @xmath182 while @xmath87 is fixed , the decays are impossible and the curves of marginal stability simply indicate that the inverse decay reaction from a @xmath264 to a @xmath224 state is possible . one might put forward the fact that states having @xmath264 but arbitrarily high electric and magnetic charges could have a mass of order @xmath87 due to a subtle compensation of the terms @xmath265 and @xmath266 in the bps mass formula . however such fine tuning , which is indeed possible , can only occur for some very special values of @xmath123 and @xmath124 and in some very special and small regions in the @xmath46-plane . nothing prevents us to avoid these special points and to cross the curve of marginal stability nearby , where the @xmath264 state again have a mass of order @xmath11 . i wish here to point out that , though we have considered the possibility of these `` fine tuning '' points , they are very likely to be unphysical . indeed , if states producing these `` fine tuning '' points would exist , the theory would not have a well defined field theory limit under the ( dual ) rg flow . we have thus obtained the following results : states with any magnetic charges , dual to the dyons @xmath267 , indeed exist , with the correct multiplicities . thus , though the positions of the branch cuts are completely different in the original and in the dual theories , the deduced spectra are perfectly compatible with duality . however , we have not yet proven that the states corresponding to @xmath268 are really present ; we simply know that they may be . this point will be investigated in appendix b. in the preceding section , we have shown using simple arguments that stable states having any magnetic charge , which are required by s duality , indeed exist . we were studying theories with non zero bare masses , but it is very tempting to think that these states can be continuously deformed while remaining stable when @xmath269 and thus are at the origin of the existence of states having any magnetic charge in the massless theories as well . however , we will not obtain in this process all the states required by duality in the massless theories . for instance in the so(3 ) ( @xmath1 ) theory s duality implies the existence of all the states @xmath201 with @xmath123 and @xmath124 relatively prime . up to now , we only have the states @xmath270 with @xmath271 , and it is not difficult to realize in the context of the semiclassical approach @xcite that once you have one of these states , then you will also have all the other states @xmath272 , for any integer @xmath212 . actually , the missing states should come from configurations where states having any @xmath35 and @xmath37 relatively prime are massless at the singularities . the fact that such configurations must exist is clearly a necessary condition for the self - duality of the theories , and does not follow straightforwardly from the sl@xmath15 invariance of the curve ( [ curve ] ) . we saw above that for any imaginary value of @xmath50 ( @xmath273 ) , the singularities are produced by states having @xmath274 or @xmath275 alone . one must then study cases where @xmath276 and find a mechanism to understand how the quantum numbers at the singularities can change . this is similar to a phenomenon which is known to occur for instance in the massive su(2 ) asymptotically free theories , where semiclassical quarks must `` transmute '' into monopoles in order to account for the singularity structure at strong coupling @xcite . however , there can be an important difference between this kind of transmutation and the one we will see at work below . we will realize that two kinds of transmutations can occur . in the first kind , the quantum numbers of the state are changed according to a matrix of the form ( 8) . we will see in the next subsections that this is the only kind of transmutation that can occur in the theories under study in this paper , because singularities in the moduli space never coincide , see ( [ singpos ] ) . in the second kind of transmutation , the state transforms according to a matrix which is not of the form ( 8) ( note that , in any case , there are many of them , since the matrices ( 8) are all conjugate to some power of @xmath277 ) . these transmutations of the second kind are directly related to the necessity for two ( or more ) singularities to coincide for some special values of the parameters @xcite . these special points are known to lead to superconformal theories @xcite . let us come back to our problem . we will soon show that the quantum numbers at the singularities can only be changed by matrices belonging to the _ monodromy _ group @xmath278 ( in the cases under study , we will see that any matrix of @xmath279 can be generated ) . it is also clear that a necessary condition for the theories to be self - dual is that any state , related to the original states becoming massless by a _ duality _ group transformation , must become massless for some value of the coupling . if the @xmath136 denote the electric and magnetic quantum numbers of the particles becoming massless in the original configuration , we see that the theory can be self - dual only if @xmath280 where @xmath281 denotes the _ duality _ group . for our purposes , @xmath282 , @xmath283 , @xmath284 and @xmath285 . we thus have to check whether any @xmath0 , @xmath3 and @xmath4 relatively prime , can be generated by acting with a @xmath5 matrix on one of the three states @xmath209 , @xmath160 or @xmath286 . this is indeed true because sl@xmath287 and that @xmath288 is generated by three transpositions whose corresponding matrices ( already described mod 2 in section 2.2.2 ) can be chosen in order to map a `` fundamental '' state , say @xmath110 , exactly into the three states @xmath110 , @xmath160 and @xmath286 we have at our disposal . let us start from a configuration , characterized by a certain value of the bare coupling @xmath290 and of the bare mass @xmath11 , and whose analytic structure is of the same type as the one described in figure 1 . we will allow the monodromy matrices @xmath291 to be of the most general form , provided they satisfy the fundamental consistency conditions @xmath292 to these monodromy matrices correspond states becoming massless , characterized by their electric and magnetic quantum numbers @xmath293 and @xmath51-charge @xmath294 . note that ( [ consistency ] ) implies via ( [ monomat ] ) , amongst other relations of the same type , @xmath295 actually , as we will only consider configurations which are sl@xmath15 transforms of the original one depicted in figure 1 , sl@xmath15 symplectic invariant relations can be checked or deduced using @xmath296 , @xmath297 , @xmath298 and @xmath299 . some useful ones are @xmath300 we must also have @xmath301 now , let us vary continuously @xmath50 from @xmath290 to @xmath302 . during this operation , the singularities , and with them the branch cuts , will move on the coulomb branch , as depicted in figure 5 . [ figure5 ] = 14.7 cm unavoidably , the singularity at @xmath161 will cross the cut @xmath303 $ ] ; at this moment its quantum numbers are changed . according to ( [ mono],[galtrans2 ] ) , we will have @xmath304 the cut originating at @xmath161 will also sweep up one of the two singularities , and then change the quantum numbers there . we choose to change the singularity at @xmath185 ( see figure 5 ) because it is the only case which leads to an analytic structure at @xmath305 which is perfectly similar to the one at @xmath290 , and this is convenient for our purposes . the ambiguity associated with the position of the branch cut originating at @xmath161 is further discussed in the next subsection . as @xmath139 sees the original cut produced by @xmath306 , whereas @xmath140 sees the cut produced by @xmath307 , we will have @xmath308 finally , by using the consistency relations ( [ consistency],[consistency2],[consistency3 ] ) , we see that at @xmath309 we are in essentially the same configuration as at @xmath310 , up to a @xmath311 transformation and to a rotation and dilatation in the @xmath46 plane . quantitatively we have @xmath312 thus we have shown that we can change the quantum numbers of the singularities with the matrix @xmath311 ( the @xmath51 charges being also changed according to ( [ galtrans2 ] ) ) . performing @xmath311 transformations is not enough to prove sl@xmath15 invariance . we need an additional transformation , which can be found by studying the analytic continuation @xmath314 exactly along the lines of the preceding subsection . we will not repeat this argument here , but we will rather introduce another idea , which is maybe more heuristic , but has the merit of showing that the structure of the low energy effective action can imply the _ equivalence _ of theories related by some transformations of the monodromy group . we will elaborate more on this kind of argument in section 8 . the point is that by sweeping up the @xmath46 plane with the branch cut originating at @xmath161 , as shown in figure 6 , we can generate the transformation : would generate @xmath315 which is equivalent to @xmath316 due to the relations ( [ consistency ] ) . ] @xmath317 [ figure6 ] = 14.7 cm doing this on the original configuration of figure 1 is innocent since it will generate a trivial @xmath318 transformation , which is interpreted as an unphysical relabelling of the electric quantum numbers @xmath35 . it simply corresponds to a shift in the @xmath16 angle , which when combined with the change in @xmath35 will leave the physical electric charge and other physical observables unchanged . this shows that the position of the branch cut originating at @xmath161 is purely conventional when @xmath319 , and by continuity this ambiguity should be preserved even when the @xmath316 matrix is general ( we will see in the next subsection that any @xmath316 conjugated to @xmath318 by a @xmath5 matrix can be generated by analytic continuation ) . note that we performed the transformation at _ fixed _ @xmath46 , @xmath50 and @xmath11 . in the original configuration where @xmath319 , this renders very explicit the fact that the transformation @xmath320 is unphysical . is unphysical . however , this is not that obvious in this framework , and we will limit ourselves to transformations of the type of @xmath318 . ] this also shows that the physical interpretation of the parameter @xmath50 is no longer the same after analytic continuation : if we choose to still use @xmath50 even after the @xmath16 angle shift , the real part of @xmath50 clearly will no longer be @xmath321 ( or @xmath322 ) . this is even more striking when @xmath316 is general . then , the transformation mixes the electric and magnetic charges in a non trivial way , still at fixed @xmath50 . this is not contradictory to the dirac quantization condition . again , this shows that the duality transformation associated with @xmath316 links together two theories which are physically equivalent and thus can be labelled by the same parameters . of course , the physical interpretation of @xmath50 in the two equivalent theories will not be the same . similar remarks also apply to @xmath46 . to close this discussion , note that the ambiguity in the interpretation of @xmath50 we deal with in this subsection is related to the fact already mentioned in subsection 3.2 that the sl@xmath323 matrix @xmath147 which acts on @xmath112 coincides with the sl@xmath15 matrix @xmath146 which acts on @xmath50 and @xmath46 , _ only up to monodromy transformations . _ this suggests that @xmath50 can only be defined modulo @xmath5 and thus that the structure of the low energy effective action implies that the theory has at least an exact @xmath5 duality symmetry . we will see in section 8 that this can be understood on physical grounds , at least heuristically . let us come back to our original problem : can we generate any @xmath5 transformation ( up to a sign ) by combining the two elementary transformations ( [ m3trans],[m1trans ] ) described in the previous subsections and that we will denote abstractly @xmath324 and @xmath325 ? suppose that we execute the following sequence of transformations , @xmath326 starting from the configuration where the monodromy matrices are given by ( [ matrices],[matrices2 ] ) . one must take care of the fact that , after each step , the monodromy matrices are conjugated by the corresponding transformation . for instance , the transformation @xmath327 will indeed generate a @xmath328 transformation on the variables @xmath86 and @xmath45 . however , the @xmath329 transformation that follows will be implemented by the matrix @xmath330 , since the monodromy at @xmath127 is no longer @xmath331 but @xmath332 . the effect of conjugating the matrices at each step is simply to reverse the order of the transformations in ( [ seq1 ] ) to which correspond , up to a sign , the sl@xmath15 matrix @xmath333 here @xmath334 and @xmath331 are fixed matrices given by ( [ matrices],[matrices2 ] ) . we are thus able to obtain any matrix of the subgroup of sl@xmath15 generated by @xmath334 and @xmath331 . this is just @xmath5 as we need to prove the self duality of the spectrum , as explained at the end of section 4.1 . the @xmath51 charges of the @xmath335 states obtained above can also be easily computed , by evaluating the product ( [ seq2 ] ) where the monodromy matrices @xmath316 and @xmath311 are replaced by the corresponding elements of the full monodromy group , which is the semi - direct product @xmath336 ( see ( [ m1trans ] ) and ( [ m3trans ] ) ) : @xmath337 if @xmath338 then @xmath339 for @xmath60 or 3 , @xmath340 for @xmath61 and @xmath341 for @xmath342 . in order to put the present work into some perspective , i will briefly discuss below some relations with other approaches . let me first explain how our line of reasoning can shed light on some aspects of the semiclassical quantization . i will illustrate this point on the following example . it is well known in the context of the semiclassical approach @xcite that the existence of dyon - monopole bound states with @xmath343 in the pure @xmath2 gauge theory would violate the self - duality of the @xmath1 theory . this is due to the fact that such bound states are in one to one correspondence with the cohomology classes of _ holomorphic _ differentials on the reduced ( @xmath66 dimensional ) multimonopole moduli space in the case of the pure gauge theory , while the differentials only need to be harmonic in the case of @xmath1 . this means that any @xmath343 bound state of the pure gauge theory corresponds to @xmath344 bound states in the @xmath1 theory , the second state corresponding to the antiholomorphic partner of the original holomorphic differential form . this is in contradiction with @xmath17 duality which predicts the existence of one and only one bound state of given @xmath35 and @xmath37 . from our point of view , this seemingly purely technical relation between the two theories comes from the fact that they can be joined together by continuously varying the bare mass and the coupling and using the renormalization group flow . stable states in the pure gauge theory will then yield stable states in the @xmath1 theory , by continuous deformation . moreover , we do not want any additional @xmath343 state in the @xmath1 theory coming from the pure gauge theory , since we know that such states originate from the existence of dual pure gauge theories configurations . i wish now to point out one surprising aspect of our derivation , and trace its profound origin in superstring dualities . all our reasoning is based on a careful study of the low energy effective action of the theories . though it is well known that the latter contains a lot of information about the _ massive _ states ( encoded in the bps mass formula ( [ bpsmass ] ) ) in the context of @xmath2 supersymmetric theories , it might seem astonishing that it also governs the _ formation _ of all the stable bound states ( coming in short multiplets ) . nevertheless , this fact , which appears for the first time in @xcite , is supported by increasing evidence . i hope that the present paper will convince the reader that it is both understandable ( if not natural ) and quite general in the field theory framework . however , a deeper understanding of this phenomenon comes from string theory . there , in the framework of the heterotic - type ii duality conjecture @xcite , you can argue that the bps states of @xmath2 supersymmetric yang - mills theories correspond to a self - dual non - critical string in six dimensions wrapped along geodesics around the corresponding seiberg - wittern curve @xcite . the metric on the curve is directly related to a particular seiberg - witten differential , which can in principle be computed unambiguously . as all these geometric data are contained in the low energy effective action , we understand why it completely determines the bps spectrum . in the case of @xmath1 , the low energy theory as well as the metric are trivial and the self - duality of the spectrum follows easily @xcite from the assumption that the type iib string gives rise to the @xmath1 gauge theory in some limit . the cases of some su(2 ) theories , including also very recently the so(3 ) theory studied in this paper , were addressed in @xcite . the case of @xmath7 should also be within the stringy approach capabilities . our results thus constitute non trivial tests of the string - string duality hypothesis . we will mainly focus in this section on the @xmath1 theory , in order to avoid tiresome repetitions . in subsection 5.3 , we will nevertheless briefly point out the peculiarities of the @xmath7 theory , which do not yield any new difficulties . as there is only _ one _ multiplet in the hilbert space of states corresponding to the fundamental @xmath1 multiplet @xmath110 , s duality predicts the existence of only _ one _ @xmath1 multiplet @xmath0 . as indicated in the table 1 of section 2 , such an @xmath1 multiplet can be decomposed into three parts characterized by their physical s charges @xmath82 , @xmath345 and @xmath346 . when @xmath12 , we can use cp invariance to show that the existence of a state having @xmath17 implies the existence of another state having @xmath347 , and that in a given monopole sector @xmath348 , @xmath349 . angle does not influence the physical @xmath17 charge @xcite . ] uniqueness of the @xmath0 state would therefore imply that @xmath350 , and the three values of @xmath17 that should be realized correspond to the cp invariant combination @xmath351 . state in the semiclassical picture must be self - dual or anti self - dual . ] the extremal values @xmath352 and @xmath353 correspond to states belonging to a @xmath2 hypermultiplet , while @xmath354 corresponds to a @xmath2 vector multiplet . the @xmath0 states we have generated in section 4 in the massive theory correspond to @xmath2 hypermultiplets , having a given charge @xmath355 . because of @xmath1 supersymmetry in the @xmath269 limit , these hypermultiplets will be associated to other states to form whole @xmath1 multiplets @xmath356 or @xmath357 . the physical @xmath17 charges of a state @xmath358 must be related to the constants @xmath51 appearing in the central charge of the supersymmetry algebra by a relation of the form @xcite @xmath359 where @xmath360 . to determine @xmath361 , one can proceed as follows . when @xmath12 , the solution for @xmath86 and @xmath45 is given by @xmath362 where @xmath363 is the sign of @xmath163 . the ( unphysical ) cut for @xmath86 comes from the choice we did in section 3 for the position of the branch cut originating at @xmath129 when @xmath364 . when @xmath364 , we will have @xmath365 however , if @xmath366 , the physical @xmath17 charge contributes to @xmath86 and @xmath45 , even at infinity , by an amount determined by @xmath361 in such a way that @xmath367 @xmath361 can then be computed straightforwardly by using the formulas for @xmath86 and @xmath45 given in appendix a. one must note , however , that the sign of @xmath361 remains arbitrary , since it can be changed by performing a weyl gauge transformation . this is related to the fact that one can choose equivalently @xmath368 or @xmath369 to be massless at @xmath129 in the configuration of figure 1 . the convention we will choose hereafter is the following : @xmath368 will be chosen to be massless at @xmath129 when @xmath370 and @xmath369 when @xmath371 ( we will only consider real values of @xmath11 ) . these conventions will allow us to use the gauge invariance in a convenient way . finally , we have @xmath372 where @xmath373 is the sign of @xmath11 . this formula implies in particular that the dyons found in section 3 by studying the dual rg flow ( [ rgdual ] ) all have @xmath374 as expected . it is not disturbing that @xmath17 may undergo a discontinuity from @xmath375 to @xmath376 . this discontinuity simply reflects the ambiguity associated with cp invariance when @xmath46 or @xmath377 are real . to be concrete , we will hereafter consider a particular configuration , obtained from the original configuration depicted in figure 1 by performing the transformation @xmath378 ( see section 4 ) , and which is described in figure 7 . this amounts to changing @xmath86 into @xmath379 and @xmath45 into @xmath380 . [ figure7 ] = 14.7 cm this choice is completely arbitrary : _ all our reasonings will be sl@xmath15 invariant and thus completely general . _ however , we think that it is more eloquent to deal with states having magnetic charges three , four or seven than simply one , for in this latter case unicity of the states is trivial from the semiclassical point of view . we already know that the @xmath1 multiplet corresponding to a state @xmath381 becoming massles is unique since the singularities are produced by only one hypermultiplet . thus , what we need to show is that , for instance , the only allowed values of @xmath51 for a state @xmath382 should be @xmath383 ( charge of the hypermultiplet becoming massless ) together with @xmath384 and @xmath385 ( this would correspond to the case where the physical @xmath17 charge of @xmath386 would be @xmath72 ) or with @xmath387 and @xmath353 ( when @xmath388 ) . suppose that a state @xmath389 exists in the @xmath1 theory . when a real mass @xmath11 is turned on , @xmath389 must produce a singularity at any point on the coulomb branch where it is massless , that is where @xmath390 which in the original variables reads @xmath391 to study this equation , at least in the large @xmath392 limit , one can use the asymptotics ( [ limite ] ) for @xmath45 at large @xmath393 . this suggest that ( [ uni1 ] ) might have solutions only when @xmath394 when @xmath370 and only when @xmath395 when @xmath371 at points @xmath396 more rigorously , using ( [ cptrans ] ) and the monodromy @xmath316 , one can show that @xmath397 for @xmath398 . moreover , one can easily show using the explicit formulas of @xmath45 and its derivative @xmath399 given in appendix a that , in the case @xmath370 , @xmath45 is a monotonically increasing function for @xmath400 , with @xmath401 ( this latter value is reached at @xmath402 ) , and @xmath403 when @xmath404 . strictly similar results are true when @xmath371 , which proves that the simple analysis using the asymptotics is valid except concerning @xmath405 , as one may have expected . as the @xmath406 exist for any @xmath370 when @xmath407 and for any @xmath371 when @xmath408 , a singularity must be associated in the massive theories to any state @xmath389 existing in the massless one , provided @xmath407 ( if @xmath370 ) or @xmath408 ( if @xmath371 ) . from this we deduce that when @xmath370 , which corresponds to the case depicted in figure 7 , the state becoming massless at @xmath161 , having @xmath409 , must have a physical @xmath17 charge @xmath72 ( it is thus associated with states having @xmath405 and @xmath410 when @xmath12 ) , and that no other @xmath389 state having @xmath411 can exist when @xmath12 . when @xmath371 , the state becoming massless at @xmath161 will have @xmath410 and @xmath388 multiplet ( see table 1 in section 2 ) which becomes massless is changed when @xmath11 changes sign with our conventions . ] and we see that no other @xmath389 state having @xmath412 can exist when @xmath12 . this proves the unicity . exactly the same reasoning can be applied to the states @xmath413 . the equation to study is in this case @xmath414 by using again ( [ limite ] ) , we can convince ourselves , and then show rigorously , that when @xmath142 ( [ uni2 ] ) will have solutions when @xmath415 ( for @xmath370 ) or when @xmath411 ( for @xmath371 ) . this is again perfectly consistent with our previous analysis of the @xmath51 and @xmath17 charges . when @xmath370 , it is the state having @xmath388 which is massless , and we showed that no state @xmath416 with @xmath417 can exist ; when @xmath371 it is the state having @xmath418 which is massless , and we showed that no state with @xmath419 can exist . thus the @xmath1 multiplet @xmath416 is indeed unique and realizes the values @xmath388 , @xmath354 and @xmath420 as it should . finally , note that the case of the states @xmath421 and @xmath422 can be easily handled for instance by using the previous results and the @xmath176 type duality transformation ( [ periodstrans ] ) , or by repeating the same analysis as above . of course , the reasoning will not change whatever configuration of the theory you may choose , and thus we have proven the unicity of the states @xmath0 for any @xmath3 and @xmath4 relatively prime . in this theory , the solution is given by @xmath423 and @xmath424 , and the bare masses of the first and second flavours are equal to @xmath425 . the uniqueness of the @xmath0 states can then be proven exactly as for the @xmath1 theory . in particular , a @xmath426 hypermultiplet becoming massless when @xmath364 will again be associated to two other sets of states having physical charges @xmath427 and @xmath428 , or @xmath429 and @xmath430 when @xmath12 , the spin(8 ) flavour symmetry playing the same rle in this theory as @xmath1 supersymmetry in the previous case . the main novelty is that one also has to study the uniqueness of the vector @xmath6 states . again one can repeat the arguments of the preceding subsection . equation ( [ uni1 ] ) will be replaced by @xmath431 from the analysis of the preceding subsection it follows that this equation has solutions for any @xmath51 such that @xmath432 . moreover , by using the formula for @xmath45 given in appendix a , one can show that @xmath433 when @xmath434 and thus the equation ( [ uni3 ] ) will also have solutions for @xmath435 at least for some values of @xmath50 . thus , the only @xmath6 states , @xmath436 mod 2 , that might exist in the @xmath12 theory must have @xmath354 , as expected . the same reasoning works for the state @xmath6 such that @xmath437 mod 2 . however , when @xmath438 mod 2 , one can not exclude directly the states having a physical @xmath17 charge @xmath439 . this seems to be a peculiarity of the vector particles : at fixed @xmath50 , these vector particles , if they existed , would not necessarily lead to an unphysical singularity , and hence their existence can not be ruled out by this argument alone . of course , this is not really an obstacle for us . by changing @xmath50 to @xmath440 in the original configuration , we exchange @xmath185 and @xmath159 and thus states @xmath437 mod 2 with states @xmath438 mod 2 , and we can then deduce that necessarily @xmath354 for the latter states . to prove the unicity , we still have to check whether only one state exists at given @xmath3 , @xmath4 and @xmath51 . the argument which worked for the @xmath0 states relied on the fact that any of these states can be related to a singularity appearing on the coulomb branch for some values of the coupling , and that we know the multiplicity of the states producing singularities . this clearly does not work for the @xmath6 states , which are never massless ( when @xmath364 ) and thus never produce a singularity . as an alternative , we will rely on a semiclassical analysis . in @xcite was shown that the states @xmath6 indeed exist and are unique _ for @xmath441 . _ the uniqueness of the @xmath442 states for _ any _ @xmath3 and @xmath4 relatively prime will then follow from the @xmath5 invariance of the theory , see section 8 . due to the bps mass formula ( [ bpsmass ] ) , a bps state is generically stable . central charge as well as mass conservation indeed impose tights constraints . to see this , suppose that a bps state of central charge @xmath42 decay into @xmath3 states of central charges @xmath443 . this is possible only if @xmath444 and @xmath445 of the @xmath3 relations ( [ cms1 ] ) , only @xmath446 are independent once one takes into account ( [ firstc ] ) , and thus they define a hypersurface of real codimension @xmath446 included in the coulomb branch . in our case where the gauge group is of rank one and thus the coulomb branch of real dimension two , this means that decays into two distinct particles can only occur along real curves , decays into three distinct particles at special points , and decays into four or more distinct particles are likely to be impossible . suppose we are studying the decay of a state @xmath447 into two states , one of them being @xmath448 . instead of using the original variables @xmath449 , it is convenient to use shifted variables @xmath450 such that @xmath183 . the equation for the curve of marginal stability can then be cast in the following sl@xmath15 invariant form : @xmath451 the left hand side of ( [ cmseq ] ) does not depend on the final state @xmath448 , and the right hand side is an a priori arbitrary rational number which we will denote by @xmath452 . we thus see that to a given particle @xmath447 is associated a whole family of curves of marginal stability , @xmath453 , indexed by a rational number @xmath452 . it is useful at this stage to summarize what we need to show in order to complete our goal , which is to demonstrate that the spectrum of bps states is in perfect agreement with @xmath17 duality both in the @xmath1 and massless @xmath102 theory , and also to prove the existence of the required states in the massive cases studied in section 3 . + we need to show that the states @xmath0 , where @xmath3 and @xmath4 are relatively prime integers , which we have generated by analytic continuation in the _ massive _ theories ( section 4 ) , still exist in the _ massless _ one . this will be established in section 7 using the curves of marginal stability . + we need to show that the states @xmath6 exist in the @xmath102 theory , since they represent the duals of the w bosons in this case . we did not give any argument in favour of the existence of such states up to now , since they are not associated with any singularity on the coulomb branch . however , one can establish their existence by using the general argument presented in section 8 , _ and _ by relying on the non trivial semiclassical results obtained in @xcite for states of magnetic charge 2 . + finally , we need to show that the dual of the w bosons ( and of the elementary quarks in the @xmath102 theory ) does exist in the dual theory studied in section 3.5 . there it was only shown that the existence of these states does not lead to any inconsistency . actually , these dual states exist in the massless theories , and in appendix b we will explain why they must still exist when the bare mass is increased and we follow the rg flow ( [ rgdual ] ) toward the dual pure gauge ( or massless @xmath119 ) theory . let us end this section by mentioning the formula @xmath454 which corresponds to a cp transformation . note that though this is _ not _ a duality transformation , it is still perfectly compatible with the bps mass formula ( [ bpsmass ] ) . this relation is useful to study the symmetries and some particular points of the curves of marginal stability . in this section , @xmath3 and @xmath4 are two relatively prime integers . in section 4 , strong evidence in favour of the existence of the states @xmath0 in the massless theories was given : depending on the value of the coupling @xmath50 , when the bare mass is turned on , the singularities on the coulomb branch may be due to any of these states . however , strictly speaking , the existence of the states was proven only when @xmath364 and @xmath455 . when @xmath12 and the singularities merge at @xmath456 , all the @xmath0 states become massless and thus degenerate in mass . though a brutal discontinuity in the spectrum between @xmath364 and @xmath12 seems very unlikely on physical grounds , we would like to argue that the @xmath0 states not only exist at the singularities they are responsible for , but also at other points in the @xmath46 plane . one way of doing this is to consider a configuration where the @xmath0 state becomes massless , and to study the family of curves of marginal stability associated with the decay of this state . if @xmath436 mod 2 , we have a family @xmath457 defined by ( cf ( [ cmseq ] ) ) @xmath458 and if @xmath438 mod 2 we have a family @xmath459 , @xmath460 the case of @xmath437 mod 2 is completely similar to the case @xmath436 mod 2 due to ( [ periodstrans ] ) . we have computed the curves @xmath461 numerically , using the analytic formulas for the periods presented in appendix a , and the result is depicted in figure 8 . [ figure8 ] = 14.7 cm it is easy to established , particularly in using ( [ cptrans ] ) , that the curves @xmath459 all intersect at @xmath127 and are symmetric with respect to the real @xmath46 axis , while the curves @xmath457 all intersect at @xmath129 and @xmath127 and @xmath462 . more important , these curves do not intersect at any other points than the particular ones mentioned above and form a dense set in the whole @xmath46 plane . this is not prohibitive : the complementary set , generated by the curves @xmath463 for _ irrational _ values of @xmath464 , is also a dense subset of the coulomb branch , where the corresponding state is stable . this shows that for any positive real number @xmath465 , there is always a point @xmath46 such that @xmath466 and which can be joined to the singularity @xmath131 by a curve @xmath463 where @xmath464 is taken to be irrational . as the state becoming massless at @xmath131 can not decay along such a curve , it must exist at @xmath46 . as @xmath465 can be chosen independently of @xmath11 , @xmath0 must exist in the @xmath12 theory for some @xmath466 and thus for any @xmath46 . to end this section , i wish to point out that the preceding analysis suggests that states which can not exist in the @xmath12 theory can not exist anymore in the @xmath364 theory . to understand this , assume that to the unwanted state @xmath467 is associated a point @xmath93 on the coulomb branch such that @xmath468 for any given @xmath50 and @xmath11 . , though it seems to be true numerically . ] . the family of curves of marginal stability @xmath469 associated with @xmath467 looks like the families depicted in figure 8 , except that all the curves will intersect at @xmath93 . now suppose that @xmath467 exist at a point on a curve @xmath470 with @xmath464 irrational ( such curves cover a dense subset @xmath471 of the coulomb branch ) . then move @xmath467 along this curve , reach @xmath93 where it is massless and find an inconsistency since @xmath93 is not a singular point . thus @xmath467 can not exist in the dense set @xmath471 : this means that it will never be a trully stable state in the theory . thus , if our hypothesis ( [ unsing ] ) is correct , the number of curves of marginal stability one must consider to study eventual decays in the massive theories is considerably decreased . we will use this fact in appendix b , see also [ 32 ] . in this section , we wish to address a very general problem : when can we expect the monodromy group of a given theory to correspond to an exact duality symmetry ? of course , this a priori naive statement is not always true , as seiberg and witten already pointed out in their original paper @xcite on the pure gauge theory . however , we would like it to be correct in the particular cases of the finite theories studied in this paper . we will give in the following some general physical arguments showing that this property can be understood simply by looking at some very general features of the structure of the low energy effective action . though we will restrict ourselves to very few examples , in the framework of su(2 ) gauge theories , and chosen in order to illuminate the cases of su(2 ) @xmath1 and @xmath2 , @xmath102 , the line of reasoning could a priori be applied to any @xmath2 theory . first , let us consider a theory with only one singularity on the coulomb branch . it could be the massless theories studied above . let us see why in this simple case , the monodromy @xmath166 associated with this singularity must correspond to an exact duality symmetry of the theory . to the singularity is associated a branch cut which extends to infinity and whose orientation clearly is arbitrary . by sweeping up the @xmath46 plane with such a branch cut , _ while keeping all the parameters fixed , _ we show that necessarily two theories connected by the transformation @xmath166 must be physically equivalent . this kind of argument was already presented in subsection 4.3 . in the case of @xmath1 or @xmath2 with four massless flavours , @xmath472 and the `` duality '' transformation is nothing but a gauge transformation . let us now go to a more interesting situation , suited for instance for the pure gauge theory , where two singularities are present at @xmath129 and @xmath126 with monodromies @xmath334 and @xmath473 . the theory now has an intrinsic scale @xmath87 such that @xmath474 . two scenarios can be imagined in this case . in the first one , for some reasons the naive arguments are true , and thus the monodromy group generated by @xmath334 and @xmath473 corresponds to an exact duality symmetry . in the second scenario , the two singularities combine : we choose the branch cut originating at @xmath161 to go through @xmath185 and then to coincide with the branch cut originating at @xmath185 . far away from the singularities , we thus only see a branch cut associated with a monodromy matrix @xmath475 which is the product of the two matrices @xmath316 and @xmath476 . of course the position of this branch cut extending to infinity can be chosen arbitrarily , and thus an exact duality symmetry must be associated with @xmath477 , at least when @xmath478 . in the case of the pure gauge theory , @xmath475 expresses the asymptotic freedom and is associated with the invariance of the theory under @xmath16 angle shifts . finally , let us study the case where three singularities @xmath137 are present on the coulomb branch , with monodromy matrices @xmath291 . the theory may now have one , two , or three intrinsic scales , corresponding to the distances @xmath479 between the singularities . in the @xmath2 theory with one flavour of quark , we have three singularities at equal distances from each other because of a @xmath480 symmetry acting on the coulomb branch @xcite . thus we have only one scale @xmath87 at our disposal . the same scenario as the one which prevailed for the pure gauge theory occurs , and we do not have any particular duality symmetry in this theory . now , the finite theories we are studying in this paper also have three singularities in the coulomb branch , but with the most important property that two independent scales can be introduced and _ any one of the three singularities can be chosen to be arbitrarily far from the other two . _ typically , we are studying a dual rg flow as in ( [ rgdual ] ) , and two of the three singularities ( say @xmath161 and @xmath185 ) are separated by an almost fixed distance @xmath481 while the third singularity ( @xmath159 ) runs to infinity with @xmath11 at a distance @xmath482 , where @xmath11 is the bare mass in the theory . in such a configuration , as in the case where only two singularities are present , we can not know a priori whether the monodromies @xmath316 and @xmath476 associated with the singularities at scale @xmath87 will correspond to an exact duality symmetry . however , the situation is different for the isolated singularity at scale @xmath11 . the latter can only be `` screened '' by the singularities at scale @xmath87 . we think that this is impossible , because it seems very unlikely that the low energy physics at scale @xmath87 could be influenced by a singularity at scale @xmath11 , produced by a particle which is ultra massive at scale @xmath87 ( if it exists ) , and vice - versa . one might argue that the branch cuts of the three singularities could merge at a new intrinsic scale @xmath483 . however , the physical origin of this new scale would be very unclear and we will rule out this possibility . we thus see that the only natural way to interpret the singularity structure is to admit that @xmath311 corresponds to an exact duality symmetry of the theory at least in the configuration where @xmath484 and @xmath485 with @xmath486 . from the point of view of the massless theory , which corresponds to @xmath487 , we have thus an invariance with respect to @xmath311 . the rle of the three singularities can of course be exchanged due to the sl@xmath15 invariance of the seiberg - witten curve ( [ curve ] ) , and we deduce that the @xmath12 theories have an exact duality group which contains @xmath5 . in particular , the existence of all the @xmath6 states in the @xmath7 theory will then follow from the existence of the three states @xmath111 ( the w ) , @xmath488 and @xmath489 , which were shown to exist in @xcite using the semiclassical method . in this paper , we proved rigorously that all the @xmath0 dyon states , where @xmath3 and @xmath4 are relatively prime integers , exist in the @xmath1 as well as in the @xmath2 , @xmath7 gauge theories with gauge group su(2 ) . these states were obtained by analytic continuation of the @xmath490 states which become massless at some points of the coulomb branch when the bare @xmath16 angle is zero . in the process of analytic continuation , we saw explicitly that the quantum numbers at the singularities can change according to a monodromy transformation . we also proved the unicity of the dyon states , in perfect agreement with an exact sl@xmath15 duality symmetry . we also provided a very general physical argument showing that @xmath1 and @xmath2 with four massless flavours must have an exact duality symmetry group which contains the monodromy group of the corresponding massive theories . this fact together with already known semiclassical results implies the existence of all the @xmath6 vector states in the @xmath7 theory . the line of reasoning used in this work , particularly the arguments of subsections 4.1 and 8 , is likely to be generalizable to theories having any gauge group , for instance to the case of su(n ) with @xmath491 . this may answer the open questions in these cases @xcite . from a very general point of view , our method may be seen as the physical counterpart of the string theory results based on the heterotic - type ii duality conjecture @xcite . it would be very interesting to prove the equivalence of the two approaches . in our approach , one is seeking for non singular points @xmath46 on the coulomb branch and for cycles @xmath492 on the seiberg - witten curve @xmath493 such that @xmath494 where @xmath495 is the seiberg - witten differential form , see appendix a. states @xmath0 corresponding to such a cycle @xmath496 can not exist , and any other ( hypermultiplet ) states are very likely to exist . in the string approach , stable states are associated with geodesics on @xmath497 endowed with the metric @xmath498 where @xmath73 is a particular element in the cohomology class of @xmath495 . for states lying in a @xmath2 hypermultiplet , these geodesics must begin and end at the branching points @xmath499 ( see ( [ roots ] ) in appendix a ) . a link between the two points of view is the following : it is clearly impossible to have a non - null geodesic with the required properties for a state @xmath496 satisfying ( [ conc1 ] ) , since the geodesic would have zero length and this requires its endpoints to coincide , that is to have a singularity . more important , the existence of a particular point @xmath46 such that ( [ conc1 ] ) is true is likely to prevent the existence of the corresponding geodesic for `` any '' @xmath46 , by continuity . of course subtleties are associated with the curves of marginal stability , and the vector particles may have a particular status . in any case , we believe that a complete understanding of the relationship between the two approaches is likely to provide a deeper understanding of @xmath2 theories . it is a pleasure to thank adel bilal for his permanent encouragements and many interesting discussions . his critical reading of the manuscript was also at the origin of many improvements . i also thank adam schwimmer for a useful discussion , and aurlia who kindly processed the figures . i present below some general formulas from the theory of elliptic integrals ( see , e.g. , @xcite ) in a form well suited for our purposes . by well suited , i mean first that they can be used straighforwardly to obtain the analytic solution for @xmath86 and @xmath45 actually , the same general formulas can be used to evaluate the periods of the asymptotically free theories having @xmath500 as well @xcite . second , i mean that they can easily be implemented on a computer system , and thus can be used to determine numerically the curves of marginal stability . let us consider a torus whose equation is @xmath501 in our particular case , it will be convenient to cast ( [ curve ] ) into this form by performing the change of variable @xmath502 we will then have @xmath503 the fundamental cycles on the torus ( [ curvebis ] ) are chosen as indicated in figure a.1 . [ figure9 ] = 14.7 cm to this choice for the homology basis , we associate the usual variables @xmath212 and @xmath504 : @xmath505 the fundamental periods are then given by @xmath506 where @xmath507 is the standard complete elliptic integral of the first kind . we will also need a formula involving elliptic integrals of the third kind . maybe the most convenient form for this formula is the following : @xmath508 where @xmath496 is any complex number which is different from the roots @xmath499 and @xmath509 is an integer . the @xmath510 are given by @xmath511 in terms of the complete elliptic integrals of the first and second kinds @xmath507 and @xmath512 . , @xmath513 . ] @xmath514 and @xmath515 are straightforward generalizations ( note that @xmath516 here ) of the weirstrass @xmath514 and @xmath515 functions , and may be expressed in terms of the standard jacobian and @xmath16 elliptic functions @xcite as @xmath517 finally , @xmath518 is such that @xmath519 , that is @xmath520.\ ] ] the indeterminate constant @xmath509 comes from the fact that the differential form @xmath521 has poles with residues @xmath522 . if , under a continuous deformation , the integration contour crosses such a pole , @xmath509 will be shifted by one unit . note also that the ambiguity that exists in the definition of @xmath518 , which stems from the fact that @xmath523 is a doubly periodic , even function of @xmath524 , can be absorbed in a redefinition of @xmath509 by using legendre s relation @xmath525 and the fact that @xmath526 for any integers @xmath527 and @xmath528 . it should be clear that such integrals can only appear in theories with non zero bare masses , since it is only here that jumps in @xmath86 and @xmath45 are allowed @xcite . changing @xmath509 will then simply amount to shifting the @xmath51 charge . we are now equipped to compute the seiberg - witten periods @xmath86 and @xmath45 rather straighforwardly . however , two points remain to be discussed . first , when applying the above formulas , we are not certain to obtain a solution with the correct analytic structure in the @xmath46 plane . this is explained in the next subsection where the effective coupling @xmath529 is computed . second , we must find the seiberg - witten differential , if possible in a form where the canonical decomposition in terms of the building blocks treated above is simple . this is done in subsection 5.3 . the solution is then given in 5.4 . the effective coupling @xmath530 is nothing but the modular parameter of the torus ( [ curve ] ) , and would thus naively be given by @xmath531 . however , this formula is a priori valid only modulo sl@xmath15 transformations , and some additional constraints must be found from physics . the first constraint comes from the evaluation of @xmath530 in the limit @xmath404 where it should coincide with @xmath50 , modulo @xmath16 angle redefinitions which correspond to trivial shifts of the electric charge . this constraint is indeed satisfied with our choices for the homology cycles @xmath532 and @xmath533 . the second constraint comes from the analytic structure of @xmath530 as a function of @xmath46 , which was described in section 3 in the case of a trivial bare @xmath16 angle . this analytic structure is most easily studied when one uses the explicit formulas @xmath534 and @xmath535 where the @xmath137 correspond to the positions of the singularities ( [ singpos ] ) . we then obtain a unique solution , given when @xmath273 by @xmath536 @xmath537 and @xmath399 are also very directly related to the fundamental periods @xmath538 and @xmath539 . before writing explicit formulas , one must note that though having the same monodromies , the solution for the so(3 ) and @xmath7 theories will differ by a global factor of @xmath540 which directly comes from the analysis of the massless theories . in the following , we will reserve the notation @xmath112 for the solution of the so(3 ) theory , and for @xmath7 we will have @xmath541 . with these conventions , @xmath542 and @xmath543 we now wish to integrate the relations ( [ solda],[soldad ] ) in order to obtain @xmath44 and @xmath45 . to do this , the standard method is to find a one form @xmath73 ( the so - called seiberg - witten differential ) so that @xmath544 up to an exact form . we will then compute @xmath44 and @xmath45 by mean of the formulas @xmath545 the simplest way to find @xmath73 certainly is to use , instead of the variables @xmath546 , the rescaled variables @xmath547 ) . we perform it in order to simplify the expressions . on the contrary , the use of @xmath548 instead of @xmath549 amounts to adding a non trivial exact differential to @xmath73 , which simplifies its form nicely . ] in terms of which the curve takes the form @xmath550 with @xmath551 by multiple integration by part we then obtain @xmath552 a very convenient and simple form for @xmath73 will then be @xmath553 we end our computation by writing @xmath73 in the form @xmath554 by noting that @xmath555 we can easily compute @xmath556 with @xmath557 and then apply ( [ fundperiods ] , [ i3 ] ) to find the desired solution : @xmath558,\ ] ] and @xmath559 \quad { \mathrm for}\enspace \im z\geq 0.\\ { \displaystyle\sqrt{2}\over\displaystyle\pi } \bigl [ \bigl ( z - z_{1 } \bigr)\ , ( \omega _ { 2}+2\omega _ { 1 } ) + { i\over 4}m\ , \bigl ( 2(\omega _ { 2}+2\omega _ { 1})\zeta ( u_{c } ) -4(\eta _ { 2}+2\eta _ { 1})u_{c } \bigr ) \bigr]\\ \end{array}\right.\nonumber\\ & & \hskip 8 cm { \mathrm for}\enspace \im z < 0.\\ \nonumber\end{aligned}\ ] ] the different prescriptions for @xmath560 or for @xmath561 correspond of course to a zero @xmath16 angle . the generalization to any @xmath16 angle should be clear in view of the discussion in section 4 . [ figure10 ] = 14.7 cm in this appendix , we complete the proof of section 3 and show that the spectrum of the dual pure gauge theory ( or dual @xmath125 massless theory ) is in perfect agreement with duality . in section 3 , we already showed that the duals of the @xmath275 monopoles indeed exist outside the curve of marginal stability @xmath562 represented in figure 4 in the theory @xmath177 . in addition to these states , duality predicts the existence of the duals of the w bosons , and also of the elementary quarks in the case of the @xmath7 theory . more precisely , we need to prove that in the so(3 ) theory the state @xmath563 indeed exists for @xmath141 outside the curve @xmath562 or equivalently that @xmath564 exists for @xmath142 . in the @xmath7 theory , in addition to the preceding state ( now interpreted as being the dual of the quarks of the massless @xmath125 theory ) , we must have the @xmath565 ( or @xmath566 ) state interpreted as being the dual of the w. our strategy will be the following : as we already know that all these states exist when @xmath12 , we will simply show that they can not decay outside the curve @xmath562 when the mass is turned on and we follow the rg flow ( [ rgdual ] ) toward the dual pure gauge ( or @xmath125 ) theory . to do this , remark first that the family of curves of marginal stability associated with @xmath563 coincides with the one for @xmath565 , and that the family for @xmath567 is the symmetric of the family for @xmath563 with respect to the real @xmath46 axis . thus , we need to study only one family of curves , whose defining equations deduced from ( [ cmseq ] ) can be cast in the form @xmath568 in this equation , @xmath569 represents the physical @xmath17 charge a state produced in the decay reaction would have when @xmath12 , and @xmath570 represents the electric charge of the same state . as argued in section 7 , we will consider only @xmath569 such that @xmath571 and thus @xmath572 or @xmath573 where @xmath107 is any integer . the corresponding curves are represented in figure b.1 . as there is no curve outside @xmath562 in the lower half - plane , the state @xmath563 ( or @xmath565 ) can not decay in this region . this completes the proof .
given a curve @xmath4 , we consider the geometric flow @xmath5 where @xmath6 is the curvature and @xmath7 the binormal component of the frenet - serret formulas @xmath8 the flow can be expressed as @xmath9 where @xmath10 is the usual cross - product , @xmath11 is the time , and @xmath12 is the arc - length parameter . it appeared for the first time in 1906 in the ph . d. thesis of da rios @xcite and was rederived in 1965 by arms and hama @xcite as an approximation of the dynamics of a vortex filament under the euler equations . this model is usually known as the vortex filament equation ( vfe ) ; we refer the reader to @xcite and @xcite for an analysis and discussion of its limitations . equations and are also known as the binormal equation and the localized induction approximation ( lia ) , respectively . some of their explicit solutions are the line , the circle , and the helix . since the tangent vector @xmath13 remains with constant length , we can assume that @xmath14 , for all @xmath11 . differentiating , we get the so - called schrdinger map equation onto the sphere : @xmath15 which is a particular case of the landau - lifshitz equation for ferromagnetism @xcite . equation can be rewritten in a more geometric way as @xmath16 where @xmath17 is the covariant derivative , and @xmath18 is the complex structure of the sphere . written in this way , can be generalized to more general definition domains and images , as in @xcite , where the hyperbolic plane @xmath19 was chosen as the target space . a relevant step forward in the understanding of was given by hasimoto in @xcite , where a transformation ( see section [ s : hasimoto ] ) that relates and with the nonlinear schrdinger ( nls ) equation was introduced : @xmath20 equation is time reversible , i.e. , if @xmath2 is a solution , so is @xmath21 . bearing in mind this fact , an important property of ( and , hence , of ) is that it has a one - parameter family of regular self - similar solutions that develop a corner - shaped singularity at finite time . this was proved in @xcite for the euclidean case , and in @xcite for the hyperbolic case . the self - similar solutions are written in the form @xmath22 as was proved by buttke in @xcite , @xmath23 is a smooth curve characterized by the geometric properties @xmath24 with @xmath25 being a constant . therefore , the shape of @xmath23 is the one of a hairpin : it looks like a segment of a circle of radius @xmath26 close to the origin , that becomes two helicoidal curves with a bigger pitch for bigger @xmath27 , and that winds around two different lines @xcite . buttke also studied numerically these self - similar solutions . later on , in @xcite , a careful numerical study of those solutions for both the euclidean and the hyperbolic cases was done . on the one hand , the authors reproduced numerically in @xcite the formation of the corner - shaped singularity , and , on the other hand , they started with a corner - shaped initial datum , recovering numerically the self - similar solutions ; in all cases , the correct choice of boundary conditions was shown to be vital . furthermore , they gave numerical evidence that fractality phenomena appeared in ( see figure 2.6 in p. 1059 of @xcite ) , if , for instance , fixed boundary conditions were imposed on the tangent vector @xmath28 . remark that the relationship between fractals and the schrdinger map equation is not new ; indeed , in @xcite , an aortic valve model was proposed , in order to study the apparent fractal character of the valve s fiber architecture . instead of @xmath2 , the authors wrote @xmath29 , where @xmath30 , which corresponds to our time , is such that the curves @xmath31 are the fibers . after imposing fixed boundary conditions at @xmath32 , they showed that the curves @xmath33 have a fractal character . besides , the fractal dimension of those curves was calculated numerically in @xcite . it is well known that this fractal behavior already occurs at the linear level , as can be easily seen when solving numerically the free schrdinger equation @xmath34 with an initial datum that is piecewise constant as the sign function in , say , @xmath35 $ ] , and then extended by periodicity . the trigonometric series associated to these problems have been extensively studied and , as we will see later , the behavior at rational times modulo @xmath36 is completely different to that at irrational ones . in @xcite , berry and goldberg made a far - reaching connection of this behavior with the so called talbot effect in optics . in @xcite and in @xcite , the authors used to model the talbot effect and showed that at rational times the solution can be obtained as a finite overlapping of translates of the initial datum , while at irrational times the images have a fractal profile . also in @xcite , a conjecture was made about the dimensions of those fractals . as far as we know , the only available results on the nonlinear setting are the recently ones obtained by erdoan and tzirakis @xcite ( see also @xcite ) in the cubic nonlinear schrdinger equation ( i.e. equation , with @xmath37 ) , for the initial data given by piecewise continuous functions . this means that the data are `` almost '' in the sobolev class @xmath38 , that is subcritical with respect to the one determined by the scaling of the equation ( i.e. @xmath39 ) . this typically implies that the nonlinear potential can be seen as an external force . then , its contribution is computed using duhamel s integral , which has a regularizing effect . as a consequence , the fractal behavior is due to the linear part of the solution . at this point , the available results for the linear theory are used ( see @xcite and @xcite ) . more recently , erdoan and tzirakis , in collaboration with chousionis , have also studied in @xcite the schrdinger map equation , but still with subcritical initial conditions . in this paper , we are interested in filaments with corners . this means to consider delta functions as the initial data for or piecewise continuous functions as the initial data for . therefore , we work with data in critical spaces and the problem turns out to be much more involved . the case of an infinite curve with just one corner and that is otherwise smooth has been studied in the sequence of papers @xcite by banica and vega . one of the conclusions is that is in fact ill - posed for this type of initial conditions , so that is also ill - posed if it is understood in a classical sense . however , after introducing appropriate function spaces and using the appropriate definition of solution , it is proved that and are well - posed for this type of initial condition . more concretely , the velocity at the point where the corner is located is determined by the self - similar solution that at time zero has a corner with the same angle . in fact , in @xcite it was proved that there exists just one self - similar solution with this property . even if the solutions of for an initial datum with a corner are well understood , nothing had been done for more general initial data with several corners , in particular , polygons . nevertheless , in two recent papers @xcite by jerrard and smets , they prove a global existence theorem that allows to consider such types of initial data . moreover , they simulate numerically the evolution of the unit square at different times , suggesting that the solution could become again polygonal ; indeed , at a certain time , the square seems to reappear , but with the axes turned by a @xmath40-angle with respect to the initial ones . in the following pages , we will show not only that these observations are correct , but that , given a regular planar polygon with @xmath41 sides as the initial datum , another polygon is obtained at times of the form @xmath42 , @xmath43 . in fact , we will give a complete description of this polygon ; in particular , and except for @xmath44 , we will see that it is a skew polygon with @xmath45 sides ( if @xmath46 odd ) , or @xmath47 sides ( if @xmath46 even ) . one could wander what , if any , is the connection of these solutions with real fluids . of course , one can not expect anything at the quantitative level , because da rios and arms and hama s approximations are based first in a truncation of the biot - savart integral and then in a renormalization of the time variable . however , using vfe as an approximation of the evolution of real filaments can not be ruled out at the qualitative level , even with data as singular as the ones we consider . in fact , in @xcite , the similitude between the self - similar solutions and a flow traversing a triangular wing is pointed out ( see figure 1.1 in p. 1051 of that paper ) . in the concrete case of the solutions considered in this article , we could ask if the macroscopic effect of the turning axes mentioned above can be seen in the dynamics of real fluids . it appears that this is something quite well documented for noncircular jets . it was first observed when the nozzle has an elliptic shape and , later on , seen for shapes given by a rectangle , a square , and an equilateral triangle among others . the interest of considering nozzles with corners started in the mid 80 s and has received considerable attention since then . we refer the reader to the survey @xcite , where the papers related to the above observations can be found , together with a detailed analysis on how sensitive the axes - turning phenomenon is to the initial geometric conditions ; and several numerical experiments are also reviewed . it has to be understood that , at the numerical level , the corners are always effectively rounded and , in real experiments , the corners are immediately smoothed out once the jet leaves the nozzle , due to viscosity effects . this is consistent with the behavior of the self - similar solutions mentioned above , because the initial corner evolves into a smooth curve with the shape of a hairpin . furthermore , the appearance of smooth polygonal structures with more angles has been appreciated for nozzles with the shape of an equilateral triangle ( see ( * ? ? ? * figure 6 ) ) . also in ( * ? ? ? * figure 10 ) , while in @xcite it is said `` ... a consistent eightfold distribution pattern is also suggested ... '' for square nozzles . similar results are obtained in the numerical simulations of @xcite ( see figure 8 in p. 9 , and p. 11 , where the @xmath48-degree `` flip - flop '' in the case of an equilateral triangle is emphasized ) . we think it would be worth studying in more detail whether or not these more complicated polygonal structures appear in noncircular jet flows . the structure of this paper is as follows . in section [ s : hasimoto ] , we review the hasimoto transformation for . in section [ s : xt ] , we gather the main theoretical arguments that support our numerical experiments in the form of a theorem . we start recalling in section [ s : formulation ] some elementary geometric symmetries that regular polygons have and that are preserved by vfe . on the one hand , those symmetries greatly simplify the numerical implementation and , on the other hand , they play a fundamental role in computing some important quantities that govern the dynamics of regular polygons . then , in section [ s : galilean ] , we pay especial attention to a group of symmetries that , as far as we know , can only be visualized through the use of the hasimoto transformation and the nls equation . they are the so - called galilean transformations , and leave invariant the set of solutions of vfe . we proceed as in @xcite , where the solution of the nls equation that has the delta function as an initial condition is characterized by the fact that it leaves invariant all the galilean transformations ; this property determines the solution , except for one function of time that is easily computed by solving a simple ode and choosing @xmath49 in appropriately . in our case , we take a planar regular polygon with @xmath41 sides as the initial datum . then , as in @xcite , the solution of is determined except , again , by a function that depends on time ; but , in this case , it is very delicate how to specify this function and does not seem to be easy . our approach in this paper is to integrate the frenet - serret frame with @xmath50 where @xmath51 , which is in this case a distribution , is obtained from the hasimoto transformation . in section [ s : rational ] , we evaluate @xmath52 at times @xmath53 that are rational multiples of @xmath54 , i.e. , @xmath42 , @xmath55 , because @xmath56 takes then the form , so the corresponding curve is a skew polygon with @xmath45 sides , for @xmath46 odd , and with @xmath47 sides , for @xmath46 even . remark that the structure of the polygon is determined by the generalized quadratic gau sums @xmath57 appearing in ( see [ s : gausssums ] ) ; in particular , when @xmath46 is even , half of the sums are zero , which ultimately explains why the polygon has @xmath47 sides in that case , instead of @xmath45 . the value @xmath58 is obtained in section [ s : recovering ] , by imposing the condition that the polygon has to be closed . as a consequence , the tangent vector @xmath28 is determined , except for a rotation that can depend on time . this rotation is computed in section [ s : rigid ] by using the extra symmetries that are available for regular polygons and that we have already mentioned . the next step is to integrate @xmath28 , in order to obtain the corresponding skew polygon @xmath59 at rational times . using again the symmetries , we are able to calculate all the necessary quantities , except for some vertical translation of @xmath59 that can be easily computed numerically ( although we do not see a simple way of doing it by using just theoretical arguments ) . in section [ s : numerical ] , we propose a numerical method for and . in section [ s : symmetries ] , we explain carefully how to take advantage of the symmetries of @xmath59 and @xmath28 for the types of solutions considered , which leads to a dramatically improvement in the computational cost , as explained in section [ s : cost ] . in section [ s : experiments ] , we perform numerical experiments , simulating the evolution of and for different numbers of initial sides @xmath41 . concerning the vertical translation of @xmath59 , we give strong evidence that the center of mass of @xmath59 moves upward with a constant speed @xmath60 that depends on @xmath41 . we compare the numerical results with the algebraic values from section [ s : xt ] , obtaining a complete agreement between both ( totally different ) approaches . in section [ s : riemann ] , we study the evolution of @xmath2 , for @xmath61 . at this point , it is important to recall the invariances of the so - called jacobi theta function , closely related to : @xmath62 this function is precisely the solution of the free schrdinger equation with a one - periodic delta as the initial condition . therefore , the galilean transformations still hold in this case ; moreover , they also imply the extra symmetry @xmath63 which , together with the periodicity property @xmath64 , generates the so - called unimodular subgroup . in fact , in @xcite , all these symmetries are used to prove the existence of a continuum range of exponents for the hlder regularity of @xmath65,\ ] ] whose real part is precisely riemann s non - differentiable function . more concretely , in @xcite , it was proved that riemann s non - differentiable function is a multifractal ( see also @xcite , where other relevant trigonometric sums are considered ) . in other words , it was proved that the set of times @xmath11 that have the same hlder exponent is a fractal with a dimension depending on the hlder exponent , in such a way that the conjecture stated by frisch and parisi in @xcite is fulfilled ( see also @xcite , for more details at this respect and the connection of this question with fully developed turbulence and intermittency ) . we could ask if something similar to the properties of riemann s non - differentiable function also holds in our case . it is very easy to find the analogous formulas to in out setting . in fact , for a given number of initial sides @xmath41 , if we write @xmath66 , bearing in mind the symmetries of the problem , we conclude that @xmath3 is a planar curve , so we identify the plane where it lives with @xmath67 and define @xmath68 this curve , or , more precisely , @xmath69 , can be seen as a nonlinear version of . is @xmath70 a multifractal ? if the answer is positive , what is its spectrum of singularities in the sense of @xcite ? we deem these two questions rather challenging from both a numerical and an analytical point of view . the results of section [ s : riemann ] strongly suggest that the answer to the first question is positive . in fact , after computing @xmath69 for different @xmath41 , and measuring the difference in the @xmath71-norm between it and an appropriately scaled and rotated version of @xmath72 , the convergence rate is rather strong . we also give some numerical evidence that the hlder exponent of @xmath70 is @xmath73 for rational times @xmath53 ; nevertheless , how the constants depend of the denominator @xmath46 , which is a fundamental ingredient in the arguments in @xcite , is unclear . this question deserves a much more detailed analysis that we plan to make in a forthcoming paper . in section [ s : fractalst ] , we study some interesting fractality phenomena associated to the tangent vector @xmath28 . in section [ s : conclusions ] , we draw the main conclusions , as well as some open questions that we postpone for the future . finally , to conclude this paper , we offer in [ s : gausssums ] a detailed study on the generalized quadratic gau sums . a central point of this paper is the natural connection between - and the nls equation . indeed , applying the hasimoto transformation @xcite to : @xmath74 the function @xmath75 satisfies the nls equation @xmath76 where @xmath49 is a certain real constant that depends on time . nevertheless , for our purposes , it is not convenient to work with the torsion @xmath77 . instead , we consider another version of the frenet - serret trihedron ; it is easy to check that all its possible generalizations have the form @xmath78 for some vectors @xmath79 and @xmath80 that form an orthonormal base with the tangent vector @xmath28 . moreover , we can make , without loss of generality , one of the coefficients @xmath81 , @xmath82 , or @xmath83 equal zero . if we choose @xmath84 , denoting @xmath85 and @xmath86 , we recover the habitual trihedron . on the other hand , in this paper , we choose @xmath87 , in order to avoid working with the torsion . in that case , the hasimoto transformation takes the form @xmath88 where @xmath89 it is quite straightforward to check that this new definition of @xmath75 also satisfies . the key point is to combine the basis vectors @xmath79 and @xmath80 into @xmath90 . since the proof closely matches that in @xcite , we omit here the details . the main idea is to work with , and , at a given @xmath11 , recover the curve @xmath2 and the tangent vector @xmath91 from @xmath52 by integrating , up to a rigid movement that we determine by the symmetries of the problem . observe that , if we define @xmath92 and integrate for @xmath93 , the new solution of is @xmath94 , where @xmath95 , and @xmath96 , i.e. , @xmath28 does not change . the aim of this paper is to understand the evolution of for polygonal initial data . in fact , we limit ourselves to studying the simplest case , i.e. , that of a regular planar polygon with @xmath41 sides . more precisely , we prove the following theorem : [ t : theorem ] assume that there is a unique solution of the initial value problem @xmath0 where @xmath98 is a regular planar polygon with @xmath41 sides . then , at a time @xmath53 that is a rational multiple of @xmath54 , i.e. , @xmath99 , with @xmath100 , @xmath101 , @xmath43 , the solution is a skew polygon with @xmath45 sides ( if @xmath46 odd ) or @xmath47 sides ( if @xmath46 even ) . all the new sides have the same length ; and the angle @xmath102 between two adjacent sides is constant . furthermore , the polygon at a time @xmath53 is the solution of the frenet - serret system @xmath103 where @xmath104 , and @xmath56 is the @xmath105-periodic function defined over the first period @xmath106 as @xmath107 with @xmath108 being a generalized quadratic gau sum ( see [ s : gausssums ] ) . the rest of this section is devoted to the proof of this theorem . moreover , we also conjecture that the angle @xmath102 is given by @xmath109 * remark . * from our point of view , uniqueness is a very challenging problem for this type of equations , because of the critical regularity that we are assuming in the initial data . as we say in the introduction , uniqueness has been established in a series of papers by banica and vega , when the initial datum has just one corner and is otherwise regular . banica and vega prove that uniqueness holds for the binormal equation and for the schrdinger map equation , but the cubic nls equation is ill - posed as an initial value problem with initial datum being the dirac delta function . therefore , the work of erdoan and tzirakis @xcite can not be used in our case , and the only general result for the binormal flow for data as singular as ours is so far the one by jerrard and smets @xcite , where , on the other hand , they just prove existence . since and are invariant with respect to rotations , given a regular planar polygon with @xmath41 sides , we can assume without loss of generality that @xmath1 , and hence @xmath110 , live in the plane @xmath111 . identifying @xmath111 with @xmath67 , and assuming without loss of generality a total length of @xmath36 , @xmath1 is the polygon parameterized by arc - length whose @xmath41 vertices , located at @xmath112 , @xmath113 , are @xmath114 and a point @xmath1 , for @xmath115 , lays in the segment that joints @xmath116 and @xmath117 . observe that it is straightforward to extend @xmath1 periodically to the whole real with period @xmath36 . then , since @xmath118 , the corresponding tangent vector is the periodic function with period @xmath36 such that @xmath119 the curve @xmath1 is continuous , while its tangent vector @xmath110 is only piecewise continuous . in fact , @xmath1 can be regarded as a curve whose @xmath105-periodic curvature is a sum of dirac deltas : @xmath120 and the constant @xmath105 has been chosen in order for the integral of the curvature over an interval of length @xmath36 to be equal to @xmath36 . at this point , it is necessary to understand the role played by symmetries in and for these initial data . both equations are invariant with respect to rotations , i.e , given a rotation matrix @xmath121 , if @xmath122 and @xmath123 are , respectively , solutions of them , so are @xmath124 and @xmath125 ( in this paper , all the vectors are represented in column form ) . therefore , if @xmath126 and @xmath127 , then , if the solution is unique , @xmath128 and @xmath129 , for all @xmath11 . in particular , since @xmath1 and @xmath110 , as defined in and , are invariant with respect to rotations of angle @xmath130 around the @xmath131-axis , for all @xmath132 , we conclude that also @xmath2 and @xmath91 are invariant with respect to that kind of rotations , for all @xmath11 . more precisely , @xmath133 consequently , at a given @xmath11 , @xmath134 lay in the same orthogonal plane to the @xmath131-axis , for all @xmath135 . this fact has important implications for implementing efficient numerical schemes , as we will see in section [ s : numerical ] . besides invariance by rotations , is also mirror invariant : if @xmath59 is a solution of , so is @xmath136 , etc . therefore , if @xmath137 , then that symmetry will be again preserved during the evolution , i.e , @xmath138 , for all @xmath11 , or , in other words , @xmath2 and @xmath139 will be symmetric with respect to the plane containing the @xmath131-axis and the @xmath140-axis . in our case , a regular polygon with @xmath41 sides like has @xmath141 of those mirror symmetries . an important corollary that will be used later is that @xmath142 is a positive multiple of the vector @xmath143 . one of the fundamental symmetries of the nls equation is the so - called galilean transformations , i.e. , if @xmath75 is a solution of , so is @xmath144 therefore , if we choose an initial datum such that @xmath145 , for all @xmath146 , i.e. , such that @xmath147 , for all @xmath146 , then , if the solution is unique , @xmath148 , for all @xmath149 . in this paper , this symmetry is enough for our purposes ; for the full symmetry group of the nls equation , see ( * ? ? ? * section 17.2 ) . in the case of a planar initial datum @xmath1 for , the torsion is equal to zero and , from , @xmath150 is the curvature of @xmath1 . hence , the @xmath150 corresponding to a regular polygon with @xmath41 sides is given by , i.e. , @xmath151 like , @xmath150 is @xmath105-periodic and , since is invariant with respect to space translations , @xmath52 is also @xmath105-periodic , for all @xmath152 . moreover , it satisfies @xmath153 , for all @xmath132 . therefore , the galilean transformation holds , i.e. , @xmath154 , for all @xmath132 , and for all @xmath152 . bearing in mind the previous arguments , we calculate the fourier coefficients of @xmath52 : @xmath155ds \cr & = \rme^{-\rmi ( mk)^2 t - \rmi m(j - k)(2mkt)}\hat\psi(j - k , t).\end{aligned}\ ] ] this identity holds for all @xmath156 and @xmath157 . in particular , evaluating both sides at @xmath158 , @xmath159 , so @xmath75 can be expressed as @xmath160 where @xmath161 is a constant that depends on time and that has to be chosen in such a way that the corresponding curve @xmath59 and tangent vector @xmath28 are periodic of period @xmath36 for a fixed @xmath11 . remark that , from the arguments following , @xmath161 can be taken up to a constant with modulus one , so , in this paper , we assume without loss of generality that @xmath161 is real for any given @xmath11 . when @xmath162 , we have trivially @xmath163 ; then , combining and , we recover the following well - known identity : @xmath164 observe that , if we solve the free schrdinger equation @xmath165 for , we get @xmath166 where @xmath167 is the well - known jacobi theta function . equation is the same as , but with @xmath168 , for all @xmath152 . let us evaluate at @xmath170 , where @xmath100 , @xmath101 , and we can suppose without loss of generality that @xmath43 . then , bearing in mind the identity , @xmath171 where @xmath57 was defined in . moreover , from , it follows that @xmath172 for a certain angle @xmath173 that depends on @xmath174 ( as well as , of course , @xmath175 and @xmath46 ) . therefore , @xmath52 in @xmath176 has evolved from @xmath41 dirac deltas at @xmath162 , to @xmath45 deltas at @xmath53 , for @xmath46 odd ; and to @xmath47 deltas at @xmath53 , for @xmath46 even . furthermore , introducing into , and restricting ourselves to @xmath177 , i.e. , @xmath106 , we conclude that @xmath178 the coefficients multiplying the dirac deltas are in general not real , except for @xmath162 and @xmath179 . therefore , @xmath56 does not correspond to a planar polygon , but to a skew polygon with @xmath45 sides , for @xmath46 odd ; and to a skew polygon with @xmath47 sides , for @xmath46 even . moreover , @xmath180 has to be determined in such a way that the polygon is closed ; this is done in the next section , where we show that its rather involved choice is unique , concluding that @xmath52 is periodic in time , with period @xmath181 . since the dirac deltas are equally spaced at @xmath182 , the length of the sides is the same . observe that , if @xmath183 is odd , i.e. , if @xmath184 , then there is no vertex at @xmath61 . for instance , if @xmath185 , @xmath44 , i.e. , @xmath186 , then @xmath187 in this case , the coefficients multiplying the dirac deltas are real , so @xmath188 is simply the curvature , and @xmath189 has been chosen in order for the integral of @xmath188 over @xmath190 to be @xmath36 . bearing in mind the symmetries of the problem , we conclude that the corresponding polygon is a regular planar polygon with @xmath41 sides , which lives at a certain plane @xmath191 , but with the axes turned by @xmath192 degrees with respect to the initial datum , as r. jerrard and d. smets predicted in @xcite . given a function @xmath193 , recovering the basis vectors @xmath28 , @xmath79 and @xmath80 from @xmath75 implies integrating . as seen in the previous section , at @xmath182 , @xmath52 is a sum of @xmath45 ( if @xmath46 odd ) or @xmath47 ( if @xmath46 even ) equally spaced dirac deltas in @xmath176 , that corresponds to a skew polygon @xmath194 of @xmath45 or @xmath47 sides . to integrate , we have to understand the transition from one side of the polygon to the next one . in order to do that , we reduce ourselves , without loss of generality , to a certain @xmath75 formed by one single dirac delta located at @xmath61 , i.e , @xmath195 , so we have to integrate @xmath196 this system corresponds to some basis vectors @xmath197 , @xmath198 and @xmath199 constant everywhere , except at @xmath61 , where there is a singularity . let us suppose that , for @xmath200 , @xmath201 , @xmath202 , @xmath203 , so we want to calculate @xmath204 , @xmath205 , @xmath206 , such that , for @xmath207 , @xmath208 , @xmath209 , @xmath210 . then , @xmath211 \cdot \left ( \begin{array}{c } { \mathbf t}(0 ^ -)^t \cr \hline { \mathbf{e}_1}(0 ^ -)^t \cr \hline { \mathbf{e}_2}(0 ^ -)^t \end{array } \right ) \cr & = \exp(\mathbf a)\cdot \left ( \begin{array}{c } { \mathbf t}(0 ^ -)^t \cr \hline { \mathbf{e}_1}(0 ^ -)^t \cr \hline { \mathbf{e}_2}(0 ^ -)^t \end{array } \right ) .\end{aligned}\ ] ] we can compute explicitly the matrix exponential . writing @xmath212 , then @xmath213c_\theta s_\theta \cr -s_\rho s_\theta & [ c_\rho - 1]c_\theta s_\theta & c_\rho s_\theta^2 + c_\theta^2 } , \end{aligned}\ ] ] where @xmath214 , @xmath215 , @xmath216 , @xmath217 . the matrix @xmath218 is a rotation matrix @xcite ; more precisely , it is the rotation matrix corresponding to a rotation about an axis @xmath219 by an angle @xmath102 . coming back to the general form of @xmath75 , we have to integrate @xmath45 or @xmath47 times to obtain a closed skew , i.e. , non - planar polygon with @xmath45 or @xmath47 sides . but , according to , in @xmath106 , @xmath220 hence , we conclude that , at @xmath221 , the angle @xmath102 between two adjacent sides is constant . @xmath222 furthermore , the structure of the polygon is completely determined by the quantities @xmath173 appearing in the generalized quadratic gau sum , where @xmath223 . let be @xmath224 the rotation matrix corresponding to @xmath225 . if @xmath226 , @xmath224 is simply an identity matrix and can be ignored . otherwise , @xmath224 is obtained after evaluating , at @xmath227 . therefore , @xmath228 or , equivalently , @xmath229 where @xmath230 , and @xmath231 is periodic modulo @xmath46 , i.e. , @xmath232 . this formula is valid for both @xmath46 odd and @xmath46 even , bearing in mind that half of the @xmath231 matrices are simply identity matrices if @xmath46 is even . this explains why the polygon has @xmath233 sides , when @xmath46 is even ; equivalently , it could be regarded as having @xmath46 sides , but half of them being indistinguishable , because when the angle between two adjacent sides is zero , they merge into one single side . in order for the polygon to be closed , we have to choose @xmath102 in in such a way that the basis vectors @xmath28 , @xmath79 and @xmath80 are periodic , i.e. , @xmath234 which is equivalent to imposing the condition @xmath235 let us define @xmath236 from , @xmath237 is an @xmath41-th root of the identity matrix . moreover , it is also a rotation matrix that induces a rotation of @xmath105 degrees around a certain rotation axis . therefore , we have to choose @xmath102 in order that any of the following properties is satisfied : @xmath238 were @xmath239 and @xmath240 denote the trace and the spectrum of @xmath237 , respectively . we worked with the trace , because it is algebraically easier . in order to understand its structure , we analyzed a few cases ( @xmath241 ) with the help of a symbolic manipulator . the obtained results strongly suggest that , for any @xmath46 ( and for any @xmath175 coprime with it ) the only possible choice for @xmath242 is given by . remark that if @xmath46 is odd , the angle @xmath102 between two adjacent sides is the same for @xmath46 and for @xmath243 . although finding a universal proof that holds for any @xmath46 goes beyond the scope of this paper , we have checked it for a few more @xmath46 . moreover , this conjecture is in agreement with our numerical simulations , as we will see in section [ s : experiments ] . therefore , we think that there is compelling evidence that is valid for all @xmath101 . once we have found the correct choice of @xmath102 , we get from , @xmath244 it is straightforward to check that @xmath245 . notice also that @xmath246 finally , from and , we conclude that @xmath56 , @xmath247 , is given by . this completes the proof of theorem [ t : theorem ] . knowing the function @xmath56 , the basis vectors @xmath28 ( and , hence , the curve @xmath59 ) , @xmath79 and @xmath80 can be completely determined up to a rigid movement , which , on the other hand , for an arbitrary polygon , can be pretty complicated to specify . however , since our initial datum is a regular planar polygon , the symmetries of the solution are very advantageous . the easiest way to understand and to implement numerically the correct rotation ( although not necessarily the simplest option for symbolic manipulation ) is to work with @xmath59 . to that aim , given a rational time @xmath53 , we compute first the associated rotation matrices @xmath224 . then , by means of , we obtain up to a rotation the piecewise constant vectors @xmath28 , @xmath79 and @xmath80 , which we denote @xmath248 , @xmath249 and @xmath250 : @xmath251 and the case @xmath177 can be assigned any value , for instance the identity matrix . again , it is not necessary to consider separately @xmath46 odd and @xmath46 even . once @xmath248 has been obtained , @xmath252 , which is @xmath59 up to a rigid movement , is computed recursively : @xmath253 where , again , @xmath254 can be given any value , for instance @xmath255 . to obtain the correct rotation for @xmath59 , we use the symmetries explained in section [ s : formulation ] . in particular , we use the fact that , at any time @xmath11 , @xmath256 , @xmath113 , have to be coplanar and lay on a plane orthogonal to the @xmath131-axis ; and that @xmath257 is a positive multiple of @xmath143 . an efficient algorithm is as follows : 1 . compute @xmath258 and @xmath259 . 2 . compute @xmath260 . 3 . compute the scalar product between @xmath261 and @xmath262 , i.e. , @xmath263 . if @xmath264 , @xmath265 is the identity matrix . if not , @xmath265 is the rotation matrix that induces a rotation of @xmath266 degrees around the axis given by the vector @xmath267 . 5 . compute @xmath268 . compute the rotation matrix @xmath269 that induces a rotation of @xmath270 degrees around the @xmath131-axis . compute the sought rotation matrix , @xmath271 . update @xmath272 and @xmath273 . once we have computed the correct rotation of @xmath59 and @xmath28 , we still have to specify a translation for @xmath59 . again , bearing in mind the symmetries of the problem , we translate @xmath59 in such a way that its mass center , given by the mean of @xmath256 , @xmath274 , is in the @xmath131-axis . thus , the position of @xmath59 is completely specified , up to a vertical translation . in this section , we simulate numerically and taking respectively and as the initial data . there have been a couple of papers devoted to reproducing numerically the self - similar solutions of the schrdinger map equation . in @xcite , a crank - nicholson scheme was consider , together with a finite - difference discretization of @xmath275 . later on , in @xcite , both finite - difference discretizations and pseudo - spectral discretizations were considered to simulate and its equivalent on the hyperbolic plane . in particular , given the asymptotical structure of the self - similar solutions of , a truncated domain @xmath276 $ ] , @xmath277 , with a grid based on the chebyshev nodes was found to be very convenient to represent those solutions . nevertheless , due to the clustering of the chebyshev nodes , an explicit scheme to advance in time implied the undesirable restriction @xmath278 , where @xmath279 is the number of nodes . in order to solve that , it was chosen to work with the stereographic projection of the tangent vector @xmath280 over @xmath67 , @xmath281 which transforms into a nonlinear schrdinger equation : @xmath282 the advantage of this equation , as opposed to , is that the higher - order term @xmath283 can be treated implicitly , therefore eliminating or at least reducing significantly the restrictions on @xmath284 . indeed , working with was a very adequate choice for the purposes of @xcite , where a second - order semi - implicit backward differentiation formula was chosen . this scheme is very stable because it imposes a very strong decay in the high frequency modes ; moreover , its low order is easily and effectively compensated by the use of an adaptive method , both in space and in time . however , in this paper , working with does not seem to be such a good option . indeed , unlike in @xcite , where an extremely high accuracy was required only for @xmath285 , we are now interested in the behaviour of @xmath28 at all times , for which a second - order scheme seems a very poor option . after many unsuccessful attempts , we admit that we have been unable to find a good higher - order scheme for . furthermore , the stereographic projection may introduce an artificial singularity at the points near the south pole of the sphere . finally , we are also interested in the evolution of the curve @xmath59 , for which working with is of limited help . bearing in mind the previous arguments , we have opted to work directly with or , more precisely , with a combination of and : @xmath286 to which we have applied a fourth - order runge - kutta scheme in time : @xmath287 the approximations @xmath288 and @xmath289 refer to the numerically obtained values of @xmath59 and @xmath28 at @xmath290 ; the last line guarantees that @xmath291 , for all @xmath292 . additionally , we can also project @xmath293 , @xmath294 , @xmath295 onto the unit sphere , but without significant improvement in the numerical results . we have combined with a pseudo - spectral discretization directly in space . more precisely , since we deal with periodic solutions in @xmath176 , we have simulated the evolution of the curve @xmath59 and of the tangent vector @xmath28 at @xmath279 equally spaced nodes @xmath296 , @xmath297 . in order to compute @xmath298 , @xmath275 , etc . , we remember that , given a periodical function @xmath299 evaluated at @xmath300 , its derivatives at @xmath300 can be spectrally approximated as @xmath301 where @xmath302 both and , which are respectively inverse and direct discrete fourier transforms ( dft ) of @xmath279 elements , can be computed efficiently by means of the fast fourier transform ( fft ) algorithm @xcite . moreover , since @xmath59 and @xmath28 are invariant with respect to rotations of @xmath105 degrees around the @xmath131-axis , as shown in , we can reduce the computational cost of and to a dft of @xmath303 elements . in what follows , we explain how to apply this idea to @xmath28 . this is valid , with no change , for any vector satisfying , i.e. , @xmath59 , @xmath298 , @xmath275 , etc . now , let us consider the first two components of @xmath28 . denoting @xmath304 , and bearing in mind that @xmath305 , @xmath306 therefore , the non - zero @xmath307 are obtained by a dft of @xmath303 elements : @xmath308\rme^{-2\pi \rmi jk/(n / m)}.\ ] ] for the third component of @xmath28 , bearing in mind that @xmath309 , we conclude @xmath310 so the non - zero @xmath311 are obtained again by a dft of @xmath303 elements : @xmath312 since @xmath313 is real , can be further simplified to a dft of @xmath314 elements . combining with a pseudo - spectral discretization in space , we obtain a scheme with a time - step restriction on @xmath284 that appears to be of the form @xmath315 , with @xmath316 . however , for a fixed @xmath41 , we only want to simulate the evolution of the curve @xmath59 and of the tangent vector @xmath28 for @xmath317 $ ] . denoting @xmath318 the biggest @xmath284 that makes stable , the smallest number of time - steps @xmath319 needed to reach @xmath320 satisfies @xmath321 from , together with and , it follows that , if the number of spatial nodes @xmath279 is of the form @xmath322 , with @xmath323 , then , the computational cost for simulating is exactly the same for any number of initial sides @xmath41 . this important fact enables us to make consistent comparisons between different @xmath41 , because we are considering an equivalent spatial resolution , i.e , we are using the same number of points @xmath303 per initial side . we take @xmath303 a power of two to take the greatest possible advantage of the fft . obviously , for @xmath303 constant , @xmath284 decreases as @xmath324 ; therefore , the most accurate results will be expected for the largest @xmath41 , as we will observe in our numerical experiments . in the following section , we simulate @xmath59 and @xmath28 for initial data with different @xmath41 . as we will see , despite its simplicity , gives surprisingly good results . in section [ s : xt ] , in order to construct algebraically the evolution of a regular polygon at rational times , we have done some very strong assumptions , the most important one being uniqueness . however , as we will see in this section , the numerical experiments are in complete agreement with their theoretical predictions . to illustrate this , we have simulated the evolution of the curve @xmath59 and of the tangent vector @xmath28 by means of , taking regular polygons with different numbers of sides @xmath41 as the initial data , and making @xmath59 and @xmath28 evolve until @xmath320 . the initial data @xmath1 and @xmath110 are given respectively by and ; in the case of @xmath59 , the non - vertex points are computed by linear interpolation . we have taken @xmath279 equally spaced nodes @xmath325 , @xmath297 ; nevertheless , from the symmetries of @xmath59 and @xmath28 , as explained in the previous section , we only have to describe the evolution of @xmath59 and @xmath28 at the first @xmath303 nodes @xmath326 , @xmath327 . we have divided the time - interval @xmath328 $ ] in @xmath319 equally spaced time steps of length @xmath329 . therefore , during the simulation of , we obtain @xmath288 and @xmath289 , at @xmath330 , @xmath331 . in our experiments , bearing in mind , we have chosen @xmath332 , for @xmath333 . remember that , in section [ s : xt ] , we had constructed algebraically the curve @xmath59 up to a vertical movement . therefore , in order to completely specify @xmath59 at a given time , we would need to give the height of one point or , more conveniently , the height @xmath334 of the mass center , which is precisely the mean of all the values @xmath335 : @xmath336 we have observed that @xmath334 can be approximated with very high accuracy by means of a constant multiplied by @xmath11 ; more precisely , @xmath337 where @xmath338 is the mean speed . in table [ t : speed ] , we give the maximum discrepancy between @xmath334 and its linear approximation @xmath339 , i.e. , @xmath340 . we have considered different numbers of initial sides @xmath41 and different numbers of nodes @xmath279 ; the errors are very small and they seem to decrease as @xmath341 . this gives very strong evidence that @xmath334 is linear of the form @xmath342 . observe that @xmath343 , which is also offered in table [ t : speed ] for @xmath344 , grows with @xmath41 , tending to @xmath345 , as @xmath41 tends to infinity , i.e. as the initial polygon @xmath1 tends to a circle . .@xmath340 , for different numbers of initial sides @xmath41 , and of nodes @xmath279 . @xmath346 ; @xmath338 . the values corresponding to @xmath343 have been calculated with @xmath344 . [ cols="^,^,^,^,^,^,^",options="header " , ] on the other hand , comparing the algebraically constructed @xmath347 with the numerically obtained @xmath348 is more problematic . for example , in figure [ f : tm3p1q3 ] , we have compared @xmath348 ( left ) with @xmath347 ( right ) , for an initial triangle , @xmath349 , at @xmath350 , @xmath344 . the exact value of @xmath28 at that time is given by @xmath347 , which , by construction , is a piecewise constant function with exactly @xmath351 pieces , that we denote @xmath352 , @xmath353 , and whose explicit values , in this case , have an easy algebraic expression : @xmath3542 - 1 \cr -\sqrt{\sqrt[3]4 - 1 } \cr 1 - \sqrt[3]4 } , \quad { \mathbf t}_2 = \pmatrix { 1 \cr 0 \cr 0 } , \quad { \mathbf t}_3 = \pmatrix { \sqrt[3]2 - 1 \cr \sqrt{\sqrt[3]4 - 1 } \cr \sqrt[3]4 - 1 } ; \ ] ] @xmath355 , @xmath356 and @xmath357 are respectively @xmath358 , @xmath359 and @xmath360 rotated @xmath361 degrees around the @xmath131-axis ; @xmath362 , @xmath363 and @xmath364 are respectively @xmath358 , @xmath359 and @xmath360 rotated @xmath365 degrees around the @xmath131-axis . observe that @xmath348 is almost identical to @xmath347 , but clearly exhibits a gibbs - type phenomenon . however , in @xmath348 , if we take the central values of each interval ( altogether 27 circles , indicated in figure [ f : tm3p1q3 ] with a black circle ) , and compare them componentwise with , we obtain a maximum error equal to @xmath366 . versus @xmath347 , for @xmath349 , at @xmath350 . @xmath367 appears in blue , @xmath368 in green , @xmath369 in red . in @xmath348 , the gibbs phenomenon is clearly visible . the black circles denote the points chosen for the comparisons.,title="fig:",scaledwidth=50.0% ] versus @xmath347 , for @xmath349 , at @xmath350 . @xmath367 appears in blue , @xmath368 in green , @xmath369 in red . in @xmath348 , the gibbs phenomenon is clearly visible . the black circles denote the points chosen for the comparisons.,title="fig:",scaledwidth=50.0% ] from the mirror symmetries of the initial data , we conclude that the curve @xmath370 , @xmath371 , lives in a plane that contains the @xmath131-axis . indeed , @xmath372 are the only values for which @xmath2 describe a planar curve . in what follows , we will describe @xmath3 , although all said here is immediately applicable to any @xmath372 . since @xmath373 is planar , bearing in mind that @xmath374 , @xmath375 , we rotate @xmath3 clockwise @xmath376 degrees around the @xmath131-axis , until it lays on the plane @xmath377 , which we identify with @xmath67 . then @xmath3 becomes @xmath378 in figure [ f : xs0 ] , we have plotted the numerically obtained @xmath70 for @xmath379 . besides the conspicuous fractal character of the curves , which immediately reminds us of the works @xcite , their most striking feature ( see figure [ f : riemann ] ) is how much @xmath380 , i.e. , z(t ) without the vertical movement , resembles the graph of @xmath381.\ ] ] , for @xmath379 . we have plotted the @xmath70 corresponding to all the @xmath382 points @xmath383 obtained during the execution of .,title="fig:",scaledwidth=33.33% ] , for @xmath379 . we have plotted the @xmath70 corresponding to all the @xmath382 points @xmath383 obtained during the execution of .,title="fig:",scaledwidth=33.33% ] , for @xmath379 . we have plotted the @xmath70 corresponding to all the @xmath382 points @xmath383 obtained during the execution of .,title="fig:",scaledwidth=33.33% ] the function @xmath72 was conveniently used in @xcite , in order to study riemann s non - differentiable function , which is precisely the real part of @xmath72 , i.e. , @xmath384 this function is non - differentiable everywhere , except at those rational points @xmath385 , with @xmath175 and @xmath46 both odd . on the other hand , when constructing algebraically @xmath3 for a given t , we have observed that the only times at which @xmath3 has no corner are of the form @xmath386 , with @xmath175 odd and @xmath184 . therefore , in order to compare @xmath72 and @xmath3 , we have to redefine slightly @xmath72 in : @xmath387,\ ] ] i.e. , we multiply @xmath388 in by @xmath389 , in order to orientate it correctly ; then we change the period from @xmath390 $ ] to @xmath391 $ ] , and the orientation of the parameterization , by doing @xmath392 . on the other hand , given a number of initial sides @xmath41 , we define @xmath393.\ ] ] in figure [ f : riemann ] , we have plotted @xmath72 ( red ) , as defined in , versus @xmath394 , for @xmath395 . since we have obtained numerically @xmath394 at @xmath396 points in @xmath397 $ ] , we have evaluated @xmath72 precisely at those points . although the figures are not identical , they are extremely similar . moreover , if we scale them accordingly , we observe that , the bigger @xmath41 grows , the more similar to @xmath72 is @xmath394 . indeed , @xmath72 and @xmath398 , except for the scaling , are visually undistinguishable . ( red ) , as defined in , against @xmath394 ( blue ) , for @xmath395 . to compute @xmath394 , we took @xmath344 ; to compute @xmath72 , we took @xmath399 . we have plotted the @xmath394 corresponding to all the @xmath382 points @xmath383 obtained during the execution of . we have computed algebraically and plotted @xmath72 at those times . observe that , except for the scaling , @xmath72 and @xmath398 are visually undistinguishable.,title="fig:",scaledwidth=33.33% ] ( red ) , as defined in , against @xmath394 ( blue ) , for @xmath395 . to compute @xmath394 , we took @xmath344 ; to compute @xmath72 , we took @xmath399 . we have plotted the @xmath394 corresponding to all the @xmath382 points @xmath383 obtained during the execution of . we have computed algebraically and plotted @xmath72 at those times . observe that , except for the scaling , @xmath72 and @xmath398 are visually undistinguishable.,title="fig:",scaledwidth=33.33% ] ( red ) , as defined in , against @xmath394 ( blue ) , for @xmath395 . to compute @xmath394 , we took @xmath344 ; to compute @xmath72 , we took @xmath399 . we have plotted the @xmath394 corresponding to all the @xmath382 points @xmath383 obtained during the execution of . we have computed algebraically and plotted @xmath72 at those times . observe that , except for the scaling , @xmath72 and @xmath398 are visually undistinguishable.,title="fig:",scaledwidth=33.33% ] in order to give some numerical evidence that an adequately scaled @xmath394 converges to @xmath72 , we study @xmath400 , for certain @xmath401 and @xmath402 . after applying a least - square fitting , we choose @xmath403}{{\operatorname{mean}}(|z_m(t ) - { \operatorname{mean}}(z_m(t))|^2)}\right ) , \\ \mu_m = { \operatorname{mean}}(\phi(t ) ) - \lambda_m{\operatorname{mean}}(z_m(t ) ) , } \ ] ] where @xmath404 denotes the mean over the @xmath319 + 1 points , etc . figure [ f : convergence ] shows the absolute and relative discrepancies between @xmath72 and the scaled @xmath394 , i.e. , @xmath405 and @xmath406 , for @xmath407 . bearing in mind that we are comparing @xmath408 different points one by one , the results give a pretty strong evidence of convergence as the number of initial sides @xmath41 tends to infinity , and strongly support the idea that @xmath409 is also a multifractal , for all @xmath410 . ( starred line ) and @xmath406 ( circled line ) , with @xmath411 and @xmath412 given by . bearing in mind that we are comparing @xmath408 points one by one , the convergence is pretty strong.,scaledwidth=50.0% ] as we mentioned in the introduction , the proof that @xmath413 , as defined in , is a multifractal was done in @xcite . in this respect , it is fundamental the identity given for @xmath388 , as defined in , which states that @xmath414 with @xmath415 , @xmath416 , @xmath43 , @xmath417 . this identity is proved in theorem 4.2 in @xcite . of particular relevance is the fact that the hlder exponent is @xmath73 , i.e. , that @xmath418 lower - order terms . in order to prove analytically that @xmath394 , is a multifractal , we would need an equivalent of . we have made some numerical experiments ( see for example figure [ f : holder ] , for @xmath349 , @xmath185 , @xmath419 ) , and all of them give strong evidence that the hlder exponent of @xmath70 is @xmath73 for rational times , i.e. , that @xmath420 . versus @xmath421 , for @xmath349 . the asymptotically linear relation between both quantities as @xmath422 is evident.,scaledwidth=50.0% ] nevertheless , how the constants depend on the denominator @xmath46 , something which is a fundamental ingredient in the arguments in @xcite and in @xcite , is unclear . this question deserves a much more detailed analysis that we plan to make in a forthcoming paper . in the previous subsection , we have given evidence of the multifractal character of the curve @xmath3 . however , in @xcite , fractal - like phenomena were observed also on the tangent vector @xmath28 when imposing fixed boundary conditions . in the following lines , we will show that similar observations are valid for periodic boundary conditions , too . as we have seen , @xmath59 can be recovered algebraically up to a vertical movement , while @xmath28 can be completely recovered algebraically at rational times . furthermore , @xmath347 is really a piecewise constant function , with no noise associated to the gibbs phenomenon . this gives us a very powerful tool to progress in the understanding of @xmath28 ( and also of @xmath59 ) , avoiding numerical simulations at all . a very interesting question is , given a rational time @xmath53 , with _ small _ @xmath46 , corresponding in @xmath59 to a polygon with @xmath45 or @xmath47 sides , and in @xmath28 to a piecewise continuous function with @xmath45 or @xmath47 jumps , what happens at a time @xmath425 , @xmath426 ? let us take @xmath427 , with @xmath428 , in order that @xmath429 . assuming that @xmath430 , then , @xmath431 , so @xmath425 would correspond to a polygon with @xmath432 or @xmath433 sides . and @xmath347 , at @xmath434,title="fig:",scaledwidth=50.0% ] and @xmath347 , at @xmath434,title="fig:",scaledwidth=50.0% ] in figure [ f : xteps ] , we have have plotted @xmath435 and @xmath347 , for @xmath349 , at @xmath436 . while visually there is no difference whatsoever between @xmath435 at @xmath437 and @xmath435 at @xmath438 , we do not have a skew polygon with 6 sides , but a skew polygon with @xmath439 sides that closely resembles a polygon with 6 sides . on the other hand , the corresponding @xmath347 are very different . indeed , in the plot of @xmath347 , we can clearly appreciate six spiral - like structures whose centers are precisely the six constant values of @xmath28 at @xmath440 . nevertheless , it is important to underline that these structures are not really spirals , but the plot of the @xmath441 different values taking by @xmath28 , that closely resembles a curve with six spirals . furthermore , these spirals remind us of the cornu spirals that appeared in @xcite . an open question that arises naturally is up to what extent the multiple - corner problem can be explained as a sum of several one - corner problems . and @xmath347 , at @xmath442 . , title="fig:",scaledwidth=50.0% ] and @xmath347 , at @xmath442 . , title="fig:",scaledwidth=50.0% ] another interesting question is what happens if we take a rational time @xmath53 with _ @xmath46 , such that there is no pair @xmath443 , with both @xmath444 and @xmath445 _ small_. in this case , the situation is very different . in figure [ f : xteps2 ] , we have plotted @xmath435 and @xmath347 for @xmath349 , at @xmath442 . while the left - hand side is not so different from the left - hand side of figure [ f : xteps ] , the right - hand side is a set of @xmath446 points in @xmath447 that creates a spectacular fractality sensation with spiral - like structures at three or four different scales . this can be better appreciated in figure [ f : tepszoom ] , where we have plotted the stereographic projection of @xmath28 onto @xmath67 . in general , a rational time @xmath53 , with @xmath424 , can be regarded as an approximation of an irrational time . trying to fully understand the phenomena exhibited by @xmath91 at those times is a challenging question that we postpone for the future . in this paper , we have studied the evolution of and , for a regular planar polygonal with @xmath41 sides as the initial datum . the algebraic calculations , backed by complete numerical simulations , prove ( under a certain uniqueness assumption ) that the curve @xmath2 is a polygon at times which are rational multiples of @xmath54 , i.e. , @xmath42 , @xmath43 , with the number of sides depending on @xmath46 , while the tangent vector @xmath91 is piecewise constant at those times . this could be seen as a nonlinear version of the talbot effect , in the spirit of the work @xcite by berry and goldberg . we have obtained a striking connection between the curve @xmath3 and the so - called riemann s non - differentiable function . in @xcite , s. jaffard proved that this function is an example of multifractal whose spectrum of singularities satisfies the frisch - parisi conjecture . although there is strong numerical evidence that @xmath3 is also a multifractal , an analytical proof seems to be challenging . in fact , a first step in this direction is to give sense to our solutions from an analytical point of view . in the future , we also plan to extend these ideas to arbitrary polygons , and to do a more detailed study on fractality ; for such purposes , a more complete algebraically description of @xmath59 and @xmath28 is no doubt required . finally , it would be interesting to prove . we are indebted to the referees and editors for the thorough , constructive and helpful comments and suggestions . we also wish to thank v. banica and f. chamizo for enlightening conversations . this work was supported by the spanish ministry of economy and competiveness , with the project mtm2011 - 24054 , and by the basque government , with the project it641 - 13 . the generalized quadratic gau sums are defined by @xmath448 for given integers @xmath449 , with @xmath450 . from now on , we assume @xmath451 , and @xmath452 , which are the cases dealt with in this paper . the value of these sums was calculated for the first time by gau for the case @xmath453 @xcite . more precisely , given two integers @xmath454 such that @xmath452 , then @xmath455 where @xmath456 denotes the jacobi symbol . as we will see in the following lines , the generalized quadratic gau sums can be reduced to normal quadratic gau sums by completing the square . since they are multiplicative , i.e. , @xmath457 we can assume without loss of generality that @xmath458 is either odd or a power of two . if @xmath458 is odd , let us find a certain @xmath459 such that @xmath460 . this @xmath459 is unique modulo @xmath458 and is precisely the inverse of @xmath461 in @xmath462 ; since @xmath463 , its existence is guaranteed by bzout s lemma , and it can be efficiently computed , for instance , by the extended euclidean algorithm . then , we have @xmath464 if @xmath458 is a power of two , we first observe that @xmath465 therefore , we have two cases : @xmath466 and @xmath467 . if @xmath466 , then @xmath25 is odd , so @xmath468 on the other hand , if @xmath467 , @xmath469 , if @xmath470 is odd . let us suppose @xmath470 is even . then , since @xmath25 is odd , we follow the previous reasoning to find a certain @xmath459 such that @xmath471 , so @xmath472 in general , given an arbitrary @xmath473 , we can factorize it as @xmath474 and use the multiplicative character of the generalized quadratic gau sums to calculate @xmath475 . for instance , if @xmath458 is even , but @xmath476 is odd , then @xmath477 i.e , @xmath478 , if @xmath470 is even , etc . to conclude this appendix , let us mention explicitly the value of @xmath479 , deduced from the previous calculations , and which is of especial relevance in this paper : @xmath480
the dynamics of micron and nanoscale cantilevers are important to a wide variety of technologies . for example , the invention of the atomic force microscope ( afm ) @xcite , which relies upon the dynamics of a cantilever a few hundred microns in length , has revolutionized surface science paving the way for direct measurements of intermolecular forces and topographical mappings with atomic precision for a broad array of materials including semiconductors , polymers , carbon nanotubes , and biological cells @xcite ( see @xcite for current reviews ) . in conventional dynamic atomic force microscopy the cantilever is used to measure the force interactions between the cantilever tip and sample . cantilevers smaller than conventional afm have also been used to unfold single protein molecules with improved force and time resolution @xcite . it has also been proposed to exploit the inherent thermal motion of small cantilevers to make dynamic measurements of single molecules @xcite . a passive undriven cantilever placed in a viscous fluid will exhibit stochastic oscillations caused by the thermal bombardment of fluid molecules by brownian motion . in fact , measuring the thermal spectra of the cantilever is a commonly used afm calibration technique @xcite . in all of these applications the ultimate force sensitivity of a particular measurement is limited by the inherent thermal noise of the experimental system . there are at least two ways to improve upon this limitation : ( i ) make the cantilevers smaller @xcite , ( ii ) use correlated measurements of multiple cantilevers @xcite . uniformly decreasing the dimensions of a cantilever results in the favorable combination of decreasing the cantilever s equivalent spring constant while increasing its resonant frequency yielding improved force sensitivity and time resolution . by measuring the cross - correlations between two cantilevers in fluid the independent fluctuations of the two do not contribute , leaving only the smaller correlated fluctuations due to coupling through the fluid . this type of approach has been used to measure femtonewton forces on millisecond time scales between two micron scale beads placed in separate optical traps ( an improvement of a hundredfold from prior measurements ) @xcite . the ability to significantly increase the force resolution by making correlated measurements has yet to be exploited for micron and nanoscale cantilevers . the magnitude of the fluid coupled noise will depend upon the spacing and exact geometries of the cantilevers . combining ( i ) and ( ii ) and measuring the correlations of multiple nanoscale cantilevers offer the potential for experimental measurements with unprecedented force and time resolution @xcite . as experimental measurement continues to push toward the stochastic limit it is important that we build a physical understanding of the stochastic dynamics of micron and nanoscale cantilevers for the precise conditions of experiment including complex cantilever geometries @xcite , the effects of nearby walls @xcite , and the fluid - coupled dynamics of multiple cantilevers in an array configuration @xcite . although micron and nanoscale cantilevers exhibit stochastic motion due to the thermal motion of matter the elastic structures are still large compared to individual fluid molecules and the equations of continuum mechanics remain valid . in what follows we are concerned with situations where the knudsen number ( the ratio of the mean free path of the fluid particles to the width of a cantilever ) remains sufficiently small so that this statement remains true . this means that the fluid is described by the usual navier - stokes equations with no - slip and stress continuity boundary conditions at the solid surfaces . the navier - stokes equations governing the motion of an incompressible fluid , and written in nondimensional form , are @xmath0 where @xmath1 is the fluid velocity , @xmath2 is the pressure , and @xmath3 is time . there are two inertial terms on the left hand side of eq . ( [ eq : ns ] ) multiplied by the nondimensional parameters @xmath4 and @xmath5 . the strouhal number @xmath6 plays the role of a frequency based reynolds number expressing the ratio between local inertia forces and viscous forces where @xmath7 and @xmath8 are characteristic length and time scales , respectively . the velocity based reynolds number @xmath9 expresses the ratio between convective inertial forces and viscous forces . micron and nanoscale cantilevers are characterized by high oscillation frequencies and small oscillation amplitudes . in this case @xmath10 so that the nonlinear convective inertial term @xmath11 is negligible and the equations become linear . however , the strouhal number , @xmath12 ( using the half - width @xmath13 as the appropriate length scale ) is often not negligible . as a result , the local inertia term must be kept in eq . ( [ eq : ns ] ) making the analysis more difficult . in addition , experimentally motivated cantilevers are often of complex geometry , near surfaces , or in an array configuration with multiple cantilevers in close proximity . these difficulties have led to the development of a thermodynamic approach to calculate the stochastic dynamics of micron and submicron scale cantilevers for the precise conditions of experiment ( discussed below ) @xcite . in the following it is assumed that @xmath5 is negligible , and we leave off the subscript denoting @xmath14 by @xmath15 . the fluid equations are coupled with the governing equations of elasticity , @xmath16 where @xmath17 is the stress tensor , @xmath18 is the cantilever density , and @xmath19 is the cantilever deflection . equations ( [ eq : ns]),([eq : mass ] ) , and ( [ eq : beam ] ) represent the governing deterministic continuum equations . if one considers lengths scales on the order of a few atomic lengths these equations would include stochastic terms @xcite . this represents a difficult fluid - solid interaction problem governing the stochastic cantilever dynamics . a brute force molecular dynamics approach which would resolve the stochastic motion of all of the fluid and solid molecules is computationally prohibitive however , the system is always near equilibrium permitting a much more accessible thermodynamic based solution strategy . the paper is organized as follows . a very general thermodynamic approach , valid for complex geometries as well as for arrays of closely spaced cantilevers , is discussed in section [ section : thermodynamic_approach ] . analytical theory based upon the thermodynamic approach , valid for long slender cantilevers , is discussed in section [ section : analytical_theory ] . in section [ section : micron_scale_beam ] a micron scale cantilever , similar to an atomic force microscope , is explored and in section [ section : nanoscale_cantileves ] the stochastic dynamics of an array of experimentally motivated nanoscale cantilevers are quantified using the thermodynamic approach with deterministic finite element numerical simulations to generate results for the precise conditions of experiment . a general discussion of the approach can be found in ref . @xcite . . in the thermodynamic approach of section [ section : thermodynamic_approach ] a step force @xmath20 is removed from the cantilever on the left and the deterministic cantilever deflections @xmath21 and @xmath22 are calculated.,width=288 ] the thermodynamic approach discussed here is based upon the fluctuation - dissipation theorem which states that for equilibrium systems the manner in which the system returns from a linear macroscopic perturbation is related to time correlations of equilibrium microscopic fluctuations @xcite . underlying this statement is the fact that equilibrium fluctuations and the dissipation of the system responding to a macroscopic perturbation are governed by the same physics . for the case of miniature elastic objects in a fluid the dissipation is dominated by the fluid viscosity and the fluctuations by the motion due to the bombardment by the fluid molecules . in what follows , we assume that all of the dissipation comes from the fluid and that elastic dissipation in the cantilever is negligible . this assumption , however , is not required and other sources of dissipation could be included if desired . the essence of our approach is to calculate the dissipation _ deterministically _ and to use the fluctuation - dissipation theorem to determine the cantilever s dynamics . the approach is exact and the only assumptions that have been made are classical dynamics and linear perturbations from equilibrium . the deterministic calculation of the dissipation can come from analytical theory , simplified models , or from detailed numerical simulations . the major benefit of this approach is that the deterministic calculations are straightforward , not computationally prohibitive , and methods of calculation are sophisticated and readily available . we will introduce the use of the thermodynamic approach for the case of two opposing cantilevers as shown in fig . [ fig : adjacent ] . consider one dynamical variable to be the stochastic displacement of the cantilever on the left @xmath23 where @xmath23 is a function of the microscopic phase space variables consisting of @xmath24 coordinates , @xmath25 , and conjugate momenta , @xmath26 , of the system and @xmath27 is the number of particles in the system . the statistical treatment that follows determines the ensemble average of the cantilever deflections @xmath23 over all experimental possibilities . for the undisturbed equilibrium system the ensemble average , and correlations , are denoted as @xmath28 . we now take the system to a prescribed macroscopic excursion from equilibrium and observe how the system returns to equilibrium . the macroscopic cantilever deflection as it returns to equilibrium from a prepared initial condition will be denoted as @xmath21 . for clarity of presentation , lower case variables are reserved for stochastic quantities and upper case variables are for deterministic quantities . the connection between the stochastic and deterministic quantities is particularly straightforward when the excursion from equilibrium is achieved through the application of a step force some time in the distant past that is removed at time @xmath29 , i.e. the force given by , @xmath30{cc}f_1 & \text{for } t \le 0\\ 0 & \text{for } t > 0 . \end{array } \right . \label{eq : step_force}\ ] ] the applied force should be conjugate to the variable for which the fluctuations are to be calculated . for the cantilever system under consideration here , to determine the fluctuations of the tip displacements , the force is applied to the tip of the left cantilever . the step force couples with the cantilever deflection @xmath23 and the full hamiltonian of the system @xmath31 is given by , @xmath32{cc}h_0 + \delta h & \text{for } t \le 0\\ h_0 & \text{for } t > 0 \end{array } \right . \label{eq : hamiltonian}\ ] ] where @xmath33 and @xmath34 is the unperturbed hamiltonian . for small perturbations , @xmath20 , the quantity @xmath35 will also be small , and the results are greatly simplified . the nonequilibrium ensemble average @xmath21 is given by , @xmath36 where @xmath37 , @xmath38 is boltzmann s constant , and @xmath8 is the absolute temperature . in eq . ( [ eq : noneq_ave ] ) the notation @xmath23 is used to represent the value of @xmath39 evaluated at the phase space coordinates that have evolved from the values @xmath40 and @xmath41 at time @xmath29 @xcite . in the limit of linear perturbations , @xmath42 , this simplifies to , @xmath43 where an equilibrium ensemble average is given by , @xmath44 if we now assume that @xmath39 and @xmath45 have the equilibrium average @xmath46 subtracted and recall that @xmath47 , we have our desired result , @xmath48 which can be rearranged to yield , @xmath49 the analysis is similar for the cross correlations of the deflections for two cantilevers in an array , @xmath50 where @xmath51 is the displacement of tip 2 arising from the step force @xmath20 applied to tip 1 . it is interesting to highlight that the cantilever motions are indeed correlated as indicated by eq . ( [ eq : cross ] ) . the correlated fluctuations appear contrary to the naive idea that random molecular impacts upon the individual cantilevers should only lead to uncorrelated motion . additionally , it should be emphasized that the correlated motion of two ( or more ) cantilevers is no more difficult a calculation than determining the autocorrelation of a single cantilever . on the left hand side of eqs . ( [ eq : auto ] ) and ( [ eq : cross ] ) are stochastic quantities and on the right hand side are deterministic quantities . this permits a significant reduction in effort in the calculations of the auto- and cross - correlations of the cantilever deflections . the macroscopic deflections @xmath21 and @xmath22 can be found from deterministic theory , model equations , or numerical simulations and the stochastic quantities are found simply through application of eqs . ( [ eq : auto ] ) and ( [ eq : cross ] ) . the deterministic problem then reduces to the fluid - solid interaction problem given by the removal of a step force on a cantilever . for the step force eq . ( [ eq : step_force ] ) , for @xmath52 , @xmath21 has a finite deflection . after the removal of the step force the cantilever returns to its equilibrium position of @xmath53 . the precise manner in which the cantilever returns to equilibrium is dominated by the dissipation in the fluid . an adjacent cantilever will exhibit time dependent deflections given by @xmath22 . in this case both the initial and final equilibrium deflections are @xmath54 . the motion of the second cantilever is a result of the fluid motion caused by the first cantilever as it returns to equilibrium . the spectral properties of the auto- and cross - correlations are determined by taking the cosine fourier transform of eqs . ( [ eq : auto ] ) and ( [ eq : cross ] ) with appropriate factors ( see eq . ( [ eq : spectral_density_definition ] ) ) . this yields the noise spectra , @xmath55 and @xmath56 , given by , @xmath57 where @xmath58 is the angular frequency . the noise spectra are precisely the measured quantity in experiment . in the above expression we have defined the spectral density of the random process @xmath59 to be , @xmath60 e^{i \omega t } dt \right|^2 , \label{eq : spectral_density_definition}\ ] ] where @xmath61 is the average of @xmath59 over time @xmath8 . in summary , the explicit steps necessary to determine the stochastic dynamics of a cantilever array are : 1 . perform the _ deterministic _ calculation . given some arrangement of cantilevers in a fluid choose one cantilever and apply a step force some time in the distant past . this will eventually result in a finite steady deflection of this cantilever , whereas the other cantilevers will have zero deflection . 2 . remove the step force and calculate the deflections of the cantilevers as a function of time as they return to equilibrium , for two cantilevers this is @xmath21 and @xmath22 ( see for example fig . [ fig : beam_fit ] for the case of one cantilever ) . this deterministic calculation can be done analytically or numerically depending upon the complexity of the particular situation . 3 . calculate the _ stochastic _ quantities . using @xmath21 and @xmath22 calculate the auto- and cross - correlations of the fluctuations from eqs . ( [ eq : auto ] ) and ( [ eq : cross ] ) . in essence , the cantilever deflections are merely rescaled to produce the stochastic response ( this is illustrated by the two ordinate axes in fig . [ fig : beam_fit ] ) . 4 . use the auto- and cross - correlations to calculate the noise spectra given by eqs . ( [ eq : noise1 ] ) and ( [ eq : noise2 ] ) . the equipartition theorem permits the calculation of the mean square mode displacements of micron and nanoscale cantilevers in vacuum @xcite . sader @xcite further developed this approach to include the damping forces of a surrounding viscous fluid and solved the equations of elasticity in the thin beam limit coupled together with the navier - stokes equations . he used the approximation that the effect of the fluid on each element of cantilever is the same for an element of a long and slender cantilever moving with the same speed . in the limit of an infinitely long cylinder the flow field around the cantilever can be assumed to be the flow field around an infinite two - dimensional cylinder to a good approximation @xcite . in addition , because the forces on each element of cantilever due to the fluid depend only on the velocity of that element , and not otherwise on its position in the cantilever , the mode structure of the damped cantilever is unchanged from the undamped limit . we follow this approach here and use the infinite cylinder approximation for the flow field . however we do not assume that the fluid noise is white as was done in @xcite but use the correct spectral density of the brownian force given by @xmath62 in eq . ( [ eq : fdt ] ) . the frequency dependence of the brownian noise @xmath62 is determined by the fluid damping @xmath63 ( illustrated in fig . [ fig : mass_damping ] and discussed further below ) . to connect with the previous work we note that the susceptibility @xmath64 connecting the tip displacement to an applied force is given by the cantilever response to a unit impulse of force , which is the derivative of the step force response given by eq . ( [ eq : auto ] ) so that @xmath65 the fourier transform of @xmath64 is the response to a deterministic sinusoidal force and the imaginary part @xmath66 , where im\ { } indicates the imaginary component is , @xmath67 where we have adopted the fourier transform convention given by , @xmath68 and @xmath69 . integrating eq . ( [ eq : dissipation ] ) by parts and using the definition of the noise spectrum eq . ( [ eq : noise1 ] ) yields , @xmath70 the susceptibility @xmath64 can be determined by solving for the deterministic cantilever response to a force impulse . in the following we present results only for the fundamental mode ; however higher modes can be included if desired . the equation governing the cantilever motion is then , @xmath71 where the amplitude of the mode displacement is characterized by the tip displacement @xmath72 , @xmath73 is the effective mass of the cantilever in vacuum , @xmath74 is the force acting on the cantilever due to the fluid , and @xmath75 is the force impulse ( @xmath76 is the dirac delta ) . the effective mass of the cantilever @xmath73 is chosen to yield the same kinetic energy as in the cantilever mode , and is related to the cantilever mass by , @xmath77 where @xmath78 is the actual cantilever mass and for the fundamental mode of oscillation for a beam @xmath79 . taking the fourier transform of this equation yields , @xmath80 the force from the fluid can be written in the form , @xmath81 where @xmath82 here @xmath83 is the effective mass of a fluid cylinder of radius @xmath13 where @xmath84 is the fluid density , @xmath85 is the cantilever width , and @xmath7 is the cantilever length . again the prefactor @xmath86 is to take into account the variation of the fluid force along the cantilever due to the varying velocity given by the mode structure . the fluid loading and damping are captured by the hydrodynamic function @xmath87 given by , @xmath88 where @xmath89 and @xmath90 are bessel functions @xcite . in this definition the frequency dependence on the right hand side appears through the frequency dependent strouhal number @xmath15 . the cantilever is loaded by the fluid which can be characterized by an effective mass , @xmath91 , larger than @xmath73 that takes into account the fluid mass that is also being moved . the fluid also damps the motion of the cantilever which can be expressed as an effective damping @xmath92 . relations for @xmath91 and @xmath92 can be found by expanding @xmath87 into its real and imaginary parts in eq . ( [ eq : eom_fourier_deterministic ] ) and rearranging such that , @xmath93 to give , @xmath94 and , @xmath95 where @xmath96 and @xmath97 are the real and imaginary parts of @xmath98 , respectively . @xmath99 is the mass loading parameter , which is the ratio of the mass of a cylinder of fluid with radius @xmath13 to the actual mass of the cantilever , and is given by , @xmath100 where @xmath18 is the density of the cantilever and @xmath101 is the cantilever thickness . it is evident from eqs . ( [ eq : fluid_loaded_mass ] ) and ( [ eq : damping ] ) that both the fluid loaded mass of the cantilever and the fluidic damping are functions of frequency . the ratio of the mass of the fluid loaded cantilever to the effective mass of the cantilever in vacuum , @xmath73 , as a function of frequency is given by @xmath102 . the factor @xmath99 is a constant for any particular choice of cantilever and fluid combination . the frequency dependence of the added mass is then given by @xmath103 and is illustrated by the solid line in fig . [ fig : mass_damping ] and uses the left ordinate axis . the added mass increases rapidly as the frequency is decreased . over the range of four orders of magnitude in frequency the added mass is seen to change by a factor of approximately 25 . the frequency dependence of the fluid damping is given by @xmath104 and is shown by the dashed line in fig . [ fig : mass_damping ] using the right ordinate axis . there is a weaker frequency dependence for the fluid damping when compared to the mass loading . the fluid damping decreases as the frequency decreases and over 4 orders of magnitude in frequency the damping changes by a factor of approximately 7 . illustrates the frequency dependence of the fluid loaded mass . ( dashed line ) @xmath105 illustrates the frequency dependence of the fluid damping.,width=288 ] solving for the cantilever response @xmath106 and taking the imaginary part yields , @xmath107 using this result in eq . ( [ eq : gxf_susceptibility ] ) and rearranging gives the desired result for the spectral density of the stochastic fluctuations in cantilever displacement , @xmath108 \nonumber},\end{aligned}\ ] ] where @xmath109 is a nondimensional reduced frequency , and @xmath110 is the strouhal number evaluated at the resonant frequency in vacuum given by , @xmath111 we emphasize that eq . ( [ eq : gxf_noise ] ) is _ not _ the same as in previous work @xcite which failed to include the frequency dependent of the brownian force . to understand the difference from the previous work we now connect eq . ( [ eq : fdt_displacements ] ) with the calculation in terms of a fluctuating force with spectral density @xmath62 . the fluctuating displacement can be described in terms of the response to this force through the susceptibility @xmath112 inserting eq . ( [ eq : gxf_susceptibility ] ) into this expression yields , @xmath113 it is useful at this point to introduce the impedence @xmath114 where @xmath115 is the velocity of the cantilever tip @xmath116 , it is then straight forward to show , @xmath117 inserting this into eq . ( [ eq : gf_susceptibility ] ) and rearranging yields the desired expression of the fluctuation dissipation theorem , @xmath118 where @xmath119 is the resistance or dissipation and re\ { } indicates the real part . for the case of oscillating cantilevers in fluid @xmath92 is the effective fluid damping . it is evident from the frequency dependence of the damping that the fluctuating force described by eq . ( [ eq : fdt ] ) is not white noise as considered previously @xcite ( see fig . [ fig : mass_damping ] for the variation of @xmath92 with frequency ) . the spectral density of fluctuations in cantilever displacement @xmath55 can also be determined by directly solving the governing stochastic equation of motion , @xmath120 where @xmath121 is the random force due to brownian motion . taking the fourier transform of this equation yields , @xmath122 solving for the magnitude of the cantilever response and using eq . ( [ eq : fdt ] ) for the spectral density of the brownian force again yields eq . ( [ eq : gxf_noise ] ) . note that integrating this result automatically leads to the equipartition result , @xmath123 we would like to point out that it is not possible to start with the equipartition result given by eq . ( [ eq : equipartition ] ) and to then calculate the spectral properties of the fluctuations , one must first start with the fluctuation - dissipation result given by eq . ( [ eq : fdt ] ) as done here . the noise spectrum of the cantilever in fluid may be used to calibrate an afm . a simple way to do this is to extract the effective spring constant from the the frequency @xmath124 giving the maximum of the noise intensity . in previous work @xcite the fluctuations of the cantilever have been estimated without the frequency dependence of the numerator in eq . ( [ eq : gxf_noise ] ) , leading to some inaccuracy in the estimate of the spring constant . once @xmath124 is known , the quality factor @xmath125 can be estimated from , @xmath126 where eqs . ( [ eq : fluid_loaded_mass ] ) and ( [ eq : damping ] ) have been used for the mass and damping . this expression for @xmath125 is only approximate again because the frequency response of the cantilever is not precisely like that of simple harmonic oscillator , since both @xmath91 and @xmath92 are frequency dependent . the error incurred by neglecting the frequency dependence in the numerator in eq . ( [ eq : gxf_noise ] ) in fitting the spectrum of a brownian driven cantilever is illustrated in fig . [ fig : mass_loading_noise_compare ] as a function of @xmath110 and @xmath99 . in fig . [ fig : mass_loading_noise_compare ] @xmath127 is the frequency at the maximum of the noise spectrum when the frequency dependence has not been included ( specifically , in eq . ( [ eq : gxf_driven ] ) @xmath62 has been assumed constant ) and @xmath124 is the correct value of the frequency using the fluctuation - dissipation theorem as discussed in section [ section : analytical_theory ] . the results indicate that as the strouhal number increases the error in this assumption becomes smaller . in particular , when the frequency dependence is neglected the approximate theory always underpredicts @xmath124 . this can be understood in light of fig . [ fig : mass_damping ] where it is clear that the magnitude of the damping decreases as the frequency of oscillation is reduced . when this decrease is not included the predicted value of @xmath124 will be unnecessarily reduced . although the fluid induced damping slowly decreases as the oscillation frequency is reduced it is important to note that the fluid loaded mass increases rapidly and it is this interplay which results in the small @xmath125 associated with oscillating micron and nanoscale cantilevers . . @xmath127 is the predicted frequency at the maximum of the noise spectrum when the frequency dependence of the brownian force is neglected as done in previous work @xcite and @xmath124 is the correct value given by the results of section [ section : analytical_theory ] . results are presented as a function of mass loading @xmath99 and the strouhal number @xmath110 based upon the cantilever s resonant frequency in vacuum . the curves are for @xmath128 where @xmath129 are labelled and the remaining curves are in sequential order.,width=312 ] figure [ fig : mass_loading_noise ] summarizes the theoretical results and allows for easy determination of the stochastic dynamics of a single cantilever of arbitrary geometry placed in an arbitrary viscous fluid . given a particular cantilever and fluid combination the procedure is : 1 . determine the cantilever spring constant @xmath130 , resonant frequency in vacuum , @xmath131 , and characteristic length scale . these can be experimental measurements or theoretical calculations . the characteristic length is a measure of the characteristic half - width of the cantilever . 2 . determine @xmath110 from eq . ( [ eq : r0 ] ) and @xmath99 from eq . ( [ eq : t0 ] ) . 3 . use @xmath110 and @xmath99 to determine @xmath132 from fig . [ fig : mass_loading_noise](a ) and @xmath125 from fig . [ fig : mass_loading_noise](b ) . of oscillation for a cantilever undergoing stochastic oscillations in a viscous fluid as a function of mass loading @xmath99 and the strouhal number @xmath110 based upon the cantilever s resonant frequency in vacuum . panel ( b ) , the quality factor @xmath125 as a function of @xmath110 and @xmath99 as given by eq . ( [ eq : qapprox ] ) . in each panel the largest value of @xmath110 is labelled and each successive curve represents @xmath133 , i.e. in panel ( a ) the last curve is for @xmath134 and in panel ( b ) the last curve is for @xmath135.,title="fig:",width=312 ] of oscillation for a cantilever undergoing stochastic oscillations in a viscous fluid as a function of mass loading @xmath99 and the strouhal number @xmath110 based upon the cantilever s resonant frequency in vacuum . panel ( b ) , the quality factor @xmath125 as a function of @xmath110 and @xmath99 as given by eq . ( [ eq : qapprox ] ) . in each panel the largest value of @xmath110 is labelled and each successive curve represents @xmath133 , i.e. in panel ( a ) the last curve is for @xmath134 and in panel ( b ) the last curve is for @xmath135.,title="fig:",width=312 ] in the case that the cantilever is driven externally the response can be found in a similar manner . if the cantilever is driven by a force @xmath136 the equation of motion becomes , @xmath137 if the driving force is @xmath138 where @xmath139 is a constant and @xmath140 is the driving frequency the amplitude of the response is given by , @xmath141 } \nonumber,\end{aligned}\ ] ] where @xmath142 is the reduced driving frequency . we emphasize that eqs . ( [ eq : eom_driven ] ) and ( [ eq : gxf_driven ] ) are for externally driven cantilevers and neglect brownian noise . ( [ eq : gxf_driven ] ) and ( [ eq : gxf_noise ] ) are similar in that they share a common denominator , whereas the additional frequency dependence in the numerator of eq . ( [ eq : gxf_noise ] ) is from the frequency dependence of the brownian noise . the stochastic dynamics of a micron scale cantilever , similar to what is commonly used in atomic force microscopy , are now quantified ( see fig . [ fig : beam ] ) . the beam is chosen from chon and sader @xcite where both theoretical and experimental values are present for comparison . the beam properties are summarized in table [ table : beam_geometry ] , , width @xmath85 , and height @xmath101.,width=240 ] [ h ] [ c]l@l@l@ l@l@l@l @xmath7 & @xmath85 & @xmath101 & @xmath143 & @xmath110 & @xmath99 + + 197@xmath144 m & 29@xmath144 m & 2@xmath144 m & 71.56khz & 109.7 & 4.89 + using classical beam theory @xcite the equivalent spring constant @xmath130 is predicted to be , @xmath145 which yields a value of @xmath146n / m and the resonant frequency of the cantilever in vacuum is @xmath147 which yields @xmath148khz where @xmath149 . using eq . ( [ eq : r0 ] ) the strouhal number in vacuum is @xmath150 and using eq . ( [ eq : t0 ] ) the mass loading factor is @xmath151 . [ h ] l@l@l@l@l@l@l & @xmath125 & @xmath152 & @xmath153 & @xmath92 ( kg / s ) & @xmath154 + + ( 1 ) & 3.24 & 8.16 & 0.34 & @xmath155 & 37.3 + ( 2 ) & 2.93 & 8.05 & 0.35 & @xmath156 & 38.7 + we now use the thermodynamic approach to calculate the stochastic dynamics of the cantilever . as previously stated this can be accomplished in a straightforward manner by determining the deterministic response of the cantilever to the removal of a step force applied to the tip . we do this using finite element numerical simulations of the full three dimensional , time - dependent , fluid - solid interaction problem ( algorithm discussed elsewhere ) . the simulation is initiated with the removal of a step force applied to the tip of the cantilever and the deterministic dynamics of the beam @xmath21 are shown by the solid line in fig . [ fig : beam_fit ] using the right ordinate axis . in order to calculate commonly used diagnostics , such as @xmath124 and @xmath125 , the deterministic cantilever deflection @xmath21 was fit to the deflections of an underdamped simple harmonic oscillator ( valid for @xmath157 ) given by the equation of motion , @xmath158 where @xmath159 and @xmath160 . the solution is , @xmath161 where , @xmath162 and @xmath163 . using a nonlinear least squares curve fit algorithm @xcite the numerical results are fit to the model to yield the values of @xmath125 , @xmath124 , @xmath92 , and @xmath91 shown in table [ table : beam_fluid_sader ] . the curve fit is nearly indistinguishable from the numerical simulation results for @xmath21 and is shown by the dashed line in fig . [ fig : beam_fit ] . using @xmath21 in eq . ( [ eq : auto ] ) yields the autocorrelation of the equilibrium deflections , @xmath164 , which are shown in fig . [ fig : beam_fit ] using the left ordinate axis . the noise spectrum is given by eq . ( [ eq : noise1 ] ) and is shown in comparison with predictions of the infinite - beam approximation given by eq . ( [ eq : gxf_noise ] ) in fig . [ fig : beam_noise ] . the numerical simulations contain all of the oscillation modes as indicated by the presence of the second mode in the figure . in the infinite beam approximation the different modes behave independently and the calculation could be extended to include higher modes if desired . in table [ table : beam_fluid_sader ] the analytical results of section [ section : analytical_theory ] are compared with results using the full thermodynamic approach and deterministic finite element numerical simulations . in general , from the table and the figures , we see that the differences between the simple model and the full calculations are small for this shape cantilever . the experimentally measured value of the resonant frequency in water is @xmath165 @xcite which agrees well with both the analytical results and the numerical predictions using the thermodynamic approach . however , table [ table : beam_fluid_sader ] suggests that the analytical results based upon the oscillating infinite cylinder model under predict the amount of damping @xmath92 . the noise spectra show differences that are small , but may be significant for quantitative applications such as calibration . in all of the deterministic finite element numerical simulations performed we were careful to ensure that the bounding no - slip surfaces of the computational domain did not affect the results . comparisons with numerical simulations performed with larger domains did not significantly alter the results . for a micron scale cantilever in water . ( solid line ) deterministic finite element numerical simulations using the thermodynamic approach . ( dashed line ) approximate analytical theory for the fundamental mode only.,width=288 ] as the cantilever dimensions become smaller the effective cantilever spring constant decreases while the resonant frequency increases . this favorable combination potentially provides access to the biologically important parameter regime characterized by 10 s of piconewtons with microsecond scale time resolution , a range that is difficult to reach using other methods . this has led to the development of nanoscale cantilevers @xcite . in what follows we quantify the stochastic dynamics of two adjacent nanoscale cantilevers immersed in water . a schematic of the nanoscale cantilever under consideration here is shown in fig . [ fig : paddle_config ] @xcite . this is an experimentally motivated cantilever whose paddle shaped geometry decreases the effective spring constant while localizing the strain for piezoresistive measurement . the configuration of the cantilever array is shown in fig . [ fig : adjacent ] with two opposing cantilevers separated by a distance @xmath166 . this configuration was chosen because of its experimental accessibility as well as its potential use in single molecule measurements with the tethering of a target biomolecule between the two cantilevers . in the work presented here we build a baseline understanding of the cantilever dynamics in the absence of target biomolecules . further consideration of the dynamics caused by a tethered biomolecule is beyond the scope of the present paper . the dimensions of the nanoscale cantilevers investigated here are summarized in table [ table : paddle - geometry ] . as discussed previously , the physical properties describing the dynamics of the cantilever in vacuum are important parameters in determining the stochastic dynamics of the cantilevers . these properties are summarized in table [ table : paddle_vacuum ] where the values have been determined from finite element simulations of the elastic cantilever structure in the absence of the surrounding fluid . it is important to emphasize that although these calculations can sometimes be performed analytically , this is not the case for many geometries of interest . however , even for cantilevers with very complex geometries , the vacuum based finite element calculations are straight forward . [ h ] [ c]l@l@l@l@l@l @xmath7 & @xmath85 & @xmath101 & @xmath167 & @xmath168 + + 3@xmath144 m & 100 nm & 30 nm & @xmath169 m & 33 nm + [ h ] [ c]l@l@l@l@l@l@l @xmath130 & @xmath131 & @xmath110 & @xmath99 + + 8.7 mn / m & @xmath170 rads / s & 0.11 & 1.12 + , width @xmath85 , and height @xmath101.,width=264 ] the stochastic dynamics of the cantilever array are determined using the thermodynamic approach discussed in section [ section : thermodynamic_approach ] with deterministic finite element numerical simulations . a series of numerical simulations have been performed to determine the variation in the cantilever dynamics as a function of the separation distance between the two cantilevers in the array . in particular we have explored @xmath171 where @xmath166 is the cantilever separation and @xmath172 nm is the cantilever thickness . the numerical simulations are initiated by the removal of a step force applied to the tip of one of the cantilevers ( in fig . [ fig : adjacent ] this is the cantilever on the left ) while the other cantilever is initially undeflected and at mechanical equilibrium . the deterministic dynamics of the nanoscale cantilever for which the step force has been removed , @xmath21 , is shown by the solid line in fig . [ fig : one_paddle_auto1 ] using the right ordinate axis . the deflection @xmath45(t ) was not significantly affected by the presence of the second cantilever for the separations considered here . as with the micron scale cantilever , we again fit the result for @xmath21 from the finite element numerical simulation to the solution of the simple harmonic oscillator equation given by eq . ( [ eq : sho_fit ] ) . for the nanoscale cantilever the dynamics are overdamped ( @xmath173 and the deterministic deflection predicted by the simple harmonic oscillator approximation is , @xmath174 where , @xmath175 the agreement with the curve fit is very good and @xmath124 and @xmath125 are shown in table [ table : paddle_fluid ] . also shown in table [ table : paddle_fluid ] are the theoretical predictions based upon the infinite cylinder approximation . for the analytical calculations of the damping @xmath92 from eq . ( [ eq : damping ] ) the fundamental mode of the nanoscale cantilever has been modelled as that of a hinge where all of the bending occurs in the short legs near the base . this is guided by numerical calculations showing the strain localized in the short legs . in this case we find the mass factor is @xmath176 . ( note that @xmath177 is not needed to determine the other quantities in table [ table : paddle_fluid ] ) . it is clear from table [ table : paddle_fluid ] that the infinite cylinder approximation is no longer valid . the analytical prediction for the added mass is an order of magnitude too large . this suggests that three - dimensional flow effects become significant for the shape of cantilever under consideration here . for example there would be flow around the tip of the cantilever , as well as the flow through the open region near the base of the cantilever . in addition , the predicted frequency shift is an order of magnitude too small . [ h ] l@l@l@l@l@l@l & @xmath125 & @xmath152 & @xmath153 & @xmath92 ( kg / s ) & @xmath154 + + ( 1 ) & 0.265 & 106.3 & 0.032 & @xmath178 & 0.0036 + ( 2 ) & 0.327 & 18.16 & 0.236 & @xmath179 & 0.026 + ( left ordinate axis ) for a single nanoscale cantilever in fluid : ( solid line ) results from finite element numerical simulations using the thermodynamic approach ; ( dashed line ) results based upon the approximate analytical theory of section [ section : analytical_theory].,width=288 ] inserting @xmath21 into eq . ( [ eq : auto ] ) yields the autocorrelation of the equilibrium fluctuations in cantilever displacement , @xmath164 . these are also shown by the solid line in fig . [ fig : one_paddle_auto1 ] using the ordinate axis on the left . the noise spectrum , @xmath55 , is found by inserting the autocorrelation in eq . ( [ eq : noise1 ] ) and the result is shown by the solid line in fig . [ fig : one_paddle_noise_comparison ] . fig . [ fig : one_paddle_noise_comparison ] also presents a comparison of the noise spectrum @xmath55 given from the theoretical predictions of eq . ( [ eq : gxf_noise ] ) and the results from the numerical simulations using the thermodynamic approach . the noise spectra are different for small frequencies , in particular the model result has a peak at finite frequency not seen in the full calculation . the inverse cosine transform of the noise spectrum from the infinite - cylinder model yields the deflection of the cantilever as a function of time shown by the dashed line in fig . [ fig : one_paddle_auto1 ] using the right ordinate axis . the model prediction for the cantilever deflection exhibits negative values that are not seen in the simulations suggesting that the response of the overdamped cantilever is not precisely that of a simple harmonic oscillator . for a single nanoscale cantilever immersed in water . ( solid line ) results from deterministic finite element simulations using the thermodynamic approach described in section [ section : thermodynamic_approach ] . ( dashed line ) results from approximate analytical theory described in section [ section : analytical_theory].,width=288 ] we now turn to the deterministic deflection of the second cantilever , @xmath22 , after removing the force on the first cantilever , evaluated using the full fluid - elasticity simulations . this is shown in fig . [ fig : two_paddles_cross_all ] using the right ordinate axis . for close separations @xmath180 the second cantilever exhibits negative deflections for all time . however , as the cantilever separation increases the second cantilever initially exhibits a positive deflection followed by a negative deflection . the flow field around an oscillating object will have both potential and non - potential components . for an incompressible fluid the potential component is instantaneous whereas the non - potential component diffuses with a diffusion coefficient given by the kinematic viscosity @xmath181 . although the deterministic flow field around simple oscillating objects is well known the fluid coupled motion of multiple elastic objects is not well understood . the fluid dynamics resulting from the motion of two adjacent cantilevers will be discussed in a forthcoming article @xcite . using eq . ( [ eq : cross ] ) the stochastic correlations between the two cantilevers can be determined from @xmath22 . the cross - correlations of the equilibrium displacement fluctuations @xmath182 are shown in fig . [ fig : two_paddles_cross_all ] using the ordinate axis on the left . for small cantilever separations @xmath180 the cross - correlations are anticorrelated for all time . this is in agreement with experimental measurements of the fluid correlations of two closely spaced micron scale beads in water @xcite . however , as the cantilever separation is increased the cross - correlations changes : for short times the cross - correlations are positively correlated whereas for larger times they are anti - correlated . the noise spectra @xmath56 are found from @xmath182 by using eq . ( [ eq : noise2 ] ) . figure [ fig : two_paddles_noise_all ] shows the variation in the noise spectra as a function of cantilever separation . the noise spectra contain both positive and negative values and for each cantilever separation there is a frequency at which correlated noise vanishes , @xmath183 . this null point could be exploited by a measurement scheme to minimize the correlated noise . for @xmath172 nm found using deterministic finite element numerical simulations with the thermodynamic approach discussed in section [ section : thermodynamic_approach ] . the curve for @xmath184 is labelled and the others follow in sequential order . the right ordinate axis illustrates the deterministic cantilever deflection as a function of time @xmath22 scaled by the deflection of the adjacent cantilever , @xmath185).,width=288 ] for two nanoscale cantilevers separated by distance @xmath186 for @xmath172 nm found using deterministic finite element numerical simulations with the thermodynamic approach discussed in section [ section : thermodynamic_approach ] . the curve for @xmath184 is labelled and the others follow in sequential order.,width=288 ] . panel ( a ) , the maximum magnitude of the force induced by fluid - coupled correlations in cantilever displacement @xmath187 . the data is empirically fit with an exponential given by @xmath188 ( in units of pn ) . the magnitude of the force felt by a single cantilever due to brownian noise , @xmath189 pn , is nearly 6 times larger than the correlated noise at close separations ( @xmath180 ) . the right ordinate axis scales @xmath187 by @xmath190 . panel ( b ) , characteristic time scales , the circles represent the time at which the maximum value of the cross correlation occurs , the squares represent the time at which the fluid induced correlations vanish . the ordinate axis on the right hand side illustrates the time scales when scaled by the period of oscillation of the cantilever in vacuum @xmath191 . the solid lines represent a linear curve fit to the data.,title="fig:",width=264 ] . panel ( a ) , the maximum magnitude of the force induced by fluid - coupled correlations in cantilever displacement @xmath187 . the data is empirically fit with an exponential given by @xmath188 ( in units of pn ) . the magnitude of the force felt by a single cantilever due to brownian noise , @xmath189 pn , is nearly 6 times larger than the correlated noise at close separations ( @xmath180 ) . the right ordinate axis scales @xmath187 by @xmath190 . panel ( b ) , characteristic time scales , the circles represent the time at which the maximum value of the cross correlation occurs , the squares represent the time at which the fluid induced correlations vanish . the ordinate axis on the right hand side illustrates the time scales when scaled by the period of oscillation of the cantilever in vacuum @xmath191 . the solid lines represent a linear curve fit to the data.,title="fig:",width=264 ] from these results it is possible to characterize the force sensitivity and time resolution of a correlation measurement technique using an array of closely spaced nanoscale cantilevers . an estimate of the force sensitivity can be found using the auto and cross correlation functions as , @xmath192 @xmath190 represents the approximate magnitude of the stochastic brownian force acting on a single cantilever and @xmath187 the force induced by fluid correlations between two cantilevers . for the case of a single nanoscale cantilever the maximum value occurs at @xmath29 and , using the equipartition theorem this yields a value of @xmath193pn . in a cross - correlation measurement between two cantilevers the brownian noise felt by the two individual cantilevers is uncorrelated and does not contribute . this leaves only the correlations due to the viscous fluidic coupling and @xmath194 represents the approximate magnitude of the force due to this hydrodynamic coupling . for example , from fig . [ fig : two_paddles_cross_all ] for the case with the closest separation @xmath184 , the magnitude of the maximum value of the cross - correlation is @xmath195nm@xmath196 which yields a force sensitivity @xmath197pn . therefore @xmath198 at @xmath184 , indicating a 6-fold reduction in thermal noise over a single cantilever measurement . the variation in @xmath187 as a function of cantilever separation is shown in fig . [ fig : force_and_time](a ) . the data is fit with an exponential curve , given by @xmath199 ( in units of pn ) , shown in fig . [ fig : force_and_time](a ) as the solid line . for very small cantilever separations @xmath180 the noise reduction possible through the use of a correlated measurement technique is approximately 6-fold . as expected , the correlated noise decreases as the cantilever separation is increased : for a separation of @xmath200 there is a 12-fold noise reduction . the ordinate axis on the right hand side of fig . [ fig : force_and_time](a ) shows @xmath201 to show this performance improvement directly . the characteristic time scales of the fluid correlated motion as a function of cantilever separation are illustrated in fig . [ fig : force_and_time](b ) . in fig . [ fig : force_and_time](b ) circles represent the time at which the maximum value in the cross correlation occur and squares represent the time at which the fluid induced correlations vanish ( i.e. , the zero - crossing in fig . [ fig : two_paddles_cross_all ] ) . as expected from the finite rate at which momentum diffuses away from an oscillating cantilever , both of these time scales increase with cantilever separation . using the ordinate axis on the right in fig . [ fig : force_and_time](b ) the ratio of these time scales to the period of a single oscillation for the cantilever in vacuum @xmath202 are given . the stochastic dynamics of micron and nanoscale cantilevers can be quantified for the precise conditions of experiment using straightforward deterministic calculations coupled with the fluctuation - dissipation theorem . this can be done for complex cantilever geometries including wall effects , and the stochastic correlated behavior of closely spaced cantilevers for which theoretical predictions currently are not available . cantilever geometries that are long and thin are well described by approximate analytical theory based upon the oscillation of an infinite cylinder in fluid . we have used the fluctuation - dissipation method to correct the previous results of this theory . for the short and wide nanocantilevers currently being proposed the infinite - cylinder approximation is no longer appropriate . the thermodynamic approach using the fluctuation - dissipation theorem provides an important means of developing a physical understanding of the stochastic dynamics of closely spaced micron and nanoscale objects that will be important as micron and nanotechnology progress and a quantitative design tool for experiment . this research has been partially supported by darpa / mto simbiosys under grant f49620 - 02 - 1 - 0085 and an aspires grant from virginia tech . this work has been carried out in collaboration with the caltech bionems effort ( m. l. roukes , pi ) and we gratefully acknowledge extensive interactions with this team .
two observations of with the rosat pspc resulted in the detection of x - ray emission from 25 members of this old open cluster ( belloni et al . 1993 , 1998 ) . the x - ray emission of many of these sources is readily understood . for example , the x - ray emission originates in deep , hot atmospheric layers in a hot white dwarf ; is due to mass transfer in a cataclysmic variable ; and is caused by magnetic activity in two contact binaries and several rs cvn - type binaries . however , belloni et al . ( 1998 ) point out several x - ray sources in for which the x - ray emission is unexplained . all but one of these objects are located away from the isochrone formed by the main sequence and the ( sub)giant branch of ( fig . [ cmd ] ) . in this paper we investigate the nature of the x - ray emission of these stars through low- and high - resolution optical spectra . in particular , we investigate whether the emission could be coronal as a consequence of magnetic activity , by looking for emission cores in the h&k lines . tidal interaction in a close binary orbit is thought to enhance magnetic activity at the stellar surface by spinning up the stars in the binary . therefore , we also derive projected rotational velocities with the crosscorrelation method . finally , we study the h@xmath0 profile as a possible indicator of activity or mass transfer . the observations and the data reduction are described in sect . 2 , and the analysis of the spectra in sect . comparison with chromospherically active binaries is made in sect . a discussion of our results is given in sect . 5 . in the remainder of the introduction we give brief sketches of the stars studied in this paper ; details on many of them are given by mathieu et al . the stars are indicated with their number in sanders ( 1977 ) , and are listed in table [ tab1 ] . l@l@l@l@l@l@l i d & @xmath1 & @xmath2-@xmath1 & sp.type & @xmath3 & @xmath4 & ctrate + & & & & ( d ) & & ( s@xmath5 ) + + + s1040 & 11.52 & 0.87 & g4iii & 42.83 & 0.027(28 ) & 0.0050(6 ) + s1063 & 13.79 & 1.05 & g8iv & 18.39 & 0.217(14 ) & 0.0047(6 ) + s1072 & 11.32 & 0.61 & g3iii iv & 1495 . & 0.32(7 ) & 0.0013(3 ) + s1082 & 11.25 & 0.42 & f5iv & & & 0.0046(6 ) + s1113 & 13.77 & 1.01 & & 2.823 & 0.031(14 ) & 0.0047(6 ) + s1237 & 10.78 & 0.94 & g8iii & 697.8 & 0.105(15 ) & 0.0010(3 ) + s1242 & 12.72 & 0.68 & & 31.78 & 0.664(18 ) & 0.0007(2 ) + s1063 and s1113 are two binaries located below the subgiant branch in the colour - magnitude diagram of . their orbital periods , 18.4 and 2.82 days respectively , are too long for them to be contact binaries ; also they are too far above the main sequence to be binaries of main - sequence stars . in principle , a ( sub)giant can become underluminous when it transfers mass to its companion , as energy is taken from the stellar luminosity to restore hydrostatic equilibrium ( e.g. kippenhahn & weigert 1967 ) . however , mass transfer through roche lobe overflow very rapidly leads to circularization of the binary orbit , whereas s1063 has an eccentricity @xmath6 . the orbit of s1113 is circular , so mass transfer could be occurring in that system . for the moment , the nature of these binaries is not understood . in both , pasquini & belloni ( 1998 ) observed emission cores in the h&k lines . s1063 is reported to be photometrically variable with @xmath7 0.10 mag ( rajamohan et al . 1988 ; kaluzny & radczynska 1991 ) , but no period is found . for s1113 , photometric variability with a period of 0.313 days and a total amplitude of 0.6 mag was claimed by kurochkin ( 1960 ) , but this has not been confirmed by kaluzny & radczynska ( 1991 ) , who find variability with only 0.05 mag . s1063 is the only star in our sample that shows significantly variable x - ray emission ( between 0.0081 and 0.0047 cts s@xmath5 ; belloni et al . 1998 ) . s1072 and s1237 are binaries with orbital periods of 1495 and 698 days , and with eccentricities @xmath8 and 0.105 , respectively . the colour and magnitude of s1072 can not be explained with the pairing of a giant and a blue straggler , since this is not compatible with its @xmath9 photometry ( nissen et al . 1987 ; mathieu & latham 1986 ) , nor with superposition of three subgiants , since this is excluded by the radial velocity correlations ( mathieu et al . the absence of the 6708 lithium feature in the spectrum of s1072 indicates that the surface material has undergone mixing ( hobbs & mathieu 1991 ; pritchett & glaspey 1991 ) . s1237 could be a binary of a giant and a star at the top of the evolved main sequence ( janes & smith 1984 ) ; high - resolution spectroscopy should be able to detect the main - sequence star in that case ( mathieu et al . 1990 ) . the wide orbits and significant eccentricities appear to exclude both mass transfer and tidal interaction as explanations for the x - ray emission . s1242 has the largest eccentricity of the binaries in our sample , at @xmath10 in an orbit of 31.8@xmath11days . its position on the subgiant branch is explained if a subgiant of 1.25@xmath12 has a secondary with @xmath13 ( mathieu et al . k line emission is reported by pasquini & belloni ( 1998 ) . photometric variability with a period of 4.88@xmath11days and amplitude of 0.0025 mag has been found by gilliland et al . we note that this photometric period corresponds to corotation with the orbit at periastron , which suggests that the x - ray emission may be due to tidal interaction taking place at periastron . the binary would then be an interesting example of a system in transition from an eccentric to a circular orbit . indeed , according to the diagnostic diagram of verbunt & phinney ( 1995 ) a giant of 1.25@xmath12 with a current radius of @xmath14 ( as derived from the location of s1242 in the colour - magnitude diagram ) can not have circularized an orbit of 31.8days . s1040 is a binary consisting of a giant and a white dwarf . the progenitor of the white dwarf circularized the orbit during a phase of mass transfer ( verbunt & phinney 1995 ) ; as a result the mass of the white dwarf is very low ( landsman et al . the white dwarf is probably too cool , at 16160@xmath11k , to be the x - ray emitter . indications for magnetic activity are h&k ( pasquini & belloni 1998 ) and ( @xmath15 2800 , landsman et al . 1997 ) emission lines . if the x - rays are due to coronal emission of the giant , this must be the consequence of the past evolution of the binary , since the giant is too small for significant tidal interaction to be taking place in the current orbit . s1082 is a blue straggler . photometric variability of 0.08 mag within a few hours was observed by simoda ( 1991 ) . goranskii et al . ( 1992 ) found eclipses with a total amplitude of 0.12 mag and a binary period of 1.07 days ; however , the radial velocities of the star do not show this period , and vary by about 2 km s@xmath5 , far too little for a 1day eclipsing binary ( mathieu et al . landsman et al . ( 1998 ) detect a significant excess at 1520 with the ultraviolet imaging telescope , and ascribe this to a hot , subluminous secondary . such a secondary was suggested already by mathys ( 1991 ) on the basis of a broad component in the d and absorption lines . optical spectra were obtained on february 28/29 , 1996 with the 4.2 m william herschel telescope on la palma , under good weather conditions ( seeing @xmath16 until 4@xmath1730 ut , @xmath18 thereafter ) . in addition to the x - ray sources in we observed two ordinary member giants of , s1288 and s1402 , for comparison . furthermore one flux standard and three velocity standards were observed . the blue high - resolution spectra of s1113 were obtained on april 7/8 , 1998 with the same telescope through a service observation ( seeing 12@xmath19 ) . a log of the observations is given in table [ tab11 ] . llrlrlr i d & & & + & & & + & ut & & ut & & ut & + & & & & & & + + + + s1040 & 02:43&60 & 22:03&600 & 00:46&300 + s1063 & 02:30&180 & 20:51&1200 & 00:11&1200 + & & & 21:15&1200 & 00:33&600 + & & & 21:37&1200 + s1072 & 02:49&60 & 22:24&360 & 01:01&240 + s1082 & 02:39&60 & 20:39&300 & 00:01&360 + & & & 23:24&600 & 01:41&360 + & & & & & 04:11&360 + s1113 & 03:01&180 & 22:37@xmath20 & 900 & 03:36&1200 + & & & 22:58@xmath20 & 900 & 03:58&600 + s1237 & 02:46&60 & 22:16&300 & 00:54&180 + s1242 & 02:55&120 & 22:33&1500 & 01:08&900 + + + s1288 & 02:58&60 & 23:12&600 & 01:33&300 + s1402 & 02:52&60 & 23:01&450 & 01:26&240 + + + & 02:21&80 & 23:42&600 & 01:57&600 + & 03:13&80 + & 04:43&480 + + + & & & 06:04 & 90 & 05:13&45 + & & & 05:55&100 & 05:42&50 + & & & & & 05:34&45 + all spectra have been reduced using the image reduction and analysis facility ( iraf ) . low - resolution spectra were taken with the isis double - beam spectrograph ( carter et al . the blue arm of isis was used with the 300 lines per mm grating and tek - ccd , resulting in a wavelength coverage of 3831 to 5404 and a dispersion of 1.54 per pixel at 4000 . the red arm , combined with the 316 lines per mm grating and eev - ccd , covered a wavelength region of 5619 to 7135 with a dispersion of 1.40 per pixel at 6500 . the format of the frames is 1124 @xmath21 200 pixels which includes the under- and overscan regions . for the object exposures the slit width was set to @xmath22 . flatfields were made with a tungsten lamp while cuar and cune lamp exposures were taken for the purpose of wavelength calibration . for the isis - spectra , basic reduction steps have been done within the iraf ccdred - package . these steps include removing the bias signal making use of the under- and overscan regions and zero frames , trimming the frames to remove the under- and overscan , and flatfielding to correct for small pixel - to - pixel gain variations . the remaining reduction has been done with iraf specred - package tasks . with the optimal extraction algorithm ( horne 1986 ) the two dimensional images are reduced to one dimensional spectra . next , the spectra are calibrated in wavelength with the arc frames . a dispersion solution is found by fitting third ( blue ) and fourth ( red ) order polynomials to the positions on the ccd of the arclamp lines . the fluxes of the spectra are calibrated with the absolute fluxes of hz44 , tabulated at 50 intervals ( massey et al . 1988 ) , and adopting the standard atmospheric extinction curve for la palma as given by king ( 1985 ) . the estimated accuracy of the flux calibration is @xmath23% . high - resolution echelle spectra were taken with the utrecht echelle spectrograph ( ues , unger et al . observations were done with a @xmath24 slitwidth . for the 1996 observations , the ues was used in combination with a 1024 @xmath21 1024 pixels tek - ccd , and the 31.6 lines per mm grating ( e31 ) , which resulted in a broad wavelength coverage , but small separation of the echelle orders on the ccd . in this setup , the ues resolving power is 49000 per resolution element ( two pixels ) , corresponding to a dispersion of 3 km s@xmath5 per pixel or 0.06 per pixel at 6000 . the frames were centered on @xmath25 and @xmath26 in order to get a blue ( 3820 to 4920 ) and red ( 4890 to 7940 ) echelle spectrum . the number of orders recorded on the ccd is 34 in the blue and 45 in the red , each covering @xmath7 45 to 80 increasing for longer wavelengths . towards the red , gaps occur between the wavelength coverage of adjacent orders . exposures of a quartz lamp were taken to make the flatfield corrections . thar exposures served as wavelength calibration frames . for the 1998 observations of s1113 , a 2048 @xmath21 2048 pixels site - ccd was used . two spectra were taken with the 79.0 lines per mm ( e79 ) grating ( @xmath27 ) . the difference between the e79 and the e31 gratings is that e79-spectra have a larger separation of the echelle - orders on the detector , which can improve the determination of the sky - background . the spectral resolution of the gratings is the same ( the central dispersion in these observations is @xmath28 per pixel ) . the specified wavelength - coverage for this combination of grating and detector is 3546 to 6163 but only the central orders were bright enough to extract spectra ( 3724 to 5998 ) . flatfield and thar exposures were made for calibration purposes . the reduction of the ues spectra has been performed using the routines available within the iraf echelle - package . first , the frames are debiased and the under- and overscan regions removed . after locating the orders on the ccd for both the quartz lamp and the object exposures , we flatfielded the frames . spectra are extracted with optimal extraction . the small order separation makes sky subtraction difficult ; however , our targets are bright , and the resulting error is negligible . in the step of wavelength calibration , the dispersion solution is derived by fitting third and fourth order polynomials leaving rms - residuals of 0.004 ( red ) and 0.002 ( blue , 0.003 for the 1998-spectra ) . to find absolute fluxes for the k ( @xmath29 ) & h ( @xmath30 ) emission lines ( sect . 3.1 ) , the fluxes of the relevant blue orders of an object have been calibrated with the calibrated isis spectrum of the same object . continuum normalization of the orders in the red spectra , required for the rotational velocity analysis , is done by fitting third to fifth order polynomials to the wavelength - calibrated spectra . we study two indicators of magnetic activity . the direct indicator is emission in the cores of the h&k lines . another indicator is the rotational speed : rapid ( differential ) rotation and convective motions are thought to generate magnetic fields through a dynamo . to estimate the amount of flux emitted in the h&k line cores , @xmath31 , we add the fluxes above the h&k absorption profiles as follows . an upper and a lower limit of the level of the absorption pseudo - continuum is estimated by eye and is marked by a straight line . for s1113 this is illustrated in fig . [ subca ] . we obtain a lower and upper limit of the emitted flux by adding the fluxes in each wavelength - bin above these levels . the value given in table [ fluxes ] is the average of these two results , the uncertainty is half their difference . use of higher order fits ( following fernndez - figueroa et al . 1994 ) to the absorption profile gives similar results . if an emission line is not clearly visible , we obtain an upper limit by estimating the minimal detectable emission flux at the h&k line centers within a 1 wide region ( typical width of the emission lines ) . six of our sources show h&k line emission ( fig . [ cahk ] ) . the profiles of s1113 appear to be double - lined , suggesting that we see activity of both stars ( fig . [ subca ] ) . the fluxes given in table [ fluxes ] are the total fluxes , i.e. no attempt was made to deblend the emission lines . no emission is visible in the spectrum of s1082 . lrrl i d & & & + & & & + + + s1040 & @xmath32 & @xmath33 & 3.0(1.0 ) + s1063 & @xmath34 & @xmath35 & 3.9(0.8 ) + s1072 & @xmath36 & @xmath37 & 8.1(1.1 ) + s1082 & @xmath38 & @xmath39 & 5.1(0.7 ) + s1113 & @xmath40 & @xmath41 & 45(6 ) + & & & 12(1 ) + s1237 & @xmath42 & @xmath43 & @xmath44 + s1242 & @xmath45 & @xmath46 & @xmath47 + in order to derive the projected rotational velocity @xmath48 of our targets , we apply the crosscorrelation technique ( e.g. tonry & davis 1979 ) . this method computes the correlation between the object spectrum and an appropriately chosen template spectrum as function of relative shift . the position of the maximum of the crosscorrelation function ( ccf ) provides the value of the radial - velocity difference between object and template . the width of the peak is indicative for the width of the spectral lines and can therefore be used as a measure of the rotational velocity of the stars . for rotational velocities not too large the line profiles may be approximated with gaussians , allowing analytical treatment of the crosscorrelation method . assuming that the binary spectrum is a shifted , scaled and broadened version of the template spectrum , the broadening can be related to the width of the ccf peak as follows ( gunn et al . with @xmath49 the dispersion in the template s and @xmath50 the dispersion in the target s spectral lines , @xmath51 the dispersion of the ccf peak and @xmath52 the dispersion of the gaussian that describes the broadening of the object s spectrum with respect to the template , one can write : @xmath53 eq . [ sigma ] applies to both components in the binary spectrum and their corresponding ccf peaks . the crosscorrelations are performed with the iraf task fxcor that uses fourier transforms of the spectra to compute the ccf . before performing the crosscorrelation , the continuum is subtracted from the normalized spectra . filtering in the fourier domain is applied to avoid undesirable contributions originating from noise or intrinsically broad lines ( see wyatt 1985 ) . the templates are chosen from the radial - velocity standards such that their spectral types resemble those of the targets . the value of @xmath49 for these stars is determined for each order separately by autocorrelation of the template spectrum adopting the same filter used for the crosscorrelations . in this case @xmath52 is zero and therefore @xmath49 is found directly from the width of the ccf peak : @xmath54 . the template spectra are correlated with our target stars order by order , where we limit ourselves to orders in the red spectra that do not suffer from strong telluric lines . most ccf peaks can be fitted well with a gaussian . as the final value for @xmath48 we give the broadening @xmath52 averaged over the different orders , and for the uncertainty we take the rms of the spread around the average of @xmath52 ( table [ fluxes ] ) . equating @xmath48 to @xmath52 implicitly assumes that @xmath49 is the width of the lines not related to rotation . an upper limit to @xmath48 is found from the other extreme in which we assume that the total width of the spectral line @xmath50 follows from rotation . this creates uncertainties of the order of @xmath52 for s1242 . in the case of s1237 we find that @xmath55 ( @xmath56 ) , i.e. the lines in s1237 are narrower than in the template . for these two stars , we give an upper limit of @xmath57 in table [ fluxes ] . s1113 is the only binary observed whose ccf shows two peaks . the ccf peaks of both stars overlap in the 1998 observation . therefore we do not use these spectra in the following analysis . for the 1996 spectra , the ccf shows two peaks , one of which is broad indicating the presence of a fast rotating star . both peaks are clearly separated with a center - to - center velocity separation of @xmath7 110 km s@xmath5 . the lines in the spectrum are smeared out and less pronounced , resulting in a noisy ccf . to improve this , we combine four sequential orders before crosscorrelating ( being constrained by the maximum number of points fxcor can handle ) . from eq . 17 in gunn et al . we derive the relative light contribution of both components to the spectrum from the height and dispersion of the crosscorrelation peaks , assuming that the binary stars have the same spectrum as the template ( @xmath58 in eq . according to this the rapidly rotating star contributes @xmath59% of the light . note that luminosity ratios derived from cross - correlations are uncertain and should be confirmed photometrically . the line profile of the fast rotating star in s1113 is not compatible anymore with a gaussian , therefore we adopt another method to determine its @xmath48 , described in piters et al . this method uses the property that the fourier - bessel transform of a spectral line that is purely rotationally broadened has a maximum at the position of the projected rotational velocity . in practice , this position is a function of the limits over which the fourier transform is performed . the local maxima of this velocity - versus - cutoff - frequency ( vcf ) function approach @xmath48 within half a percent , the result growing more accurate for maxima at higher frequencies . this error is negligible when compared to errors arising from noise , other line broadening mechanisms , etc . ( see piters et al . 1996 ) . in our determination of @xmath48 of the primary of s1113 , we have used the first local maximum in the vcf - plot of the transformation of four isolated lines at @xmath15 6265.14 , 6400.15 , 6408.03 and 6411.54 . the @xmath48 in table [ fluxes ] is an average of the resulting values ; the @xmath48 of the secondary is found from crosscorrelation . the application of the fourier - bessel transformation method is limited on the low velocity side by the spectral resolution : the fourier transform can not be performed beyond the nyquist frequency which for slow rotators lies at a frequency that is lower than the cutoff - frequency at which the first maximum occurs in the vcf - plot . for our spectra this means that the method can not be used for @xmath60 5.2 km s@xmath5 . indeed , for every star for which the crosscorrelation method gives a @xmath48 smaller than this value , the vcf - plot does not reach the first local maximum , except for s1082 for which we find @xmath48 = 9.5(1.6 ) . for s1072 , the fourier - bessel transform gives a @xmath48 of 12.7(1.0 ) km s@xmath5 . spectral lines were selected from those used in groot et al . ( 1996 ) and from the additional lines used for s1113 . the results of our search for emission cores in the h&k lines are displayed in fig . [ cahk ] and in table [ fluxes ] . the emission lines are strong in s1063 , s1113 and s1040 ; and still detectable in s1242 which indicates chromospheric activity in these stars . in s1072 and s1237 the emission cores are marginal , and on s1082 we can only determine an upper limit . the ( projected ) rotational velocities of all of our stars are relatively small @xmath61 km s@xmath5 , with the exception of s1113 . in sect . 4.1 we investigate whether the relations between x - ray emission , strength of the emission cores in the h&k lines , and the rotational velocities of the unusual x - ray emitters in are similar to the relations found for well - known magnetically active stars , the rs cvn binaries . in sect . 4.2 we briefly discuss the behaviour of the h@xmath0 line and spectral lines other than h&k that are indicators of chromospheric activity . individual systems are discussed in sect . 4.3 . to investigate whether the x - rays of the stars studied in this paper are related to magnetic activity , we compare their optical activity indicators and x - ray fluxes with those of a sample of rs cvn binaries . in particular , we select rs cvn binaries for which fluxes of the emission cores in the h&k lines have been determined from high - resolution spectra by fernndez - figueroa et al . ( 1994 ) . to obtain x - ray countrates for these binaries , we searched the rosat data archive for pspc observations of them . we then analyzed all these observations , and determined the countrates , in the same bandpass as used in the analysis of , using the standard procedure described in zimmermann et al . all pointings that we have analyzed actually led to a positive detection of the rs cvn system : even when not the target of the observation , the rs cvn system is usually the brightest object in the field of view . the results of our analysis are listed in table [ tabros ] . lllrl@r name & d & jd & & ctrate & + & ( pc ) & & & ( s@xmath5 ) & + + + + & 107@xmath6212 & 8420.575 & 16156 & 0.322(5 ) & 36 + & & 8775.123 & 8345 & 0.126(5 ) & 36 + & 21.7@xmath620.5 & 8864.531 & 4267 & 6.072(15 ) & 0 + & & 9003.872 & 5776 & 6.228(11 ) & 0 + & & 8864.595 & 3651 & 6.84(2 ) & 24 + & & 9004.871 & 3151 & 5.41(2 ) & 24 + & 37.7@xmath621.9 & 9049.117 & 14531 & 1.206(4 ) & 0 + & 16.4@xmath620.2 & 9247.809 & 9895 & 1.009(7 ) & 0 + @xmath63 & 21.4@xmath620.5 & 9085.714 & 2140 & 1.10(2 ) & 0 + & 49.9@xmath622.1 & 9148.099 & 1209 & 1.23(3 ) & 0 + & 25.3@xmath624.9 & 8604.712 & 5249 & 0.888(16 ) & 36 + + + & 29.0@xmath627 & 8648.446 & 3098 & 6.29(2 ) & 0 + & 168@xmath6251 & 8426.486 & 21428 & 0.146(3 ) & 33 + & 108@xmath6212 & 8810.279 & 2526 & 0.336(17 ) & 44 + & & 8991.040 & 6214 & 0.372(11 ) & 44 + & & 8796.874 & 5076 & 0.305(12 ) & 51 + & & 8966.506 & 2904 & 0.366(18 ) & 51 + & 44.5@xmath621.2 & 8431.422 & 71803 & 2.3152(14 ) & 44 + & & 9158.435 & 37658 & 2.403(3 ) & 44 + @xmath64 & 202@xmath6257 & 9030.359 & 10327 & 0.059(2 ) & 0 + @xmath64 & 193@xmath6239 & 8789.392 & 8668 & 0.198(5 ) & 0 + @xmath65 & 42.0@xmath621.0 & 8620.767 & 13460 & 1.717(4 ) & 0 + & & 9136.873 & 4892 & 2.902(13 ) & 0 + + + & 192@xmath6234 & 8322.477 & 3516 & 0.574(13 ) & 3 + @xmath66 & 37.5@xmath621.1 & 8346.927 & 4745 & 5.269(14 ) & 0 + & & 8904.569 & 1438 & 3.43(4 ) & 0 + & & 8735.159 & 7940 & 4.753(8 ) & 0 + & 138@xmath6210 & 9272.484 & 4156 & 1.32(2 ) & 41 + & 106@xmath627.6 & 8328.919 & 14690 & 0.624(6 ) & 40 + @xmath67 & 103.@xmath628.5 & 8863.981 & 10576 & 1.554(8 ) & 31 + @xmath67 & 323@xmath6253 & 9088.631 & 24889 & 0.112(2 ) & 0 + & 96.8@xmath627.1 & 8589.200 & 6335 & 1.692(15 ) & 44 + & & 8769.630 & 8150 & 1.422(12 ) & 44 + & & 8973.263 & 22143 & 1.909(4 ) & 44 + @xmath68 & 25.8@xmath620.5 & 8448.155 & 31165 & 4.077(2 ) & 0 + to compare systems at different distances , we multiply the rosat countrate and the flux of the emission cores for each system with the square of the distance listed in table [ tabros ] ; for we adopt a distance of 850pc ( twarog & anthony - twarog 1989 ) . no corrections are made for interstellar absorption . the choice of the 0.42.4kev bandpass minimizes the effects of interstellar absorption , which are severe at energies @xmath69kev . as it is unknown which component of the binary emits the x - rays , we plot the total x - ray and fluxes , adding the contributions of both components where these are given separately by fernndez - figueroa et al . ( 1994 ) . the resulting absolute countrates and fluxes are shown in fig . the systems with h&k emission clearly visible in fig . [ cahk ] , viz . s1063 , s1113 , s1040 and s1242 lie on the relation between x - ray and h&k emission defined by the rs cvn systems , in agreement with the hypothesis that the x - ray flux of these objects is related to the magnetic activity . it is also seen that the upper limits or marginally detected emission cores in s1082 , s1072 and s1237 are high enough that we can not exclude the hypothesis that the x - ray emission in these systems is related to magnetic activity . the rotational velocity is another indicator of magnetic activity . we investigate the relation between rotational velocity and ca emission by selecting those stars from the sample of fernndez - figueroa for which a value of @xmath48 is given in the catalogue of chromospherically active binary stars ( strassmeier et al . 1993 ) . in fig . [ vsinica ] the h&k emission of these stars is compared with their @xmath48 . in this figure we do discriminate between the separate contributions of both stars to @xmath31 , with the exception of s1113 for which we combine the total flux @xmath31 with the @xmath48 of the primary . the stars are found within the range occupied by chromospherically active stars . we note that the correlation between the observed h&k flux and @xmath48 is not tight . in particular , high and low h&k emission flux is found at low values of @xmath48 . some of the scatter may be due to the use of @xmath48 instead of the stellar rotation period . parameters depending on the spectral type ( e.g. properties of the convective region ) have been used to reduce the scatter in the activity - rotation relation ; whereas this is successful for main - sequence stars with @xmath70 , it fails for other main - sequence stars and for giants ( see discussion in stpie 1994 ) . for example , the three giants ( k0 iii ) , ( k0 iii ) and ( k0 - 2 iii ) have @xmath48 values of 10 , 10 and 8 km s@xmath5 , respectively , but differ in @xmath71 by three orders of magnitude ( see fig . [ vsinica ] ) . the h@xmath0 lines ( @xmath72 6562.76 ) of only two stars , s1063 and s1113 , show clear evidence of emission , as shown in fig . [ subha ] . this is described in more detail in their individual subsection in sect . 4.3 . for the other stars , we have used the few ( sub)giants in the library of ues - spectra of montes & martn ( 1998 ) to investigate the behaviour of h@xmath0 . we have chosen library spectra of stars that match the spectra of the stars as closely as possible ( see fig . [ halpha ] ) . s1072 and s1237 show no evidence for filling in of the h@xmath0 profile compared to a g0iv v and a k0iii star , respectively . in s1040 h@xmath0 seems slightly filled in compared to a g8iv and a k0iii star . this is also the case for s1242 compared to a g0iv - v star , but we note that no classification for this star is found in literature . for s1082 , no matching spectrum is available in this wavelength region . filling in of the lines in the b triplet ( @xmath15 5167.33 , 5172.70 and 5183.62 ) and in the d doublet ( @xmath15 5889.95 and 5895.92 ) is visible in some active stars . the presence of a d@xmath73 ( @xmath72 5876.56 ) absorption or emission feature can also indicate activity ( see discussion in montes & martn ( 1998 ) and references therein ) . however , in none of the stars we see filled in b and d lines . neither do we see a clear d@xmath73 feature . for s1082 ( b and d ) and s1113 ( b , d and d@xmath73 ) we find no suitable library stars for these features . the two stars below the subgiant branch , s1063 and s1113 , both show relatively strong h&k emission , and are the only two stars in our sample showing h@xmath0 in emission , shown in fig . [ subha ] . we use the orbital solutions for both objects to try and identify the star responsible for these emission lines . the velocities of both components in s1113 , and of one component in s1063 are indicated in fig . [ subha ] . the h@xmath0 line profile of s1063 is asymmetric , showing emission which is blue - shifted with respect to the absorption . the location of the absorption line is compatible with the velocity of the primary , which dominates the flux ; the emission is probably due to the secondary . remarkably , the h&k emission peak is at the velocity of the primary . this suggests that the h@xmath0 emission is not chromospheric in nature . the h@xmath0 emission of s1063 does not show the double peak that is known to indicate accretion disk emission ( horne & marsh , 1986 ) . in s1113 the h@xmath0 emission profile is symmetric and broad , with full width at continuum level of 15 . the emission peak is centered on the more massive star , which contributes 82% of the total flux ( table [ fluxes ] ) . this suggests that the h@xmath0 emission is due to the primary . the h&k emission shows marginal evidence for a double peak , suggesting that both stars contribute to the chromospheric emission . in fig . [ subca ] we indicate the expected position of the h&k lines for both stars . for the phase observed , their peaks overlap in the crosscorrelation function . in the figure we use the velocities resulting from fitting the order that gives the cleanest crosscorrelation . the h&k emission in the wide binaries s1072 and s1237 is only marginally significant . the level of their x - ray and emission is more appropriate for active main - sequence stars ( fig.[xca ] ) . one might speculate that it is due to the invisible companion of the giant detected in the crosscorrelation ; even if this were the case , we would not understand why this companion would be chromospherically active . we conclude that we do not understand why these two stars are x - ray sources . we find no indication for a faint secondary in the crosscorrelation profile of s1237 ; at the time of observation , the spectra of two equally massive stars as suggested by janes & smith ( 1984 ) would be separated by 4 to 7 km s@xmath5 ( derived from the ephemeris in mathieu et al . since the secondary is 1.6 magnitude fainter in v than the primary , we think that this small separation is compatible with finding a single peak in the crosscorrelation . s1242 is chromospherically active , as shown by its h&k emission . we suggest that this activity , which also explains the x - rays , is due to rapid rotation induced by tidal interaction at periastron , which tries to bring the subgiant into corotation with the orbit at periastron . if we assume that the observed period of photometric variability is the rotation period we derive an inclination of @xmath7 9@xmath74 using our maximum value of @xmath48 and the estimated radius . this would be in agreement with a companion at the high end of the range 0.140.94 @xmath75 allowed by the mass function ( mathieu et al . 1990 ) . our detection of clear chromospheric emission indicates that the x - ray emission of s1040 is due to the giant . the white dwarf has a low temperature , and is unlikely to contribute to the x - ray flux . we find a rather slow rotational velocity for the giant , 3 km s@xmath5 . gilliland et al . ( 1991 ) detected a periodicity of 7.97 days in the visual flux ( b and v bandpasses ) of s1040 , with an amplitude of 0.012mag . if this is the rotation period of the giant , the radius of 5.1@xmath76 ( landsman et al . 1997 ) implies an equatorial rotation velocity of @xmath77 km s@xmath5 . this is compatible with the velocity measured with our crosscorrelation , @xmath48 , for an inclination @xmath78 . this inclination has an a priori probability less than 0.5% ; and it implies an unacceptably high mass for the white dwarf , from the measured mass function @xmath79 . we conclude that the 8days period can not be the rotation period of the giant . it is doubtful that the white dwarf can be responsible , as its contribution to the @xmath2 and @xmath1 flux is small . the h@xmath0 absorption profile of the blue straggler s1082 is variable . if we consider the most symmetric spectrum profile , that of 00:01 ut , as the unperturbed profile of the primary , we find that the changes are due to extra emission . this is illustrated in fig . [ s1082 ha ] . we suggest that this variation is due to the subluminous companion , possibly to a wind of that star . we have also investigated the presence of a broad shallow depression underlying the d lines ( near @xmath72 5895 ) and the triplet ( near @xmath72 7775 ) as found by mathys ( 1991 ) . we find that this broad component is variable , as illustrated in fig . [ s1082na ] . mathys ( 1991 ) suggests that the broad component originates in the subluminous companion . this companion outshines the primary by a factor six at @xmath72 1500 and thus is presumably hot ( landsman et al . 1998 ) . we note that the star can not be too hot or it would not show neutral lines . in this paper we have tried to find an explanation for the x - ray emission of seven sources in . for s1242 and s1040 we have concluded from the h&k emission cores that magnetic activity is responsible for the x - rays . this is supported by filling in of h@xmath0 ( see e.g. montes et al . 1997 ; eker et al . 1995 ) . in s1242 , activity is likely to be triggered by interaction at periastron in the eccentric orbit . this is also reflected in the period of photometric variability . for s1040 , the reason for activity is less clear . the explanation could involve mass transfer from the precursor of the white dwarf to the giant and the latter s subsequent expansion during the giant phase . as was already noted by landsman et al . ( 1997 ) , a similar system is , a binary of a white dwarf of @xmath80 k ( simon et al . 1985 ) and a g5iii giant in an 56.8 days circular orbit . the @xmath48 of that giant is also low , 4 km s@xmath5 , and the long photometric period of 77.2 days implies asynchronous rotation ( strassmeier et al . the x - ray luminosity for is @xmath81 erg s@xmath5 in the 0.24 kev band as measured with einstein by walter & bowyer ( 1981 ) , somewhat higher than the luminosity of s1040 . ( with the coronal model discussed by belloni et al . ( 1998 ) , the countrate for s1040 corresponds to @xmath82 erg s@xmath5 in the 0.24 kev band . ) walter & bowyer attribute the x - rays to coronal activity of the giant . the h&k emission cores in s1063 and s1113 are very strong . in s1113 we might even see emission of both stars . due to the shape of the h@xmath0 emission we can not conclude with certainty that the x - rays arise in an active corona and not in a disk or stream . the wings in the emission peak of s1113 are very broad . however , montes et al . ( 1997 ) have demonstrated that the excess emission in the h@xmath0 lines of the more active binaries is sometimes a composite of a narrow and broad component , the latter having a full width at half maximum of up to 470 km s@xmath5 . they ascribe this broad component to microflaring accompanied by large scale motions . we note the similarity between s1113 and , a well known extremely active binary of a g5iv ( @xmath48 = 13 km s@xmath5 ) and k1iv ( @xmath83 km s@xmath5 ) star in a 2.84 days circular orbit and a mass ratio 0.79 ( strassmeier et al . 1993 ) ; the mass ratio of s1113 is 0.70 ( mathieu et al . 1998 , in preparation ) . from the countrate of in table [ tabros ] we find @xmath84 erg s@xmath5 in the 0.12.4 kev band using the same model as in belloni et al . ( 1998 ) with @xmath85 , which is comparable to the luminosity of s1113 in the same band @xmath86 erg s@xmath5 ( belloni et al . the h@xmath0 emission of s1063 is more difficult to explain . as this system is not double - lined , h@xmath0 emisson by the ( invisible ) secondary star would have to be strong to rise above the continuum of the primary . in the binaries s1072 and s1237 we see no h@xmath0 emission while the level of h&k emission is low in comparison with active stars of the same luminosity class . we have no explanation for this . for s1072 , an option is a wrong identification of the x - ray source with an optical counterpart . belloni et al . ( 1998 ) give a probability of 43% that one or two of their twelve identifications of an x - ray source with a binary in is due to chance . no h&k emission is seen in the spectrum of the blue straggler s1082 . possibly , the x - ray emission has to do with the hot , subluminous secondary that could also cause the photometric variability and whose signature we might have seen in the h@xmath0 line . the authors wish to thank g. geertsema for her help during our observations , p. groot for providing his program to compute projected rotational velocities with the fourier - bessel transformation method and m. van kerkwijk for comments on the manuscript . mvdb is supported by the netherlands organization for scientific research ( nwo ) . carter , d. , benn , c. r. , rutten , r. g. m. , breare , j. m. , rudd , p. j. , king , d. l. , clegg , r. e. s. , dhillon , v. s. , arribas , s. , rasilla , j .- l . , garcia , a. , jenkins , c. r. , charles , p. a. 1993 , isis users manual , http://www.ing.iac.es zimmermann , h. , becker , w. , belloni , t. , dbereiner , s. , izzo , c. , kahabka , p. , schwentker , o. 1994 , exsas user s guide : extended scientific analysis system to evaluate data from the astronomical x - ray satellite rosat , technical report 257 , mpe
since any experiment on a superfluid fermi gas is done in trap potential@xcite , it is interesting to explore physical phenomena originating from this spatial inhomogeneity . an example is surface oscillations observed in a @xmath2li superfluid fermi gas@xcite . another example is the phase separation observed in a @xmath2li fermi gas with spin imbalance@xcite , where the spin - balanced superfluid region in the trap center is spatially surrounded by excess atoms . in addition to these macroscopic phenomena , the spatial inhomogeneity can also affect microscopic superfluid properties . noting that a trap potential breaks the spatial inversion symmetry when the inversion center is taken to be away from the trap center , we expect that the parity becomes no longer a good quantity to classify the spatial structure of a cooper pair , leading to the admixture of even and odd parity symmetry . since a pair wavefunction is always antisymmetric with respect to the exchange of two fermions , this naturally leads to the mixing of spin - singlet and spin - triplet state . when this phenomenon occurs , the @xmath0-wave superfluid state is accompanied by a triplet cooper pair amplitude , in addition to the ordinary singlet component . ( the cooper pair amplitude is symbolically written as @xmath3 , where @xmath4 is an annihilation operator of a fermi atom with pseudospin @xmath5 . ) the purpose of this paper is to theoretically explore this possibility in a trapped @xmath0-wave superfluid fermi gas . using symmetry considerations , we prove that this phenomenon may occur , when a spin rotation symmetry of this system is also broken , in addition to the broken inversion symmetry by a trap potential . in a two - component fermi gas , this additional condition is realized , when two species feel different trap potentials or chemical potentials , or when they have different atomic masses . although this is a necessary condition , we numerically confirm that a triplet pair amplitude is really induced under this condition , within the mean - field theory for a model two - dimensional lattice fermi superfluid in a harmonic trap . in considering a triplet pair amplitude , one should note that the appearance of this quantity does not immediately mean the realization of a triplet superfluid state . actually , the system is still in the @xmath0-wave superfluid state , as far as the system only has an @xmath0-wave interaction . this is simply because the symmetry of a fermi superfluid is fully determined by the symmetry of the superfluid order parameter , which is essentially given by the product of a pairing interaction and a pair amplitude . for example , an @xmath0-wave superfluid fermi gas with a contact type @xmath0-wave pairing interaction @xmath6 ( @xmath7 is characterized by the ordinary @xmath0-wave superfluid order parameter , @xmath8 which is finite when the pair amplitude @xmath9 has the @xmath0-wave component . the odd - parity component does not contribute to @xmath10 in eq . ( [ eq.1 ] ) . however , for an @xmath0-wave superfluid fermi gas with both the singlet and triplet pair amplitude , when one suddenly changes the @xmath0-wave pairing interaction to a triplet ( and odd parity ) one @xmath11 , while the @xmath0-wave superfluid order parameter in eq . ( [ eq.1 ] ) immediately vanishes due to the vanishing @xmath0-wave interaction ( @xmath12 ) , the product of the triplet interaction and the triplet component in the pair amplitude @xmath9 ( which is assumed to have already existed in the @xmath0-wave state ) immediately gives a finite triplet superfluid order parameter , @xmath13 when the triplet interaction @xmath11 is chosen so that the momentum summation in eq . ( [ eq.2 ] ) can be finite . in an ultracold fermi gas , the change of the interaction is possible by using a tunable interaction associated with a feshbach resonance@xcite . then , by definition , one obtains a triplet superfluid fermi gas characterized by the superfluid order parameter @xmath14 in eq . ( [ eq.2 ] ) , at least just after this manipulation . this makes us expect that , when one can induce a @xmath1-wave pair amplitude in an @xmath0-wave superfluid fermi gas , a @xmath1-wave superfluid fermi gas may be realized . this possibility has recently been discussed by one of the authors@xcite , where a @xmath1-wave pair amplitude is induced by a synthetic spin - orbit interaction@xcite . the present paper provides another source of @xmath1-wave pair amplitude , without using an artificial gauge field . the admixture of singlet and triplet cooper pairs has recently attracted much attention in the field of non - centrosymmetric superconductivity@xcite , where a crystal lattice with no inversion center causes this phenomenon . in this field , it has been pointed out that this admixture may be the origin of the anomalous temperature dependence of the penetration depth observed in li@xmath15pt@xmath16b@xcite . thus , an @xmath0-wave superfluid fermi gas with a triplet pair amplitude would be also helpful to the study of this electron system . this paper is organized as follows . in sec.ii , we clarify the necessary condition for a triplet cooper pair amplitude to appear in a trapped @xmath0-wave superfluid fermi gas . in sec.iii , we numerically evaluate how large a triplet pair amplitude is induced under the condition obtained in sec.ii . in this section , we treat a superfluid fermi gas loaded on a two - dimensional square lattice , within the mean - field theory . throughout this paper , we take @xmath17 , for simplicity . we consider a three - dimensional @xmath0-wave superfluid fermi gas , described by the hamiltonian , @xmath18 . \label{eq.3}\ ] ] here , @xmath19 is a fermion field operator with pseudospin @xmath5 , describing two atomic hyperfine states . @xmath6 @xmath20 is a contact - type @xmath0-wave pairing interaction . @xmath21 is a one - particle hamiltonian density , consisting of a kinetic term and a potential term , detailed expression of which will be given later . we assume that the system is in the ordinary @xmath0-wave superfluid state with the @xmath0-wave superfluid order parameter , @xmath22 we also assume that any other spontaneous symmetry breaking is absent ( such as the triplet superfluid state ) . in this model superfluid , we consider the spin - triplet cooper - pair amplitude , given by @xmath23 & ~~(s_z=0 ) , \\ \langle\psi_\downarrow({\bm r})\psi_\downarrow({\bm r}')\rangle & ~~(s_z=-1 ) , \end{array } \right . \label{eq.6}\end{aligned}\ ] ] where @xmath24 denotes the @xmath25-component of the total spin of each pair amplitude . the triplet pair amplitude in eq . ( [ eq.6 ] ) does not contribute the @xmath0-wave superfluid order parameter @xmath26 in eq . ( [ eq.4 ] ) , because @xmath27 . the spin - singlet pair amplitude , @xmath28 , \label{eq.5}\ ] ] only contributes to eq . ( [ eq.4 ] ) . we first prove that the broken spatial inversion symmetry is necessary for a triplet pair amplitude to appear in an @xmath0-wave superfluid fermi gas . for this purpose , we conveniently introduce the inversion operator @xmath29 with respect to the inversion center @xmath30 . the field operator is transformed under this operation as @xmath31 where @xmath32 . the inverted hamiltonian @xmath33 is then given by @xmath34 . \label{eq.8}\end{aligned}\ ] ] when the one - particle hamiltonian density @xmath21 has the symmetry @xmath35 , this system is invariant ( @xmath36 ) under this symmetry operation . on the other hand , the triplet pair amplitude @xmath37 in eq . ( [ eq.6 ] ) with the center of mass position @xmath38/2 $ ] is transformed as , @xmath39 where @xmath40 is the relative coordinate . we also find @xmath41 . that is , the triplet pair amplitude @xmath42 vanishes , when the system has the inversion symmetry ( @xmath36 ) with respect to the inversion center @xmath38/2 $ ] . thus , the broken inversion symmetry is necessary for a triplet pair amplitude to appear . for the singlet pair amplitude in eq . ( [ eq.5 ] ) , this symmetry operation simply gives @xmath43 . as expected , this quantity may be finite . the one - particle hamiltonian density @xmath21 in the ordinary uniform fermi gas has the form @xmath44\delta_{\alpha,\alpha ' } , \label{eq.10}\ ] ] where @xmath45 , @xmath46 is an atomic mass , and @xmath47 is the fermi chemical potential . equation ( [ eq.10 ] ) has the symmetry property , @xmath35 , with respect to @xmath48 for an arbitrary @xmath30 . to conclude , any triplet pair amplitude is not induced . in the presence of a harmonic trap , the one - particle hamiltonian density becomes inhomogeneous as @xmath49 \delta_{\alpha,\alpha ' } , \label{eq.11}\ ] ] so that it does not have the inversion symmetry except at @xmath50 . however , when we consider the @xmath0-wave superfluid state in this trapped case , any triplet pair amplitude is not actually induced ( although we do not explicitly show the result here ) . of course , since the condition obtained from the inversion symmetry is a _ necessary _ condition , the broken inversion symmetry does not guarantee the appearance of a triplet pair amplitude . in this regard , we point out that the vanishing triplet pair amplitude in the trapped case is due to the fact that this system still has a rotation symmetry in spin space . to see this , we next consider the spin rotation of the field operator , given by @xmath51 here , @xmath52 describes the spin rotation around the unit vector @xmath53 with the angle @xmath54 , and @xmath55 , where @xmath56 ( @xmath57 are pauli matrices . ( as usual , we take the spin quantization axis in the @xmath58-direction . ) for the three @xmath59 rotations " specified by @xmath60 , @xmath61 , @xmath62 , eq . ( [ eq.12 ] ) can be written as , @xmath63 where we have used the formula @xmath64 . under the @xmath59-rotation , the hamiltonian in eq . ( [ eq.3 ] ) is transformed as @xmath65 \nonumber \\ & = & \int d{\bm r } \bigl [ \sum_\alpha \psi^\dagger_\alpha({\bm r } ) \left ( \sigma_j{\hat h}({\bm r})\sigma_j \right)_{\alpha,\alpha ' } \psi_{\alpha'}({\bm r } ) -u_s \psi^\dagger_\uparrow({\bm r } ) \psi^\dagger_\downarrow({\bm r } ) \psi_\downarrow({\bm r } ) \psi_\uparrow({\bm r } ) \bigr ] . \label{eq.14}\end{aligned}\ ] ] here , @xmath66 . thus , one has @xmath36 , when @xmath67=0 \label{eq.15}\ ] ] is satisfied . while the singlet pair amplitude in eq . ( [ eq.5 ] ) remains unchanged under these @xmath59-rotations , the triplet component is transformed as @xmath68 for example , when we set @xmath69 , eq . ( [ eq.16 ] ) means that @xmath70 , when @xmath71=0 $ ] . ( the other two components with @xmath72 are not excluded in this case . ) when the system invariant under all the @xmath59-rotations ( @xmath73 ) , any triplet pair amplitude is not induced . to conclude , the broken spin rotation symmetry characterized by @xmath74 is necessary for a triplet pair amplitude to be induced in a trapped @xmath0-wave superfluid fermi gas . this is the reason why the model case described by eqs . ( [ eq.3 ] ) and ( [ eq.11 ] ) is not accompanied by any triplet pair amplitude . the broken spin rotation symmetry is realized , when the strength of trap potential @xmath75 in eq . ( [ eq.11 ] ) depends on spin ( @xmath76 ) . in this case , the one - particle hamiltonian @xmath66 can be written as @xmath77 + { k_\uparrow - k_\downarrow \over 4}r^2\sigma_z . \label{eq.17}\ ] ] equation ( [ eq.15 ] ) is then satisfied only when @xmath78 . thus , we find from the last line in eq . ( [ eq.16 ] ) that only the triplet pair amplitude with @xmath79 may be induced . since the last term in eq . ( [ eq.17 ] ) works as an external magnetic field , this phenomenon is also expected in the presence of spin imbalance@xcite , where two species feel different fermi chemical potentials @xmath80 . another possibility is a trapped hetero superfluid fermi gas@xcite , where two species have different atomic masses @xmath81 . in sec . iii , we will numerically examine these cases . before ending this section , we briefly note that the broken inversion symmetry , as well as the broken spin rotation symmetry , are also realized in a spin - orbit coupled @xmath0-wave superfluid fermi gas@xcite . indeed , refs.@xcite predict that a @xmath1-wave pair amplitude is induced in this case . although we do not deal with this case in sec.iii , we explain in appendix a how to apply the present symmetry consideration to this case . to examine whether or not a triplet pair amplitude is induced under the condition obtained in sec . ii , we consider a model @xmath0-wave fermi superfluid loaded on a @xmath82 two - dimensional square lattice , within the mean - field approximation . although this simple model can not be directly applied to a real continuum fermi gas , it is still helpful to grasp basic characters of this phenomenon . the hamiltonian is given by @xmath83 + \sum_i\delta_s({\bm r}_i ) \bigl [ c^\dagger_{{\bm r}_i,\uparrow}c^\dagger_{{\bm r}_i,\downarrow}+{\rm h.c . } \bigr ] \nonumber \\ & + & \sum_{i,\alpha } \bigl [ v_\alpha({\bm r}_i)-\mu_{\alpha}-u_s n_{-\alpha}({\bm r}_i ) \bigr]c_{{\bm r}_i,\alpha}^\dagger c_{{\bm r}_i,\alpha}. \label{eq.18}\end{aligned}\ ] ] here , @xmath84 is a creation operator of a fermi atom at the lattice site @xmath85 , with pseudospin @xmath5 , and the fermi chemical potential @xmath86 . @xmath87 describes a particle hopping between the nearest neighbor sites , and @xmath88 means the summation over the nearest neighbor pairs . in eq . ( [ eq.18 ] ) , the @xmath0-wave superfluid order parameter @xmath89 , as well as the hartree potential @xmath90 , are obtained from the mean - field approximation for the on - site hubbard interaction @xmath91 , where @xmath6 ( @xmath92 ) is the interaction strength . @xmath93 is a harmonic trap potential , where @xmath94 is the strength of a trap potential which @xmath95-spin atoms feel . here , the spatial position @xmath96 is measured from the center of the @xmath82 square lattice , and @xmath97 is the distance between the trap center and the edge of the system . for simplicity , we take the lattice constant @xmath98 to be unity . in the present model , the spatial inversion symmetry is broken by the trap potential except at the trap center . for the spin rotation symmetry , it is broken when one of @xmath99 , @xmath86 , and @xmath94 , depends on pseudospin @xmath5 . since all these cases satisfy eq . ( [ eq.15 ] ) only when @xmath78 , eq . ( [ eq.16 ] ) indicates that one may only consider the possibility of the triplet pair amplitude with @xmath79 . in the present model , this component is given by @xmath100 . \label{eq.25}\end{aligned}\ ] ] for comparison , we also consider the ordinary singlet component , given by @xmath101 . \label{eq.26}\end{aligned}\ ] ] besides the superfluid order parameter , the condensate fraction is also a fundamental quantity in the superfluid phase@xcite , which physically describes the number of bose - condensed cooper pairs . in an ordinary @xmath0-wave superfluid state , it has the form , @xmath102 where the local condensate fraction @xmath103 is directly related to the singlet pair amplitude in eq . ( [ eq.26 ] ) as @xmath104 in addition to the singlet component of the condensate fraction @xmath105 in eq . ( [ eq.26b ] ) , the present system may also have the spin - triplet component@xcite , @xmath106 , where @xmath107 the total condensate fraction is given by @xmath108 . in what follows , we simply call @xmath105 and @xmath109 the singlet and triplet condensate fraction , respectively . along the @xmath110-axis ( @xmath111 ) in an @xmath0-wave superfluid fermi gas with trap - potential difference . ( b ) @xmath0-wave superfluid order parameter @xmath112 . ( c ) density profile @xmath113 . we take @xmath114 , @xmath115 , and @xmath116 . this parameter set is also used in figs . [ fig2]-[fig4 ] . ] we note that the square lattice in our model does not affect the symmetry consideration in sec . this is because the square lattice is invariant under the spatial inversion with respect to the center of mass position @xmath117/2 $ ] of a triplet pair amplitude @xmath118 . in addition , since the spin rotation symmetry is also unaffected by the crystal lattice , the necessary condition obtained in sec . ii is still valid for the present case . as usual , we diagonalize the mean - field hamiltonian @xmath119 in eq . ( [ eq.18 ] ) by the bogoliubov transformation@xcite . since this is a standard procedure@xcite , we do not explain the detail here , but summarize the outline in appendix b. we numerically carry out the bogoliubov transformation , to self - consistently determine @xmath120 , @xmath121 , and @xmath86 . we then evaluate the triplet pair amplitude @xmath122 in eq . ( [ eq.25 ] ) . in numerical calculations , we take the lattice size @xmath123 , and @xmath124 , where @xmath125/2 $ ] . to avoid lattice effects , we consider the low density region , by setting @xmath126 in the absence of spin imbalance ( where @xmath127 is the number of fermi atoms in the @xmath95-spin component ) . the total number @xmath128 of fermi atoms then equals @xmath129 . in this case , the particle density is at most @xmath130 even in the trap center . we take a small but finite temperature @xmath131 , in order to suppress effects of discrete energy levels associated with the finite system size . in an singlet superfluid fermi gas with trap - potential difference . ( a ) @xmath132 . ( b ) @xmath133 . ( c ) @xmath134 . ( d ) singlet pair amplitude @xmath135 . ] figure [ fig1](a ) shows the evidence that the triplet pair amplitude with @xmath79 is induced in the @xmath0-wave superfluid state , when both the spatial inversion symmetry and the spin rotation symmetry are broken by the trap potential @xmath136 . from the comparison of this figure with figs.[fig1](b ) and ( c ) , one finds that @xmath137 appears everywhere in the gas cloud where the @xmath0-wave superfluid order parameter @xmath138 , as well as the atom density @xmath139 , are finite , except at the trap center . since the system still has the spatial inversion symmetry at the trap center , the node structure seen in fig . [ fig1](a ) agrees with the symmetry consideration in sec . we emphasize that the triplet pair amplitude is not induced when @xmath140 , although we do not explicitly show the result here . figure [ fig2](a ) shows that the point node seen in fig.[fig1](a ) is actually a line node along the @xmath141-axis . this node structure comes from the symmetry property that , while the present square - lattice model has the reflection symmetry with respect to the @xmath141 axis , the triplet pair amplitude @xmath118 behaves as , @xmath142 when @xmath117/2=(0,r_y)$ ] and @xmath143 . since the present lattice model is also invariant under the reflections with respect to the @xmath110 axis , as well as the lines along @xmath144 , the triplet pair amplitude @xmath118 , with the relative vector @xmath145 being perpendicular to one of them , has the line node along the reflection line . ( see figs . [ fig2](b ) and ( c ) . ) in a continuum system with no lattice , the triplet pair amplitude is expected to always have the line node , which is perpendicular to the relative vector of the pair amplitude . we briefly note that such a node is not obtained in the singlet component , as shown in fig . [ fig2](d ) . , as a function of the relative coordinate @xmath145 . we take @xmath114 . for each center of mass position @xmath117/2 $ ] ( open square ) , the pair amplitude is plotted inside the region between two dashed line , by taking @xmath30 as the origin . ( b ) spatial variation of synthesized @xmath1-wave superfluid order parameter @xmath146 . in this panel , the @xmath147-dependence of @xmath146 is schematically shown , being centered at @xmath30 ( open square ) . ] figure [ fig3](a ) shows the spatial structure of the triplet pair amplitude @xmath122 with respect to the relative coordinate @xmath145 . noting that the pairing symmetry is determined by the angular dependence in relative - momentum space , we find that the induced pair amplitude has the @xmath1-wave symmetry . that is , the pair amplitude has the @xmath148-wave ( @xmath149-wave ) symmetry , when the center of mass position is on the @xmath110-axis ( @xmath141-axis ) . an advantage of the cold fermi gas system is that one can tune the pairing interaction by adjusting the threshold energy of a feshbach resonance . although this technique is usually used to adjust the interaction strength for a fixed interaction channel , one may also use this technique to change the interaction channel from the @xmath0-wave one to a @xmath1-wave one . in an @xmath0-wave superfluid fermi gas with triplet pair amplitude shown in fig . [ fig3](a ) , when one suddenly change the @xmath0-wave interaction to the @xmath1-wave one@xcite , @xmath150 the @xmath1-wave superfluid order parameter , @xmath151 would immediately become finite . here , @xmath152 is the fourier - transformed triplet pair amplitude with respect to the relative coordinate @xmath153 . we emphasize that this triplet pair amplitude has already existed before the change of the interaction . thus , just after this manipulation , we expect the spatial structure of the induced @xmath1-wave superfluid order parameter schematically shown in fig . [ fig3](b ) . the @xmath0-wave superfluid order parameter immediately disappears because of the vanishing @xmath0-wave interaction ( @xmath12 ) , and the @xmath0-wave pair amplitude @xmath154 only remains . thus , at least just after this manipulation , by definition , the system is in the @xmath1-wave superfluid state with the synthesized @xmath1-wave superfluid order parameter in eq . ( [ eq.26e ] ) . this unconventional superfluid phase would be in the non - equilibrium state , so that we need further analyses on the time evolution of this state . however , the combined feshbach technique with the induced triplet pair amplitude is an interesting idea to realize a @xmath1-wave superfluid fermi gas . -wave superfluid fermi gas with trap - potential difference . ( a ) triplet component @xmath155 . the intensity is magnified to ten times . ( b ) singlet component @xmath103 . ] figure [ fig4 ] shows the local condensate fraction @xmath156 in an @xmath0-wave superfluid fermi gas with trap - potential imbalance . in panel ( a ) , the triplet component @xmath155 is enhanced around @xmath157 , as well as the region near the trap center ( except at @xmath50 , where the triplet condensate fraction vanishes ) . on the other hand , panel ( b ) shows that the singlet component @xmath103 monotonically decreases , as one goes away from the trap center . the latter behavior is consistent with the spatial variation of the @xmath0-wave superfluid order parameter @xmath120 shown in fig . [ fig1](b ) . in an @xmath0-wave superfluid fermi gas with trap - potential difference . ( a ) triplet component @xmath109 . ( b ) singlet component @xmath105 . ( c ) @xmath109 and @xmath105 , as functions of the interaction strength @xmath158 , when @xmath114 . in panels ( a ) and ( b ) , the region above the solid line is in the superfluid state within our numerical accuracy . ( note that we take @xmath159 in our numerical calculations . ) the region between the solid line and the dashed line is in the fflo phase , being characterized by a spatially oscillating superfluid order parameter @xmath120 . these lines are also drawn in fig.[fig6 ] . ] the large triplet condensate fraction @xmath155 near the trap center seen in fig . [ fig4](a ) is due to the spin imbalance ( @xmath160 ) in the trap center . ( see fig . [ fig1](c ) . ) this naturally leads to the broken spin rotation symmetry through the fermi chemical potential @xmath86 , as well as the hartree potential @xmath161 in eq . ( [ eq.18 ] ) . thus , although two spin components feel almost the same trap potential ( @xmath162 ) around the trap center , the triplet condensate fraction is enhanced there ( except at @xmath50 ) . while the difference @xmath163 becomes remarkable as one goes away from the trap center , the spin imbalance ( @xmath164 ) becomes small to eventually vanish at @xmath165 . ( see fig . [ fig1](c ) . ) in the outer region , the spin imbalance again occurs as @xmath166 . these enhance @xmath167 , as shown in fig . [ fig4](a ) . in a trapped @xmath0-wave superfluid fermi gas in the presence of mass imbalance @xmath168 . ( a ) triplet component @xmath109 . ( b ) singlet component @xmath105 . ( c ) @xmath109 and @xmath105 , as functions of the interaction strength @xmath169 , when @xmath170 . ] summing up the local condensate fraction @xmath156 in the gas cloud , one obtains the condensate fraction @xmath171 in fig . as expected , panel ( a ) shows that the triplet component @xmath109 is enhanced when @xmath172 . we also find that @xmath109 becomes large in the intermediate coupling regime but becomes small when @xmath173 . in the strong coupling regime , most fermi atoms form singlet molecules , which suppresses effects of broken inversion and spin rotation symmetry . indeed , [ fig5](b ) shows that the singlet component @xmath105 monotonically increases with increasing the interaction strength @xmath158 . to clearly see the difference between @xmath109 and @xmath105 , fig . [ fig5](c ) shows these quantities as functions of the interaction strength @xmath158 . in figs . [ fig5](a ) and ( b ) , one sees the fflo ( fulde - ferrell - larkin - ovchinnikov ) phase@xcite . in this regard , since we are dealing with a two - dimensional lattice model within the simple mean - field theory , it is unclear whether or not the fflo phase still remains in a realistic three - dimensional continuum fermi gas@xcite . however , since fig . [ fig5](a ) indicates that the triplet condensate fraction is also induced in the ordinary bcs region , we find that the fflo state is not necessary for the triplet pair amplitude to appear . in a trapped @xmath0-wave superfluid fermi gas with spin imbalance @xmath174 . ( a ) triplet component @xmath109 . ( b ) singlet component @xmath105 . ( c ) @xmath109 and @xmath105 , as functions of the interaction strength @xmath169 , when @xmath175/n=0.17 $ ] ( @xmath176 ) . in panels ( a ) and ( b ) , the region above the solid line is in the superfluid phase . in the superfluid region shown in this figure , the superfluid order parameter in the outer region of the gas cloud always exhibits a fflo - type oscillation in the radial direction . ] and the triplet condensate fraction @xmath177 in a trapped @xmath0-wave superfluid fermi gas with spin imbalance . ( a ) @xmath178 . ( b ) @xmath179 . we set @xmath180 . ] figure [ fig6 ] confirms that the triplet pair amplitude @xmath122 is also induced , when the spin rotation symmetry is broken by mass imbalance ( @xmath168 ) . in addition , fig . [ fig7 ] shows that this phenomenon also occurs in a trapped @xmath0-wave superfluid fermi gas with spin imbalance ( @xmath174)@xcite . in the latter case , the fermi chemical potential @xmath86 depends on pseudospin @xmath5 , which breaks the spin rotation symmetry@xcite . in the presence of spin imbalance , the phase separation is known to occur@xcite , where the superfluid region in the trap center is surrounded by excess atoms . figure [ fig8](a ) shows this case . in this panel , since the spin imbalance is almost absent around the trap center , the triplet condensate fraction is suppressed there , compared to the case when the phase separation does not occur ( panel ( b ) ) . in addition , the region around the edge of the gas cloud is highly spin polarized , so that the triplet pair amplitude is also suppressed there . as a result , when the phase separation occurs , the triplet pair amplitude is localized around the edge of the gas cloud of the minority component ( @xmath181 ) , as shown in fig . [ fig8](b ) . to summarize , we have discussed the possibility of inducing a triplet pair amplitude in a trapped @xmath0-wave superfluid fermi gas . using symmetry considerations , we clarified that both the broken spatial inversion symmetry and a broken spin rotation symmetry are necessary for this phenomenon to occur . we numerically confirmed that a triplet pair amplitude is induced when this condition is satisfied , within the mean - field theory for a two - dimensional lattice model . in this confirmation , we considered the three cases with ( 1 ) trap - potential difference , ( 2 ) mass imbalance , and ( 3 ) spin imbalance . in the first case , we showed that the induced triplet pair amplitude is dominated by a @xmath1-wave symmetry . among the above three cases , a trapped @xmath0-wave superfluid fermi gas with spin imbalance has been realized@xcite . thus , our results imply that a triplet pair amplitude is already present in this system , although there is no experimental evidence yet . since the symmetry of a fermi superfluid is fully determined by the symmetry of the superfluid order parameter , the induction of a triplet pair amplitude does not immediately mean the realization of a triplet fermi superfluid . in the present case , the system is still in the @xmath0-wave superfluid state which is characterized by the @xmath0-wave superfluid order parameter , even in the presence of a triplet pair amplitude . under this situation , however , when one suddenly changes the @xmath0-wave pairing interaction to an appropriate @xmath1-wave one , the product of the @xmath1-wave interaction and the triplet pair amplitude that has been induced before this manipulation may immediately give a finite @xmath1-wave superfluid order parameter . since the @xmath0-wave superfluid order parameter vanishes , by definition , we have a @xmath1-wave superfluid state , characterized by this @xmath1-wave superfluid parameter . the change of the interaction would be possible by using the feshbach resonance technique . in cold fermi gas physics , although the realization of a @xmath1-wave superfluid state is a crucial challenge , current experiments are facing various difficulties originating from a @xmath1-wave interaction , such as three particle loss@xcite , as well as the dipolar relaxation@xcite . in this regard , the above idea may avoid these difficulties to some extent , because the triplet pair amplitude is prepared in an @xmath0-wave superfluid fermi gas with no @xmath1-wave interaction . in addition , since we can start from a finite value of the @xmath1-wave superfluid order parameter , the system would be in the @xmath1-wave superfluid state for a while , until it is strongly damaged by the particle loss and dipolar relaxation after the @xmath1-wave interaction is introduced . in this sense , the induction of a triplet pair amplitude discussed in this paper is important , not only as a fundamental physical phenomenon , but also on the viewpoint of the challenge toward the realization of a @xmath1-wave superfluid fermi gas . in this paper , we have treated a lattice model to simply confirm the induction of a triplet pair amplitude . to quantitatively evaluate this quantity , we need to extend the present analyses to a realistic continuum fermi superfluid . to assess the idea that one produces a @xmath1-wave superfluid fermi gas from the induced triplet pair amplitude , it is also important to clarify the time evolution of the @xmath1-wave superfluid order parameter after the @xmath0-wave interaction is replaced by a @xmath1-wave one . these problems remain as our future problems . since the pair amplitude always exists in a fermi superfluid , our results would be useful for the study of this fundamental quantity in cold fermi gas physics . yo thanks n. yamamoto for useful comments . this work was supported by kipas project in keio university . ye and rh were supported by a grant - in - aid for jsps fellows . yo was supported by grant - in - aid for scientific research from mext in japan ( 25105511 , 25400418 , 15h00840 ) . we consider a uniform @xmath0-wave superfluid fermi gas with a spin - orbit interaction . the model hamiltonian is given by @xmath182 here , @xmath183 is a creation operator of a fermi atom with the kinetic energy @xmath184 , measured from the chemical potential @xmath47 . the antisymmetric spin - orbit interaction @xmath185 has the form@xcite , @xmath186 where @xmath187 is assumed as @xmath188+\lambda_z\sigma_z . \label{eq.a3}\ ] ] here , @xmath189 and @xmath190 are spin - orbit couplings . we assume that the system is in the @xmath0-wave superfluid state with the superfluid order parameter @xmath191 , we also assume that any other spontaneous symmetry breaking is absent . in momentum space , the spatial inversion @xmath192 is described as @xmath193 . under this operation , each term in eq . ( [ eq.a1 ] ) is invariant , except for the spin - orbit interaction , which is transformed as @xmath194 thus , the spin - orbit interaction @xmath185 in eq . ( [ eq.a2 ] ) breaks the inversion symmetry . when @xmath208 , the spin - orbit interaction @xmath185 is not invariant under any @xmath59 rotations @xmath209 . within this analysis , one concludes that all the triplet pair amplitudes in eq . ( [ eq.a7 ] ) may be induced . however , within the mean - field theory , ref.@xcite shows that the component with @xmath79 is not induced when @xmath208 and @xmath210 . this is because , in this two - component case , the mean - field bcs hamiltonian , @xmath211 , \label{eq.a9}\ ] ] is invariant under the momentum dependent @xmath59 spin - rotation which is followed by the @xmath212 gauge transformation , given by @xmath213 where @xmath214 . in this case , the triplet pair amplitude with @xmath79 is transformed as @xmath215 , so that one finds @xmath216 , as obtained in ref.@xcite . the mean - field bcs hamiltonian in eq . ( [ eq.18 ] ) can be diagonalized by the bogoliubov transformation in real space , given by @xmath217 here , @xmath218 is a @xmath219 orthogonal matrix , which is chosen so that @xmath119 in eq . ( [ eq.18 ] ) can be diagonalized as @xmath220 where @xmath221 is a bogoliubov single - particle excitation energy . after the diagonalization , the superfluid order parameter @xmath120 , as well as the number density @xmath222 , are evaluated as , respectively @xmath223 @xmath224 @xmath225 where @xmath226 $ ] is the fermi distribution function . the number @xmath127 of fermi atoms in the @xmath95-spin component is given by @xmath227 we numerically calculate eqs . ( [ eq.19 ] ) , and ( [ eq.21])-([eq.24 ] ) , to self - consistently determine @xmath120 , @xmath121 , and @xmath86 . the triplet pair amplitude @xmath122 in eq . ( [ eq.25 ] ) , as well as the singlet pair amplitude @xmath154 in eq . ( [ eq.26 ] ) , are then calculated as , respectively , @xmath228 , \label{eq.25a}\end{aligned}\ ] ] @xmath229 . \label{eq.26a}\end{aligned}\ ] ] 99 c. a. regal , m. greiner , and d. s. jin , phys 92 * , 040403 ( 2004 ) . m. w. zwierlein , c. a. stan , c. h. schunck , s. m. f. raupach , a. j. kerman , and w. ketterle , phys . lett . * 92 * , 120403 ( 2004 ) . j. kinast , s. l. hemmer , m. e. gehm , a. turlapov , and j. e. thomas , phys . lett . * 92 * , 150402 ( 2004 ) . m. bartenstein , a. altmeyer , s. riedl , s. jochim , c. chin , j. h. denschlag , and r. grimm , phys . lett . * 92 * , 203201 ( 2004 ) . m. w. zwierlein , a. schirotzek , c. h. schunck , and w. ketterle , science , * 311 * , 492 ( 2006 ) . g. b. partridge , w. li , r. i. kamar , y .- a . liao , and r. g. hulet , science * 311 * , 503 ( 2006 ) . y. shin , c. h. schunck , a. schirotzek , and w. ketterle , nature ( london ) * 451 * , 689 ( 2008 ) . for a review , see , c. chin , r. grimm , p. julienne , and e. tiesinga , rev . phys . * 82 * , 1225 ( 2010 ) . c. a. regal , c. ticknor , j. l. bohn , and d. s. jin , phys . * 90 * , 053201 ( 2003 ) . c. ticknor , c. a. regal , d. s. jin , and j. l. bohn , phys . a * 69 * , 042712 ( 2004 ) . j. zhang , e. g. m. van kempen , t. bourdel , l. khaykovich , j. cubizolles , f. chevy , m. teichmann , l. tarruell , s. j. j. m. f. kokkelmans , and c. salomon , phys . a * 70 * , 030702(r ) ( 2004 ) . t. nakasuji , j. yoshida , and t. mukaiyama , phys . a * 88 * , 012710 ( 2013 ) . j. p. gaebler , j. t. stewart , j. l. bohn , and d. s. jin , phys . * 98 * , 200403 ( 2007 ) . y. inada , m. horikoshi , s. nakajima , m. kuwata - gonokami , m. ueda , and t. mukaiyama , phys . * 101 * , 100401 ( 2008 ) . j. fuchs , c. ticknor , p. dyke , g. veeravalli , e. kuhnle , w. rowlands , p. hannaford , and c. j. vale , phys . a * 77 * , 053616 ( 2008 ) t. yamaguchi and y.ohashi , arxiv:1504.03835 . for a review , see , j. dalibard , f. gerbier , g. juzeli@xmath230nas , and p. hberg , rev . mod . phys . * 83 * , 1523 ( 2011 ) . lin , r. l. compton , a. r. perry , w. d. phillips , j. v. porto , and i. b. spielman , phys . lett . * 102 * , 130401 ( 2009 ) . lin , r. l. compton , k. jimnez - garca , j. v. porto , and i. b. spielman , nature * 462 * , 628 ( 2009 ) . lin , r. l. compton , k. jimnez - garca , w. d. phillips , j. v. porto , and i. b. spielman , nat . * 7 * , 531 ( 2011 ) . lin , k. jimnez - garca , and i. b. spielman , nature * 471 * , 83 ( 2011 ) . l. w. cheuk , a. t. sommer , z. hadzibabic , t. yefsah , w. s. bakr , and m. w. zwierlein , phys . rev . lett . * 109 * , 095302 ( 2012 ) . p. wang , z .- q . yu , z. fu , j. miao , l. huang , s. chai , h. zhai , and j. zhang , phys . lett . * 109 * , 095301 ( 2012 ) . j. d. sau , r. sensarma , s. powell , i. b. spielman , s. dassarma , phys . b * 83 * , 140510(r ) ( 2011 ) . b. m. anderson , i. b. spielman , and g. juzelinas , phys . lett . * 111 * , 125301 ( 2013 ) . e. bauer , g. hilscher , h. michor , ch . paul , e. w. scheidt , a. gribanov , yu . seropegin , h. nol , m. sigrist , and p. rogl , phys . * 92 * , 027003 ( 2004 ) . v. p. mineev and m. sigrist , in _ non - centrosymmetric superconductors _ , edited by e. bauer and m. sigrist ( springer - verlag , berlin , 2012 ) chapter 4 . h. q. yuan , d. f. agterberg , n. hayashi , p. badica , d. vandervelde , k. togano , m. sigrist , and m. b. salamon , phys . . lett . * 97 * , 017006 ( 2006 ) . s. nascimbne , n. navon , s. pilati , f. chevy , s. giorgini , a. georges , and c. salomon , phys . lett . * 106 * , 215303 ( 2011 ) . t. kashimura , r. watanabe , and y. ohashi , phys . a * 86 * , 043622 ( 2012 ) , _ ibid _ , * 89 * , 013618 ( 2014 ) . m. taglieber , a .- c . voigt , t. aoki , t. w. hnsch , and k. dieckmann , phys . lett . * 100 * , 010401 ( 2008 ) . e. wille , f. m. spiegelhalder , g. kerner , d. naik , a. trenkwalder , g. hendl , f. schreck , r. grimm , t. g. tiecke , j. t. m. walraven , s. j. j. m. f. kokkelmans , e. tiesinga , and p. s. julienne , phys . lett . * 100 * , 053201 ( 2008 ) . f. m. spiegelhalder , a. trenkwalder , d. naik , g. hendl , f. schreck , and r. grimm , phys . * 103 * , 223203 ( 2009 ) . n. takemori and a. koga j. phys . . jpn . * 81 * , 063002 ( 2012 ) . lin , w. yi and l. -m . duan , phys . a * 74 * , 031604(r ) ( 2006 ) . dao , a. georges , and m. capone , phys . rev . b * 76 * , 104517 ( 2007 ) . m. a. cazalilla , a. f. ho , and t. giamarchi , phys . 95 * , 226402 ( 2005 ) . r. hanai , t. kashimura , r. watanabe , d. inotani , and y. ohashi , phys . a * 88 * , 053621 ( 2013 ) . r. hanai and y. ohashi , phys . a * 90 * , 043622 ( 2014 ) . l. jiang , x. j. liu , h. hu , and h. pu , phys . rev . a * 84 * , 063618 ( 2011 ) . h. hu , l. jiang , x .- j . liu , and h. pu , phys . * 107 * , 195304 ( 2011 ) . c. n. yang , rev . phys . * 34 * , 694 ( 1962 ) . l. salasnich , n. manini , and a. parola , phys . rev . a * 72 * , 023621 ( 2005 ) . g. e. astrakharchik , j. boronat , j. casulleras , and s. giorgini , phys . 95 * , 230405 ( 2005 ) . n. fukushima , y. ohashi , e. taylor , and a. griffin , phys . a * 75 * , 033609 ( 2007 ) . p. g. de gennes , _ superconductivity of metals and alloys _ ( addison - wesley , new york , 1989 ) . y. ohashi , phys . rev . a * 83 * , 063611 ( 2011 ) . y. ohashi , phys . * 94 * , 050403 ( 2005 ) . ho and r. b. diener , phys . rev . * 94 * , 090402 ( 2005 ) . p. fulde , and r. a. ferrell , phys . rev . * 135 * , a550 ( 1964 ) . a. i. larkin , and yu . n. ovchinnikov , sov . jetp * 20 * , 762 ( 1965 ) . s. takada , and t. izuyama , prog . . phys . * 41 * , 635 ( 1969 ) . t. mizushima , m. ichioka , and k. machida , j. phys . 76 * , 104006 ( 2007 ) . y. ohashi , j. phys . . jpn . * 71 * , 2625 ( 2002 ) . the oscillating behavior of the phase boundary ( solid line ) in figs . [ fig7](a ) and ( b ) is due to discrete energy levels in a harmonic trap potential . to explain this in a simple manner , when we consider a two - dimensional harmonic potential @xmath231 , one particle energy is given by @xmath232 $ ] ( @xmath233 ) . in this case , the degeneracy of an eigen - energy @xmath234 $ ] equals @xmath235 . when degenerate energy levels below @xmath236 are fully occupied by @xmath237-spin atoms or @xmath238-spin atoms , the energy gap @xmath239 suppresses the superfluid phase transition to some extent , leading to the oscillation of the phase boundary in figs . [ fig7](a ) and ( b ) . the situation , @xmath80 , actually occurs also in the case with trap - potential difference , as well as the case with mass imbalance . figure [ fig7 ] indicates that a triplet pair amplitude is induced even when the difference of the chemical potentials only exists . j. levinsen , n. r. cooper , and v. gurarie , phys . lett . * 99 * , 210402 ( 2007 ) . m. jona - lasinio , l. pricoupenko , and y. castin , phys . a * 77 * , 043611 ( 2008 ) . j. levinsen , n. r. cooper , and v. gurarie , phys . a * 78 * , 063616 ( 2008 ) . j. p. gaebler , j. t. stewart , j. l. bohn , and d. s. jin , phys . 98 * , 200403 ( 2007 ) .
the radio - loud gravitational lenses b1600 + 434 and b1608 + 656 were discovered in the cosmic lens all sky survey @xcite . b1608 + 656 was also discovered independently by @xcite . b1600 + 434 is a two image system separated by 14 with source redshift @xmath9 @xcite . the lens galaxy of b1600 + 434 is an edge - on early - type spiral galaxy at @xmath10 @xcite . b1608 + 656 is a four image system about 21 across with a source redshift at @xmath11 @xcite and a lens redshift at @xmath12 @xcite . the lens of b1608 + 656 consists of two interacting elliptical galaxies within the einstein ring . in addition , _ hst _ observations have detected extended images of the agn host galaxies in both systems @xcite . one of the important applications of gravitational lenses is to constrain the hubble constant @xcite or alternatively , the dark matter distribution in the lens galaxy , through time - delay measurements between the lensed images . the time - delays in both b1600 + 434 and b1608 + 656 have been measured @xcite . modeling of the two gravitational lenses has been carried out by several studies ( e.g. , koopmans & fassnacht 1999 ; koopmans et al . 2003 ; kochanek 2002 , 2003 ) . in most studies , there are model degeneracy problems in that a number of models can reproduce the limited number of constraints and predict different hubble constant values based on each time delay measurement . therefore , it is important to map the mass distribution close to the lens galaxy with other independent methods to better constrain the lens models . in many lenses , galaxy groups or clusters close to the lenses provide non - negligible contribution to the lens potential @xcite . x - ray observations are particularly important in this respect because they can accurately constrain the positions and masses of the galaxy groups or clusters . in particular , the centroid of the x - ray emission from the intracluster gas provides a more accurate measurement of the cluster s position than optical studies . x - ray clusters or groups have been observed in the lenses rx j0911.4 + 0551 , q0957 + 561 , b1422 + 231 , and pg 1115 + 080 @xcite . the superb angular resolution of _ chandra _ also allows us to resolve the lensed quasar images if they are separated by more than 05 . thus , besides constraining the properties of galaxy groups or clusters , x - ray observations of gravitational lenses can also be used to constrain the sizes of quasar x - ray emission regions @xcite , to measure ultra - short time delays between lensed images @xcite , and to study the interstellar medium of the lens galaxies ( e.g. , dai et al . in addition , the flux magnification provided by gravitational lensing facilitates studies of objects that are intrinsically x - ray faint , such as broad absorption line quasars @xcite and high redshift quasars @xcite . in this paper , we present results from _ chandra _ observations of b1600 + 434 and b1608 + 656 . we discuss the data acquisition and processing in @xmath132 , the spectral properties of the lensed quasars in @xmath133 , our limits on the presence of x - ray luminous groups or clusters in @xmath134 , and the serendipitous sources in the fields in @xmath135 . we assume cosmology with @xmath14 , @xmath15 , and @xmath16 throughout the paper . we observed b1600 + 434 and b1608 + 656 with the advanced ccd imaging spectrometer ( acis ; garmire et al . 2003 ) onboard _ chandra _ @xcite for @xmath17 ks and @xmath18 ks on 2003 october 7 , and 2003 september 21 , respectively . the data were taken continuously with no interruptions during each observation , and there were no significant background flares during the observations . both gravitational lenses were placed on the back - illuminated acis - s3 chip , where the aim points of the observations are located , and the data were taken in the timed / vfaint mode . b1600 + 434 was placed near the aim point of the observation ( @xmath198 off - axis ) and b1608 + 656 was placed much further from the aim point about @xmath1975 off - axis . the latter lens was shifted off - axis to ensure a galaxy clump in the _ hst _ images would stay in the field . this configuration was used to maximize the sensitivity for detecting nearby galaxy groups or clusters that might contribute to the lensing and affect estimates of the hubble constant . the data were reduced with the ` ciao 3.1 ` software tools provided by the _ chandra x - ray center _ ( cxc ) . we improved the image quality of the data by removing the pixel randomization applied to the event positions in the cxc processing and by applying a subpixel resolution technique @xcite . in the data analysis , only events with standard _ asca _ grades of 0 , 2 , 3 , 4 , and 6 were used . we first discuss the x - ray properties of the lensed quasars in both lens systems . the imaging and spectral analysis are discussed in @xmath13[sec : img ] and @xmath13[sec : spec ] and the x - ray flux ratios of the lensed images are estimated in @xmath13[sec : fr ] . we analyzed images limited to the 0.28 kev band where we detected a total of 332 and 84 net events for b1600 + 434 and b1608 + 656 , respectively . the raw and smoothed _ chandra _ images of b1600 + 434 and b1608 + 656 are shown in figure [ fig : img ] . both of the raw images are binned with a bin size of 015 . the smoothed images are generated by binning the images with a bin size of 01 for both lenses and then were smoothed with gaussians of width @xmath20 02 and @xmath20 03 for b1600 + 434 and b1608 + 656 , respectively . in b1600 + 434 we detected both lensed images . the measured image separation ( 135@xmath21005 ) is consistent with the separation measured in the cfa - arizona space telescope lens survey ( castles ) _ hst _ observations . the situation is more complicated in b1608 + 656 due to the large number of images with smaller minimum separations , but also because the count rate of the lens is lower . the relatively larger off - axis angle does not degrade the point spread function ( psf ) very much . we simulated the psf using the ` chart ` @xcite and ` marx ` @xcite tools and found the 50% encircled energy contour is of @xmath1905 radius as opposed to @xmath1904 on axis . three out of four lensed images are clearly detected . image b is well separated , but images a and c are merged . we fit the image assuming there are three point images of a , b , and c constraining the relative quasar position to match the castles @xmath22-band relative position and using the above mentioned psf model . however , we are unable to obtain a stable solution . this may be due to the low count - rate of the source or it may indicate that the images are not point sources . we note that there are some x - ray events about 1 northwest to image a and we are unable to determine whether these events belong to image a or from other unidentified sources . the _ hst _ images do not show any source at the corresponding position other than part of the einstein ring image of the host galaxy . image d is not detected by the _ chandra _ observation . the non - detection of image d could be caused by small number statistics additional absorption is not required to mask the image . we extracted spectra for each quasar using the ` ciao ` tool ` psextract ` and fit them using ` xspec v11.3.1 ` @xcite within the 0.48 kev observed energy range . we note that in ` ciao 3.1 ` the recently observed decline in the low energy quantum efficiency of acis , possibly caused by molecular contamination of the acis filters , is already accounted for by the ` psextract ` tool . therefore , we did not perform any other corrections beyond that . we extracted both the combined spectrum of both images within a circle with 5 radius about the centroid of the combined image and individual spectra of images within 07 radius circles about the centroids of images a and b , respectively . we extracted the background spectrum within an annulus with inner and outer radii of 10 and 30 . we fitted the spectra of b1600 + 434 with a series of models summarized in table [ tab : spec ] . we first modeled the spectrum as a power - law modified by galactic absorption . this model gives a reasonable fit with @xmath23 for 18 degrees of freedom ( dof ) . however , the column density obtained for galactic absorption , @xmath24 @xmath25 , is much higher than the expected @xmath26 @xmath25 @xcite . this exercise indicates that additional absorption components are needed to interpret the spectrum . these additional absorption components can be either intrinsic to the quasar , located in the quasar host galaxy , or in the lens galaxy . we found that adding a neutral absorption component at the redshift of either the quasar or the lens can produce an acceptable fit with @xmath27 for 18 dof and @xmath28 for 18 dof , respectively , for absorption at the quasar and lens redshifts . the absorption column densities obtained are @xmath29 @xmath25 and @xmath30 @xmath25 for absorption at the quasar or lens redshift , respectively . the combined spectrum and the best fit model with absorption at the lens redshift are shown in figure [ fig : spec]a . the estimate of the intrinsic photon index is around @xmath31 for both models with observed 0.48 kev band flux of @xmath3210@xmath33 , and lensed , unabsorbed x - ray luminosities in the 210 kev and 120 kev band rest frame of @xmath3410@xmath35 and @xmath410@xmath5 , respectively . we note that the lensing magnification must be included in order to obtain the unlensed luminosity . @xcite interpreted the wavelength dependence of the image flux ratios as due to @xmath8 of differential extinction obscuring image b. we also see in the spectra of the image that image b has proportionally less soft x - ray emission so that differential absorption must be intrinsic to the lens . to estimate this differential absorption we simultaneously fit the individual spectra of images a and b constrained to have the same intrinsic power - law photon index and galactic absorption column density but allowed to have different absorption column densities at the redshift of the lens . the best fit model is shown in figure [ fig : spec]b . this model also results in acceptable fit with a @xmath36 for 12 dof as compared to @xmath37 for 13 dof if we do not allow for differential absorption . the improvement of the fit by allowing for differential absorption is significant at the 90% confidence level according to the f - test indicating that the x - ray data are consistent with , but do not require , column density variations . the photon index obtained in this fit is slightly steeper than for the previous fits in part because the small extraction regions used to obtain individual spectra of the images to the loss of some hard x - ray photons . the absorption column densities at the lens for images a and b are @xmath38 @xmath25 and @xmath39 @xmath25 , respectively , and the differential absorption between the two images is @xmath7 as shown in figure [ fig : con ] . this indicates that the determined from the two spectra are different at the 90% confidence level with higher for image b. this is consistent with the lens configuration , where the lens galaxy is much closer to image b than to image a. comparing with the estimated differential extinction of @xmath8 mag between images a and b @xcite , we obtained a dust - to - gas ratio of @xmath40 within a factor of five from the galactic value , @xmath41 @xcite . we extracted the combined spectrum of all the images of b1608 + 656 within a 5 radius circle centered on the source . we extracted the background spectrum within an annulus with inner and outer radii of 10 and 18 . the background subtracted spectrum consists of only 86 net events with a count - rate of 2.9@xmath4210@xmath43 count s@xmath44 , making a spectral analysis of the individual images impossible . we modeled the spectrum with a power - law modified by galactic absorption . this simple model fits the spectrum quite well , with @xmath45 for 3 dof . the parameters are presented in table [ tab : spec ] and the spectrum and the best fit model are shown in figure [ fig : spec]c . the power - law photon index and galactic are constrained to be @xmath3 and @xmath46 @xmath25 consistent with the expected galactic column density of @xmath47 @xmath25 @xcite . if we fix the galactic absorption to this value the fits are consistent with no additional absorption at either the lens or quasar redshift with upper limits on absorption column of @xmath48 and @xmath49 @xmath25 at the redshift of lens or quasar , respectively . there is differential extinction in b1608 + 656 as well , but it is most significant for image d which we failed to detect . our upper limit on the flux of d is not strong enough to estimate a dust - to - gas ratio . the x - ray flux in the observed 0.48 kev band is @xmath5010@xmath33 , and the lensed , unabsorbed x - ray luminosities in the 2 - 10 kev and 119 kev band rest frame are @xmath5110@xmath35 and @xmath5210@xmath35 , respectively . cccccccc 1 & b1600 + 434 & total & @xmath53 & @xmath54 & & @xmath55 & 16.3(18 ) + 2 & b1600 + 434 & total & @xmath56 & 1.29 ( fixed ) & @xmath57 @xmath58 & @xmath59 & 16.1(18 ) + 3 & b1600 + 434 & total & @xmath56 & 1.29 ( fixed ) & @xmath60 @xmath61 & @xmath62 & 15.8(18 ) + 4 & b1600 + 434 & a & @xmath63 & 1.29 ( fixed ) & @xmath64 @xmath61 & @xmath65 & 13.1(12 ) + & & b & & & @xmath66 @xmath61 & @xmath67 & + 5 & b1608 + 656 & total & @xmath68 & @xmath69 & & @xmath70 & 3.3(3 ) + we estimated the absorption corrected flux ratio between images a and b of b1600 + 434 using ` xspec ` in the 0.48 kev band and obtained a flux ratio of b / a = @xmath71 . we compare this to the time delay corrected flux ratios obtained for the optical and radio bands @xcite in table [ tab : frzero ] . for b1608 + 656 we estimated the flux ratios based on the image count rates because of the low s / n of individual images . in addition , we estimated the count rate of combined images a and c since it is difficult to resolve the two images . the result for b1608 + 656 is listed in table [ tab : freight ] . the flux ratios in x - rays are consistent with the time delay corrected flux ratio in the radio band in b1608 + 656 @xcite . cc x - ray ( 0.48 kev ) & @xmath71 + optical ( i ) & 0.67 + radio ( 8.5 ghz ) & 0.825 + ccc x - ray ( 0.48 kev ) & @xmath72 & @xmath73 + radio ( 8.5 ghz ) & 0.325 & 0.114 + it is important to constrain the mass and central location of any nearby groups or clusters to construct a better lens model . we performed source detection on the two _ chandra _ observations over the full acis fields using the ` ciao ` tool ` wavdetect ` @xcite provided by cxc and did not find significant extended x - ray emission in either field . we also smoothed the image with the ` csmooth ` @xcite software tool to detect any extended sources within 4 of the lens galaxies in the acis - s3 and s2 fields in both observations . we removed the point sources detected by ` wavdetect ` before the smoothing process . however , the smoothed images do not show significant ( above @xmath74 ) extended emission close to the lens galaxies in either field . we also smoothed the soft x - ray band images between ( 0.42 kev ) to increase the signal - to - noise ratio , as the x - ray emission from the cluster or group is expected to be stronger in the soft x - ray band , and did not find significant extended emission above @xmath74 . we note that there are two low significance extended sources ( about @xmath75 ) present close the lenses at r.a . = 16@xmath76 1@xmath77 359 and decl . = 43 17 27 ( j2000 ) with a @xmath78 contour radius of @xmath1911 in the field of b1600 + 434 , and at r.a . = 16@xmath76 9@xmath77 113 and decl . = 65 32 7 ( j2000 ) with a @xmath78 contour radius of @xmath195 in the field of b1608 + 656 . however , these sources have low signal - to - noise ratios and could be simply be background fluctuations . in addition , many events in these peaks have energies below 0.4 kev where the calibration uncertainty is large . a longer observation would be required to have any confidence in their existence , so we proceed on the assumption that they are background fluctuations . we used the background count rates to estimate upper limits on the fluxes of any group / cluster in the two fields . we assumed gas temperature of @xmath191 kev when converting from count rates to flux . this temperature consistent with the upper limits of x - ray luminosities obtained from our analysis . the background count rate in the 0.42 kev band for b1600 + 434 is @xmath79 . in the analysis , we used an aperture size of 100 kpc radius to limit the effects of any systematic errors in the background estimate . this aperture size is large enough for the groups with luminosities below @xmath80 @xcite . for a 100 kpc radius at the redshift of the lens ( @xmath10 ) , the @xmath74 upper limit of the x - ray flux is @xmath81 , which corresponds to an x - ray luminosity of @xmath82 the background flux in the 0.42 kev band for b1608 + 656 is @xmath83 . for a 100 kpc radius at the redshift of the lens ( @xmath12 ) , the @xmath74 upper limit of the x - ray flux is @xmath84 , which corresponds to an x - ray luminosity of @xmath85 we used the standard @xmath86@xmath87 , @xmath88@xmath87 , and @xmath88@xmath86 relations ( e.g. , girardi et al . 1996 ; helsdon & ponman 2000 ; xue & wu 2000 ; mahdavi & geller 2001 ; kochanek et al . 2003 ; osmond & ponman 2004 ) for groups and clusters to estimate upper limits for the velocity dispersion of any cluster of @xmath89 and @xmath90 for b1600 + 434 and b1608 + 656 respectively . the large scatter of the relations were considered when estimating the upper limits . considering a singular isothermal sphere model for the group or cluster , the lens potential , @xmath91 , from the group / cluster is @xmath92 where @xmath93 and @xmath94 are angular diameter distances from the lens to source and observer to source , and @xmath95 is the angular distance from the lens to the group / cluster . thus , the lens strength , b , is constrained to have upper limits of @xmath9630 and @xmath9732 for b1600 + 434 and b1608 + 656 respectively . we also cataloged the serendipitous point sources in the two fields using the ` ciao ` tool ` wavdetect ` . the source detection was performed on each ccd separately with a detection threshold of @xmath98 . the detection threshold was set as the inverse of the total number of pixels in the binned images . we detected 75 and 82 serendipitous sources in the b1600 + 434 and b1608 + 656 fields , respectively . the detection threshold corresponds to a flux limit of roughly @xmath99 and the numbers of sources are consistent with standard estimates for the source numbers at this flux limit ( e.g. , * ? ? ? * and references there in ) . the _ chandra _ field of b1600 + 434 is included in dr3 of the sloan digital sky survey ( sdss , * ? ? ? we searched for optical counterparts of the x - ray sources within a 2 radius , which corresponds to the largest error bars for the _ chandra _ positions , and allows the inclusion of counterparts that have slight offsets between the optical and x - ray centroids , such as the ultra - luminous x - ray sources seen in nearby galaxies . we found counterparts for 27 of the 75 x - ray sources , 13 of which are flagged as quasar candidates in the sdss catalogs . the properties of the serendipitous sources and the optical counterparts are listed in table [ tab : szero ] . b1608 + 656 field is not covered by sdss , and so we used the palomar digital sky survey ( dposs , * ? ? ? * ) survey to identify the counterparts . the lens position in the _ chandra _ and dposs images are offset by approximately 15 , so we used a slightly larger radius to identify optical counterparts . we found 27 optical counterparts out of the 82 x - ray sources . the properties of the serendipitous sources and the optical counterparts in b1608 + 656 field are listed in table [ tab : seight ] . we present results from _ chandra _ observations of the gravitational lenses b1600 + 434 and b1608 + 656 . we resolved the two lensed images in b1600 + 434 and three of four images in b1608 + 656 . we did not detect the faintest image d in b1608 + 656 . the analysis of b1608 + 656 was complicated by the small minimum image separation and the low count - rate of the lensed image d. we could not find a stable solution for the x - ray flux ratios even if we fixed the image positions to those inferred from _ hst _ observations . a longer observation would be needed to constrain the flux ratios of the individual x - ray images of b1608 + 656 . we also performed a spectral analysis of the lensed quasar images and detected differential absorption column density of @xmath7 between the two images of b1600 + 434 . the improvement of the fit by allowing differential absorption is significant at the 90% confidence level indicating that the x - ray data are consistent with , but do not require , column density variations . comparing with the estimated differential extinction of @xmath8 mag between images a and b @xcite , we obtained a dust - to - gas ratio of @xmath40 within a factor of five from the galactic value . considering the patchiness of the interstellar medium , a factor of five difference may not be significant enough to conclude that the dust - to - gas ratio of the lens galaxy in b1600 + 434 is higher than that of the milky way . however , this trend is consistent with the cases in b0218 + 357 ( @xmath100 ) and pks 1830211 ( @xmath101 , * ? ? ? * ) , where the estimated dust - to - gas ratios in the lens galaxies were higher than the galactic value . it is , however , the opposite of the situation found in q2237 + 0305 ( @xmath102 , * ? ? ? a larger sample is needed to investigate the average dust - to - gas ratios in high redshift ( @xmath103 ) galaxies . there is differential extinction in b1608 + 656 as well , but it is most significant for image d which we failed to detect . our upper limit on the flux of d is not significant enough to estimate a limit on the dust - to - gas ratio . the differential absorption measured between different images in gravitational lenses may serve as a convenient probe to study the gas content in high redshift galaxies since it can separate the absorbing column originated from the lens galaxy and those intrinsic to quasars . we did not detect significant x - ray emission from nearby galaxy groups or clusters associated with the lens galaxies . we found upper limits on the x - ray luminosity of any cluster at the lens redshift and within 4 from the lenses of @xmath0 and @xmath1 for b1600 + 434 and b1608 + 656 , respectively . considering standard cluster relations between x - ray luminosity , temperature , and velocity dispersion and assuming a singular isothermal sphere model for the group / cluster , we obtained upper limits of the lens strength parameter , @xmath104 , of 30 and 32 for b1600 + 434 and b1608 + 656 respectively . with such a tight limit , only the poor groups to which each lens belongs could be contributing to the lens potential . cccccccccccc cxo j160022.5 + 431521 & 240.09376 & 43.256002 & 0.7 & 0.5 & 152@xmath21 16 & 13.6 & & & & & + cxo j160041.2 + 431544 & 240.17168 & 43.262303 & 1.0 & 0.8 & 10@xmath21 4 & 2.8 & & & & & + cxo j160044.4 + 430842 & 240.18509 & 43.145258 & 0.6 & 0.2 & 26@xmath21 7 & 4.4 & & & & & + cxo j160045.3 + 431433 & 240.18909 & 43.242736 & 1.1 & 0.8 & 13@xmath21 4 & 3.8 & 240.18867 & 43.242565 & 1.3 & 22.7490 & `` + cxo j160047.0 + 431354 & 240.19615 & 43.231796 & 0.5 & 0.6 & 7@xmath21 3 & 2.4 & & & & & + cxo j160049.2 + 431433 & 240.20518 & 43.242667 & 0.4 & 1.0 & 10@xmath21 4 & 3.3 & & & & & + cxo j160051.2 + 431431 & 240.21355 & 43.242076 & 1.0 & 0.7 & 26@xmath21 6 & 6.3 & & & & & + cxo j160053.8 + 430932 & 240.22429 & 43.159003 & 1.2 & 0.7 & 28@xmath21 7 & 5.3 & & & & & + cxo j160054.8 + 431528 & 240.22854 & 43.258031 & 0.9 & 0.4 & 14@xmath21 4 & 4.6 & 240.22871 & 43.257572 & 1.7 & 15.9220 & `` + cxo j160055.4 + 431052 & 240.23113 & 43.181337 & 0.8 & 0.5 & 6@xmath21 3 & 2.4 & & & & & + cxo j160056.4 + 431151 & 240.23515 & 43.197578 & 1.0 & 0.4 & 9@xmath21 3 & 3.1 & & & & & + cxo j160100.8 + 431055 & 240.25354 & 43.182145 & 0.7 & 0.6 & 11@xmath21 4 & 3.5 & & & & & + cxo j160101.6 + 430108 & 240.25677 & 43.019044 & 2.0 & 0.8 & 12@xmath21 4 & 3.5 & & & & & + cxo j160104.8 + 431115 & 240.27041 & 43.187527 & 0.9 & 0.3 & 8@xmath21 3 & 2.9 & & & & & + cxo j160105.0 + 431850 & 240.27114 & 43.314013 & 0.5 & 0.9 & 4@xmath21 2 & 1.7 & & & & & + cxo j160110.8 + 431139 & 240.29526 & 43.194329 & 0.2 & 0.1 & 299@xmath21 18 & 51.1 & 240.29534 & 43.193954 & 1.4 & 17.1220 & ` target_galaxy ` + cxo j160113.2 + 431803 & 240.30505 & 43.300870 & 0.3 & 0.2 & 8@xmath21 3 & 3.4 & & & & & + cxo j160114.4 + 430717 & 240.31025 & 43.121493 & 0.8 & 0.9 & 9@xmath21 4 & 2.7 & & & & & + cxo j160116.3 + 431740 & 240.31819 & 43.294634 & 0.4 & 0.2 & 11@xmath21 3 & 4.3 & & & & & + cxo j160116.8 + 431834 & 240.32011 & 43.309628 & 0.2 & 0.3 & 14@xmath21 4 & 5.7 & & & & & + cxo j160116.9 + 431949 & 240.32073 & 43.330454 & 0.5 & 0.2 & 10@xmath21 3 & 4.1 & 240.32050 & 43.330372 & 0.7 & 21.8080 & `` + cxo j160117.1 + 432129 & 240.32156 & 43.358059 & 0.4 & 0.2 & 53@xmath21 8 & 14.9 & & & & & + cxo j160120.8 + 431827 & 240.33669 & 43.307585 & 0.1 & 0.1 & 59@xmath21 8 & 22.9 & 240.33658 & 43.307536 & 0.3 & 20.2470 & ` target_qso_faint ` + cxo j160121.0 + 430936 & 240.33777 & 43.160101 & 0.3 & 0.5 & 5@xmath21 2 & 2.1 & & & & & + cxo j160122.8 + 431522 & 240.34521 & 43.256365 & 0.3 & 0.2 & 6@xmath21 2 & 2.7 & & & & & + cxo j160125.6 + 431337 & 240.35703 & 43.227171 & 0.2 & 0.2 & 3@xmath21 2 & 1.5 & & & & & + cxo j160125.9 + 431832 & 240.35828 & 43.308986 & 0.2 & 0.1 & 24@xmath21 5 & 11.4 & 240.35824 & 43.308991 & 0.1 & 21.6330 & `` + cxo j160126.5 + 431704 & 240.36068 & 43.284556 & 0.2 & 0.1 & 8@xmath21 3 & 3.9 & & & & & + cxo j160126.6 + 431313 & 240.36091 & 43.220496 & 0.3 & 0.2 & 11@xmath21 3 & 5.0 & & & & & + cxo j160128.9 + 430519 & 240.37078 & 43.088780 & 1.0 & 0.6 & 11@xmath21 4 & 3.4 & & & & & + cxo j160129.2 + 430748 & 240.37185 & 43.130045 & 0.7 & 0.5 & 17@xmath21 4 & 5.9 & & & & & + cxo j160129.5 + 431940 & 240.37294 & 43.327784 & 0.2 & 0.2 & 7@xmath21 3 & 3.4 & 240.37304 & 43.327680 & 0.5 & 14.1040 & `` + cxo j160129.8 + 431533 & 240.37433 & 43.259386 & 0.2 & 0.1 & 9@xmath21 3 & 4.3 & & & & & + cxo j160130.3 + 431711 & 240.37637 & 43.286424 & 0.0 & 0.2 & 2@xmath21 1 & 1.0 & & & & & + cxo j160131.1 + 431814 & 240.37993 & 43.304018 & 0.2 & 0.2 & 9@xmath21 3 & 4.1 & & & & & + cxo j160132.7 + 431309 & 240.38644 & 43.219338 & 0.1 & 0.2 & 5@xmath21 2 & 2.4 & 240.38645 & 43.219297 & 0.2 & 22.9010 & `` + cxo j160135.3 + 431333 & 240.39740 & 43.226067 & 0.1 & 0.0 & 330@xmath21 18 & 114.4 & 240.39737 & 43.226019 & 0.2 & 20.4010 & ` target_qso_faint ` + cxo j160136.2 + 431353 & 240.40113 & 43.231525 & 0.1 & 0.3 & 4@xmath21 2 & 2.0 & & & & & + cxo j160136.4 + 432107 & 240.40199 & 43.352020 & 0.1 & 0.1 & 71@xmath21 9 & 28.2 & 240.40162 & 43.352124 & 1.0 & 22.3800 & ` target_qso_reject ` + cxo j160137.2 + 431609 & 240.40509 & 43.269321 & 0.1 & 0.0 & 39@xmath21 6 & 18.5 & & & & & + cxo j160137.3 + 432236 & 240.40543 & 43.376842 & 0.2 & 0.2 & 23@xmath21 5 & 8.5 & & & & & + cxo j160138.6 + 432204 & 240.41088 & 43.367895 & 0.4 & 0.3 & 6@xmath21 3 & 2.6 & & & & & + cxo j160141.8 + 431453 & 240.42440 & 43.248288 & 0.4 & 0.1 & 5@xmath21 2 & 2.5 & 240.42461 & 43.248106 & 0.9 & 20.9150 & ` target_qso_faint ` + cxo j160141.9 + 431453 & 240.42459 & 43.248073 & 0.1 & 0.1 & 23@xmath21 5 & 11.8 & 240.42458 & 43.248080 & 0.0 & 20.9290 & ` target_qso_faint ` + cxo j160143.5 + 431858 & 240.43147 & 43.316195 & 0.2 & 0.1 & 7@xmath21 3 & 3.3 & 240.43154 & 43.316113 & 0.4 & 25.7410 & `` + cxo j160143.7 + 431126 & 240.43234 & 43.190683 & 0.3 & 0.4 & 18@xmath21 4 & 7.1 & 240.43238 & 43.190730 & 0.2 & 23.3230 & `` + cxo j160144.6 + 431936 & 240.43594 & 43.326831 & 0.1 & 0.1 & 9@xmath21 3 & 4.4 & & & & & + cxo j160146.4 + 431419 & 240.44346 & 43.238879 & 0.2 & 0.3 & 7@xmath21 3 & 3.5 & & & & & + cxo j160146.8 + 431334 & 240.44529 & 43.226177 & 0.1 & 0.1 & 223@xmath21 15 & 79.9 & 240.44533 & 43.226154 & 0.1 & 21.1140 & `` + cxo j160147.5 + 430907 & 240.44803 & 43.152111 & 0.6 & 0.9 & 12@xmath21 4 & 3.7 & & & & & + cxo j160147.9 + 432006 & 240.44998 & 43.335108 & 0.1 & 0.1 & 19@xmath21 4 & 8.8 & & & & & + cxo j160149.6 + 430836 & 240.45684 & 43.143382 & 0.6 & 0.3 & 24@xmath21 5 & 7.0 & & & & & + cxo j160152.6 + 430915 & 240.46947 & 43.154196 & 0.5 & 0.5 & 49@xmath21 8 & 10.9 & 240.46913 & 43.153922 & 1.3 & 20.5530 & ` target_qso_faint ` + cxo j160153.7 + 431817 & 240.47391 & 43.304920 & 0.0 & 0.0 & 510@xmath21 23 & 173.3 & 240.47392 & 43.304845 & 0.3 & 18.8240 & ` target_qso_faint ` + cxo j160154.5 + 431519 & 240.47735 & 43.255280 & 0.3 & 0.1 & 3@xmath21 2 & 1.5 & 240.47692 & 43.255471 & 1.3 & 20.6500 & ` target_qso_faint ` + cxo j160154.6 + 431147 & 240.47791 & 43.196539 & 0.5 & 0.5 & 6@xmath21 3 & 2.5 & & & & & + cxo j160154.7 + 431019 & 240.47811 & 43.172163 & 0.2 & 0.6 & 3@xmath21 2 & 1.6 & 240.47841 & 43.172037 & 0.9 & 22.8610 & `` + cxo j160156.8 + 431125 & 240.48683 & 43.190556 & 0.2 & 0.2 & 128@xmath21 12 & 32.4 & 240.48681 & 43.190463 & 0.3 & 19.7180 & ` target_qso_faint ` + cxo j160158.0 + 431145 & 240.49198 & 43.195840 & 0.5 & 0.4 & 13@xmath21 4 & 5.3 & & & & & + cxo j160200.1 + 431228 & 240.50070 & 43.207905 & 0.5 & 0.3 & 7@xmath21 3 & 3.3 & & & & & + cxo j160203.4 + 431808 & 240.51433 & 43.302496 & 0.2 & 0.3 & 3@xmath21 2 & 1.5 & & & & & + cxo j160203.9 + 431733 & 240.51654 & 43.292745 & 0.2 & 0.3 & 6@xmath21 2 & 3.0 & & & & & + cxo j160204.2 + 431558 & 240.51789 & 43.266260 & 0.2 & 0.2 & 41@xmath21 6 & 16.8 & 240.51805 & 43.265916 & 1.3 & 20.7600 & `` + cxo j160204.3 + 431556 & 240.51823 & 43.265618 & 0.2 & 0.1 & 26@xmath21 5 & 12.2 & 240.51809 & 43.265900 & 1.1 & 20.9400 & `` + cxo j160208.0 + 431357 & 240.53357 & 43.232582 & 0.4 & 0.3 & 9@xmath21 3 & 4.1 & & & & & + cxo j160209.8 + 431208 & 240.54096 & 43.202248 & 0.5 & 0.5 & 9@xmath21 3 & 3.3 & & & & & + cxo j160210.8 + 430500 & 240.54508 & 43.083499 & 0.7 & 0.6 & 96@xmath21 15 & 7.5 & 240.54492 & 43.083590 & 0.5 & 21.1700 & ` target_qso_faint ` + cxo j160213.0 + 431212 & 240.55434 & 43.203367 & 0.7 & 0.5 & 19@xmath21 5 & 5.4 & 240.55386 & 43.203749 & 1.9 & 21.2800 & ` target_qso_faint ` + cxo j160217.9 + 430659 & 240.57464 & 43.116569 & 0.5 & 0.4 & 62@xmath21 12 & 6.4 & & & & & + cxo j160218.1 + 430928 & 240.57562 & 43.157988 & 0.8 & 0.6 & 62@xmath21 9 & 11.5 & & & & & + cxo j160221.4 + 430800 & 240.58931 & 43.133401 & 0.7 & 0.6 & 58@xmath21 12 & 5.9 & 240.58890 & 43.133187 & 1.3 & 22.0480 & `` + cxo j160221.5 + 431227 & 240.58993 & 43.207556 & 0.8 & 0.6 & 21@xmath21 5 & 5.9 & & & & & + cxo j160221.6 + 430319 & 240.59001 & 43.055445 & 0.3 & 0.2 & 903@xmath21 41 & 30.6 & 240.59003 & 43.054918 & 1.9 & 17.2630 & ` target_galaxy target_qso_cap ` + cxo j160232.1 + 431039 & 240.63379 & 43.177725 & 0.9 & 0.6 & 24@xmath21 7 & 4.0 & & & & & + cxo j160235.5 + 430813 & 240.64792 & 43.136956 & 0.6 & 0.5 & 162@xmath21 20 & 9.9 & 240.64794 & 43.137232 & 1.0 & 20.8500 & ` target_qso_faint ` + ccccccccccc cxo j160713.3 + 652929 & 241.80581 & 65.491391 & 2.0 & 0.9 & 32@xmath21 8 & 4.4 & 241.80481 & 65.491737 & 1.9 & 21.67 + cxo j160719.3 + 652459 & 241.83052 & 65.416517 & 0.9 & 0.4 & 311@xmath21 22 & 22.3 & 241.83031 & 65.416038 & 1.8 & 19.44 + cxo j160727.2 + 653314 & 241.86370 & 65.553915 & 1.9 & 1.0 & 21@xmath21 6 & 4.2 & & & & + cxo j160739.8 + 652040 & 241.91592 & 65.344505 & 2.2 & 1.1 & 20@xmath21 6 & 3.9 & & & & + cxo j160740.4 + 652817 & 241.91870 & 65.471505 & 0.9 & 0.4 & 121@xmath21 12 & 19.8 & 241.91879 & 65.470879 & 2.3 & 21.82 + cxo j160741.5 + 653017 & 241.92300 & 65.504803 & 2.0 & 0.6 & 14@xmath21 5 & 3.6 & & & & + cxo j160747.1 + 652610 & 241.94660 & 65.436200 & 3.1 & 0.8 & 26@xmath21 7 & 5.0 & 241.94743 & 65.435570 & 2.6 & 21.08 + cxo j160803.3 + 652522 & 242.01387 & 65.422904 & 1.5 & 0.3 & 7@xmath21 3 & 2.4 & & & & + cxo j160804.4 + 652540 & 242.01872 & 65.427796 & 2.0 & 0.6 & 14@xmath21 5 & 3.8 & & & & + cxo j160806.5 + 652204 & 242.02720 & 65.367868 & 2.3 & 0.4 & 15@xmath21 5 & 4.2 & & & & + cxo j160807.6 + 652828 & 242.03167 & 65.474558 & 0.4 & 0.2 & 7@xmath21 3 & 2.7 & & & & + cxo j160812.0 + 652054 & 242.05015 & 65.348453 & 1.7 & 0.6 & 12@xmath21 4 & 3.7 & 242.05020 & 65.347809 & 2.3 & + cxo j160813.0 + 652319 & 242.05442 & 65.388855 & 1.8 & 0.6 & 16@xmath21 5 & 4.8 & 242.05438 & 65.389442 & 2.1 & 21.08 + cxo j160816.7 + 652528 & 242.06963 & 65.424496 & 1.3 & 0.7 & 13@xmath21 4 & 4.1 & & & & + cxo j160816.9 + 652453 & 242.07071 & 65.414763 & 1.7 & 0.9 & 10@xmath21 4 & 3.3 & & & & + cxo j160822.0 + 652245 & 242.09175 & 65.379180 & 1.6 & 0.7 & 20@xmath21 5 & 5.2 & & & & + cxo j160823.1 + 652328 & 242.09637 & 65.391192 & 1.2 & 0.4 & 35@xmath21 7 & 8.3 & 242.09776 & 65.390694 & 2.8 & + cxo j160826.7 + 653546 & 242.11160 & 65.596178 & 1.0 & 0.2 & 46@xmath21 8 & 10.5 & 242.11134 & 65.595993 & 0.8 & 21.84 + cxo j160827.3 + 652445 & 242.11378 & 65.412762 & 0.5 & 0.2 & 8@xmath21 3 & 3.4 & & & & + cxo j160831.3 + 652423 & 242.13047 & 65.406416 & 1.2 & 0.7 & 16@xmath21 4 & 5.2 & & & & + cxo j160832.8 + 652255 & 242.13706 & 65.381987 & 1.3 & 0.5 & 11@xmath21 4 & 3.6 & & & & + cxo j160835.1 + 653605 & 242.14658 & 65.601453 & 0.7 & 0.3 & 11@xmath21 4 & 3.9 & & & & + cxo j160835.8 + 653137 & 242.14931 & 65.527038 & 0.7 & 0.2 & 6@xmath21 2 & 2.7 & 242.15018 & 65.527252 & 1.5 & 16.37 + cxo j160836.5 + 652236 & 242.15214 & 65.376812 & 1.2 & 0.7 & 12@xmath21 4 & 3.9 & & & & + cxo j160840.9 + 653523 & 242.17070 & 65.589992 & 0.7 & 0.3 & 9@xmath21 3 & 3.3 & & & & + cxo j160843.8 + 653245 & 242.18292 & 65.546039 & 0.2 & 0.1 & 89@xmath21 10 & 35.4 & 242.18413 & 65.545868 & 1.9 & 19.86 + cxo j160844.1 + 653246 & 242.18413 & 65.546163 & 0.2 & 0.1 & 95@xmath21 10 & 39.3 & 242.18413 & 65.545868 & 1.1 & 19.86 + cxo j160849.1 + 653358 & 242.20462 & 65.566266 & 0.3 & 0.1 & 30@xmath21 6 & 13.4 & & & & + cxo j160850.7 + 653707 & 242.21163 & 65.618791 & 0.8 & 0.4 & 8@xmath21 3 & 3.4 & & & & + cxo j160850.8 + 653220 & 242.21187 & 65.539142 & 0.1 & 0.0 & 617@xmath21 25 & 207.9 & 242.21214 & 65.538887 & 1.0 & 19.78 + cxo j160854.0 + 653338 & 242.22506 & 65.560669 & 0.4 & 0.1 & 12@xmath21 3 & 5.7 & & & & + cxo j160855.0 + 652738 & 242.22933 & 65.460712 & 0.3 & 0.3 & 3@xmath21 2 & 1.5 & & & & + cxo j160858.9 + 652816 & 242.24561 & 65.471199 & 0.6 & 0.2 & 9@xmath21 3 & 4.5 & & & & + cxo j160859.4 + 653225 & 242.24789 & 65.540349 & 0.2 & 0.0 & 69@xmath21 8 & 30.0 & 242.24838 & 65.540154 & 1.0 & 21.07 + cxo j160900.4 + 653445 & 242.25186 & 65.579230 & 0.4 & 0.1 & 19@xmath21 4 & 8.7 & & & & + cxo j160901.6 + 653744 & 242.25685 & 65.629054 & 1.3 & 0.3 & 21@xmath21 5 & 6.5 & & & & + cxo j160902.1 + 653216 & 242.25881 & 65.537796 & 0.7 & 0.1 & 4@xmath21 2 & 1.9 & & & & + cxo j160903.9 + 653040 & 242.26625 & 65.511286 & 0.4 & 0.2 & 8@xmath21 3 & 3.6 & & & & + cxo j160904.8 + 652737 & 242.27018 & 65.460319 & 0.2 & 0.1 & 25@xmath21 5 & 11.8 & & & & + cxo j160905.2 + 653655 & 242.27176 & 65.615366 & 1.0 & 0.2 & 13@xmath21 4 & 4.6 & & & & + cxo j160905.5 + 652400 & 242.27325 & 65.400242 & 1.6 & 0.4 & 6@xmath21 3 & 2.3 & & & & + cxo j160906.1 + 652700 & 242.27560 & 65.450274 & 0.7 & 0.4 & 14@xmath21 4 & 5.7 & & & & + cxo j160907.1 + 653126 & 242.27985 & 65.524103 & 0.3 & 0.0 & 6@xmath21 2 & 2.9 & 242.28047 & 65.523933 & 1.1 & 20.14 + cxo j160907.9 + 652626 & 242.28297 & 65.440657 & 0.5 & 0.3 & 16@xmath21 4 & 7.0 & 242.28284 & 65.440109 & 2.0 & 21.62 + cxo j160909.1 + 652559 & 242.28817 & 65.433131 & 0.8 & 0.3 & 6@xmath21 2 & 2.7 & & & & + cxo j160909.6 + 653252 & 242.29026 & 65.547984 & 0.3 & 0.1 & 17@xmath21 4 & 8.1 & & & & + cxo j160911.0 + 653008 & 242.29611 & 65.502295 & 0.4 & 0.1 & 5@xmath21 2 & 2.4 & & & & + cxo j160913.0 + 653432 & 242.30456 & 65.575761 & 0.2 & 0.1 & 92@xmath21 10 & 38.9 & 242.30478 & 65.575493 & 1.0 & + cxo j160913.1 + 652457 & 242.30483 & 65.415846 & 1.3 & 0.3 & 5@xmath21 2 & 2.3 & & & & + cxo j160913.1 + 653020 & 242.30487 & 65.505795 & 0.3 & 0.2 & 6@xmath21 2 & 2.8 & & & & + cxo j160913.4 + 652827 & 242.30590 & 65.474394 & 0.1 & 0.1 & 82@xmath21 9 & 36.7 & 242.30664 & 65.474052 & 1.7 & 20.42 + cxo j160914.1 + 653700 & 242.30883 & 65.616860 & 0.7 & 0.3 & 18@xmath21 5 & 6.5 & & & & + cxo j160914.4 + 653206 & 242.31012 & 65.535131 & 0.2 & 0.1 & 39@xmath21 6 & 17.8 & 242.31021 & 65.534958 & 0.6 & + cxo j160916.8 + 652852 & 242.32012 & 65.481128 & 0.2 & 0.1 & 8@xmath21 3 & 4.1 & & & & + cxo j160917.4 + 653021 & 242.32254 & 65.506092 & 0.2 & 0.1 & 29@xmath21 5 & 13.9 & & & & + cxo j160917.4 + 652925 & 242.32268 & 65.490353 & 0.0 & 0.0 & 2@xmath21 1 & 1.0 & & & & + cxo j160917.7 + 653104 & 242.32390 & 65.517999 & 0.3 & 0.1 & 7@xmath21 3 & 3.4 & & & & + cxo j160919.6 + 652915 & 242.33176 & 65.487717 & 0.5 & 0.1 & 7@xmath21 3 & 3.5 & 242.33199 & 65.487556 & 0.7 & 18.87 + cxo j160919.9 + 653552 & 242.33310 & 65.597953 & 0.4 & 0.1 & 29@xmath21 6 & 11.2 & & & & + cxo j160921.4 + 652243 & 242.33930 & 65.378676 & 1.4 & 0.6 & 11@xmath21 4 & 3.4 & 242.34024 & 65.378616 & 1.4 & 17.14 + cxo j160925.2 + 653421 & 242.35506 & 65.572583 & 0.1 & 0.1 & 36@xmath21 6 & 17.0 & & & & + cxo j160926.6 + 653844 & 242.36086 & 65.645732 & 0.6 & 0.2 & 180@xmath21 14 & 32.6 & 242.36060 & 65.645737 & 0.4 & 20.59 + cxo j160927.2 + 653247 & 242.36361 & 65.546655 & 0.3 & 0.1 & 5@xmath21 2 & 2.4 & 242.36388 & 65.546600 & 0.4 & 21.39 + cxo j160930.8 + 652810 & 242.37839 & 65.469570 & 0.3 & 0.0 & 4@xmath21 2 & 2.1 & & & & + cxo j160936.4 + 652906 & 242.40200 & 65.485094 & 0.3 & 0.1 & 6@xmath21 2 & 3.0 & & & & + cxo j160938.3 + 653447 & 242.40970 & 65.579868 & 0.2 & 0.1 & 79@xmath21 9 & 30.2 & 242.40987 & 65.579842 & 0.3 & 20.33 + cxo j160939.7 + 653235 & 242.41543 & 65.543118 & 0.1 & 0.1 & 105@xmath21 10 & 44.3 & 242.41559 & 65.542908 & 0.8 & 20.73 + cxo j160940.1 + 653313 & 242.41720 & 65.553738 & 0.3 & 0.0 & 4@xmath21 2 & 1.9 & & & & + cxo j160941.5 + 652813 & 242.42317 & 65.470465 & 0.4 & 0.3 & 7@xmath21 3 & 3.3 & & & & + cxo j160943.6 + 653317 & 242.43203 & 65.554966 & 0.4 & 0.2 & 16@xmath21 4 & 7.6 & 242.43243 & 65.554749 & 1.0 & + cxo j160943.8 + 652700 & 242.43287 & 65.450271 & 0.8 & 0.0 & 2@xmath21 1 & 1.0 & & & & + cxo j160948.2 + 653330 & 242.45110 & 65.558338 & 0.6 & 0.1 & 6@xmath21 2 & 2.7 & & & & + cxo j160950.9 + 652922 & 242.46227 & 65.489607 & 0.3 & 0.0 & 4@xmath21 2 & 2.0 & & & & + cxo j160953.0 + 653155 & 242.47116 & 65.531966 & 0.4 & 0.2 & 4@xmath21 2 & 1.8 & & & & + cxo j160956.5 + 652824 & 242.48574 & 65.473340 & 0.9 & 0.1 & 7@xmath21 3 & 3.1 & & & & + cxo j160957.6 + 652129 & 242.49040 & 65.358242 & 0.6 & 0.2 & 18@xmath21 7 & 2.8 & & & & + cxo j161001.4 + 652845 & 242.50600 & 65.479256 & 0.4 & 0.2 & 71@xmath21 9 & 24.3 & 242.50522 & 65.479378 & 1.3 & 18.89 + cxo j161004.8 + 652454 & 242.52003 & 65.415231 & 0.6 & 0.9 & 3@xmath21 2 & 1.5 & & & & + cxo j161010.8 + 652719 & 242.54515 & 65.455479 & 0.4 & 0.0 & 5@xmath21 2 & 2.1 & & & & + cxo j161016.6 + 652354 & 242.56936 & 65.398384 & 1.6 & 0.8 & 8@xmath21 3 & 2.7 & & & & + cxo j161029.9 + 652405 & 242.62477 & 65.401536 & 0.9 & 0.5 & 142@xmath21 14 & 17.7 & 242.62561 & 65.401230 & 1.7 & 20.03 + cxo j161052.3 + 651833 & 242.71824 & 65.309300 & 1.2 & 0.8 & 83@xmath21 14 & 7.5 & & & & + chartas , g. , dai , x. , & garmire , g. p. 2004 , in measuring and modeling the universe , edited by w. l. freedman ( pasadena : carnegie obs . ) , http://www.ociw.edu/ociw/symposia/series/symposium2/proceedings.html jackson , n. , de bruyn , a. g. , myers , s. , bremer , m. n. , miley , g. k. , schilizzi , r. t. , browne , i. w. a. , nair , s. , wilkinson , p. n. , blandford , r. d. , pearson , t. j. , & readhead , a. c. s. 1995 , , 274 , l25 kochanek , c. s. , falco , e. e. , impey , c. d. , lehr , j. , mcleod , b. a. , & rix , h .- w . 1999 , in aip conf . 470 , after the dark ages : when galaxies were young ( the universe at @xmath105 ) , ed . s. holt & e. smith ( new york : aip ) , 163 mori , k. , tsunemi , h. , miyata , e. , baluta , c. , burrows , d. n. , garmire , g. p. , & chartas , g. 2001 , in asp conf . 251 , new century of x - ray astronomy , ed . h. inoue & h. kunieda ( san francisco : asp ) , 576 myers , s. t. , fassnacht , c. d. , djorgovski , s. g. , blandford , r. d. , matthews , k. , neugebauer , g. , pearson , t. j. , readhead , a. c. s. , smith , j. d. , thompson , d. j. , womble , d. s. , browne , i. w. a. , wilkinson , p. n. , nair , s. , jackson , n. , snellen , i. a. g. , miley , g. k. , de bruyn , a. g. , & schilizzi , r. t. 1995 , , 447 , l5
the central topic of this paper is the discussion of pseudocontinuability for schur functions in the unit disk . pseudocontinuability is a particular type of meromorphic continuation for functions from the meromorphic nevanlinna class in the unit disk . this concept originated in shapiro s papers @xcite , @xcite and was then systematically discussed a little bit later in the context of invariant subspaces for the backward shift in the landmark paper douglas / shapiro / shields @xcite . ( regarding modern treatments of this area we refer the reader to the monographs cima / ross @xcite and ross / shapiro @xcite . ) important applications of pseudocontinuability are contained in the work of d.z . arov on darlington synthesis , @xmath4-inner functions , and related topics ( see , e.g. , @xcite and @xcite ) . a further domain of application of pseudocontinuable functions is rational approximation . this theme which originated from a series of papers by g.ts . tumarkin ( see @xcite , @xcite , and @xcite ) was treated by katsnelson in @xcite . the paper @xcite is recommended for some other reasons too . it contains an extensive historical overview on the investigation of pseudocontinuable functions which also takes into account matrix - valued functions . moreover , the paper @xcite is very well written from the pedagogical point of view . it strongly influenced our approach to introducing pseudocontinuable functions ( see section [ s2 ] ) . the present paper continues recent investigations on several questions of pseudocontinuability of schur functions in the unit disk ( see @xcite and @xcite ) . our main aim is to establish a direct relation between two well - known criteria for pseudocontinuability of non - inner schur functions in the unit disk ( see theorem [ t3.9 ] and theorem [ t4.2 ] ) . the main result of this paper is a new characterization of pseudocontinuability of a non inner schur function @xmath0 in the unit disk . this characterization is expressed in terms of a special orthogonal basis in an @xmath3 space on the unit circle , where @xmath1 is some probability measure on the unit circle which is canonically associated with @xmath0 ( see theorem [ 4.3 ] ) . the paper is organized as follows . in section [ s1 ] , we recall that the set of schur functions in the unit disk stands in bijective correspondences to the set of normalized carathodory functions in the unit disk and to the set of probability measures on the borel @xmath5-algebra of the unit circle . this will give us the possibility to study the pseudocontinuability of a given schur function in terms of the associated normalized carathodory function and in terms of the associated probability measure , respectively . in section [ s2 ] ( influenced by katsnelson @xcite ) , we summarize essential facts on meromorphic functions and the concept of pseudocontinuability which is due to h.s . shapiro . section [ s3 ] contains a function theoretic approach to the study of pseudocontinuability of non - inner schur functions in the unit disk . we will recognize that a function of this class admits a pseudocontinuation if and only if the associated probability measure satisfies the szeg condition and the corresponding szeg function is pseudocontinuable ( see theorem [ t3.9 ] ) . the central theme of section [ s4 ] is an operator theoretic approach to the investigation of pseudocontinuability of non - inner schur functions in the unit disk . the starting point there is the observation that an arbitrary schur function can be represented as a characteristic function of some completely non unitary contraction in a separable complex hilbert space . then the pseudocontinuability of a non - inner schur function @xmath0 can be characterized in terms of the contraction @xmath2 having @xmath0 as its characteristic function . more precisely , the maximal unilateral shifts @xmath6 and @xmath7 contained in @xmath2 and @xmath8 , respectively , have to fulfill certain conditions of mutual interrelations ( see theorem [ t4.2 ] ) . in section [ s5 ] , we start from a probability measure @xmath1 on the borel @xmath5-algebra of the unit circle . the main aim is to construct a unitary colligation @xmath9 , the characteristic function @xmath10 of which coincides with the schur function @xmath11 associated with the measure @xmath1 according to section [ s1 ] . in section [ s6 ] , we consider a probability measure @xmath1 on the borel @xmath5-algebra of the unit circle for which the polynomials are non complete in the space @xmath12 . in this case the question arises as to the existence of a natural completion of the system of orthonormal polynomials in @xmath12 to a complete orthonormal system in @xmath12 . the primary concern of section [ s6 ] is to construct such a natural completion . in the particular case that the measure @xmath1 is associated with a pseudocontinuable non - inner schur function @xmath0 , we will show that the functions obtained by completing the orthonormal system of polynomials are boundary values of functions belonging to the meromorphic nevanlinna class in the unit disk ( see propoisition [ p6.14a ] ) . the goal of section [ s7 ] is to reformulate the characterization of the pseudocontinuability of a non - inner schur function given by theorem [ t4.2 ] in terms of the complete orthonormal system which was created in section [ s6 ] ( see theorem [ 4.3 ] ) . finally , section [ s8 ] is aimed at determining and further clarifying a direct connection between the criteria of pseudocontinuability of a non - inner schur function which were pointed out in theorem [ t3.9 ] and theorem [ t4.2 ] , respectively . let @xmath13 and @xmath14 be the unit disk and the unit circle in the complex plane @xmath15 , respectively . the central object in this paper is the schur class @xmath16 of all functions @xmath17 which are holomorphic in @xmath18 and satisfy @xmath19 . our main aim is to study the phenomenon of pseudocontinuability for functions belonging to @xmath16 . in order to allow for a more effective treatment of this question , we first consider certain important objects that relate bijectively to the class @xmath16 . this is the main content of the present section . let @xmath20 . then the function @xmath21 defined by @xmath22 is holomorphic in @xmath18 and satisfies @xmath23 > 0,\quad \zeta\in\mathbb d,\ ] ] and @xmath24 let @xmath25 be the carathodory class of all functions @xmath26 which are holomorphic in @xmath18 and satisfy @xmath27 \geq 0 $ ] and let @xmath28 in view of , , , and we have @xmath29 it can be easily verified that via a bijective correspondence between the classes @xmath16 and @xmath30 is established . note that from it follows that @xmath31 the class @xmath25 is intimately related with the class @xmath32 of all finite nonnegative measures on the borel @xmath5-algebra @xmath33 of @xmath34 . according to the riesz - herglotz theorem ( see , e.g. , ( * ? ? ? * theorem 2.2.2 ) ) : for each function @xmath35 there exists a unique measure @xmath36 and a unique number @xmath37 such that @xmath38 obviously , @xmath39 $ ] . on the other hand , it can be easily checked that , for arbitrary @xmath36 and @xmath40 , the function @xmath41 , which is defined by the right hand side of ( [ nr.0.3 ] ) , belongs to @xmath25 . if we consider the riesz - herglotz representation ( [ nr.0.3 ] ) for a function @xmath42 , then @xmath43 and @xmath1 belongs to the set @xmath44 of all probability measures belonging to @xmath32 ( i.e. @xmath45 . actually , in this way we obtain a bijective correspondence between the classes @xmath30 and @xmath44 . in the result of the above considerations we obtain special ordered triples @xmath46 $ ] consisting of a function @xmath20 , a function @xmath47 , and a measure @xmath48 which are interrelated in such way that each of these three objects uniquely determines the other two . for that reason , if one of the three objects is given , we will say that the two others are associated with it . based on this procedure , in section [ s3 ] we will characterize the pseudocontinuability of a function @xmath20 in terms of the associated objects @xmath47 and @xmath48 . in this section , we summarize some facts on several classes of meromorphic functions which will be used later . ( a detailed treatment of this subject can be found , e.g. , in duren @xcite and nevanlinna @xcite . ) in particular , we will recall the concept of pseudocontinuation . we will use @xmath49 and @xmath50 to denote the extended complex plane and the exterior of the closed unit disk , respectively , i.e. @xmath51 and @xmath52 . furthermore , the symbol @xmath53 stands for the normalized linear lebesgue - borel measure on @xmath34 . so we have @xmath54 , i.e. @xmath55 . assume that @xmath56 is one of the domains @xmath18 or @xmath50 . let @xmath57 be the nevanlinna class of all functions which are meromorphic in @xmath56 and which can be represented as a quotient of two bounded holomorphic functions in @xmath56 . if @xmath58 ( resp . @xmath59 ) , then a well - known theorem due to fatou implies that there is a set @xmath60 with @xmath61 and a borel measurable function @xmath62 such that @xmath63 for all @xmath64 . in the following , we will continue to use the symbol @xmath65 to denote such a boundary function of a function @xmath66 which belongs to @xmath67 or @xmath68 . there is a standard bijective correspondence between the sets @xmath67 and @xmath68 . in order to describe this we introduce some further notation . let @xmath56 be a nonempty subset of @xmath49 . then @xmath69 denotes that subset of @xmath49 which is symmetric to @xmath56 with respect to the unit circle , i.e. @xmath70 with the usual conventions @xmath71 and @xmath72 . here @xmath73 means the complex conjugate of @xmath74 . if @xmath75 , then @xmath76 stands for the complex - valued function which is defined on @xmath69 by @xmath77 now the bijective correspondence between the sets @xmath67 and @xmath68 can be expressed as follows . [ r2.1 ] if @xmath58 ( resp . @xmath59 ) , then the function @xmath78 belongs to @xmath68 ( resp . to @xmath67 ) and @xmath79 is the boundary function of @xmath78 . [ r2.2 ] let @xmath80 and let @xmath81 not vanish identically in @xmath56 . then the function @xmath82 is @xmath53-integrable and @xmath83 . the set @xmath67 ( resp . @xmath68 ) is obviously an algebra over @xmath84 . the subalgebra of all @xmath58 which are holomorphic in @xmath85 will be designated by @xmath86 . the class @xmath86 can be characterized as the set of all functions @xmath66 which are holomorphic in @xmath85 and which satisfy @xmath87 where @xmath88 for each @xmath89 . if a function @xmath90 admits a representation @xmath91 \,m(dt ) \right\ } , \quad \zeta\in{{\mathbb d}},\ ] ] with some @xmath92 and some borel measurable function @xmath93 satisfying @xmath94 | \,m(dt ) < + \infty,\ ] ] then @xmath95 belongs to @xmath86 and @xmath96 holds @xmath53-a.e . on @xmath97 such functions @xmath95 are called outer functions in @xmath86 . [ r2.3 ] let @xmath95 be an outer function in @xmath86 . then @xmath98 for all @xmath99 and the function @xmath100 is an outer function in @xmath86 as well . moreover , if @xmath101 and @xmath102 are outer functions in @xmath86 , then @xmath103 is an outer function in @xmath86 . an outer function @xmath95 in @xmath86 is called normalized if @xmath104 . [ r2.4 ] let @xmath105 be a function which does not vanish identically in @xmath85 . then @xmath41 is an outer function in @xmath86 . if @xmath47 , then @xmath41 is a normalized outer function in @xmath86 . in view of remark [ r2.3 ] and remark [ r2.4 ] one can conclude the following . [ r2.4b ] let @xmath105 . then @xmath106 is a function belonging to @xmath25 which vanishes nowhere in @xmath85 . thus , @xmath106 and @xmath107 are outer functions in @xmath86 . [ p2.6 ] let @xmath108 be an @xmath53-integrable function which satisfies the szeg condition @xmath109 \,m(dt ) > -\infty .\ ] ] then there is a unique normalized outer function @xmath95 in @xmath86 such that the relation @xmath110 holds @xmath53-a.e . on @xmath97 this function @xmath90 is given by @xmath111 \,m(dt)\right\ } . \ ] ] the function @xmath95 which appears in proposition [ p2.6 ] is called the szeg function associated with @xmath112 ( see also nikishin / sorokin ( * ? ? ? * chapter 5 in 5 ) ) . clearly , the schur class @xmath16 is a subclass of @xmath86 . a function @xmath113 is called an inner function if the relation @xmath114 holds @xmath53-a.e . on @xmath97 . we denote by @xmath115 the subclass of @xmath16 consisting of all inner functions . [ p2.8 ] ( v. i. smirnov ) + let @xmath58 . then there exists an outer function @xmath116 in @xmath86 and some functions @xmath117 such that @xmath118 in the case of @xmath119 , @xmath120 can be chosen as constant function with value @xmath121 . [ r2.9 ] let @xmath116 be an outer function in @xmath86 which also belongs to @xmath115 . then @xmath116 is a constant function with unimodular value . now we recall the concept of pseudocontinuability . let @xmath58 . then one says that @xmath66 admits a pseudocontinuation into @xmath122 if there exists a function @xmath123 such that the boundary functions @xmath65 and @xmath124 coincide @xmath53-a.e . on @xmath97 . from remark [ r2.2 ] it follows that a function @xmath58 admits at most one pseudocontinuation into @xmath122 . note that if @xmath58 admits a pseudocontinuation @xmath125 into @xmath122 and if @xmath66 is analytically continuable through some open arc of @xmath97 , then the analytic continuation coincides with @xmath125 . in the following , the notation @xmath126 stands for the set of all functions belonging to @xmath67 which admit a pseudocontinuation into @xmath122 . obviously , the restriction of a rational function onto @xmath85 belongs to @xmath126 . if @xmath127 , then the symbol @xmath125 will be used to denote the pseudocontinuation of @xmath66 into @xmath122 . based on remark [ r2.1 ] one can draw the following conclusion . [ r2.10 ] if @xmath127 , then @xmath128 and @xmath129 @xmath53-a.e . on @xmath97 . moreover , if @xmath58 such that there is some @xmath130 fulfilling the relation @xmath131 @xmath53-a.e . on @xmath97 , then @xmath127 and @xmath132 . remark [ r2.10 ] shows , in particular , that a function @xmath58 belongs to @xmath126 if and only if there exists some @xmath130 fulfilling @xmath131 @xmath53-a.e . on @xmath97 . thus , the pseudocontinuability of @xmath66 is determined by @xmath79 . the following properties of the class @xmath126 can be easily checked . [ r2.11 ] let @xmath133 . if @xmath134 , then @xmath135 . moreover , the function @xmath136 and , in the case that @xmath66 does not not vanish identically in @xmath85 , also @xmath137 and @xmath138 belong to @xmath126 . a particular situation is met in the class @xmath115 of inner functions belonging to @xmath16 . in this case we not only have pseudocontinuability , but can write down the corresponding pseudocontinuation , explicitly . [ r2.12 ] the inclusion @xmath139 holds . moreover , for some @xmath140 the function @xmath141 does not vanish identically in @xmath122 and @xmath142 . a combination of remarks [ r2.11 ] and [ r2.12 ] yields the following . [ r2.13 ] let @xmath58 and let the outer function @xmath116 in @xmath86 be chosen according to proposition [ p2.8 ] . then @xmath127 if and only if @xmath143 . remark [ r2.13 ] indicates the importance of the study of pseudocontinuability of outer functions in @xmath86 . this phenomenon can be characterized in the following way in terms of inner functions belonging to @xmath16 . [ p2.14 ] let @xmath95 be an outer function in @xmath86 . then @xmath144 if and only if there exist functions @xmath117 such that the identity @xmath145 is satisfied @xmath53-a.e . on @xmath97 . let @xmath144 . then the pseudocontinuation @xmath146 of @xmath95 into @xmath122 is well - defined and we set @xmath147 . from remark [ r2.10 ] we get @xmath130 and @xmath148 because of @xmath130 the canonical factorization theorem of smirnov ( see proposition [ p2.8 ] ) provides us with the existence of an outer function @xmath116 in @xmath86 and two functions @xmath149 such that @xmath150 in view of remark [ r2.3 ] the function @xmath116 vanishes nowhere in @xmath85 and @xmath151 is an outer function in @xmath86 . because of we obtain @xmath152 thus , using we infer @xmath153 @xmath53-a.e . on @xmath97 . since @xmath154 and @xmath155 are both unimodular @xmath53-a.e . on @xmath97 from one can see that @xmath156 is unimodular @xmath53-a.e . on @xmath97 . furthermore , since we already know that @xmath157 is an outer function in @xmath86 the maximum modulus principle of smirnov ( see , e.g. , duren ( * ? ? ? * theorem 2.11 on page 25 ) ) implies that @xmath157 belongs to @xmath115 as well . hence , remark [ r2.9 ] shows that @xmath157 is a constant function with unimodular value . consequently , the settings @xmath158 and @xmath159 yield functions belonging to @xmath115 such that holds @xmath53-a.e . on @xmath97 . conversely , we assume that there exist functions @xmath117 such that is satisfied @xmath53-a.e . on @xmath97 . let @xmath160 . then @xmath130 and in view of we get @xmath161 this leads , in combination with remark [ r2.10 ] , to @xmath144 . it should be mentioned that in the above proof of the fact that the validity of @xmath53-a.e . on @xmath97 implies @xmath144 we do not use the assumption that @xmath95 is an outer function in @xmath86 . in the present section we investigate the pseudocontinuability of a non - inner function @xmath20 in terms of the function @xmath47 and the measure @xmath48 which is generated by @xmath0 as explained in section [ s1 ] . the following result provides us with further insight into the connection between functions @xmath0 and @xmath41 . [ p3.1 ] let @xmath20 and let @xmath47 be defined via . then @xmath163 if and only if @xmath164 . taking into account and an application of remark [ r2.11 ] yields the asserted equivalence . the rest of this section is devoted to the study of pseudocontinuability of a non - inner function @xmath20 in terms of its associated measure @xmath48 . our approach to that question will be based on an analysis of various aspects of the boundary behavior of @xmath0 . in particular , the @xmath53-integrability of the function @xmath165 $ ] will be of importance . [ p3.2 ] let @xmath166 and let @xmath167 . then : 1 . the function @xmath66 belongs to @xmath86 and satisfies @xmath168 @xmath53-a.e . on @xmath97 . if @xmath169 , then the function @xmath170 is @xmath53-integrable . using some elementary properties on the class @xmath67 with a view to remark [ r2.10 ] , assertion ( a ) follows . now let @xmath169 . because of ( a ) one can see that the function @xmath65 does not vanish @xmath53-a.e . on @xmath97 . therefore , the function @xmath66 does not vanish identically in @xmath85 . consequently , remark [ r2.2 ] provides finally assertion ( b ) . let @xmath20 be embedded in the triple @xmath46 $ ] as in section [ s1 ] . motivated by proposition [ p3.2 ] our next considerations are aimed at characterizing the @xmath53-integrability of the function @xmath165 $ ] in terms of the function @xmath41 or the measure @xmath1 . [ r3.3 ] let @xmath20 and let @xmath47 be defined by . then a straightforward calculation yields that the identity @xmath171}{|\underline{\phi}(t)+1|^2}\ ] ] holds @xmath53-a.e . on @xmath97 . [ p3.4 ] let @xmath20 and let @xmath47 be defined by . then the function @xmath165 $ ] is @xmath53-integrable if and only if @xmath172 $ ] is @xmath53-integrable . because of remark [ r3.3 ] we get @xmath173 = \ln 4 + \ln [ { \rm re}\,\underline{\phi } ] - 2 \ln_{\!}|\underline{\phi}+1|\ ] ] @xmath53-a.e . on @xmath97 . in view of remark [ r2.4b ] we know that @xmath106 is an outer function in @xmath174 . thus , remark [ r2.3 ] implies in combination with remark [ r2.2 ] that @xmath175 is @xmath53-integrable . hence , the assertion is an immediate consequence of . following d.z . arov we denote by @xmath176 the subclass of pseudocontinuable functions of @xmath16 . [ c3.5 ] let @xmath177 and let @xmath47 be defined by . then the function @xmath172 $ ] is @xmath53-integrable . combine part ( b ) of proposition [ p3.2 ] with proposition [ p3.4 ] . now we start from a function @xmath105 and look to characterize the @xmath53-integrability of @xmath172 $ ] in terms of its riesz - herglotz measure @xmath1 associated with @xmath41 , subject to . in order to realize this goal we will apply the following result which is an immediate consequence of a theorem due to fatou on the boundary behavior of poisson integrals ( see , e.g. , rosenblum / rovnyak ( * ? ? ? * theorem 1.18 ) ) . [ p3.6 ] let @xmath105 and let @xmath178 be associated with @xmath41 by the riesz - herglotz theorem , i.e. via with @xmath179 $ ] . furthermore , let the lebesgue decomposition of @xmath1 with respect to @xmath53 be given by @xmath180 where @xmath181 stands for the singular part of @xmath1 with respect to @xmath53 . then the relation @xmath182 holds @xmath53-a.e . on @xmath97 . proposition [ p3.6 ] leads us to a particular subclass of @xmath183 . let @xmath178 and let the lebesgue decomposition of @xmath1 with respect to @xmath53 be given by , where @xmath181 stands for the singular part of @xmath1 with respect to @xmath53 . then @xmath1 is said to satisfy the szeg condition if the function @xmath184 is @xmath53-integrable or equivalently if holds . in this case , the szeg function @xmath185 associated with @xmath112 is given by . if we start from a measure @xmath178 which satisfies the szeg condition , we will call @xmath95 also the szeg function associated with @xmath1 . an application of proposition [ p3.6 ] immediately provides the following result . [ c3.7 ] let @xmath186 and let @xmath178 be associated with @xmath41 by the riesz - herglotz theorem via with @xmath179 $ ] . then the function @xmath172 $ ] is @xmath53-integrable if and only if the measure @xmath1 satisfies the szeg condition . [ l3.8 ] let @xmath41 and @xmath95 be functions belonging to @xmath67 which do not vanish identically in @xmath85 , where the relation @xmath187 is satisfied @xmath53-a.e . on @xmath97 . then @xmath164 if and only if @xmath144 . first suppose @xmath164 . then we define @xmath188 \cdot d^{-1}\ ] ] in view of the choice of @xmath41 and @xmath95 , remark [ r2.1 ] , and the fact that the set @xmath67 is an algebra over @xmath84 it follows that @xmath130 . moreover , an application of remark [ r2.10 ] yields the equality @xmath189 @xmath53-a.e . on @xmath97 . hence , @xmath190 @xmath53-a.e . on @xmath97 . taking this and @xmath130 into account from remark [ r2.10 ] we get @xmath144 . conversely , we suppose now that @xmath144 . we then set @xmath191 similar to the above , using remark [ r2.1 ] , one can reason that @xmath192 , where remark [ r2.10 ] results firstly in @xmath193 @xmath53-a.e . on @xmath97 . thus , from remark [ r2.10 ] we can conclude @xmath164 . the following criterion for the pseudocontinuability of a non - inner function @xmath20 is the main result of this section . it is formulated in terms of the probability measure @xmath194 which is associated with @xmath0 as explained in section [ s1 ] . [ t3.9 ] let @xmath195 and let @xmath194 be the measure associated with @xmath0 according to section [ s1 ] . then the following statements are equivalent : 1 . the measure @xmath1 satisfies the szeg condition and the szeg function @xmath95 associated with @xmath1 belongs to @xmath197 . let @xmath47 be defined by . assume that the measure @xmath1 satisfies the szeg condition . hence , from proposition [ p2.6 ] we infer that the szeg function @xmath95 associated with @xmath1 belongs to @xmath174 . moreover , the combination of proposition [ p2.6 ] and proposition [ p3.6 ] yields the equality @xmath198 @xmath53-a.e . on @xmath97 . suppose now that ( i ) holds . then from proposition [ p3.1 ] we get @xmath199 moreover , in view of ( i ) and corollary [ c3.5 ] , the function @xmath172 $ ] is @xmath53-integrable . thus , corollary [ c3.7 ] shows that @xmath1 satisfies the szeg condition . consequently , is satisfied . by virtue of , , and lemma [ l3.8 ] it follows that @xmath200 . so we have verified that ( i ) implies ( ii ) . conversely , we suppose now that ( ii ) holds . from ( ii ) we obtain and @xmath200 . therefore , lemma [ l3.8 ] yields . accordingly , an application of proposition [ p3.1 ] supplies @xmath196 , i.e. ( i ) . the starting point of this section is the observation that a given schur function @xmath20 can be represented as characteristic function of some contraction in a hilbert space . that means that there exists a separable complex hilbert space and bounded linear operators @xmath201 , @xmath202 , @xmath203 , and @xmath204 such that the block operator @xmath205 is unitary and moreover that the equality @xmath206 is fulfilled . note that in the complex plane @xmath84 is considered as the one - dimensional complex hilbert space with the usual inner product @xmath207 the unitarity of the operator @xmath208 implies that the operator @xmath2 is contractive ( i.e. @xmath209 ) . thus , for all @xmath99 the operator @xmath210 is boundedly invertible . the unitarity of the operator @xmath208 means that the ordered tuple @xmath211 is a unitary colligation . in view of the function @xmath0 is the characteristic operator function of the unitary colligation @xmath212 . the following subspaces of @xmath213 will later play an important role @xmath214 by the symbol @xmath215 we mean the smallest closed subspace generated by the subsets @xmath216 of the corresponding vector spaces . the spaces @xmath217 and @xmath218 are called the subspaces of controllability and observability , respectively . we note that the unitary operator @xmath208 can be chosen such that @xmath219 holds . in this case the unitary colligation @xmath212 is called simple . the simplicity of a unitary colligation means that there does not exist a nontrivial invariant subspace of @xmath213 on which the operator @xmath2 induces a unitary operator . such contractions , @xmath2 , are called completely non unitary . in the language of unitary colligations the pseudocontinuability of a non - inner schur function @xmath20 has the following consequence . [ p4.1 ] let @xmath177 and let @xmath212 be a simple unitary colligation , the characteristic operator function of which coincides with @xmath0 . then the subspaces @xmath220 are nontrivial . because of and it follows that the subspace @xmath221 @xmath222resp . @xmath223 is invariant with respect to @xmath2 @xmath222resp . @xmath224 . it can be shown ( see ( * ? ? ? * chapter 1 ) ) that @xmath225 are unilateral shifts . more precisely , @xmath6 ( resp . @xmath226 ) is just the maximal unilateral shift contained in @xmath2 @xmath222resp . this means that an arbitrary invariant subspace with respect to @xmath2 @xmath222resp . @xmath224 on which @xmath2 @xmath222resp . @xmath224 induces a unilateral shift is contained in @xmath221 @xmath222resp . @xmath223 . in terms of unitary colligations the pseudocontinuability of a non - inner schur function @xmath20 was characterized in ( * ? ? ? * theorem 3.17 ) as follows . [ t4.2 ] let @xmath20 and let @xmath212 be a simple unitary colligation , the characteristic operator function of which coincides with @xmath0 . then the following statements are equivalent : 1 . 2 . @xmath228 . 3 . @xmath229 . a comparison of theorem [ t3.9 ] and theorem [ t4.2 ] shows that they contain rather different characterizations of the pseudocontinuability of a non - inner schur function @xmath20 . our subsequent considerations are aimed at establishing direct connections between the two different criteria . let @xmath230 . then our subsequent considerations are concerned with the investigation of the unitary operator @xmath231 which is defined for @xmath232 by @xmath233 denote by @xmath234 the embedding operator of @xmath15 into @xmath235 , i.e. @xmath236 and for each @xmath237 the image @xmath238 of @xmath239 is the constant function on @xmath34 with value @xmath239 . denote by @xmath240 the subspace of @xmath235 which is generated by the constant functions and denote by @xmath241 the constant function on @xmath34 with value @xmath121 . then obviously @xmath242 and @xmath243 . we consider the subspace @xmath244 denote by @xmath245 the block representation of the operator @xmath231 with respect to the orthogonal decomposition @xmath246 . then ( see ( * ? ? ? * section 2.8 ) ) the following result holds . [ 2.1 ] let @xmath48 . define @xmath247 , @xmath248 , @xmath249,and @xmath250 . then @xmath251 is a simple unitary colligation , the characteristic function @xmath252 of which coincides with the schur function @xmath0 associated with @xmath1 . in view of theorem [ 2.1 ] the operator @xmath253 is a completely non unitary contraction and if the function @xmath41 is given by ( [ nr.0.3 ] ) with @xmath254 , then from ( [ nr.0.1f ] ) it follows that @xmath255 let @xmath48 . then the simple unitary colligation given by is called the unitary colligation associated with @xmath1 . let @xmath48 and suppose that the measure @xmath1 has infinitely many points of growth . furthermore , we use , for all integers @xmath256 , the mapping @xmath257 defined by @xmath258 thus , we have @xmath259 , where @xmath231 is the operator defined by ( [ nr.2.0 ] ) . we now consider the system @xmath260 . applying the gram - schmidt orthogonalization method in the space @xmath235 we get a unique sequence @xmath261 of polynomials , where @xmath262 such that the conditions @xmath263 are satisfied . here and in the following , if @xmath264 is some family of elements of @xmath235 , then the symbol @xmath265 stands for the smallest closed subspace in @xmath235 which contains all elements of this family . we note that the second condition in ( [ nr.3.2 ] ) is equivalent to @xmath266 and @xmath267 in particular , since @xmath268 from the construction of @xmath269 we see that @xmath270 we consider the case that the system @xmath261 is non complete in the space @xmath235 . in this case the question arises as to the existence of a natural completion of this system to a complete orthonormal system in @xmath235 . the main goal of this section is to construct such a natural completion . with a view to the lebesgue decomposition , where @xmath181 stands for the singular part of @xmath1 with respect to @xmath53 , it is well known ( see , e.g. , rosenblum / rovnyak ( * ? * chapter 4 ) ) that the system of polynomials ( [ nr.3.1 ] ) is non complete in @xmath235 if and only if the szeg condition is satisfied . in particular , in the case studied below , the szeg function @xmath185 which is given by is well - defined . [ 3.1 ] denote by @xmath271 a support of the measure @xmath181 on @xmath34 and denote by @xmath272 the set of all points of @xmath34 , where the function @xmath95 does not have non tangential boundary values . . then @xmath274 . thus @xmath222see hoffman ( * ? ? ? * exercise 3 in chapter 4)@xmath275 , we can choose a boundary function @xmath276 of @xmath95 which satisfies @xmath277 and @xmath278 to construct the above - mentioned completion we consider a simple unitary colligation @xmath279 of type ( [ nr.2.2 ] ) associated with the measure @xmath1 . here the completion of the orthonormal system ( [ nr.3.1 ] ) of polynomials to a complete orthonormal system in @xmath235 will be determined in a natural way by properties of the operator @xmath253 . we note that the controllability space ( cf . ) associated with the unitary colligation @xmath279 has the form @xmath280 let the sequence of functions @xmath281 be defined by @xmath282 in view of the formula @xmath283 it can be seen that the sequence @xmath284 can be obtained by applying the gram - schmidt orthonormalization procedure to @xmath281 with additional consideration of the normalization condition ( [ nr.3.3 ] ) . thus , we obtain the following result . [ 3.2 ] the system @xmath284 of orthonormal polynomials is a basis in the space @xmath285 . this system can be obtained by applying the gram - schmidt orthogonalization procedure to the sequence , taking into account the normalization condition . [ 3.3 ] the orthonormal system of polynomials @xmath286 is non complete in @xmath235 if and only if @xmath287 . if @xmath2 is a contraction acting on some hilbert space @xmath213 , then we set @xmath288 where @xmath289 ( resp . @xmath290 ) is the closure of the range of the defect operator @xmath291 ( resp . @xmath292 ) . let @xmath293 ( cf . ) . then the space @xmath294 is invariant with respect to @xmath295 and moreover , if @xmath296 , then the restriction @xmath297 is the maximal unilateral shift contained in @xmath295 ( see ( * ? ? ? * theorem 1.6 ) ) . we now suppose that @xmath296 . in view of @xmath298 , the multiplicity of the unilateral shift @xmath299 is equal @xmath121 and coincides with @xmath300 . this is equivalent to the operator @xmath253 containing a maximal unilateral shift @xmath301 of multiplicity @xmath121 ( see ( * ? ? ? * remark 1.11 ) ) . thus , we obtain the following result . [ 3.4 ] the orthonormal system of polynomials @xmath286 is non complete in @xmath235 if and only if the contraction @xmath253 @xmath222resp . @xmath302 contains a maximal unilateral shift @xmath301 @xmath222resp . @xmath303 of multiplicity @xmath121 . we consider the orthogonal decomposition @xmath304 denote by @xmath305 the wandering subspace which generates the subspace associated with the unilateral shift @xmath299 . then @xmath306 and since @xmath299 is an isometric operator we have @xmath307 consequently , @xmath308^n ( \tilde{{{\mathfrak l}}}_0).\ ] ] there exists ( see ( * ? ? ? * corollary 1.10 ) ) a unique unit vector @xmath309 which fulfills @xmath310 because of ( [ nr.3.10 ] ) , ( [ nr.3.11 ] ) , and ( [ nr.3.12 ] ) it follows that the sequence @xmath311 , where @xmath312^{k-1 } \psi_1 , \quad k\in\{1,2,\ldots\},\ ] ] is the unique orthonormal basis of the space @xmath294 which satisfies the conditions @xmath313 or equivalently @xmath314 as in the paper @xcite we introduce the following notion . [ 3.5 ] the constructed orthonormal basis @xmath315 in the space @xmath235 which satisfies the conditions and is called the canonical orthonormal basis in @xmath235 . obviously , the canonical orthonormal basis in @xmath235 is uniquely determined by the conditions and . here the sequence @xmath286 is an orthonormal system of polynomials ( depending on @xmath316 ) . the orthonormal system @xmath311 is built with the aid of the operator @xmath231 from the function @xmath317 @xmath222see @xmath275 in similar to the way the system @xmath286 was built from the function @xmath269 @xmath222see and @xmath275 . the only difference is that the system @xmath318^k \psi_1 \right)_{k=0}^\infty$ ] is orthonormal , whereas in the general case the system @xmath319 is not orthonormal . in this respect the sequence @xmath311 can be considered as a natural completion of the system of orthonormal polynomials @xmath286 to an orthonormal basis in @xmath235 . it should be mentioned that the part @xmath311 of ( [ nr.3.16 ] ) has a clear interpretation in terms of prediction theory of stationary sequences . a closer look at the papers wiener / masani @xcite and @xcite shows that @xmath311 is the spectral image of the sequence of so - called normalized innovations corresponding to some stationary sequence which is naturally associated with @xmath235 . we hope to discuss this and related matters in a separate paper dedicated to a prediction - theoretical analysis of pseudocontinuability of schur functions . [ 3.6 ] the orthonormal system @xmath320 is an orthonormal basis in the space @xmath321 , which takes the orthogonal decomposition of @xmath321 into account . we will call it the canonical orthonormal basis in @xmath321 . in view of ( [ nr.3.15 ] ) , to obtain a description of the sequence @xmath311 it suffices to determine @xmath317 . the above considerations lead us to the following result . [ 3.7 ] the function @xmath322 is completely characterized by the following four conditions : 1 . @xmath323 . 2 . @xmath324 . @xmath325 is an orthonormal basis in @xmath294 . 4 . @xmath326 . [ 3.8 ] in view of theorem [ 3.2 ] condition ( a ) in lemma [ 3.7 ] is equivalent to the orthogonality conditions in @xmath235 which are expressed by @xmath327 or equivalently @xmath328 now we are going to prove another auxiliary result . here and in the following , for any function @xmath329 , we simply write @xmath330 instead of @xmath331 . [ 3.9 ] assume that @xmath332 satisfies the following three conditions : 1 . 2 . @xmath334 . @xmath335 is an orthonormal system in @xmath294 . suppose that @xmath1 has the lebesgue decomposition and let @xmath90 be the szeg function given by . denote by @xmath116 the borel subset of @xmath34 which was introduced in remark [ 3.1 ] . then there is a function @xmath336 such that the function @xmath337 has the form @xmath338 in view of remark [ 3.8 ] from condition ( @xmath339 ) we infer @xmath340 moreover , from ( @xmath341 ) and @xmath342 we also have @xmath343 . thus , in view of we get @xmath344 . this yields @xmath345 . therefore , it follows that @xmath346 by using the riesz brothers theorem ( see , e.g. , hoffman ( * ? ? ? * chapter 4 ) ) we obtain from and we obtain that the complex measure @xmath347 is absolutely continuous with respect to @xmath53 . according to the decomposition ( [ nr.3.4 ] ) this implies @xmath348 the function @xmath337 is determined @xmath53-a.e . on the set @xmath349 . for this reason , taking into account remark [ 3.1 ] and , we can assume that @xmath350 because of ( [ nr.3.4 ] ) , ( [ nr.3.18 ] ) , and the condition ( @xmath341 ) for each @xmath351 it follows @xmath352 consequently , taking complex conjugates for each non zero integer @xmath256 , we get @xmath353 hence , there is a constant @xmath237 such that the identity @xmath354 holds @xmath53-a.e . on @xmath34 . combining ( @xmath355 ) with ( [ nr.3.4 ] ) , ( [ nr.3.18 ] ) , and ( [ nr.3.19 ] ) we can conclude @xmath356 therefore , from ( [ nr.3.19 ] ) it follows that @xmath357 @xmath53-a.e . on @xmath358 . from ( [ nr.3.20 ] ) and proposition [ p2.6 ] we see that @xmath359 . using ( [ nr.3.4 ] ) , ( [ nr.3.18 ] ) , , and we obtain @xmath360 thus , by setting @xmath361 we get @xmath362 in view of ( [ nr.3.20 ] ) , the identity @xmath363 holds @xmath53-a.e . on @xmath34 . since @xmath95 is an outer function in @xmath364 , it follows from ( [ nr.3.21 ] ) , ( [ nr.3.22 ] ) and the inner - outer factorization ( see proposition [ p2.8 ] ) that the function @xmath365 admits the representation @xmath366 @xmath53-a.e . on @xmath97 for some @xmath336 . hence , the identity @xmath367 holds @xmath53-a.e . on @xmath97 . combining this with ( [ nr.3.18b ] ) we get the representation ( [ nr.3.17 ] ) . [ 3.10 ] let @xmath90 be the szeg function given by . denote by @xmath116 the borel subset of @xmath34 which was introduced in remark [ 3.1 ] . then the unit vector @xmath309 which is uniquely determined via is given by @xmath368 the conditions ( a ) , ( b ) , and ( c ) in lemma [ 3.7 ] lead in combination with lemma [ 3.9 ] to the relation @xmath369 because of condition ( c ) in lemma [ 3.7 ] , the sequence @xmath325 is an orthonormal basis in @xmath294 , where @xmath370 , @xmath371 . now we are going to prove that this implies that the function @xmath372 is constant with unimodular value . we prove this by contradiction . assume that @xmath372 is not a unimodular constant function . then beurling s theorem ( see , e.g. , garnett ( * ? ? ? * theorem 7.1 in chapter 2 ) implies that the system @xmath373 is not closed in @xmath374 . thus , there exists an element @xmath375 which is orthogonal in @xmath374 to the sequence @xmath376 . now we show that the function @xmath377 has the following properties @xmath378 @xmath379 indeed , since the functions @xmath380 and @xmath276 belong to @xmath374 from ( [ nr.3.24 ] ) , ( [ nr.3.23 ] ) , and the fact that @xmath336 for each @xmath381 we obtain @xmath382 this implies ( [ nr.3.25 ] ) . moreover , taking into account ( [ nr.3.24 ] ) , ( [ nr.3.23 ] ) , and the choice of @xmath380 for each @xmath371 one can conclude @xmath383 thus , ( [ nr.3.26 ] ) is proved . however , condition ( c ) in lemma [ 3.7 ] shows that from ( [ nr.3.25 ] ) and ( [ nr.3.26 ] ) it follows that @xmath384 . combining this with ( [ nr.3.23 ] ) and ( [ nr.3.24 ] ) we get @xmath385 @xmath53-a.e . on @xmath358 . therefore , @xmath380 vanishes @xmath53-a.e . on @xmath358 . because @xmath274 , this implies that @xmath380 vanishes @xmath53-a.e . on @xmath34 this contradicts the assumption that @xmath380 belongs to @xmath386 . hence , the function @xmath372 is constant . since @xmath372 is an inner function , there is an @xmath387 such that @xmath372 is the constant function with value @xmath388 . thus , formula ( [ nr.3.23 ] ) can be rewritten in the form @xmath389 denote by @xmath390 the orthoprojector in @xmath235 onto the closed subspace @xmath240 . using , ( [ nr.3.27 ] ) , and ( [ nr.3.12 ] ) we obtain @xmath391 \,m ( dt ) \right\ } .\end{aligned}\ ] ] consequently , it follows @xmath392 . [ c3.9 ] for each @xmath393 , the function @xmath394 of the canonical orthonormal basis ( [ nr.3.16b ] ) in the space @xmath321 is given by @xmath395 [ c3.10 ] the canonical orthonormal basis in the space @xmath235 consists of the system @xmath286 of orthonormal polynomials @xmath222depending on @xmath396 given by and the orthonormal system @xmath311 given by . [ 3.11 ] if @xmath1 coincides with the lebesgue - borel measure @xmath53 on @xmath34 , then the szeg condition for @xmath1 is satisfied and @xmath95 is the constant function with value @xmath121 . thus , @xmath397 and @xmath398 in other words , the canonical orthonormal basis in that case is given by @xmath399 where for each integer @xmath256 the function @xmath257 is defined as in . [ 3.12 ] the space @xmath294 consists of all functions @xmath400 of the form @xmath401 where @xmath402 is some function belonging to @xmath374 . thereby , @xmath403 i.e. the mapping @xmath404 establishes a metric isomorphism between @xmath294 and @xmath374 . now we consider the case that the measure @xmath48 is generated by the pseudocontinuable non inner function @xmath0 . [ p6.14a ] let @xmath405 and @xmath48 be the measure associated with @xmath0 according to section [ s1 ] . then : 1 . [ p6.14a - ta ] the measure @xmath1 satisfies the szeg condition and the szeg function @xmath95 associated with @xmath1 belongs to @xmath197 . [ p6.14a - tb ] denote by @xmath146 the pseudocontinuation of @xmath95 . let @xmath406 . for all @xmath407 such that @xmath146 is holomorphic at @xmath408 and satisfies @xmath409 , we define @xmath410^ * } \ ] ] then @xmath411 and its boundary values @xmath412 satisfy @xmath413 @xmath1a.e . on @xmath414 , where @xmath394 belongs to the canonical orthonormal basis in @xmath3 which was introduced in definition [ 3.5 ] and where @xmath116 stands for the borel subset of @xmath414 , which was introduced in remark [ 3.1 ] . 3 . [ p6.14a - tc ] let @xmath415 and let @xmath416 then @xmath417 and its boundary values @xmath418 satisfy @xmath419 and @xmath420 the mapping @xmath421 is a metric isomorphism between @xmath422 and @xmath294 . [ p6.14a - ta ] this follows from theorem [ t3.9 ] . [ p6.14a - tb ] in view of remark [ r2.10 ] we have @xmath423 and @xmath424 thus , in view of @xmath274 , the combination of , theorem [ 3.10 ] and shows that @xmath413 @xmath1a.e . on @xmath414 . [ p6.14a - tc ] using corollary [ 3.12 ] , the assertion of [ p6.14a - tc ] can be obtained analogously . [ 3.13 ] in @xcite a detailed description of the matrix representations of the operators @xmath253 , @xmath425 , @xmath426 , and @xmath427 @xmath222and , consequently , of the operator @xmath428 with respect to the canonical basis was given @xmath222see 2.8 comment , chapter a , and theorems 2.13 , 2.15 , and 2.17 there@xmath275 . these matrix representations are expressed in terms of schur parameters of the characteristic function @xmath252 of the unitary colligation which is associated with the probability measure @xmath1 . we note that in chapter 4 of the monograph @xcite by b. simon some historical remarks concerning these matrix representations are presented . in particular , the representation , abbreviated by @xmath429 , is used in @xcite . here @xmath429 stands for geronimus , gragg , and teplyaev . it is well known ( see , e.g. , brodskii @xcite ) that one can consider simultaneously together with the simple unitary colligation ( [ nr.2.2 ] ) the adjoint unitary colligation @xmath430 which is also simple . for @xmath431 , its characteristic function @xmath432 is given by @xmath433 we note that the unitary colligation ( [ nr.3.29 ] ) is associated with the operator @xmath434 . it can be easily checked that the action of @xmath434 is given for each @xmath435 by @xmath436 ( t ) = t\cdot f(t),\quad t\in\mathbb t .\ ] ] if we replace the operator @xmath437 by @xmath434 in the preceding considerations , which have lead to the canonical orthonormal basis ( [ nr.3.16 ] ) , we obtain an orthonormal basis of the space @xmath235 which consists of two sequences @xmath438 of functions . from our treatments above it follows that the orthonormal basis ( [ nr.onb ] ) is uniquely determined by the following conditions : 1 . the sequence @xmath439 arises from the result of the gram - schmidt orthogonalization procedure of the sequence @xmath440^n { \bf 1}\right)_{n=0}^\infty$ ] while taking into account the normalization conditions @xmath441^n { \bf 1 } , \tilde{\varphi}_n \bigr)_{l_\mu^2 } > 0 , \quad n\in \{0,1,2,\ldots \}.\ ] ] 2 . the relations @xmath442 hold . it can be easily checked that @xmath443 and @xmath444 as in the paper @xcite , we introduce the following notion . [ 3.14 ] the orthogonal basis @xmath445 is called the conjugate canonical orthonormal basis with respect to the canonical orthonormal basis . similar to ( [ nr.3.8 ] ) , the identity @xmath446^k { \bf 1 } = \left ( \bigvee_{k=0}^{n-1 } ( t_\mu^\ast)^k g_\mu^\ast ( 1 ) \right ) \oplus \mathbb c_\mathbb t\ ] ] can be verified . thus , @xmath447 @xmath448 in ( * ? ? ? * chapter 3 ) the unitary operator @xmath208 was introduced which maps the elements of the canonical basis ( [ nr.3.16 ] ) onto the corresponding elements of the conjugate canonical basis ( [ nr.3.31 ] ) . more precisely , @xmath449 the operator @xmath208 is related to the conjugation operator in @xmath235 . namely , if @xmath232 and @xmath450 then @xmath451 in ( * ? ? * theorem 3.6 ) the matrix representation of the blocks of the operator @xmath208 with respect to the canonical basis ( [ nr.3.16 ] ) was expressed in terms of schur parameters of the characteristic function @xmath452 . a finer analysis of this matrix representation ( see ( * ? ? ? * chapters 4 and 5 ) ) yields effective criteria for the pseudocontinuability of a non - inner function belonging to the schur class @xmath16 . these criteria are formulated in terms of schur parameters . let @xmath453 the relations ( [ nr.0.1f ] ) and ( [ nr.0.3 ] ) allow us to associate a measure @xmath454 with the function @xmath0 . furthermore , we associate the simple unitary colligation @xmath279 given by ( [ nr.2.2 ] ) with the same measure @xmath1 . then theorem [ 2.1 ] implies @xmath455 . the unitary colligation @xmath279 suggests that we consider the canonical basis ( [ nr.3.16 ] ) and the conjugate canonical basis ( [ nr.3.31 ] ) in @xmath235 . the equations ( [ nr.3.39 ] ) and ( [ nr.3.40 ] ) allow us to reformulate theorem [ t4.2 ] into the following criterion of pseudocontinuability of non - inner schur functions in terms of the canonical orthonormal basis of @xmath235 introduced in definition [ 3.5 ] . [ 4.3 ] let @xmath195 and let @xmath48 be the measure associated with @xmath0 according to section [ s1 ] . let @xmath279 be the simple unitary colligation which satisifies @xmath455 . furthermore , let the canonical orthogonal basis in @xmath235 be given by . then @xmath0 admits a pesudocontinuation if and only if @xmath456 ( where the symbol , , @xmath457 is used in the context of @xmath235 ) . the combination of proposition [ p2.14 ] and theorem [ 4.3 ] shows that in order to establish a direct connection between the criteria of pseudocontinuability of functions belonging to @xmath162 which are contained in theorems [ t3.9 ] and [ t4.2 ] it is sufficient to verify the following statement . [ 4.5 ] let @xmath195 and let @xmath194 be the measure associated with @xmath0 according to section [ s1 ] . then condition in theorem [ 4.3 ] is fulfilled if and only if the measure @xmath1 satisfies the szeg condition and there exist functions @xmath117 such that the identity holds @xmath53-a.e . on @xmath97 . let condition ( [ nr.4.2 ] ) be satisfied . then the system @xmath261 is not complete in @xmath235 . thus , as mentioned above , condition ( [ nr.3.4a ] ) is fulfilled . from the definition of the subspaces @xmath458 and @xmath459 ( see ( [ nr.d1.5c ] ) and section [ s5 ] ) it follows that the subspaces @xmath459 and @xmath294 are invariant with respect to the operator @xmath460 . moreover , since the unitary colligation @xmath279 is simple we get @xmath461 taking into account ( [ nr.3.39 ] ) , ( [ nr.3.40 ] ) , and ( [ nr.4.2 ] ) it follows @xmath462 hence , the space @xmath463 is also invariant with respect to @xmath460 and the restriction of @xmath460 is a unilateral shift of multiplicity @xmath121 . let @xmath337 be a basis function of the generating wandering subspace of the unilateral shift @xmath464 . then the function @xmath337 satisfies the conditions in lemma [ 3.9 ] . thus , there is a function @xmath465 such that @xmath466 on the other hand , in view of @xmath467 there exists a sequence @xmath468 of polynomials such that the limit relation @xmath469 holds . this implies @xmath470 combining this with ( [ nr.4.5 ] ) we obtain @xmath471 therefore , the unimodular function @xmath472 which is defined by @xmath473 is the @xmath474-limit of the sequence @xmath475 . thus , since the sequence @xmath476 belongs to the hardy space @xmath364 and since @xmath477 is a unimodular function , we see that there is a function @xmath478 such that its boundary function @xmath479 coincides @xmath53-a.e . on @xmath97 with @xmath477 . hence , the identity @xmath480 holds @xmath53-a.e . on @xmath97 this implies that @xmath481 holds @xmath53-a.e . on @xmath97 therefore , if we define the function @xmath482 by @xmath483 then @xmath484 is a function belonging to @xmath115 such that holds @xmath53-a.e . on @xmath97 . conversely , we assume now that the conditions and are satisfied . let @xmath116 be the set introduced in remark [ 3.1 ] . in view of @xmath274 we can assume that the boundary functions @xmath485 and @xmath486 of the inner functions @xmath484 and @xmath120 vanish on @xmath116 . thus , the functions @xmath485 and @xmath486 vanish @xmath181-a.e . on @xmath97 . from it follows that @xmath487 holds @xmath53-a.e . on @xmath97 . we are going to prove that the function @xmath337 defined by @xmath488 belongs to the intersection of the sets described on the left - hand side of formula ( [ nr.4.2 ] ) . indeed , in view of corollary [ 3.12 ] and ( [ nr.3.40 ] ) we infer @xmath489 on the other hand , using ( [ nr.4.8 ] ) , ( [ nr.4.7 ] ) , ( [ nr.3.28 ] ) , and @xmath274 for each @xmath490 one can conclude @xmath491 thus , we get @xmath492 in view of @xmath493 we have @xmath494 consequently , by using ( [ nr.4.11 ] ) , ( [ nr.4.8 ] ) , and @xmath495 we obtain @xmath496 since @xmath497 is an orthonormal basis of @xmath235 from ( [ nr.4.10 ] ) and ( [ nr.4.13 ] ) it follows that @xmath498 combining this with ( [ nr.4.9 ] ) we see that @xmath499 by using @xmath500 , @xmath274 , and ( [ nr.4.8 ] ) we obtain @xmath501 therefore , from ( [ nr.4.14 ] ) we see @xmath502 thus , ( [ nr.4.2 ] ) is fulfilled . we now look to cast some light on an interesting aspect of pseudocontinuability of schur functions by considering our main result under the light of the following well - known douglas - rudin theorem ( see , e.g. , garnett ( * ? ? ? * theorem 2.1 in chapter v ) ) . thus , the combination of theorem [ 4.3 ] with theorem [ 4.5 ] shows that the phenomenon of pseudocontinuability of a non - inner function @xmath20 is related to the fact that for the unimodular function @xmath507 an extremal situation can be met in theorem [ 4.7 ] . namely , there exist functions @xmath505 satisfying the equation @xmath508 . the authors are grateful to professor v.e . katsnelson , who carefully read an older version of this manuscript . his comments and remarks very much influenced and improved the conception and presentation of the content of this paper . : _ regular j - inner matrix functions and related continuation problems _ , in : linear operators in function spaces ( eds . : g. arsene , h. helson , b. sz .- nagy , f .- h . vasilescu ) , oper . theory . adv . 43 , birkhuser , basel , 1990 , pp . 6587 . : _ shift operators contained in contractions , schur parameters and pseudocontinuable schur functions _ , in : interpolation , schur functions and moment problems ( eds . : d. alpay , i. gohberg ) , oper . theory adv . 165 , birkhuser , basel , 2006 , pp . 175250 . : _ description of a class of functions which admit an approximation by rational functions with preassigned poles i _ , in : matrix and operator valued functions - the vladimir petrovich potapov memorial volume ( eds . i. gohberg , l.a . sakhnovich ) , oper . theory adv . 72 , birkhuser , basel , 1994 , pp . 87132 . : _ weighted polynomial approximation and boundary behavior of analytic functions _ ( in russian ) , in : contemporary problems in theory anal . functions ( internat . conf . , erevan , 1965 ) , nauka , moscow , 1966 , pp . 326335 . : _ orthogonal polynomials on the unit circle . classical theory ; part 2 . spectral theory _ , american mathematical society colloquium publications , vol . 54 , part 1 + 2 . soc . , providence , r.i . : _ approximation with respect to various metrics of functions defined on the circumference to sequences of rational functions with fixed poles _ ( in russian ) , izv nauk sssr , ser . mat . 30 ( 1966 ) , 721766 ; english transl . in : amer . series 2 , volume 177 , 1968 , pp . 183233 . : _ necessary and sufficient conditions for the possibility of approximating a function on a circumference by rational functions , expressed in terms directly connected with the distribution of poles of the approximating functions _ ( in russian ) , izv nauk sssr , ser . mat . 30 ( 1966 ) , 969980 ; english transl . in : amer . transl . , series 2 , volume 177 , 1968 , pp . 235248 .
in the present - day view , mergers are a common and important way of building up massive elliptical galaxies . with the presence of supermassive black holes in the centers of most galaxies ( @xcite ; @xcite ; @xcite ) , merger processes are thought to have left signatures on the central structure of the remnant galaxies . @xcite suggest that a binary black hole that is formed by the merger of two galaxies scours the stars from the center of the newly created system as the binary shrinks . the energy liberated by the hardening of the binary evacuates the central part through the ejection of surrounding stars , causing less light in the center . _ hubble space telescope _ ( _ hst _ ) observations of galaxy nuclei see two kinds of behavior in the central light profiles of ellipticals , traditionally classified as `` power - law '' and `` core '' galaxies @xcite . the power - law profiles took their name from their approximation by a single power law at small radii ( @xmath6 or 20 ) . more modern interpretations have emphasized that these profiles can be better understood as the inward continuation of the galaxy s overall srsic profile , usually modified by an additional , nuclear - scale stellar component @xcite . the core galaxy , in contrast , displays a surface brightness profile with a distinct break from a steep outer slope to a shallower inner cusp . core profiles mainly occur in very luminous elliptical galaxies and are considered the result of dissipationless mergers of two galaxies that have central black holes . much numerical work has been done to explore the binary black hole idea . the work of @xcite suggests that the stellar mass ejected from the core is comparable to the mass of the central black hole @xmath0 . @xcite find that core radius roughly doubles after each major merger . @xcite , using isothermal king models , suggest that the core size should scale with the final @xmath0 . simulations performed by @xcite show for the first time that merging two galaxies with steep cusps can result in a merger remnant with a shallow power - law cusp in the inner part ( core ) . @xcite is the first to quantify the magnitude of the mass deficit ( or the mass ejected by the binary ) with respect to the galaxy s merger history . by following the binary evolution up to the stalling radius , he formulates the mass deficit to be @xmath7@xmath0 with @xmath8 is the number of major merger events . an investigation into the post - coalescence stage of the binary suggests that the mass deficit could still be larger . this is because the coalescing binary emits gravitational waves which impart a kick on the merged black hole and eject it from the center . due to dynamical friction , the resulting oscillations of the black hole would gradually damp , returning it to the center of the galaxy and ejecting more stars from the core . this process could increase the mass deficit existing before the binary coalescence by up to 5@xmath0 for a single merger @xcite . several publications in the past have performed function fitting to galaxy profiles in order to quantify the core parameters such as the core size , the deficit in light or mass , and to analyse the correlation between these core properties and @xmath0 or the global properties of the galaxies ( @xcite ; @xcite ; @xcite , kormendy et al . ( 2009 , hereafter kfcb09 ) , kormendy & bender ( 2009 , hereafter kb09 ) , @xcite ; @xcite ; see also @xcite ) . these studies are based on different samples , datasets , approaches , parametrizations , and assumptions , which apparently lead to different , partly systematic , core quantities . the most important difference is the amount of central mass deficit with respect to the black hole mass that can be expected in core galaxies . @xcite and @xcite find an average mass deficit of around twice the black hole mass , @xcite derive central mass deficits of about 0.54 @xmath0 while find their mean mass deficit to be 10.5 @xmath0 . aside from that , each of these studies has a drawback in at least one of three aspects crucial to achieve high accuracy . the first is the radial range of the surface brightness profiles . @xcite , @xcite , and @xcite fit galaxy profiles that extend typically to only 1020 . moreover , it appears that some galaxy profiles in @xcite , which are taken from lauer et al . ( 2005 , hereafter l05 ) , suffer from oversubtraction of the sky that biases the core - srsic fits ( see section [ previousfits ] ) . a full galaxy profile is necessary to properly measure the outer profile , to optimally remove the sky background , and thus to provide better constraints on the light and mass deficit . the second is the determination of the stellar mass - to - light ratio ( @xmath9 ) to convert the light deficit to the mass deficit . in the past studies , either different values are assigned to the galaxies according to their luminosities ( @xcite , ) or a constant value is assumed for all the galaxies in the sample ( @xcite , @xcite ) . this of course introduces additional uncertainties and biases in deriving the mass deficits . the last important aspect lies in the black hole masses , which are mostly derived from outdated black hole - bulge relations , especially the ones between @xmath0 and the velocity dispersion of the host bulge ( @xmath0@xmath3relation ) . the majority of core galaxies are bright massive elliptical galaxies . for this class of galaxies @xcite find that the @xmath10-@xmath11 relation predicts a higher density of massive black holes ( @xmath12 @xmath13 ) compared to the @xmath0@xmath3 relation . furthermore , the @xmath0@xmath3 relation used in these papers also fails to predict the existence of the largest black holes ( @xmath14@xmath13 ) that are found in distant quasars @xcite in the sample of local galaxies . the recent finding of black holes of 10 billion solar masses @xcite , which might be the missing link between the local black holes and the massive distant quasars , further supports the indications that the previous @xmath0@xmath3 relations for massive bright early type galaxies need to be reevaluated ( e.g. , * ? ? ? in addition , most previous studies have used @xmath0 measurements ( or @xmath0@xmath3relations based on @xmath0 measurements ) which did not include dark - matter halos in the modeling . values typically increase by @xmath15% @xcite or 30% ( * ? ? ? * hereafter paper i ) when dark matter is included in the modeling , but the increase can be as large as a factor of 6 when the black hole s sphere of influence is not well resolved @xcite . in the particular sample that we use in this paper , which includes bcgs , we find that the measured black hole mass is on average more than 3 times higher than that predicted by the @xmath0@xmath3 relation ( e.g. @xcite , @xcite , @xcite . in this paper , we combine improved core - property measurements with new measurements and present the analysis of possible relations between the core and black hole mass or the galaxy properties . we use only directly measured black hole masses and consequently avoiding scatter and possible biases inherent in using proxy measurements . we use dynamically - determined , individual stellar mass - to - light ratios ( @xmath2 ) to estimate the mass deficit , instead of relying on a scaling function or a constant value . we also use full galaxy profiles in the fitting and the quantification of the core parameters . our analysis is based on a sample of 20 galaxies ; seven of these galaxies are taken from , in which the black hole masses and @xmath2 are derived taking into account the presence of dark matter halos . for completeness , we add a further 3 galaxies to the sample ; they are not included in the main sample due to the lack of reliable literature sources for the black hole masses . in order to derive the mass deficit , we assume that the original ( pre - scouring ) profile is the extrapolated srsic profile that fits the light distribution at large radii . in section [ themethod ] , we describe the method that we use to identify the core galaxies and to fit the observed surface brightness profiles . in section [ corefitting ] , we present the best - fitting models for each galaxy ; cases where galaxies are best fit with the addition of an outer envelope are discussed in section [ multi - component - fits ] . we compare the fitting results with previous studies in section [ previousfits ] . section [ coredeficit ] presents the details of how the luminosity and mass deficit in the core are computed based on the best - fitting models . the core - related quantities derived in section [ corefitting ] and [ coredeficit ] are compared with the black hole mass , velocity dispersion and galaxy luminosity in section [ bhcore ] . the last section discusses and summarizes the results . all of the photometric profiles examined in this work are quite extended , reaching out to at least 9 kpc . therefore , we base our analysis on a function that is designed to fit not just the inner or the outer part of galaxies , but rather the galaxy as a whole . to identify core galaxies , we follow the criteria in @xcite and @xcite , who define the core as `` a downward deviation from the inward extrapolation of the outer ( srsic ) profile '' . this involves fitting the galaxy in question with both the srsic and the core - srsicfunctions . the srsic profile @xcite is written as : @xmath16 \right\}. \label{sersiceq}\ ] ] @xmath17 is the intensity at @xmath18 , the projected half - light radius . @xmath19 is called the srsic index which describes the shape or curvature of the light profile . the quantity @xmath20 is a function of @xmath19 , defined in such a way that @xmath18 encloses half of the total luminosity . we approximate @xmath21 by using the asymptotic expansion of @xcite , eq . 18 for @xmath22 . many other approximation formulas are available and are summarized in @xcite . for @xmath23 , the srsic function reduces to an exponential function and for @xmath24 , it becomes the de vaucouleurs profile @xcite . the core - srsic function introduced by @xcite and @xcite is expressed as : @xmath25^{\gamma/\alpha}\,{\rm exp}\left[-b_n\left(\frac{r^\alpha+r_b^\alpha}{r_e^\alpha}\right)^{1/n\alpha}\right ] \label{coresersiceq1}\ ] ] with @xmath26 . \label{coresersiceq2}\ ] ] this profile consists of a srsic profile in the outer part , specified by the projected half - light radius @xmath18 and the srsic index @xmath19 , and a power - law profile in the inner part with a slope of @xmath27 . the change from one to another regime occurs at the break radius @xmath28 and the sharpness of the transition is specified by the parameter @xmath29 . higher values of @xmath29 mean sharper transitions . @xmath20 is approximated in the same way as for the srsic function . we summarize four criteria in @xcite that we use to identify core galaxies below . in section [ corefitting ] we examine each criterion . 1 . a characteristic pattern should be visible in the residuals when fitting a srsic profile to an idealized core galaxy . this serves as a qualitative evidence . 2 . a core - srsic function gives a significantly better fit than a srsicmodel . quantitatively , it is expected that the reduced @xmath30 value for the srsic fit @xmath31 is larger than twice that of the core - srsicfit @xmath32 . 3 . to avoid an ambiguous core detection , the potential core must be well - resolved by the data . quantitatively , @xmath28 should be @xmath33 where @xmath34 is the second innermost datapoint in the observed profile . 4 . the power - law slope has to be consistently less than the slope of the srsic fit inside @xmath28 . we decided to adopt an additional criterion based on resolution limits . trujillo et al . did not include such a limit , partly because many of their profiles did not probe radii smaller than the _ hst _ resolution limit due to dust , etc . ; however , they did note that one of their `` possible core '' galaxies ( ngc 3613 ) had a break radius smaller than the resolution limit of 0.16 originally suggested by @xcite . the latter limit is @xmath35 times the fwhm of most of the pc data used by @xcite ; generalizing this , we suggest the following additional criterion : 1 . the potential core should have @xmath36 the fwhm of the imaging data used for the innermost part of the profile . for most of the galaxies considered in this paper , which use optical wfpc2 or acs / wfc data , this corresponds to @xmath37 ; for two galaxies using nicmos2 data ( ngc 4261 and ngc 4374 ) , the limit would be @xmath38 , and for ngc 1550 , where we use sinfoni data with a fwhm of 0.17 , the limit would be 0.34 . we construct a one - dimensional light profile along the circularized radius ( mean profile ) , instead of along the semi major radius as is commonly done . the circularized radius is defined as @xmath39 where @xmath40 and @xmath41 are the semi major and semi minor axes radius of the isophote . this way , we take into account the variation of ellipticity with radius , giving a fair representation of the isophotes . the use of this mean profile is consistent with the implementation of the psf ( point spread function ) correction ( section [ psfconv ] ) . the fitting is done using a one - dimensional profile fitting software written by p. erwin . an initial model is generated based on user - given starting values . it is compared to the observed profile and the @xmath30 is calculated . the best - fitting solution is achieved by minimizing the @xmath30 of the fit , using the levenberg - marquardt algorithm @xcite . the user provides a starting value and its range of allowed values for each free parameter of the model . there are three free parameters in the srsic model , i.e. @xmath19 , @xmath18 , @xmath17 and there are six free parameters in the core - srsic model , i.e. @xmath19 , @xmath28 , @xmath18 , @xmath42 , @xmath29 , @xmath27 . since we are mainly interested in the inner region of the galaxy , the blurring of the central light profile by the psf becomes important and has to be taken into account . for several galaxies , the photometric profiles are derived from deconvolved images ( see table [ tab : coresample ] ) . we fit these profiles as described in the previous paragraph . for the other galaxies we incorporate a psf correction routine , described in the next section , into the fitting program such that the initial model is convolved first before being compared to the observed light profile . in this case , the minimised @xmath30 represents the difference between the convolved model and the data . the psf convolution is done at every iteration . equal to 1/40 of the half - light radius of the srsic model . dashed lines and diamonds show the case of a circular srsic image ; the solid lines and rectangles are for an elliptical srsic image ( ellipticity = @xmath43 ) . the red lines are the unconvolved srsic models . the black symbols are the results of one - dimensional convolution ( equation [ equ5]-[equ7 ] ) and the blue lines are the radial profiles of the convolved images , derived using ellipse - fitting . * bottom panel : * the flux ratio or the difference in surface brightness ( @xmath44 ) between the one - dimensionally convolved srsic profile ( @xmath45 or @xmath46 ) and the radial profile of the convolved srsic image ( @xmath47 or @xmath48 ) for both the circular and elliptical cases . + ] for galaxies whose light profiles are not seeing - corrected , we take the psf into account in the fitting by means of convolution . we use the psf corresponding to the galaxy image used to derive the innermost galaxy isophotes . for the galaxies from , the psf image is the same one which was used in the deprojection . for the other galaxies , the psf image is generated using tinytim by supplying the appropriate filter and camera . from the image , we generate a circularized profile of the psf by averaging its radial profile at different azimuthal angles ( done using midas ) . we further normalize the psf profile such that the total flux is one . the implementation of the convolution follows @xcite ( their appendix a ) . below , we review the method . the psf convolution is carried out by multiplying the galaxy surface brightness profile and the psf in the fourier space . on the sky , the galaxy surface brightness is a function of a two - dimensional position * r * and so is the psf . if @xmath49 and @xmath50 denote the surface brightness and the psf distribution , then the convolved surface brightness profile becomes : @xmath51 where @xmath52 since we use a one - dimensional , circularized surface brightness profile , i or p are a function of only the radius @xmath53 . in this case , @xmath54 where @xmath55 is the bessel function of order zero . then , equation [ eqq ] becomes : @xmath56 with @xmath57 @xmath58 in this way , although the surface - brightness and the psf profiles are one dimensional , the convolution is implemented such that the light coming from different angles is also taken into account . this is important because the light is scattered , due to the seeing , not only in the radial but also in the azimuthal direction . equations [ equ5]-[equ7 ] are also called the hankel transform . we compute the convolved surface brightness profile numerically , making use of integration routines in the gnu scientific library . fig . [ testconv ] shows how well this convolution works in the case of circular ( upper profiles ) and non - circular ( lower profiles ) images . for the former , we generate a galaxy image which follows a srsic law with @xmath24 and zero ellipticity everywhere using the ` makeimage ` program from the two - dimensional image - fitting package imfit ( erwin , in preparation ) . for the psf , we create a circular gaussian image with @xmath3 = 1/40 of the half - light radius of the galaxy model . we extract the radial profile of both images and use these to perform the one - dimensional convolution as described above . as an independent check , we separately convolve the galaxy image with the gaussian image in imfit . the radial profile of this convolved image ( @xmath47 ) is compared to the result of the 1d convolution ( @xmath45 ) . the ratio is shown in the lower panel of the figure by the dashed line . this process is then repeated for a srsic image with an ellipticity of 0.4 . the circularized profile of the convolved elliptical image is shown by the solid blue line and this is compared to the 1d convolution result ( the rectangles ) ; the ratio of both is displayed as the solid line in the lower panel . the method works well in the case of circular galaxy and shows a deviation of less than 0.02 mag arcsec@xmath59 for an elliptical image ; we have repeated this exercise for different srsic indices , resulting in similar conclusions . since the ellipticity of 0.4 is larger than the ones measured in the majority of galaxies in our sample , we expect smaller errors in the actual galaxy case . the list of galaxies examined in this section is given in table [ tab : coresample ] , along with the data sources . we compiled this list taking into account all galaxies which had both dynamically measured central black hole mass and evidence for cores , but in the end retained only those with clear evidence for a core _ and _ good enough data for measuring the core properties . the recent compilation of @xcite lists 9 more galaxies which we do not include : a1836-bcg , a3565-bcg ( ic 4296 ) , ngc 524 ngc3607 , ngc4473 , ngc5077 , ngc5128 , ngc5576 , ngc7052 . strong nuclear dust in a1836-bcg , ic 4296 , ngc3607 , ngc 5128 and ngc7052 makes it very difficult to robustly compute the core properties , so these galaxies are not included . ngc 4473 has been shown not to be a core galaxy by @xcite , , and @xcite , while the same is true for ngc 5576 @xcite ; see also @xcite . in addition , @xcite were only able to classify ngc 5077 as a `` possible core '' galaxy , so we exclude it as well . ngc 524 is a particularly problematic case , since there is evidence for very complex morphology e.g. , @xcite fit a ground - based image using a srsic bulge , an exponential disk , and two `` lens '' components . this means that a simple core - srsic fit or even core - srsic + outer envelope , as we use for some galaxies is probably not appropriate . in addition , the best available high - resolution data is _ hst_-nicmos2 , and the fit to this data by @xcite suggests a core break radius @xmath60 , which is smaller than our adopted resolution limit for nicmos2 data ( see section [ identifycore ] ) . ( wfpc2 images are not usable due to the extensive circumnuclear dust lanes . ) we reluctantly exclude this galaxy as well . ngc1374 represents a case where the core detection is ambiguous , so it is discussed independently in a separate section . this galaxy was classified as a non - core galaxy in @xcite using a limited dataset from , and also by @xcite using a more extended acs - wfc profile from @xcite ; in section [ n1374 ] we show the fit using our surface brightness profile which is slightly more extended in radius and which does not suffer the possible sky - subtraction uncertainties of the acs - only turner et al . table [ tab : csparams ] summarizes the list of galaxies for which the presence of core is confirmed . the best - fitting core - srsic parameters are also given in the same table . the best - fit srsic and core - srsic functions of these galaxies are shown in fig . [ fittingplots ] and the criteria which lead to the core identification are discussed below . when possible , we use all the available datapoints in the fitting . for all galaxies , the core - srsic function gives a much better fit to the whole profile than the srsic . in several galaxies , however , we find that the fit residuals become rather large at several outermost datapoints . since these datapoints do not seem to be well - described by a srsicfunction , they are excluded from the fit . we keep only the datapoints where the fit residual is less than 0.15 mag . the deviation from the srsic model at large radii by more than 0.15 mag may be caused by imperfect sky subtraction or it may represent a real departure from the srsic profile , for example when where an outer halo is present ; see section [ multi - component - fits ] for details on fits using extra components . the fit is then repeated using only the remaining datapoints , after excluding the outliers . the fit residuals , their rms and the @xmath30 values are presented in fig . [ fittingplots ] . in this figure , the surface brightness profiles shown are not corrected for galactic extinction . in all the galaxies in fig . [ fittingplots ] we see a pattern in the residuals of the srsic fit that qualitatively suggests the presence of the core compare to fig . 3 in @xcite . this residual pattern disappears when the galaxy is fitted with the core - srsic profile . so the first criterion in section [ identifycore ] is secured . in each case , the rms of the residuals for the best srsic fit is much larger than that for the core - srsic . the @xmath30 values for each galaxy are calculated by adopting the residual rms of the best core - srsic fit as the error on the surface brightness measurements . as discussed in , this estimate for the real error is reasonable as long as the rms is of order of the profile measurement error , i.e. , a few hundredths of a mag arcsec@xmath59 , which is indeed the case for all the galaxies that we examine . using this error approximation , the @xmath30 of the core - srsic fit is artificially set to 1.0 . note that the rms and @xmath30 values are dependent on the radial extent and the radial sampling ( logarithmic sampling is used here ) . we find that the ( reduced ) @xmath30 value for each srsic fit is much more than twice that of the core - srsic . the average ratio of @xmath61 is around 49 , with a minimum of 6.8 , easily fulfilling the second criterion . to assess the third criterion , one can compare the position of the second innermost datapoint to the position of the dotted vertical line which marks the break radius . it is thus clear that we have cases of resolved cores . the fourth point is also satisfied , since inside the break radius the core - srsic function shows a consistent deficit in light compared to the extrapolated srsicprofile ( shown by the blue line ) . for the cases where we exclude some datapoints , the core is already well identified when fitting all the datapoints , i.e. fulfilling the four criteria simultaneously . limiting the fitting range only improves the distinction . lastly , the break radius for each galaxy is larger than twice the fwhm of the imaging data used to construct the innermost part of the surface brightness profile , described more quantitatively in section [ identifycore ] . among the 23 galaxies , several exhibit more complex structures . inside @xmath62 , the surface - brightness profile of ic1459 seems to show a small excess of light . we model this nuclear excess as an extra gaussian component in the center on top of the core - srsic function , which after psf convolution provides a good fit to the data . ( in the cases of ngc1407 and ngc4552 , it is not clear that their inner excesses can be modeled with a gaussian , so we exclude the data at @xmath63 for ngc1407 and @xmath64 for ngc4552 . ) for six galaxies ( including ngc1407 ) , we find that significantly better fits are achieved by including one or two extra srsic or exponential components , representing outer halos . this approach is discussed in more detail in section [ multi - component - fits ] . the parameter values of the best - fit core - srsic model for each galaxy are provided in table [ tab : csparams ] . the uncertainties of the core - srsicparameters are determined through a monte carlo simulation ( see next section ) . the srsic indices for these galaxies are almost all larger than 4 , something which is common for giant ellipticals ( e.g. , @xcite , ) . the exceptions are four of the galaxies which we model with multiple components ( see section [ multi - component - fits ] ) . the best - fit @xmath29 parameter for ngc7768 is found to be 16.6 . this large value indicates a very sharp transition between the srsic and the power - law profile . when the fit is repeated with @xmath65 , which is a good approximation for @xmath66 @xcite , the best - fit values for the other five parameters hardly change . since for large @xmath29 ( @xmath67 10 ) the profile becomes insensitive to the exact value of @xmath29 , we prefer to use the latter fit ( fixing @xmath29 to 100 ) to the light profile of ngc7768 . this choice prevents wildly varying values of @xmath29 that are not physically meaningful in the monte carlo simulations . for six galaxies whose surface brightness profiles are taken from ( i.e. , ngc4261 , ngc4374 , ngc4472 , ngc4486 , ngc4552 , and ngc4649 ) , the best - fitting srsic model of is also overplotted on the right - hand panels of fig . [ fittingplots ] . our inward - extrapolated srsic components and their srsic fits do not coincide , reflecting the differences in the way the core is defined . note that use major - axis profiles while we use circularized profiles , resulting in a seemingly poorer fit of kfcb s srsic models in fig . [ fittingplots ] . for a more thorough comparison between our and their fitting procedure , see appendix [ comparisonkfcb09 ] . ngc3091 and ngc7619 show small values in @xmath29 , i.e. @xmath68 . @xcite produce a best - fitting core - srsic model of ngc7619 by keeping @xmath29 constant at 5 , which enforces a rather sharp transition from srsic to power - law profiles . doing this exercise using our circularized profile of ngc7619 results in a smaller @xmath19 ( 7.71 ) , larger @xmath28 ( 0.73 ) and @xmath27 ( 0.47 ) and a larger rms ( 0.033 ) . it is worth emphasizing that @xcite s profile for ngc7619 is not as extended as ours and stops short of @xmath69 . ngc1374 is the one galaxy in the sample of with conflicting arguments about the existence of a core . classify ngc1374 as a core galaxy , based on a nuker profile fit to their _ hst _ imaging data . @xcite , however , do not confirm the core in this galaxy using the same data , and @xcite come to the same no - core conclusion using a more extended acs - wfc based profile from @xcite . in contrast , @xcite argues for the presence of the core . from fitting core - srsic models to our data , we find no solid evidence that a core is present in this galaxy . we tested different fitting ranges by varying the maximum radii of the fitted datapoint , but there is no strong visibility of a core in any case . in the top row of fig . [ fitting1374 ] , we show the srsic and core - srsic fits to the whole profile and to a limited profile . for the latter , we limit the fitting range to @xmath70 ( for a fair comparison to the fitting using data , see below ) . judging by eye , there is no significant difference in fit quality between the srsic and the core - srsicprofiles , although the overall residual rms and @xmath30 of the core - srsic are smaller and @xmath71 . when fitting the whole range , the srsic profile actually provides a better fit in the center and instead of seeing a core , the core - srsic fit indicates an extra light on top of the srsic component ( this might be consistent with the extra nuclear component used by @xcite in their fits to this galaxy ) . there is a slight improvement and a vague indication of a core when we fit only up to 7 . however , if we base our core classification on @xcite s criteria , this fit fails at the second criterion . our next attempt is to fit the data of to confirm or reject their finding . their profile has been corrected for the psf effect , so we directly fit it with the models without convolution . as shown in fig . [ fitting1374 ] , we detect a core in this galaxy , albeit weak , by discarding the two outermost datapoints ( @xmath72 ) . extending the fitting range of data does not give the same results , i.e the existence of the core becomes inconclusive and the core disappears altogether when all the datapoints are included in the fit . fit their light profile with a nuker model and it appears that they also made the compromise of neglecting the 34 last datapoints in the fit , in order to detect the core . moreover , our best - fitting break radius is smaller than 0.1 , less than the inner resolution limit that we adopt as an additional criterion of core detection in section [ identifycore ] . one might conclude that our data lack spatial resolution and that this situation is aggravated by the psf convolution in the fitting process . it is , however , important to note that the core disappears when extended datapoints are included in either our profile or the profile . graham et al . ( 2003 ) pointed out that fitting a nuker model to a limited light profile , such as the case of ngc1374 in , could lead to a false conclusion that a core exists . having no definite evidence of a depleted core in ngc1374 , we exclude it from the core sample . [ cols= " < , < , < , < , < , < , < , < " , ] taking the break radius ( @xmath28 ) as a measure of the core size , we examine its relationships with galaxy velocity dispersion , luminosity and the black hole mass . these are displayed in fig . [ rbplots ] . the fit to the datapoints ( also for all the other fits in this section ) is based on a bayesian method with gaussian errors and intrinsic scatter @xmath73 , described in @xcite , and the results are shown by the black lines . for each diagram we perform non - symmetrical regression using either of the parameter to predict the other . we further compute the bisector of the two fitting results ( with flipped @xmath74 and @xmath75-axes ) to give a symmetrical representation of the correlation . the coefficients of the fits involving @xmath28 are shown in table [ tab : core - relations ] . all the fits are linear fits to the logarithm of the relevant quantities expressed in the following form : @xmath76 where @xmath75 is @xmath77 ( in kpc ) and @xmath74 is @xmath78 , @xmath79 , or @xmath80 . for the `` direct '' fits @xmath81 , @xmath79 , and @xmath82 are the independent variables in the fit , while in the `` reverse '' fits @xmath28 is the independent variable . the errors on the coefficient values are taken as the rms of the a - posteriori parameter probability distribution . we also list the spearman correlation coefficients and the corresponding probabilities for each relation in table [ tab : core - relations ] . @xcite analysed the same relationships using bisector fits . in the @xmath28@xmath81 diagram , our fit predicts a slightly higher @xmath28 for a given velocity dispersion , though the fits are compatible within the errors . in the @xmath28@xmath83 diagram , our fit coincides with their fit for high - luminosity galaxies ( @xmath84 ) . in the case of the @xmath28@xmath0 relation , graham & dullo performed fits using both measured black hole masses ( their equation 12 ) and proxy black hole masses estimated from the @xmath85@xmath3 and @xmath0@xmath86 relations ( their equation 13 and 14 ) ; only the former is actually plotted ( their fig . 20 ) . since our fits use only measured black hole masses , we compare our results with their equation 12 ( lower right panel of figure 6 ) . for this relation , our fit suggests larger @xmath0 for the same value of break radius ( or , conversely , smaller break radius for a given @xmath0 ) . we note that the fit in dullo & graham s equation 12 is based on only 7 galaxies , all but one of which have black hole masses @xmath87 , versus the 20 galaxies used in our fit , with 14 having @xmath88 . the measured rms scatter is almost identical for the @xmath28@xmath81 and @xmath28@xmath0 fits ( 0.24 dex versus 0.23 dex ) . the tightest relationship turns out to be between @xmath28 and @xmath83 , where the rms is 0.20 dex . the correlation coefficients indicate that all three relations are strong and statistically significant , with the strongest being that between @xmath89 and @xmath0 ( @xmath90 , @xmath91 ) . both @xcite and @xcite use a larger number of datapoints than ours in examining the @xmath89@xmath92 and @xmath89@xmath93 relationships , and they find opposing results as to which parameter ( @xmath92 or @xmath93 ) is better correlated with @xmath89 ; our results suggest that @xmath93 is a better predictor of @xmath89 than @xmath92 . ( note that the @xmath89@xmath93 correlation uses the _ total _ galaxy luminosity ; if for the multi - component galaxies we use the luminosity of just the core - srsic component instead , the correlation is weaker and only marginally significant , with @xmath94 and @xmath95 . ) despite the moderately large scatter , the @xmath89@xmath0 relation is apparently the strongest one among the three , suggesting that @xmath85 is the most relevant parameter for determining @xmath89 . [ deficitplots ] explores relationships between the light or mass deficit in the core and the black hole mass or the stellar velocity dispersion . the upper left panel shows the plot of the light deficit in the @xmath96-band versus the stellar velocity dispersion calculated within the effective radius @xmath81 . the correlation between stellar velocity dispersion of host galaxies and @xmath97 determined in is overplotted in red . note that the velocity dispersion used to construct the latter correlation is defined differently : it is based on the velocity dispersion averaged within a slit aperture with a length of two half - light radii . although there is probably no significant difference between these two methods of determining @xmath81 , this means that the individual datapoints can not be strictly compared to the relation . for seven galaxies that overlap with s sample , we indicate their values of light deficit using gray diamonds . on average , our values are more negative ( i.e. more luminous ) than those of by 0.30 mag , and the rms of the difference is 1.15 mag . in the upper right panel of fig . [ deficitplots ] , the plot between @xmath97 and the black hole masses is shown . similarly , we overplot the correlation between these two quantities , as derived in , in red . coefficients of the fits plotted in fig . [ deficitplots ] are listed in table [ tab : mdef - relations ] . the meaning of the coefficients is the same as in equation [ eqn : basic ] , except now @xmath75 corresponds to either @xmath98 or @xmath99 . in the direct fits , the independent variables are either @xmath100 or @xmath4 . compared to , our @xmath97@xmath81 correlation is steeper and our @xmath97@xmath0 exhibits a shift towards a larger @xmath0 at a given value of light deficit . this highlights the difference between estimates based on the @xmath0@xmath3 relation and the more reliable dynamical estimates of @xmath0 that we use . the scatter is notably higher in @xmath97@xmath0 diagram , opposite to what is found in . the most interesting relation from a theoretical point of view is that between the _ mass _ deficit and the black hole mass , since this is the quantity which simulations predict ( see references in the introduction ) . the lower panels of fig . [ deficitplots ] show this relation , and also the relation between mass deficit and velocity dispersion ; the coefficients of the fits are in the bottom half of table [ tab : mdef - relations ] . the large intrinsic scatter ( and the large rms values ) that we derive indicate that the linear fit is not representing the data well . of the seven correlations we have examined ( figs . [ rbplots ] and [ deficitplots ] ) , those between @xmath28 and @xmath0 and between @xmath97 and @xmath81 appear to be the strongest ( @xmath101 and @xmath102 , respectively ) . moreover , the rms scatter in @xmath103 for the fit between @xmath0 and @xmath28 is only 0.30 dex ( it is 0.35 dex for the correlation with @xmath104 and 0.33 dex for that with the light deficit ) . this is comparable to or even lower than the typical rms scatter for the @xmath0@xmath3 relation ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? this suggests that measuring @xmath89 can allow one to estimate @xmath0 ( in core galaxies ) with a precision similar to that obtained using the @xmath85@xmath3 relation , without the need for spectroscopy . we give here below the corresponding equation ( derived from the fits in table [ tab : core - relations ] ) : @xmath105 the ratio between @xmath104 and @xmath0 seems to vary strongly , but is biased toward the low end of the range . most galaxies have a mass deficit between 1 and 5 @xmath0 , and the median value is 2.2 @xmath0 . in three galaxies ( ic1459 , ngc3842 and ngc5813 ) , the mass deficit is just below the black hole mass . ngc3608 has a remarkably low mass deficit : only 0.21 @xmath0 . the largest mass deficit that we find is just below 40 @xmath0 , occurring in ngc1399 . this unusually large value should perhaps be taken with a grain of salt , however , given the large uncertainty in the black hole mass of ngc1399 . the two publications which have presented dynamical measurements of the black hole mass @xcite provide values which differ by almost a factor of 3 , and neither measurement included the dark matter halo in the modelling , which could likely result in underestimating @xmath0 and thereby overestimating the ratio between the mass deficit and @xmath0 . the next - largest mass deficit is 18 times @xmath0 , in ngc4261 , with the one other galaxy having a mass deficit larger than 10@xmath0 being ngc4486 . the broad range of @xmath106 ( between 0.2 and 18 if we exclude ngc1399 ) is in better agreement with @xcite than with . the former also shows a number of galaxies having mass deficits of 14 @xmath0 while the latter has no galaxies in their sample with mass deficits below 5 @xmath0 . considering that and @xcite use mostly outdated @xmath0values and an old @xmath0@xmath3 relationship for their analysis , it is likely that their ratios of @xmath106 are simply too high . ( for the seven of our galaxies in common with those in @xcite , we find that their mass deficits are somewhat higher than ours median ratio @xmath107 while their black hole masses are only @xmath108% of ours . ) the simulations of @xcite indicate that mass deficits up to @xmath109 can be generated from single dry mergers . therefore , our results can probably be explained with a reasonably low number of major dry mergers . low @xmath106 ratios could also be obtained if gas to refill the core via new star formation is available ( i.e. , the merger is `` wet '' ) . why does @xmath28 correlate so well with @xmath0 , while the light and the mass deficit do not ? we note that our estimate of @xmath110 , from which the mass deficit is in turn derived , is based on the assumption that the inward extrapolation of the outer srsic fit is the correct representation of the inner projected density profile of the galaxy before the binary black hole scouring takes place . this might not be the case for real galaxies , and the large scatter about the relations of fig . [ deficitplots ] could simply reflect this ( see , e.g. * ? ? ? the size of the core may be a more robust quantity in this respect . it will be interesting to investigate whether the internal kinematics of the galaxies change at this radius . we have presented measurements for cores in 23 elliptical galaxies , where the cores are clearly detected based on the criteria suggested in @xcite . these detections and measurements were made using core - srsic fits to extended surface - brightness profiles , the majority of which go out significantly beyond the half - light radius ; for a quarter of the sample , we found that the best fits were with the sum of core - srsic and an outer envelope , the latter represented by an exponential or srsic component . for 20 of these galaxies , reliable @xmath0 measurements are available from the literature or from our own work ; the majority of these are derived from recent stellar dynamical modelling . we computed the light deficit in the core ( assumed to be the result of supermassive black hole mergers which follow galaxy mergers ) assuming that the original profile , before the core was formed , followed a srsic model matching the outer part of the galaxy , based on the core - srsic fit ( or the outer part of the core - srsic component in the case of multi - component fits ) . we then derived the corresponding stellar mass deficit using stellar mass - to - light ratios from dynamical models , available for all of the galaxies in the sample . we find that the size of the core ( break radius @xmath89 ) is most strongly correlated with the black hole mass , while the second strongest correlation ( and the linear fit with the lowest scatter ) is with the luminosity of the host galaxy ; the weakest correlation is that between the break radius and the galaxy s stellar velocity dispersion . the low scatter of the @xmath28@xmath0 relation ( 0.30 dex when @xmath89 is the independent variable ) opens up the possibility of accurately estimating black hole masses in core galaxies without the need for spectroscopy . the stellar mass deficits we derive range between 0.2 and 39 times the black hole mass ; for about 75% of the sample , the range is 15 @xmath0 , and the median value for the whole sample is 2.2 @xmath0 . given the results of recent simulations , these values can be explained by a reasonably small number of ( dry ) mergers . we thank the anonymous referee for a number useful comments that helped us significantly improve the paper . spr was supported by the dfg cluster of excellence `` origin and structure of the universe '' . pe was supported by the deutsche forschungsgemeinschaft through the priority programme 1177 `` galaxy evolution '' . some of the data presented in this paper were obtained using sinfoni at the very large telescope vlt and from the mikulski archive for space telescopes ( mast ) . the vlt is operated by the european southern observatory on cerro paranal in the atacama desert of northern chile . stsci is operated by the association of universities for research in astronomy , inc . , under nasa contract nas5 - 26555 . this research is based in part on data from the isaac newton group archive , which is maintained as part of the casu astronomical data centre at the institute of astronomy , cambridge . funding for the creation and distribution of the sdss archive has been provided by the alfred p. sloan foundation , the participating institutions , the national aeronautics and space administration , the national science foundation , the u.s . department of energy , the japanese monbukagakusho , and the max planck society . the sdss web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium ( arc ) for the participating institutions . the participating institutions are the university of chicago , fermilab , the institute for advanced study , the japan participation group , the johns hopkins university , the korean scientist group , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . finally , this work 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 . we constructed the circularized @xmath96-band photometry of ic1459 by matching _ hst _ and ground based data . inside 12 we used an _ hst _ wfpc2 f814w profile ( proposal i d : 5454 , pi : marijn franx ) derived using the software of @xcite while masking the central dusty features . at larger radii we used ccd data from @xcite ( see also @xcite ) , calibrated with photoelectric photometry available in hyperleda . the circularized @xmath111-band photometry of ngc1399 was constructed by matching the extended profile considered in @xcite to the profile derived using the iraf software from an _ hst _ acs f606w image ( proposal i d : 10129 , pi : thomas puzia ) of the galaxy in the inner 12 . the light profile for this galaxy is based on that used in for the bh mass measurement . to facilitate the exploration of possible outer - envelope models ( section [ multi - component - fits ] ) , we replaced the inner 7.5 of the previous profile with data from a deconvolved version of the _ hst_-acs f435w image ( 20 iterations of lucy - richardson deconvolution ) . this allowed us to fit the profile without psf convolution . the light profile was constructed using wfpc2 f814w images from the _ hst _ archive ( proposal i d : 5512 , pi : sandra faber ) and sdss @xmath112-band images . the pc chip image of _ hst _ was deconvolved using 20 rounds of lucy - richardson deconvolution and then the profile was measured through ellipse fitting . that profile was used for @xmath113 . at larger radii , we used an ellipse - fit profile measured using a mosaic of sdss @xmath112-band images , constructed using swarp @xcite . the sdss @xmath112-band magnitudes were converted to cousins @xmath114 using an @xmath115 color of 0.44 ( measured from ellipse fits to single sdss @xmath53 and @xmath112 images ) and the transformation of @xcite . the surface - brightness profile for ngc3608 is a combination of ellipse fits to an f555w wfpc2 image ( proposal i d : 5454 , pi : marijn franx ) for @xmath116 and ellipse fits to an sdss @xmath117-band image for @xmath118 . the sdss image was calibrated to @xmath96-band using the sdss @xmath117 and @xmath53-band images and the color transformations of lupton . we matched the two datasets using fixed - ellipse - fit profiles to the sdss image and the wfpc2 mosaic , using the overlap between 685 to determine the best scaling and background subtraction to apply to the wfpc2 data . the surface brightness profile for ngc3842 is calibrated to the @xmath96-band and is assembled from three pieces . the innermost part ( @xmath119 ) comes from the published f555 wfpc2 pc1 profile of . the profile in the intermediate radii is derived from an ellipse fit to a f555w wfpc2 mosaic image ( galaxy center on wf3 chip ; proposal i d : 265 , pi : douglas geiser ) . this is matched to an ellipse fit to an sdss g - band image , including determining the level of sky background of the wfpc2 image . the combination then consists of wfpc2 data at @xmath120 and sdss data at @xmath121 . the final surface brightness profile is constructed by matching the combined profile and the profile . for ngc4291 , the profile is a combination of ellipse fits to an wfpc2 f702w image [ proposal i d : 6357 , pi : walter jaffe ; see @xcite ] and ellipse fits to a sloan @xmath53-band isaac newton telescope wide field camera ( int - wfc ) image retrieved from the isaac newton group ( ing ) archive . the latter was a 120s exposure originally obtained on 2004 january 24 . the int - wfc image was calibrated to @xmath96-band using aperture photometry from @xcite . we matched the two datasets using fixed - ellipse - fit profiles to the int - wfc image and the wfpc2 mosaic , using the overlap between 1080 to determine the best scaling and background subtraction to apply to the wfpc2 data . the final profile used data from the wfpc2 pc chip for @xmath70 and data from the int - wfc image for @xmath72 . the profile for this galaxy was constructed by combining data from wfpc2 f814w images from the _ hst _ archive ( proposal i d : 6579 , pi : john tonry ) and an @xmath122-band image from the wide field imager of the eso - mpi 2.2 m telescope . the sky background for the wfpc2 data was determined by matching ellipse - fit profiles of the full wfpc2 mosaic with a profile from the wfi image . the pc chip image of _ hst _ was deconvolved using 20 rounds of lucy - richardson deconvolution and then the profile was measured through ellipse fitting ; the resulting profile was used for @xmath123 . at larger radii , we used an ellipse - fit profile from the wfi image . the profile for this galaxy is a combination of ellipse fits to an f814w wfpc2 image ( proposal i d : 5454 , pi : marijn franx ) and an sdss @xmath112-band image . since patchy dust was present in the nucleus , we combined the f814w image with the corresponding f555w image to perform a simple dust extinction correction on the f814w image . sky backgrounds for the two filters were determined by matching ellipse - fit profiles of the full wfcp2 mosaic image with ellipse - fit profiles from the sky - subtracted sdss @xmath117 and @xmath112 images . the f555w and f814w images were also used , along with the prescriptions in @xcite , to generate @xmath96-band and @xmath114-band calibrations for each , iterating from an intial guess for the central ( @xmath113 ) @xmath124 color until the measured color converged . we then formed a @xmath124 pc - chip image and used this to color - correct the f814w pc image , assuming a simple screen model for the dust . the final profile used the ( extinction - corrected ) wfpc2 pc - chip data for @xmath125 and sdss @xmath112-band data at larger radii . since ngc5846 had available the same kind of data as ngc5813 _ and _ similar patchy dust extinction in the circumnuclear region , we followed an essentially identical process for this galaxy . the final profile used the ( extinction - corrected ) wfpc2 pc - chip data for @xmath126 ( proposal i d : 5920 , pi : jean brodie ) and sdss @xmath112-band data at larger radii ( the difference in transition radius compared with ngc5813 is due to the larger extent of circumnuclear dust in ngc5846 ) . we used a wfpc2 f814w image ( proposal i d : 7281 , pi : roberto fanti ) and an sdss @xmath112-band image to construct the surface brightness profile based on combined ellipse fits . the part of the profile inside 11 comes from the pc chip of wfpc2 ; and the innermost 1.5 is from the pc chip of wfpc2 after performing 20 rounds of lucy - richardson deconvolution . the light profile is calibrated to the @xmath122-band . this was done by first converting the sdss @xmath112-band profile to sdss @xmath53 by matching with a profile from the sdss @xmath53-band image . the sdss @xmath53-band magnitudes were then converted to cousins @xmath122 using a @xmath127 color of 0.95 ( measured from the sdss g and r images ) and the conversion of @xcite . the light profile of ngc7768 is mostly based on wfpc2 f814w images ( proposal i d : 8184 , pi : p. ct ) , which were combined to make both a pc image and a mosaic . the pc image was corrected for dust extinction using the f814w image and f450w wfpc2 images , but there was still considerable residual extinction inside 1 , so an _ hst_-nicmos2 f160w image was used for the inner region . the final profile is obtained through ellipse fits to the following : ( 1)nicmos f160w image for @xmath128 , ( 2)wfpc2 f814w pc image for @xmath40 between 0.93 and 10 and ( 3)wfpc2 f814w mosaic for @xmath129 . the f814w image was corrected for sky background by matching a fixed - ellipse - fit profile with a similar fixed - ellipse - fit profile from an @xmath53-band image with the isaac newton telescope wide field camera ( 100s , 3 dec 2002 ) , taken from the isaac newton group archive . the profiles were calibrated to @xmath96-band using aperture photometry from @xcite ( specifically , the int - wfc @xmath53-band image was calibrated to @xmath96 , and then the wfpc2 f814w profile was matched to the int - wfc @xmath53-band profile . ) of all the galaxies presented in fig . [ rbplots ] , there are five galaxies which are also part of sample , i.e. ngc4472 , ngc4486 ( m87 ) and ngc4649 , ngc4261 and ngc4374 . using the photometric data available in , we derive the missing light using the core - srsic function as described in sections [ themethod ] and [ coredeficit ] . from these five measurements , we find that some of our @xmath110 fall outside the uncertainties provided by . the difference is largest for ngc4486 and ngc4261 ( by @xmath1300.7 magnitude or a factor of two in luminosity ) and it is not systematic . the disagreement probably reflects the different details that were involved in the fitting and the derivation of the missing light : \1 . we use the circularized radius ; uses semi - major axis radius . galaxies are not perfectly round or circular in projection . non - zero ellipticity anywhere in the galaxy , not just in the center , will induce a deviation in @xmath110 since the light deficit in both and this work depends also on the srsic fit to the outer part of the galaxy . the magnitude of the deviation is difficult to estimate , since this depends very much on the characteristics of each galaxy . a different fitting range introduces more discrepancies . because we use circularized radius , we can use only the datapoints for which the ellipticity is given . in the photometric profiles of , the ellipticity is often not specified for a few innermost or outermost radii ; therefore we can not use those points in our fits . in addition , in the particular case of ngc4486 , used a srsic fit ( from , using semi - major axis ) which excluded data outside 400 , because it apparently gave rise to a srsic index which was too large ( almost 12 ) due to the cd halo . however , we found no difficulties in fitting a core - srsic function to the entire profile ( out to @xmath131 ) , using the circularized radius ; we obtain a srsic index of 8.9 ( almost identical to s value ) , albeit with different values of @xmath18 and @xmath132 . we fit the galaxy profile with a core - srsic profile , fit the outer part of the galaxy with a srsic profile . the critical point is the transition region between the srsic and the core . in the way defines the core , there is no transition region : the core immediately starts when the srsic model no longer fits the datapoints in the inner part . we use a core - srsic parametrization and since we fit the @xmath29 parameter , instead of fixing it to infinity , we allow some part of the galaxy to be partly srsic , partly power - law . this leads to a different inner limit of the radial range that is fitted only by the srsic model , which in the end results in a different best - fitting srsic profile . the light deficit in is obtained by calculating the difference of the integrated intensity between the extrapolated srsic function and the actual ( observed ) profile after psf deconvolution . the ellipticity of the srsic model is fixed to the ellipticity at the inner limit of the fitted radial range . since some of our light profiles are not psf - deconvolved , we instead compute the difference between the best - fitting core - srsic function and the extrapolated srsic component . we also implicitly assume an ellipticity of zero when integrating the extrapolated srsic function , since we have used a circularized profile in the first place .
minimizers of a submodular function form a distributive lattice , and are compactly represented by a poset ( partially ordered set ) via birkhoff representation theorem . this fact reveals a useful hierarchical structure of the minimizers , and is applied to the dm - decomposition of matrices and further refined block - triangular decompositions @xcite . in this paper , we address such a birkhoff - type representation for minimizers of a _ @xmath0-submodular function_. here @xmath0-submodular functions , introduced by huber kolmogorov @xcite , are functions on @xmath2 defined by submodular - type inequalities . this generalization of ( bi)submodular functions has recently gained attention for algorithm design and modeling @xcite . our main result is to establish a compact representation for minimizers of a @xmath0-submodular function . this can be viewed as a generalization of the above poset representation for submodular functions and ando fujishige s signed - poset representation for bisubmodular functions @xcite . a feature of our representation is to utilize a _ poset with inconsistent pairs ( pip ) _ @xcite , which is a discrete structure having a stronger power of expression than that of a signed - poset . actually a pip is a poset endowed with an additional binary relation ( _ inconsistent relation _ ) , and is viewed as a poset reformulation of 2-cnf . this concept , also known as an _ event structure _ , was first introduced by nielsen plotkin winskel @xcite as a model of concurrency in theoretical computer science , and was independently considered by barthelemy constantin @xcite to establish a birkhoff - type representation theorem for a _ median semilattice_a semilattice generalization of a distributive lattice . a pip was recently rediscovered by ardila sullivant @xcite to represent nonpositively - curved cube complexes ; the term `` pip '' is due to them . our results consist of structural and algorithmic ones , summarized as follows : [ [ structural - results . ] ] structural results . + + + + + + + + + + + + + + + + + + + we show that minimizers of a @xmath0-submodular function form a median semilattice ( lemma 3 ) . by a birkhoff - type representation theorem @xcite for median semilattices , the minimizer set is represented by a pip , where minimizers are encoded into special ideals in the pip , called _ consistent ideals_. pips arising from @xmath0-submodular functions are rather special . we completely characterize such pips ( theorem [ thm : closed_set_and_elementary_pip ] ) , which we call _ elementary_. this representation is actually compact . we show that the size of the elementary pip for a @xmath0-submodular function of @xmath3 variables is @xmath4 ( proposition [ prop : number_of_irreducibles ] ) . [ [ algorithmic - results . ] ] algorithmic results . + + + + + + + + + + + + + + + + + + + + we present algorithms to construct the elementary pip of the minimizers of a @xmath0-submodular function @xmath1 under the following three situations : * a minimizing oracle of @xmath1 is given . * @xmath1 is network - representable . * @xmath1 arises from a potts energy function . for ( i ) , we show that the pip is obtained by calling the minimizing oracle @xmath5 time ( theorem [ thm : alg_oracle ] ) . notice that a polynomial time algorithm to minimize @xmath0-submodular functions is not known for the value - oracle model but is known for the valued - csp model @xcite . our result for ( i ) is applicable to such a case . for ( ii ) ( and ( iii ) ) , we consider an efficiently minimizable class of @xmath0-submodular functions considered in @xcite , where a @xmath0-submodular function in this class is represented by the cut function in a network of @xmath4 vertices and can be minimized by a minimum - cut computation . we show that the pip is naturally obtained from the residual graph of a maximum flow in the network ( theorems [ thm : network_algo ] and [ thm : network_time ] ) . for ( iii ) , we deal with a @xmath0-submodular function @xmath6 obtained from a @xmath0-label potts energy function @xmath7 by adding the @xmath8-label ( meaning `` non - labeled '' ) . such a @xmath0-submodular function , called _ potts @xmath0-submodular _ , is particularly useful in vision applications . indeed , via the _ persistency property _ @xcite , a minimizer of @xmath9 ( an optimal labeling ) is partly recovered from a minimizer of the relaxation @xmath10 . kolmogorov @xcite showed that a minimizer of a potts @xmath0-submodular function can be obtained by @xmath11 calls of a max - flow algorithm performed on a network of @xmath12 vertices . we show that the pip is also obtained in the same time complexity ( theorem [ thm : potts ] ) . in showing this result , we reveal an intriguing structure of the pip for a potts @xmath0-submodular function ( theorem [ thm : pip_potts ] ) , and utilize results @xcite from undirected multiflow theory . we also discuss enumeration aspects for minimizers . maximal minimizers , which are minimizers with a maximal number of nonzero components , are of particular interest from the view of partial optimal labeling . for a potts @xmath0-submodular function , we show that the problem of enumerating all maximal minimizers reduces to the problem of enumerating all ideals of a single poset ( theorem [ thm : enumeration_potts ] ) . this enables us to use an existing fast enumeration algorithm , and leads to a practical algorithm enumerating all maximal partial optimal labeling in actual computer vision problems . we present experimental results for real instances of stereo matching problems . [ [ organization . ] ] organization . + + + + + + + + + + + + + the rest of this paper is organized as follows . in section [ sec : preliminaries ] , we give preliminaries including a birkhoff - type representation theorem between pips and median semilattices . in section [ sec : structure ] , we prove the above structural results . in section [ sec : algorithms ] , we prove the algorithmic results . finally , in section [ sec : application ] , we describe applications and present experimental results . for a nonnegative integer @xmath3 , we denote @xmath13 by @xmath14}}$ ] ( with @xmath15 } } \defeq \varnothing$ ] ) . for a subset @xmath16 of an ordered set , let @xmath17 denote the minimum element in @xmath16 ( if it exists ) . let @xmath18 be the set of real numbers and @xmath19 . for a function @xmath1 from a set @xmath20 to @xmath21 , a _ minimizer _ of @xmath1 is an element @xmath22 that satisfies @xmath23 for all @xmath24 . the set of minimizers of @xmath1 is simply called the _ minimizer set _ of @xmath1 . we assume that posets are always finite , and assume the standard notions of lattice theory , such as join @xmath25 and meet @xmath26 . let @xmath0 be a positive integer . let @xmath27 denote @xmath28 . the partial order @xmath29 on @xmath27 is defined by @xmath30 if and only if @xmath31 for each @xmath32 . consider the @xmath3-product @xmath33 of @xmath27 , where the partial order on @xmath33 is defined as the direct product of @xmath29 and is also denoted by @xmath29 . in this way , @xmath33 and its subsets are regarded as posets . for @xmath34 , the _ support _ of @xmath35 is the subset of indices @xmath36}}$ ] with nonzero @xmath37 , and is denoted by @xmath38 : @xmath39}}}\mathrel{}\middle|\mathrel{}{x_i \neq 0}\right\}}.\end{aligned}\ ] ] a _ @xmath0-submodular function _ @xcite is a function @xmath40 satisfying the following inequality @xmath41 for all @xmath42 . here the binary operation @xmath43 on @xmath33 is given by @xmath44 for every @xmath42 and @xmath36}}$ ] . the operation @xmath45 in is dually defined by changing @xmath46 to @xmath47 in . besides its recent introduction , a @xmath0-submodular function seems to be recognized when bouchet @xcite introduced _ multimatroids_. indeed , a @xmath0-submodular function is a direct generalization of the rank function of a multimatroid , and was suggested by fujishige @xcite in 1995 as a _ multisubmodular function_. it is not known whether @xmath0-submodular functions for @xmath48 can be minimized in polynomial time on the standard oracle model . however , some special classes of @xmath0-submodular functions are efficiently minimizable . for example , kolmogorov thapper ivn @xcite showed that a sum of low - arity @xmath0-submodular functions can be minimized in polynomial time , where the _ arity _ of a function is the number of variables . a nonnegative combination of binary _ basic @xmath0-submodular functions _ , introduced by iwata wahlstrm yoshida @xcite , can be minimized by computing a minimum @xmath49-cut on a directed network ; see section [ sec : network ] . a nonempty subset of @xmath33 is said to be _ @xmath50-closed _ if it is closed under the operations @xmath43 and @xmath45 . from , the following obviously holds . the minimizer set of a @xmath0-submodular function is @xmath50-closed . a key tool for providing a compact representation for @xmath50-closed sets is a correspondence between median semilattices and pips , which was established by barthlemy constantin @xcite . a recent paper @xcite also contains an exposition of this correspondence . a _ median semilattice _ @xcite is a meet - semilattice @xmath51 satisfying following conditions : ms every principal ideal is a distributive lattice . for all @xmath52 , if @xmath53 and @xmath54 exist , then @xmath55 exists in @xmath56 . note that every distributive lattice is a median semilattice . an element of @xmath56 is said to be _ join - irreducible _ if it is not minimum and is not represented as the join of other elements . let @xmath57 denote the set of join - irreducible elements of @xmath56 . next we introduce a _ poset with inconsistent pairs ( pip)_. a pip @xcite is a poset @xmath58 endowed with an additional symmetric relation @xmath59 satisfying the following conditions : ic for all @xmath60 with @xmath61 , there is no @xmath62 with @xmath63 and @xmath64 . for all @xmath65 , if @xmath66 and @xmath67 , then @xmath61 . a pip is also denoted by a triple @xmath68 . the relation @xmath59 is called an _ inconsistent relation_. each unordered pair @xmath69 of @xmath70 is called _ inconsistent _ if @xmath61 . note that every inconsistent pair of @xmath70 is incomparable and has no common upper bound in @xmath70 . an inconsistent pair @xmath69 of @xmath70 is said to be _ minimally inconsistent _ if @xmath71 , @xmath72 and @xmath67 imply @xmath73 and @xmath74 for all @xmath75 . if @xmath69 is minimally inconsistent , the @xmath61 is particularly denoted by @xmath76 . we can easily check the following properties of the minimally inconsistent relation : mic for all @xmath60 with @xmath76 , there is no @xmath62 with @xmath63 and @xmath64 . for all @xmath65 with @xmath71 and @xmath72 , if @xmath77 and @xmath76 , then @xmath78 and @xmath79 . actually , pips can also be defined as a triplet @xmath80 , where @xmath81 is a binary symmetric relation on a poset @xmath58 satisfying the conditions ( mic1 ) and ( mic2 ) . in this definition , the inconsistent relation @xmath59 on @xmath70 is obtained by @xmath82 for every @xmath60 . since both definitions of pips are equivalent , we will use a convenient one . for a pip @xmath70 , an ideal of @xmath70 is said to be _ consistent _ if it contains no ( minimally ) inconsistent pair . let @xmath83 denote the family of consistent ideals of @xmath70 . regard @xmath83 as a poset with respect to the inclusion order @xmath84 . \(a ) at ( 0 , 0 ) ; ( b ) at ( 2 , 0 ) ; ( c ) at ( 1 , 1 ) ; ( p ) at ( 0.4 , 1 ) @xmath85 ; ( d ) at ( 4 , 0 ) ; ( e ) at ( 4 , 1 ) ; ( q ) at ( 4.6 , 1 ) @xmath86 ; ( f ) at ( 0 , 2 ) ; ( g ) at ( 2.5 , 2 ) ; ( r ) at ( 2.5 , 2.4 ) @xmath87 ; / n a / c , b / c , d / e , c / f , c / g , e / g ( ) ( ) ; ( c ) to ( e ) ; ( e ) to ( f ) ; \(a ) at ( 0 , 0 ) ; ( b ) at ( 2 , 0 ) ; ( c ) at ( 1 , 1 ) ; ( d ) at ( 4 , 0 ) ; ( p ) at ( 4.6 , 0 ) @xmath85 ; ( e ) at ( 4 , 1 ) ; ( qr ) at ( 5.1 , 1 ) @xmath88 ; ( f ) at ( 0 , 2 ) ; / n a / c , b / c , d / e , c / f ( ) ( ) ; ( c ) to ( e ) ; ( d ) to [ bend right=40 ] ( e ) ; ( e ) to ( f ) ; \(a ) at ( 0 , 0 ) ; ( b ) at ( 2 , 0 ) ; ( c ) at ( 1 , 1 ) ; ( d ) at ( 4 , 0 ) ; ( e ) at ( 4 , 1 ) ; ( f ) at ( 0 , 2 ) ; / n a / c , b / c , d / e , c / f ( ) ( ) ; ( c ) to ( e ) ; ( e ) to ( f ) ; \(a ) at ( 0 , 0 ) ; ( b ) at ( 2 , 0 ) ; ( p ) at ( 1.4 , 0 ) @xmath89 ; ( c ) at ( 1 , 1 ) ; ( p ) at ( 0.4 , 1 ) @xmath85 ; ( d ) at ( 4 , 0 ) ; ( p ) at ( 4.65 , 0 ) @xmath90 ; ( e ) at ( 4 , 1 ) ; ( p ) at ( 4.65 , 1 ) @xmath86 ; ( f ) at ( 0 , 2 ) ; / n a / c , b / c , d / e , c / f ( ) ( ) ; ( b ) to ( d ) ; \(a ) at ( 0 , 0 ) ; ( b ) at ( 2 , 0 ) ; ( c ) at ( 1 , 1 ) ; ( d ) at ( 4 , 0 ) ; ( e ) at ( 4 , 1 ) ; ( f ) at ( 0 , 2 ) ; / n a / c , b / c , d / e , c / f ( ) ( ) ; / n b / e , d / c , d / f , c / e , e / f ( ) ( ) ; ( b ) to ( d ) ; figure [ fig : pip ] shows examples of pips and non - pip structures . solid arrows indicate the orders between elements ( drawn from higher elements to lowers ) . dotted lines and dashed lines indicate the inconsistent relations . in ( a ) , ( b ) and ( d ) , labeled elements indicate where the violations of ( ic1 ) and ( ic2 ) are . in ( c ) and ( e ) , the minimally inconsistent relations are drawn by dashed lines . the following theorem establishes a one - to - one correspondence between median semilattices and pips . [ thm : median_semilattice_and_pip ] 1 . let @xmath51 be a median semilattice and @xmath59 a symmetric binary relation on @xmath56 defined by @xmath91 for every @xmath92 . then @xmath93 forms a pip with inconsistent relation @xmath59 . the consistent ideal family @xmath94 is isomorphic to @xmath56 , and an isomorphism is given by @xmath95 . 2 . let @xmath70 be a pip . the consistent ideal family @xmath83 forms a median semilattice . the pip @xmath96 obtained in the same way as ( 1 ) is isomorphic to @xmath70 . the latter part of theorem [ thm : median_semilattice_and_pip ] ( 2 ) is implicit in @xcite , and follows from theorem [ thm : median_semilattice_and_pip ] ( 1 ) and the fact that for pips @xmath70 and @xmath97 , if @xmath83 and @xmath98 are isomorphic , then @xmath70 and @xmath97 are also isomorphic @xcite . [ lem : pip_and_cnf ] a pip is an alternative expression of a satisfiable boolean 2-cnf , where consistent ideals correspond to true assignments . indeed , for a pip @xmath68 with @xmath99}}$ ] , consider the following 2-cnf of boolean variables @xmath100 : @xmath101 then an assignment @xmath102 is true if and only if the set of elements @xmath103 with @xmath104 is a consistent ideal . the reverse construction of a pip from a 2-cnf satisfiable at @xmath105 is also easily verified . the starting point for a compact representation for @xmath50-closed sets is the following . [ lem : closed_set_is_median_semilattice ] every @xmath50-closed set is a median semilattice . let @xmath106 be a @xmath50-closed set . then @xmath107 is a semilattice since @xmath33 is a semilattice with minimum element @xmath108 , and the operator @xmath43 coincides with @xmath26 on @xmath33 . we show that @xmath107 satisfies the conditions ( ms1 ) and ( ms2 ) . let @xmath109 be the principal ideal of @xmath110 . for all @xmath111 and @xmath36}}$ ] , @xmath112 is equal to either 0 or @xmath37 . therefore , for all @xmath113 , the join @xmath114 exists and it holds @xmath115 . next let @xmath116}}}}$ ] be an injection defined by @xmath117 for every @xmath111 . one can easily see that @xmath118 and @xmath119 for every @xmath113 . in other words , @xmath120 is an isomorphism from @xmath121 to @xmath122 . since any nonempty subset of @xmath123}}}$ ] closed under @xmath124 and @xmath125 is a distributive lattice ordered by inclusion , @xmath109 is also distributive . let @xmath126 be such that the join of any two of them exists in @xmath107 . since @xmath127 and @xmath128 are comparable for any @xmath36}}$ ] , the join @xmath55 exists in @xmath33 , and coincides with @xmath129 . finally since @xmath107 is closed under @xmath45 , the join @xmath55 belongs to @xmath107 . let @xmath59 be a symmetric binary relation on @xmath33 defined by @xmath130 for every @xmath42 . note that if @xmath131 then @xmath132 is equal to @xmath133 for every @xmath134 . from theorem [ thm : median_semilattice_and_pip ] ( 1 ) and lemma [ lem : closed_set_is_median_semilattice ] , we obtain the following . [ thm : closed_set_and_pip ] let @xmath107 be a @xmath50-closed set . then @xmath135 forms a pip with inconsistent relation @xmath59 . the consistent ideal family @xmath136 is isomorphic to @xmath107 , and the isomorphism is given by @xmath137 . \(000 ) at ( 0 , 0 ) 00002 ; ( 100 ) at ( -1 , 1 ) 13002 ; ( 001 ) at ( 1 , 1 ) 00012 ; ( 110 ) at ( -1.5 , 2 ) 13102 ; ( 101 ) at ( -0.5 , 2 ) 13012 ; ( 120 ) at ( 0.5 , 2 ) 13202 ; ( 031 ) at ( 1.5 , 2 ) 00312 ; ( 111 ) at ( -1.5 , 3 ) 13112 ; ( 121 ) at ( -0.5 , 3 ) 13212 ; ( 131 ) at ( 0.5 , 3 ) 13312 ; ( 231 ) at ( 1.5 , 3 ) 22312 ; / n 000/100 , 000/001 , 100/110 , 100/101 , 100/120 , 001/101 , 001/031 , 110/111 , 101/111 , 101/121 , 101/131 , 120/121 , 031/131 , 031/231 ( ) ( ) ; \(100 ) at ( -1 , 0.5 ) 13002 ; ( 110 ) at ( -1.5 , 1.5 ) 13102 ; ( 120 ) at ( -0.5 , 1.5 ) 13202 ; ( 001 ) at ( 1 , 0 ) 00012 ; ( 031 ) at ( 1 , 1 ) 00312 ; ( 231 ) at ( 1 , 2 ) 22312 ; / n 100/110 , 100/120 , 001/031 , 031/231 ( ) ( ) ; / n 110/120 , 120/031 , 100/231 ( ) ( ) ; ( 110 ) to [ bend right=17 ] ( 031 ) ; figure [ fig : example_of_closed_set_and_pip ] shows an example of a @xmath138-closed set and the corresponding pip . in ( a ) , elements surrounded by double - lined frames are join - irreducible . in ( b ) , only minimally inconsistent relations are drawn . from theorem [ thm : closed_set_and_pip ] , it will turn out that the set @xmath139 of join - irreducible elements of every @xmath50-closed set @xmath107 does not lose any information about the structure of @xmath107 . that is , non - minimum elements in @xmath107 can be obtained as the join of one or more join - irreducible elements of @xmath107 ( notice that we can not obtain the minimum element of @xmath107 in this way ) . therefore we call @xmath135 a _ pip - representation _ of @xmath107 . furthermore , the following proposition , which will be proved in section [ sec : proof_of_theorem_1 ] , says that this representation is actually compact . [ prop : number_of_irreducibles ] let @xmath107 be a @xmath50-closed set on @xmath33 . the number of join - irreducible elements of @xmath107 is at most @xmath140 . theorem [ thm : closed_set_and_pip ] states that any @xmath50-closed set can be represented by a pip . however , not all pips correspond to some @xmath50-closed sets . a natural question then arises : _ what class of pips represents @xmath50-closed sets ? _ the main result ( theorem [ thm : closed_set_and_elementary_pip ] ) of this section answers this question . [ def : elementary ] a pip @xmath58 is called _ elementary _ if it satisfies the following conditions : ep @xmath70 is the disjoint union of @xmath141 such that every pair @xmath142 of distinct elements is minimally inconsistent if and only if @xmath143 for some @xmath36}}$ ] . for any distinct @xmath144}}$ ] , if @xmath145 and @xmath146 , there is no element @xmath147 with @xmath148 . for any distinct @xmath144}}$ ] , if @xmath145 and @xmath149 , either of the following holds : ep every pair of @xmath147 and @xmath150 is not comparable . there exist @xmath151 and @xmath152 such that @xmath153 and @xmath154 for all @xmath155 and @xmath156 . \(01 ) at ( 0 , 0 ) ; ( 02 ) at ( 1 , 0 ) ; ( 03 ) at ( 2 , 0 ) ; ( 10 ) at ( 0 , 1 ) ; ( 10 ) ( 01 ) ; ( 01 ) ( 02 ) ; ( 02 ) ( 03 ) ; ( 01.south east ) to [ bend right=25 ] ( 03.south west ) ; \(01 ) at ( 0 , 1 ) ; ( 02 ) at ( 1 , 1 ) ; ( 03 ) at ( 2 , 1 ) ; ( 10 ) at ( 0 , 0 ) ; ( 01 ) ( 10 ) ; ( 01 ) ( 02 ) ; ( 02 ) ( 03 ) ; ( 01.north east ) to [ bend left=25 ] ( 03.north west ) ; \(10 ) at ( 0 , 0 ) ; ( 20 ) at ( 1 , 0 ) ; ( 30 ) at ( 2 , 0 ) ; ( 01 ) at ( 0 , 1 ) ; ( 02 ) at ( 1 , 1 ) ; ( 03 ) at ( 2 , 1 ) ; ( 01 ) ( 10 ) ; ( 02 ) ( 20 ) ; ( 03 ) ( 30 ) ; ( 01 ) ( 02 ) ; ( 02 ) ( 03 ) ; ( 01.north east ) to [ bend left=25 ] ( 03.north west ) ; ( 10 ) ( 20 ) ; ( 20 ) ( 30 ) ; ( 10.south east ) to [ bend right=25 ] ( 30.south west ) ; \(10 ) at ( 0 , 0 ) ; ( x ) at ( -0.5 , 0.05 ) @xmath157 ; ( 02 ) at ( -0.5 , 1 ) ; ( 03 ) at ( 0.5 , 1 ) ; ( 01 ) at ( 2 , 0 ) ; ( x ) at ( 2.5 , 0.05 ) @xmath158 ; ( 20 ) at ( 1.5 , 1 ) ; ( 30 ) at ( 2.5 , 1 ) ; ( 02 ) ( 10 ) ; ( 03 ) ( 10 ) ; ( 20 ) ( 01 ) ; ( 30 ) ( 01 ) ; ( 01 ) ( 03 ) ; ( 02 ) ( 03 ) ; ( 02 ) to [ bend right=10 ] ( 01 ) ; ( 10 ) ( 20 ) ; ( 20 ) ( 30 ) ; ( 10 ) to [ bend right=10 ] ( 30 ) ; figure [ fig : example_of_elementary_pip ] shows examples of elementary pips and non - elementary pips . in all diagrams , the drawn pips satisfy the condition ( ep0 ) with @xmath159 . each element is colored black or white according to the corresponding part @xmath160 . non - minimally inconsistent relations are not drawn in each diagram . [ thm : closed_set_and_elementary_pip ] 1 . for every @xmath50-closed set @xmath107 , the pip @xmath135 is elementary . 2 . for every elementary pip @xmath70 , there is a @xmath50-closed set @xmath107 isomorphic to @xmath83 . in case of @xmath161 , since the condition ( ep2 ) in definition [ def : elementary ] is equivalent to the skew - symmetricity of the corresponding hasse diagram , the pip - representation for @xmath50-closed sets coincides with the one by ando fujishige @xcite . the following corollary of theorem [ thm : median_semilattice_and_pip ] ( 2 ) and theorem [ thm : closed_set_and_elementary_pip ] ( 1 ) will be used in section [ sec : algorithms ] . [ cor : elementary_pip ] let @xmath70 be a pip . if @xmath83 is isomorphic to some @xmath50-closed set , then @xmath70 is elementary . the remaining part of this section is devoted to proving theorem [ thm : closed_set_and_elementary_pip ] . the proof of theorem [ thm : closed_set_and_elementary_pip ] ( 1 ) is outlined as follows : 1 . first we define the _ differential _ of a join - irreducible element @xmath35 as the difference between @xmath35 and the unique lower cover @xmath157 of @xmath35 . next we introduce a _ @xmath50-closed set , which is a @xmath50-closed set such that every differential has exactly one nonzero component . we show that every @xmath50-closed set is isomorphic to some normalized @xmath50-closed set . this set gives us a natural partition of join - irreducible elements . finally we construct an elementary pip from the partition . a @xmath50-closed set @xmath107 is said to be _ simple _ if @xmath162 . any @xmath50-closed set can be converted to a simple @xmath50-closed set without any structural change . [ def : differential ] let @xmath106 be a simple @xmath50-closed set . for @xmath134 , we say that @xmath163 is a _ lower cover _ of @xmath35 , or _ @xmath35 covers @xmath163 _ , if @xmath164 and there is no @xmath165 such that @xmath166 . for a join - irreducible element @xmath167 , there uniquely exists @xmath168 covered by @xmath35 . the _ differential _ @xmath169 of @xmath35 is defined by @xmath170 if @xmath171 and @xmath172 if @xmath173 for each @xmath36}}$ ] . \(000 ) at ( 0 , 0 ) 0000 ; ( 100 ) at ( -1 , 1 ) 1300 ; ( 001 ) at ( 1 , 1 ) 0001 ; ( 110 ) at ( -1.5 , 2 ) 1310 ; ( 101 ) at ( -0.5 , 2 ) 1301 ; ( 120 ) at ( 0.5 , 2 ) 1320 ; ( 031 ) at ( 1.5 , 2 ) 0031 ; ( 111 ) at ( -1.5 , 3 ) 1311 ; ( 121 ) at ( -0.5 , 3 ) 1321 ; ( 131 ) at ( 0.5 , 3 ) 1331 ; ( 231 ) at ( 1.5 , 3 ) 2231 ; / n 000/100 , 000/001 , 100/110 , 100/101 , 100/120 , 001/101 , 001/031 , 110/111 , 101/111 , 101/121 , 101/131 , 120/121 , 031/131 , 031/231 ( ) ( ) ; \(100 ) at ( -1 , 0.5 ) 1300 + _ 1300 _ ; ( 110 ) at ( -1.5 , 1.5 ) 1310 + _ 0010 _ ; ( 120 ) at ( -0.5 , 1.5 ) 1320 + _ 0020 _ ; ( 001 ) at ( 1 , 0 ) 0001 + _ 0001 _ ; ( 031 ) at ( 1 , 1 ) 0031 + _ 0030 _ ; ( 231 ) at ( 1 , 2 ) 2231 + _ 2200 _ ; / n 100/110 , 100/120 , 001/031 , 031/231 ( ) ( ) ; / n 110/120 , 120/031 , 100/231 ( ) ( ) ; ( 110 ) to [ bend right=20 ] ( 031 ) ; \(000 ) at ( 0 , 0 ) 000 ; ( 100 ) at ( -1 , 1 ) 100 ; ( 001 ) at ( 1 , 1 ) 001 ; ( 110 ) at ( -1.5 , 2 ) 110 ; ( 101 ) at ( -0.5 , 2 ) 101 ; ( 120 ) at ( 0.5 , 2 ) 120 ; ( 031 ) at ( 1.5 , 2 ) 031 ; ( 111 ) at ( -1.5 , 3 ) 111 ; ( 121 ) at ( -0.5 , 3 ) 121 ; ( 131 ) at ( 0.5 , 3 ) 131 ; ( 231 ) at ( 1.5 , 3 ) 231 ; / n 000/100 , 000/001 , 100/110 , 100/101 , 100/120 , 001/101 , 001/031 , 110/111 , 101/111 , 101/121 , 101/131 , 120/121 , 031/131 , 031/231 ( ) ( ) ; \(100 ) at ( -1 , 0.5 ) 100 + _ 100 _ ; ( 110 ) at ( -1.5 , 1.5 ) 110 + _ 010 _ ; ( 120 ) at ( -0.5 , 1.5 ) 120 + _ 020 _ ; ( 001 ) at ( 1 , 0 ) 001 + _ 001 _ ; ( 031 ) at ( 1 , 1 ) 031 + _ 030 _ ; ( 231 ) at ( 1 , 2 ) 231 + _ 200 _ ; / n 100/110 , 100/120 , 001/031 , 031/231 ( ) ( ) ; / n 110/120 , 120/031 , 100/231 ( ) ( ) ; ( 110 ) to [ bend right=20 ] ( 031 ) ; figure [ fig : example_of_differential_and_normalizing ] ( a ) and ( b ) show examples of a simple @xmath50-closed set and the corresponding pip , respectively . in ( b ) , the differential of each join - irreducible element is written in italics . we show some properties about differentials . in what follows , we denote the subset @xmath174 by @xmath175 for @xmath36}}$ ] and @xmath176}}$ ] . note that @xmath175 also forms a @xmath50-closed set if @xmath177 . [ lem : differential ] let @xmath106 be a simple @xmath50-closed set . the following hold : 1 . for every @xmath36}}$ ] and @xmath176}}$ ] with @xmath177 , @xmath178 is join - irreducible in @xmath107 and @xmath179 holds . 2 . for every @xmath167 and @xmath36}}$ ] with @xmath180 , it holds @xmath181 . 3 . for every @xmath36}}$ ] and @xmath176}}$ ] , there is at most one join - irreducible element @xmath167 such that @xmath179 . 4 . for every @xmath167 , the differential @xmath182 of @xmath35 has at least one nonzero components . the map @xmath183 is an injection from @xmath139 to @xmath33 . let @xmath36 } } { , \;}\alpha \in { { \left[{k}\right]}}$ ] and @xmath178 . suppose to the contrary that @xmath184 . then there exist @xmath185 such that @xmath186 and @xmath187 . since @xmath188 , either @xmath112 or @xmath128 is equal to @xmath189 . this contradicts the assumption that @xmath181 and @xmath186 . hence @xmath35 is join - irreducible . moreover , from definition [ def : differential ] , it holds @xmath179 . let @xmath167 and @xmath36}}$ ] such that @xmath180 . then @xmath190 . let @xmath163 be the minimum element of @xmath175 . if @xmath191 , definition [ def : differential ] contradicts @xmath192 . hence we have @xmath193 . suppose that @xmath107 has a join - irreducible element @xmath167 such that @xmath179 . from ( 2 ) , it holds @xmath181 . this lemma follows from the uniqueness of the minimum element of @xmath175 . assume that @xmath107 has a join - irreducible element @xmath35 such that @xmath194 . then for every @xmath195 , there exists @xmath196 such that @xmath197 . thus the join of @xmath198 is equal to @xmath35 . this contradicts the assumption that @xmath35 is join - irreducible . let @xmath199 such that @xmath200 . since @xmath201 from ( 4 ) , there exists @xmath36}}$ ] such that @xmath202 . then @xmath203 follows from ( 3 ) . now proposition [ prop : number_of_irreducibles ] is a consequence of lemma [ lem : differential ] . it suffices to consider the case where @xmath107 is simple . from lemmas [ lem : differential ] ( 3 ) and ( 4 ) , it holds @xmath204 . furthermore , since the map @xmath205 is injective , we have @xmath206 . hence @xmath207 is at most @xmath140 . a simple @xmath50-closed set @xmath107 is said to be _ normalized _ if it satisfies @xmath208 for all @xmath167 . examples of a normalized @xmath50-closed set and the corresponding pip are shown in figure [ fig : example_of_differential_and_normalizing ] ( c ) and ( d ) , respectively . [ lem : normalized_is_general ] for any @xmath50-closed set @xmath107 , there exists a normalized @xmath50-closed set that is isomorphic to @xmath107 . we can suppose that @xmath107 is simple . we first show : 1 . for @xmath199 , it holds that @xmath209 or @xmath210 . assume that @xmath107 has a join - irreducible element @xmath35 such that @xmath211 and @xmath212 for distinct @xmath144}}$ ] . if there exists @xmath213 such that @xmath214 and @xmath215 , it holds @xmath216 from lemma [ lem : differential ] ( 3 ) . hence we have @xmath217 since @xmath218 . however , @xmath35 and @xmath133 belong to @xmath219 since @xmath220 , thus it holds @xmath221 from lemma [ lem : differential ] ( 2 ) . this is a contradiction . by ( 1 ) , we can define an equivalence relation @xmath222 over the index set @xmath223}}}\mathrel{}\middle|\mathrel{}{\text{there exists $ x \in { { m}^\mathrm{ir}}$ such that $ x_i \neq 0$}}\right\ } } \end{aligned}\ ] ] as follows : @xmath224 then each equivalence class can be `` contracted '' into a single index without any structural change of @xmath107 as follows . let @xmath225 be the set of equivalence classes . for @xmath226}}$ ] , let @xmath227 be the set of join - irreducible elements having nonzero support @xmath228 . then , by lemma [ lem : differential ] , 1 . for every @xmath110 and @xmath229}}$ ] , either @xmath230 for all @xmath231 or there uniquely exists @xmath232}}$ ] such that @xmath233 for all @xmath231 . let @xmath234 } k_j$ ] . define @xmath235 by @xmath236 if @xmath230 for @xmath231 , and @xmath237 if @xmath238 for @xmath231 . it is easily verified ( from lemma [ lem : differential ] ) that the map @xmath120 is injective and preserves @xmath29 . an irreducible element of @xmath239 is the image of an irreducible element of @xmath107 , and , by construction , has the differential of a singleton support . thus @xmath239 is a normalized @xmath50-closed set . now we are ready to prove theorem [ thm : closed_set_and_elementary_pip ] ( 1 ) . by lemma [ lem : normalized_is_general ] , it suffices to consider the case where @xmath107 is normalized . for every @xmath36}}$ ] , let @xmath240 . from the definition of normalized @xmath50-closed sets , @xmath241 forms a partition of @xmath139 ( note that @xmath242 may be empty ) . we show that the pip @xmath243 satisfies the axiom of elementary pips with @xmath244 for every @xmath36}}$ ] . ( ep0 , `` only if '' part ) . let @xmath245 be a minimally inconsistent pair . then there exists @xmath36}}$ ] such that @xmath246 . we show @xmath247 . suppose that @xmath248 . from definition [ def : differential ] , there exists @xmath249 such that @xmath250 and @xmath251 . now we have @xmath250 and @xmath252 , which contradict the assumption that @xmath35 and @xmath163 are minimally inconsistent . therefore @xmath211 holds , and we can show @xmath214 in the same way . thus @xmath253 holds . ( ep1 ) is an immediate consequence of the following property : * let @xmath144}}$ ] be distinct . if there exist @xmath254 and @xmath255 such that @xmath256 , then for all @xmath257 , there exists @xmath258 such that @xmath259 ; in particular @xmath260 if @xmath261 . we show ( @xmath262 ) . let @xmath254 and @xmath255 such that @xmath256 . now it holds @xmath263 and @xmath264 . let @xmath257 . we have @xmath265 since @xmath266 . thus @xmath267 does not hold . we show that @xmath268 . if not , @xmath269 is equal to @xmath270 , hence @xmath133 belongs to @xmath271 . therefore it holds @xmath267 since @xmath163 is the minimum element of @xmath271 from lemma [ lem : differential ] ( 2 ) . we have a contradiction here . let @xmath272 . this @xmath158 belongs to @xmath273 from lemma [ lem : differential ] ( 1 ) , and it holds @xmath274 . let @xmath144}}$ ] be distinct indices such that @xmath275 and @xmath276 . we can assume that ( ep2 - 1 ) does not hold , i.e. , there exist @xmath254 and @xmath255 such that @xmath256 . consider @xmath277 . by ( @xmath262 ) , there exist @xmath278 such that @xmath279 and @xmath280 . we show @xmath281 . suppose not . since @xmath279 and @xmath282 , we can take @xmath283 such that @xmath284 by ( @xmath262 ) with changing the role of @xmath285 and @xmath286 . now we have @xmath287 , which contradicts @xmath288 . therefore @xmath158 and @xmath289 are same elements . consequently , the required element @xmath290 in ( ep2 - 2 ) is given by @xmath158 . by changing the role of @xmath285 and @xmath286 , we see that @xmath291 is given by @xmath157 . ( ep0 , `` if '' part ) . let @xmath292 be distinct with @xmath36}}$ ] . now since @xmath293 , there exists a minimally inconsistent pair @xmath294 such that @xmath295 and @xmath296 . from the `` only if '' part of ( ep0 ) , @xmath157 and @xmath158 belong to @xmath273 for some @xmath297}}$ ] . if @xmath298 , then we have @xmath292 , @xmath299 , @xmath250 and @xmath300 , which contradict ( ep2 ) . hence @xmath301 and it must hold @xmath302 . let @xmath68 be an elementary pip with partition @xmath303 of condition ( ep0 ) . for every @xmath36}}$ ] , let @xmath304 , where @xmath305 . let @xmath306 } } } k_i$ ] . for a consistent ideal @xmath109 , let @xmath307 be defined by @xmath308}}}\right)}.\end{aligned}\ ] ] now @xmath309 is well - defined since every consistent ideal @xmath109 of @xmath70 has at most one element in each @xmath160 by ( ep0 ) . . then @xmath83 and @xmath107 are clearly isomorphic . therefore the rest of the proof of theorem [ thm : closed_set_and_elementary_pip ] ( 2 ) is to show that @xmath107 forms a @xmath50-closed set . we define binary operations @xmath43 and @xmath45 on @xmath83 as @xmath311 and @xmath312 for every @xmath313 . theorem [ thm : closed_set_and_elementary_pip ] ( 2 ) follows immediately from : [ lem : consistent_ideal_family_of_p_is_closed ] for every @xmath313 , it holds @xmath314 , @xmath315 and @xmath316 . let @xmath313 . since the consistent ideal family of a pip is closed under the intersection , @xmath317 also forms a consistent ideal of @xmath70 . in addition , we can easily check that @xmath315 holds . next we consider @xmath318 . we show that @xmath318 is a consistent ideal . suppose that @xmath318 is not an ideal of @xmath70 . there exist @xmath319 and @xmath320 such that @xmath321 . without loss of generality , we assume @xmath322 . now @xmath109 also contains @xmath85 since @xmath109 is an ideal . let @xmath144}}$ ] such that @xmath323 and @xmath324 . we can take @xmath325 since @xmath326 . thus @xmath327 is greater than 1 , and @xmath328 is also greater than 1 since @xmath321 contradicts the condition ( ep1 ) if @xmath329 . from ( ep2 - 2 ) , there exists @xmath330 such that @xmath331 . it holds @xmath332 since @xmath333 . now we have @xmath334 , @xmath322 , @xmath332 and @xmath335 . this contradicts @xmath319 . therefore @xmath318 is an ideal . finally suppose that @xmath318 includes an inconsistent pair @xmath69 . since @xmath318 is an ideal , it also includes the minimally inconsistent pair @xmath336 with @xmath71 and @xmath72 . from ( ep0 ) , @xmath89 and @xmath90 belong to the same part @xmath160 of the partition . this contradicts the fact that @xmath337 , and thus @xmath318 is a consistent ideal . @xmath316 follows from the definitions of @xmath45 on @xmath33 and on @xmath83 . in this section , we study algorithmic aspects of constructing pip - representations for the minimizer sets of @xmath0-submodular functions . let @xmath338 denote the minimizer set of a function @xmath1 . let @xmath339 denote the time complexity of an algorithm of a maximum flow ( and a minimum cut ) in a network of @xmath3 vertices and @xmath340 edges . we assume a standard max - flow algorithm , such as preflow - push algorithm , and hence assume that @xmath339 is not less than @xmath341 ; notice that the current fastest one is an @xmath341 algorithm by orlin @xcite . we also assume @xmath342 for dealing with @xmath0-submodular functions on @xmath33 . we can obtain the pip - representation for the minimizer set of a @xmath0-submodular function @xmath40 by using a minimizing oracle @xmath0-sfm , which returns a minimizer of @xmath1 and its restrictions . let @xmath343 be the minimum value of @xmath1 . for @xmath36}}$ ] and @xmath344 , we define a new @xmath0-submodular function @xmath345 from @xmath1 by @xmath346 namely , @xmath347 is a function obtained by fixing the @xmath285-th variable of @xmath1 to @xmath348 . before describing the main part of our algorithm , we present a subroutine in algorithm [ alg : get_minimum_minimizer ] . this subroutine returns the minimum minimizer of a @xmath0-submodular function . the validity of this subroutine can be checked by the fact that @xmath349 is equal to @xmath343 if @xmath350 and otherwise it holds @xmath351 . this subroutine calls @xmath0-sfm at most @xmath352 times . a @xmath0-submodular function @xmath40 the minimum minimizer @xmath353 of @xmath1 algorithm [ alg : get_irreducibles_minimizers ] shows a procedure to collect all join - irreducible minimizers of a @xmath0-submodular function . let @xmath35 be the minimum minimizer of @xmath1 . the function @xmath354 in algorithm [ alg : get_irreducibles_minimizers ] is defined as @xmath355 for every @xmath356 . since @xmath357 is equal to @xmath112 if @xmath230 and to @xmath37 if @xmath358 , we can regard @xmath359 as a @xmath0-submodular function obtained by fixing each @xmath285-th variable of @xmath1 to @xmath37 if @xmath358 . note that the minimum values of @xmath1 and @xmath359 are the same . the correctness of this algorithm is based on lemma [ lem : differential ] ( 1 ) and ( 2 ) . namely , the set of join - irreducible minimizers of @xmath1 coincides with the set @xmath360 } } \setminus { \mathop{\mathrm{supp}}}{x } { , \;}\alpha \in { { \left[{k}\right ] } } { , \;}\min { { \tilde{f}}_{{i},{\alpha } } } = \min f}\right\}}.\end{aligned}\ ] ] the algorithm collects each join - irreducible minimizer according to by calling at most @xmath361 times . consequently , if a minimizing oracle is available , the minimizer set can also be obtained in polynomial time . [ thm : alg_oracle ] the pip - representation for the minimizer set of a @xmath0-submodular function @xmath40 is obtained by @xmath5 calls of @xmath0-sfm . a @xmath0-submodular function @xmath40 the set @xmath362 of all join - irreducible minimizers of @xmath1 iwata wahlstrm yoshida @xcite introduced _ basic @xmath0-submodular functions _ , which form a special class of @xmath0-submodular functions . they showed a reduction of the minimization problem of a nonnegative combination of binary basic @xmath0-submodular functions to the minimum cut problem on a directed network . we describe their method and present an algorithm to obtain the pip - representation for the minimizer set . let @xmath3 and @xmath0 be positive integers . we consider a directed network @xmath363 with vertex set @xmath364 , edge set @xmath365 and nonnegative edge capacity @xmath366 . suppose that @xmath364 consists of source @xmath367 , sink @xmath368 and other vertices @xmath369 , where @xmath36}}$ ] and @xmath176}}$ ] . let @xmath370 for @xmath36}}$ ] . an @xmath49-cut of @xmath371 is a subset @xmath16 of @xmath364 such that @xmath372 and @xmath373 . we call an @xmath49-cut @xmath16 _ legal _ if @xmath374 for every @xmath36}}$ ] . there is a natural bijection @xmath375 from @xmath33 to the set of legal @xmath49-cuts of @xmath371 defined by @xmath376 for an @xmath49-cut @xmath16 of @xmath371 , let @xmath377 denote the legal @xmath49-cut obtained by removing vertices in @xmath378 from @xmath16 for every @xmath36}}$ ] with @xmath379 . the _ capacity _ @xmath380 of @xmath16 is defined as sum of capacities @xmath381 of all edges @xmath382 from @xmath16 to @xmath383 . we say that a network @xmath371 _ represents _ a function @xmath40 if it satisfies the following conditions : nr there exists a constant @xmath384 such that @xmath385 for all @xmath386 . it holds @xmath387 for all @xmath49-cuts @xmath16 of @xmath371 . from ( nr1 ) , the minimum value of @xmath388 is equal to the capacity of a minimum @xmath49-cut of @xmath371 . for every minimum @xmath49-cut @xmath16 of @xmath371 , @xmath377 is also a minimum @xmath49-cut since @xmath371 satisfies the condition ( nr2 ) . therefore @xmath389 is a minimizer of @xmath1 , and a minimum @xmath49-cut can be computed by maximum flow algorithms . indeed , wahlstrm yoshida @xcite showed that nonnegative combinations of basic @xmath0-submodular functions are representable by such networks . now we shall consider obtaining the pip - representation for the minimizer set of a @xmath0-submodular function @xmath40 represented by a network @xmath371 . the minimizer set of @xmath1 is isomorphic to the family of legal minimum @xmath49-cuts of @xmath371 ordered by inclusion , where the isomorphism is @xmath375 . it is well - known that the family of ( not necessarily legal ) minimum @xmath49-cuts forms a distributive lattice . thus , by birkhoff representation theorem , the family is efficiently representable by a poset . queyranne @xcite showed an algorithm to obtain the poset from the residual graph corresponding to a maximum @xmath49-flow of @xmath371 . we describe their theorem briefly . for an @xmath49-flow @xmath120 of @xmath371 , the _ residual graph _ corresponding to @xmath120 is a directed graph @xmath390 , where @xmath391 [ thm : picard ] let @xmath363 be a directed network with @xmath392 and @xmath393 the residual graph corresponding to a maximum @xmath394-flow of @xmath371 . let @xmath395 be the set of strongly connected components ( sccs ) of @xmath393 other than the following : sccs reachable from @xmath367 . sccs reachable to @xmath368 . let @xmath396 be a partial order on @xmath395 defined by @xmath397 for every @xmath398 . the ideal family of the poset @xmath399 is isomorphic to the family of minimum @xmath49-cuts of @xmath371 ordered by inclusion . the isomorphism @xmath400 is given by @xmath401 , where @xmath402 is the set of vertices reachable from @xmath367 . our result is the following . [ thm : network_algo ] let @xmath371 be a network representing a @xmath0-submodular function @xmath40 and @xmath393 the residual graph corresponding to a maximum @xmath394-flow of @xmath371 . let @xmath395 be the set of sccs of @xmath393 other than the following : sccs reachable from @xmath367 . sccs reachable to @xmath368 . sccs reachable to an scc containing two or more elements in @xmath403 for some @xmath36}}$ ] . sccs reachable to sccs @xmath16 and @xmath404 such that @xmath405 and @xmath406 for some @xmath36}}$ ] . a partial order @xmath396 on @xmath395 is defined in the same way as theorem [ thm : picard ] . let @xmath59 be a symmetric binary relation on @xmath395 defined as @xmath407}}$}. \end{aligned}\ ] ] then @xmath395 forms an elementary pip with inconsistent relation @xmath59 . the consistent ideal family of @xmath395 is isomorphic to the minimizer set of @xmath1 , where the isomorphism is @xmath408 . first we prove that @xmath395 is a pip . we can see that @xmath59 satisfies the condition ( ic1 ) since for every @xmath398 with @xmath409 , an scc @xmath410 reachable to @xmath16 and @xmath404 does not belong to @xmath395 according to the above exclusion rule ( 4 ) . the condition ( ic2 ) is also satisfied from the definition of the relation @xmath59 . thus @xmath395 forms a pip . next we show @xmath411 for every consistent ideal @xmath109 of @xmath395 . let @xmath412 be the poset given in theorem [ thm : picard ] . note that @xmath395 is a subposet of @xmath412 . we show that @xmath109 is an ideal of @xmath412 . suppose not . then there exist @xmath413 and @xmath414 such that @xmath404 is reachable from @xmath16 and meets the above exclusion rules ( 3 ) or ( 4 ) . now since @xmath16 also satisfies the same exclusion rule , @xmath16 does not belong to @xmath395 . this is a contradiction . hence @xmath109 is an ideal of @xmath412 , and @xmath415 is a minimum @xmath49-cut ( theorem [ thm : picard ] ) . moreover , from the exclusion rule ( 3 ) and the definition of @xmath59 , we can see that @xmath415 is legal . therefore @xmath416 is a minimizer of @xmath1 . conversely , let @xmath386 be a minimizer of @xmath1 . since @xmath417 is a minimum @xmath49-cut , @xmath418 is an ideal of @xmath412 ( theorem [ thm : picard ] ) . suppose that @xmath419 . then there exists @xmath420 which meets the exclusion rule ( 3 ) or ( 4 ) . suppose that @xmath16 meets the rule ( 3 ) . then @xmath16 is reachable to an scc @xmath404 such that @xmath421 for some @xmath36}}$ ] . now @xmath404 is not reachable to @xmath368 since @xmath16 is not reachable to @xmath368 . thus @xmath404 meets the rule ( 1 ) or belongs to @xmath109 otherwise . in either case it holds @xmath422 . this contradicts the fact that @xmath417 is legal . a similar argument can also be applied in the case where @xmath16 meets the rule ( 4 ) . therefore @xmath423 holds , and @xmath109 is an ideal of @xmath395 since @xmath395 is a subposet of @xmath412 . the consistency of @xmath109 is an immediate consequence of the fact that @xmath417 is legal . now we have shown that @xmath408 is a bijection from @xmath424 to @xmath338 . in addition , @xmath408 clearly preserves the orders , hence it is an isomorphism . finally from corollary [ cor : elementary_pip ] , @xmath395 is elementary . consider the time complexity to obtain @xmath395 from the residual graph @xmath393 . first we can obtain the sccs of @xmath393 in @xmath425 time , where @xmath426 . additionally , the exclusion rules ( 1 ) , ( 2 ) and ( 3 ) can be applied to the sccs in the same time complexity . hence it is only the exclusion rule ( 4 ) that we should carefully take account of . an efficient algorithm is as follows : for each remaining scc @xmath16 , memorize the union of sccs reachable from @xmath16 , and if the union set has two or more elements in @xmath403 for some @xmath36}}$ ] , remove all sccs reachable to @xmath16 . by executing this process in reverse topological order , it can be done in @xmath427 time since the size of each memorized set is at most @xmath3 . therefore the time complexity for obtaining @xmath395 from @xmath393 is much less than the one for computing @xmath393 from the network @xmath371 . consequently , we obtain the following theorem : [ thm : network_time ] let @xmath40 be a @xmath0-submodular function represented by a network @xmath371 with @xmath428 edges . the pip - representation for the minimizer set of @xmath1 is obtained in @xmath429 time . let @xmath430 be a connected undirected graph on vertex set @xmath431}}$ ] with @xmath432 , where each edge @xmath433 has a positive edge weight @xmath434 . let @xmath435}}$ ] be the set of labels . potts @xmath0-submodular function _ is a @xmath0-submodular function @xmath436 of the following form : @xmath437 where @xmath438 is any @xmath0-submodular function on @xmath27 for each @xmath36}}$ ] and @xmath439 is a @xmath0-submodular function on @xmath440 defined by @xmath441 for each @xmath442 . a potts @xmath0-submodular function is naturally associated with a _ potts energy function _ @xmath443}}^n}{\setr}$ ] : @xmath444}}^n}\right)},\end{aligned}\ ] ] where @xmath438 is any function on @xmath435}}$ ] for each @xmath36}}$ ] and @xmath445}}^2}{\setr}$ ] is defined by @xmath446 if @xmath447 and @xmath448 if @xmath449 . finding a labeling @xmath450}}^n$ ] of the minimum potts energy is np - hard for @xmath48 but particularly important in computer vision applications . useful information of optimal labelings of the potts energy can be extracted from a minimizer of a potts @xmath0-submodular function with appropriate @xmath0-submodular functions @xmath451 . define each @xmath451 by @xmath452 for @xmath176}}$ ] and @xmath453}}}\::\:{\beta \neq \gamma } } } { \left({{{g_i}{{\mathopen{}\mathclose\bgroup\left}({\beta}{\aftergroup\egroup\right } ) } } + { { g_i}{{\mathopen{}\mathclose\bgroup\left}({\gamma}{\aftergroup\egroup\right})}}}\right)}/2 $ ] . in this case , @xmath10 is a _ @xmath0-submodular relaxation _ of @xmath9 , and an optimal labeling of @xmath9 is a partially recovered from a minimizer of @xmath9 ; see the next section . another choice of @xmath454 is : @xmath455 } } \setminus { \left\{{\alpha}\right\ } } } { { g_i}{{\mathopen{}\mathclose\bgroup\left}({\beta}{\aftergroup\egroup\right})}}}\right)}/2 $ ] for @xmath176}}$ ] and @xmath456 . also in this case , a part of an optimal labeling is obtained from a minimizer of @xmath10 , and coincides with kovtun s partial labeling @xcite . the goal of this section is to develop a fast algorithm to construct the pip of a potts @xmath0-submodular function @xmath10 . notice that @xmath10 is network - representable with @xmath457 edges @xcite . therefore we can obtain a minimizer as well as the pip - representation for @xmath10 in @xmath458 time by the network construction in the previous section . however it is hard to apply this algorithm to the vision application with large @xmath459 in @xcite . kolmogorov @xcite developed an @xmath460-time algorithm to find a minimizer of @xmath10 . the main theorem in this section is a stronger result that the pip - representation is also obtained in the same time complexity . [ thm : potts ] the pip - representation for the minimizer set of @xmath10 is obtained in + @xmath460 time . the rest of this subsection is devoted to proving this theorem . first we construct a network @xmath371 , different from the one in the previous section . for each @xmath36}}$ ] , decompose @xmath451 as follows . let @xmath461 be defined by @xmath462 if @xmath463 and @xmath464 otherwise . choose a minimizer @xmath465 of @xmath451 . then @xmath451 is represented as @xmath466 } } } \sigma_{i , \alpha } { { 1_{=}}{{\mathopen{}\mathclose\bgroup\left}({\alpha , x_i}{\aftergroup\egroup\right})}},\end{aligned}\ ] ] where @xmath467 and @xmath468 for @xmath176}}$ ] . we remark that @xmath469 is nonnegative by @xmath0-submodularity , and that @xmath470 implies @xmath471 . let us construct @xmath371 . starting from @xmath430 , define the edge - capacity @xmath472 of each edge @xmath433 by @xmath434 . next add new vertices @xmath473 , called _ terminals_. for each @xmath36}}$ ] , if @xmath470 with @xmath474}}$ ] , add a new edge @xmath475 of capacity @xmath476 . an edge @xmath475 is called a _ terminal edge_. for each @xmath176}}$ ] with @xmath477 , add a new vertices @xmath478 and new edge @xmath479 of capacity @xmath480 . a vertex @xmath478 is called the _ @xmath189-fringe _ of @xmath285 . let @xmath481 . let @xmath482 be the set of all fringes , @xmath483 the set of all edges incident to fringes , and @xmath484 the set of all terminal edges . let @xmath485 and @xmath486 . let @xmath487 be the resulting network . second we show that @xmath10 is represented as a certain multicut function in @xmath371 . for a vertex subset @xmath16 , the cut capacity @xmath380 of @xmath16 is the sum of @xmath381 of all edges @xmath382 between @xmath16 and @xmath383 . for @xmath176}}$ ] , a vertex subset @xmath16 is called an _ @xmath488-isolating cut _ or _ @xmath189-cut _ if @xmath489 , @xmath490 for @xmath491 } } \setminus { \left\{{\alpha}\right\}}$ ] , and @xmath16 contains no @xmath189-fringe . semi - multicut _ is an ordered partition @xmath492 of @xmath493 such that @xmath494 is an @xmath189-cut for each @xmath176}}$ ] . the _ capacity _ @xmath495 of a semi - multicut @xmath496 is defined by @xmath497 } } } { { c}{{\mathopen{}\mathclose\bgroup\left}({x_\alpha}{\aftergroup\egroup\right})}}.\end{aligned}\ ] ] an _ admissible _ semi - multicut is a semi - multicut @xmath498 such that for each @xmath176}}$ ] , each @xmath189-fringe @xmath478 belongs to @xmath402 if @xmath499 and belongs to @xmath500 if @xmath501 for @xmath491 } } \setminus { \left\{{\alpha}\right\}}$ ] . observe that the part to which a fringe of @xmath36}}$ ] belongs is uniquely determined from the part which @xmath285 belongs . for an admissible semi - multicut @xmath496 , define @xmath502 by @xmath503 if and only if @xmath504 . this map @xmath505 is a bijection from the family of all admissible semi - multicuts to @xmath33 . the next lemma says that a @xmath0-submodular function @xmath10 is actually represented by capacities of admissible semi - multicuts . for any admissible multicut @xmath506 , it holds @xmath507 } } } { { \tilde{g}_i}{{\mathopen{}\mathclose\bgroup\left}({\gamma_i}{\aftergroup\egroup\right})}}. \end{aligned}\ ] ] let @xmath508 . the capacity @xmath509 of a fringe edge @xmath479 contributes to @xmath495 by @xmath469 if @xmath510 and by @xmath8 otherwise . thus the contribution is equal to @xmath511 . the capacity @xmath512 of a terminal edge @xmath475 contributes to @xmath495 by @xmath8 if @xmath510 and by @xmath513 if @xmath514 , and by @xmath512 if @xmath501 with @xmath515 . thus the contribution is equal to @xmath516 . similarly , we verify that the contribution of the capacity @xmath434 of @xmath433 is equal to @xmath517 . thus the claimed equality holds . third we show that a minimum admissible semi - multicut is easily obtained by @xmath0 max - flow computations , where `` minimum '' is with regard to the cut capacity . an _ admissible _ @xmath189-cut is an @xmath189-cut @xmath16 such that for each @xmath491 } } \setminus { \left\{{\alpha}\right\}}$ ] , each @xmath518-fringe @xmath519 belongs to @xmath16 if @xmath520 and @xmath521 otherwise . then an admissible semi - multicut @xmath498 is exactly a partition of @xmath493 such that @xmath494 is an admissible @xmath189-cut for each @xmath176}}$ ] . [ lem : minimum_semimultiway_cut ] 1 . any minimum @xmath189-cut is admissible . 2 . for @xmath176}}$ ] , let @xmath522 be the inclusion - minimal minimum @xmath189-cut . then @xmath523 is a minimum admissible semi - multicut . in particular , a minimum admissible semi - multicut is exactly a partition @xmath524 of @xmath493 such that @xmath494 is a minimum @xmath189-cut for each @xmath176}}$ ] . let @xmath16 be an @xmath189-cut . for @xmath491 } } \setminus { \left\{{\alpha}\right\}}$ ] , if the @xmath518-fringe @xmath519 of @xmath520 is outside of @xmath16 , then include @xmath519 into @xmath16 to decrease the cut capacity . similarly , for @xmath491 } } \setminus { \left\{{\alpha}\right\}}$ ] , if the @xmath518-fringe @xmath519 of @xmath525 belongs to @xmath16 , then remove @xmath519 from @xmath16 to decrease the cut capacity . ( 2 ) is immediate from the standard uncrossing argument ; see the proof of lemma [ lem : uncrossing ] . in particular , a minimum admissible multicut is obtained by computing a minimal minimum @xmath189-cut for each @xmath176}}$ ] . the network @xmath371 has at most @xmath526 vertices and @xmath527 edges . when computing a minimum @xmath189-cut , all @xmath518-fringes with @xmath515 can be removed , and a max - flow algorithm is performed on a network of @xmath528 vertices and @xmath529 edges ( after combining @xmath530 for @xmath515 and all @xmath189-fringes into a single vertex ) . thus we obtain the following , which was essentially shown in @xcite . a minimizer of @xmath10 can be obtained in @xmath531 time . fourth we explain how to obtain the pip representation from maximum @xmath189-flows for @xmath176}}$ ] , where by an _ @xmath189-flow _ we mean a flow from @xmath488 to the union of @xmath532 and @xmath189-fringes . to construct the pip , we use the following intersecting properties of minimum isolating cuts . here a minimum @xmath189-cut is simply called an _ @xmath189-mincut_. [ lem : uncrossing ] 1 . for distinct @xmath533}}$ ] , if @xmath16 is an @xmath189-mincut and @xmath404 is a @xmath518-mincut , then @xmath534 is an @xmath189-mincut and @xmath535 is a @xmath518-mincut . 2 . for distinct @xmath536}}$ ] , if @xmath16 is an @xmath189-mincut , @xmath404 is a @xmath518-mincut and @xmath410 is a @xmath537-mincut , then @xmath538 . in the theory of minimum cuts on undirected networks , the following inequalities are well - known : @xmath539 for every @xmath540 . then ( 1 ) is an immediate consequence of the first inequality and the fact that any subset of an @xmath189-cut containing @xmath488 is again an @xmath189-cut . ( 2 ) is also immediate from the second inequality and the condition that @xmath430 is connected and each edge of @xmath371 has a positive capacity . by lemma [ lem : uncrossing ] ( 2 ) , each vertex belongs to at most two minimum isolating cuts . let @xmath541 denote the directed network obtained from @xmath371 by replacing each undirected edge @xmath542 by two directed edge @xmath543 and @xmath544 of capacity @xmath545 . for each @xmath176}}$ ] , consider the network obtained from @xmath541 by removing all @xmath518-fringes with @xmath515 and contracting terminals @xmath530 with @xmath515 and @xmath189-fringes into a single terminal @xmath546 , and consider the residual graph @xmath547 corresponding to a maximum @xmath548-flow in this network . let @xmath549 be the poset obtained from @xmath547 in the same way as defined in theorem [ thm : picard ] . here each element in @xmath550 is a subset of @xmath364 ( not including terminals and fringes ) . the ideal family of each @xmath550 is isomorphic to the family of minimum @xmath189-cuts . the intersecting part in @xmath550 and @xmath551 is described as follows . [ lem : intersection_of_isolating_cuts ] let @xmath533}}$ ] with @xmath447 . the following hold : 1 . for every @xmath552 and @xmath553 , it holds either @xmath554 or @xmath555 . 2 . for every @xmath556 , if @xmath557 , then it holds @xmath558 . assume that @xmath559 and @xmath560 . we can assume @xmath561 . consider an @xmath189-mincut @xmath16 with @xmath562 and consider a @xmath518-mincut @xmath404 with @xmath563 . take @xmath404 minimal . then every scc of @xmath564 included in @xmath404 is less than or equal to @xmath565 with respect to @xmath566 . suppose that @xmath404 contains @xmath365 . there is a @xmath518-mincut @xmath410 containing @xmath567 and disjoint from @xmath565 . by lemma [ lem : uncrossing ] ( 1 ) , @xmath568 is an @xmath189-mincut and properly intersects @xmath365 . however this is impossible since the set of non - fringe vertices in each @xmath189-mincut is a disjoint union of sccs of @xmath547 . suppose that @xmath404 does not contain @xmath365 . then @xmath534 is an @xmath189-mincut and properly intersects @xmath365 again . this is a contradiction . assume that @xmath557 and @xmath569 . there is a @xmath518-mincut @xmath404 containing @xmath365 and disjoint with @xmath565 . consider an @xmath189-mincut @xmath16 containing @xmath565 . then @xmath534 is an @xmath189-mincut , contains @xmath565 , and does not contain @xmath365 . however this contradicts the assumption that @xmath557 . by lemma [ lem : intersection_of_isolating_cuts ] , we obtain the elementary pip representing minimum admissible cuts just by `` gluing '' @xmath570 along the intersections . let @xmath571 } } } \sigma_\alpha \times { \left\{{\alpha}\right\}}$ ] and @xmath396 a partial order on @xmath70 defined by @xmath572 for every @xmath573 . in addition , let @xmath81 be a symmetric binary relation on @xmath70 defined by @xmath574 for every @xmath575 . [ thm : pip_potts ] the triplet @xmath576 is an elementary pip with minimally inconsistent relation @xmath81 . the consistent ideal family @xmath83 and the family of minimum admissible semi - multicuts of @xmath371 are in one - to - one correspondence by the map @xmath577 where , for each @xmath176}}$ ] , @xmath578 is the admissible @xmath189-cut containing all vertices @xmath579 such that @xmath285 is reachable from @xmath488 in @xmath547 or belongs to @xmath16 for some @xmath580 . it is easy to see from lemma [ lem : intersection_of_isolating_cuts ] ( 2 ) that @xmath70 is a pip with minimally inconsistent relation @xmath81 . we next show that @xmath83 represents the family of minimum semi - multiway cuts . let @xmath581 . then @xmath578 is a minimum @xmath488-isolating cut ( by theorem [ thm : picard ] ) . by consistency , it necessarily holds @xmath582 for @xmath447 . thus @xmath583 is a minimum semi - multiway cut . conversely , let @xmath498 be a minimum semi - multiway cut . each @xmath494 is a minimum @xmath488-isolating cut , and is represented by an ideal @xmath584 of @xmath550 . now @xmath585 } } } i_\alpha \times { \left\{{\alpha}\right\}}$ ] is a consistent ideal of @xmath70 since @xmath586 are pairwise disjoint . then it holds @xmath587 for @xmath176}}$ ] . now @xmath70 represents a @xmath588-closed set in @xmath33 , and is necessarily elementary by corollary [ cor : elementary_pip ] . therefore the pip - representation is obtained by computing a maximum @xmath189-flow for each @xmath176}}$ ] . the pip - representation for the minimizer set of @xmath10 is obtained in @xmath589 time . finally we present an improved algorithm of time complexity @xmath460 . the key is the existence of a single `` multiflow '' that includes all maximum @xmath189-flows . let @xmath590 denote the set of @xmath189-paths over all @xmath176}}$ ] , where an _ @xmath189-path _ is a path connecting @xmath488 and an @xmath189-fringe or a terminal @xmath530 with @xmath515 . a _ multiflow _ is a nonnegative - valued function @xmath1 on @xmath590 satisfying the capacity constraint : @xmath591 let @xmath592 denote the total - flow value of @xmath1 : @xmath593 for @xmath176}}$ ] , let @xmath594 be the submultiflow of @xmath1 defined by @xmath595 if @xmath596 is an @xmath189-path and @xmath597 otherwise . although the set @xmath590 is exponential , we can efficiently handle multiflow @xmath1 by keeping @xmath1 as @xmath0 flows of node - arc form in @xmath541 , as in @xcite . the following is a special case of ( * ? ? ? * theorem 1.2 ) ( a version of _ multiflow locking theorem _ ) . there exists a multiflow @xmath1 such that @xmath598 is equal to the minimum capacity of an @xmath189-cut for each @xmath176}}$ ] . thus the submultiflow @xmath594 of @xmath1 turns into a maximum @xmath189-flow in @xmath541 . we call such a multiflow _ locking_. our goal is to show that a locking multiflow @xmath1 is obtained in @xmath460 time . this , consequently , yields @xmath460-time algorithm to obtain posets @xmath570 and the desired pip @xmath599 } } } \sigma_\alpha \times { \left\{{\alpha}\right\}}$ ] . in the case where there are no fringes , the problem of finding a locking multiflow is nothing but the _ maximum free multiflow problem _ , which is a well - studied problem in multiflow theory . karzanov nagamochi developed a divide - and - conquer @xmath600-time algorithm ( _ ikn - algorithm _ ) to obtain a locking multiflow . we remark that gridchyn algorithm @xcite is also designed by a similar divide - and - conquer idea in `` cut side . '' we adapt ikn - algorithm to the case where fringes exist . the pseudo code is shown in algorithm [ alg : locking ] . a network @xmath371 with terminal set @xmath601 a locking multiflow in @xmath371 let us explain the detail of the algorithm . consider the case @xmath602 . as in ikn - algorithm , our algorithm divides terminal set @xmath601 into two sets @xmath603 and @xmath604 such that @xmath605 and @xmath606 , and find a minimum cut @xmath16 with @xmath607 and @xmath608 , where fringes are removed in computation . two networks @xmath609 and @xmath610 are constructed as follows . the network @xmath609 is obtained from @xmath371 by contracting @xmath521 into a single terminal @xmath546 and by removing all @xmath189-fringes for @xmath611 . similarly , the network @xmath610 is obtained from @xmath371 by contracting @xmath16 into a single terminal @xmath612 and by removing all @xmath189-fringes for @xmath613 . suppose that we have a locking multiflow @xmath614 in @xmath609 and locking multiflow @xmath615 in @xmath610 . then a locking multiflow @xmath1 in @xmath371 is obtained by `` aggregating '' @xmath614 and @xmath615 as follows . an @xmath189-path @xmath596 in @xmath609 not connecting @xmath546 is regarded as an @xmath189-path in @xmath371 . set @xmath616 for such a path @xmath596 . similarly , set @xmath617 for each @xmath189-path @xmath596 in @xmath610 not connecting @xmath612 . next consider paths connecting @xmath546 in @xmath609 and @xmath612 in @xmath610 . observe that @xmath618 is a minimum @xmath546-cut in @xmath609 and @xmath619 is a minimum @xmath612-cut in @xmath610 . an edge @xmath382 in @xmath371 joining @xmath16 and @xmath620 becomes an edge connecting @xmath546 in @xmath609 and an edge connecting @xmath612 in @xmath610 . then @xmath621 necessarily holds . consider an @xmath546-path @xmath622 in @xmath609 and @xmath612-path @xmath623 in @xmath610 containing @xmath382 . the two paths @xmath622 and @xmath623 are concatenated along @xmath382 into an @xmath624-path @xmath596 in @xmath371 for @xmath625 , and set @xmath626 . decrease @xmath614 by @xmath627 on @xmath622 ( no @xmath612-flows in @xmath610 ) , and decrease @xmath615 by @xmath627 on @xmath623 . repeating this process until there are no @xmath546-flows in @xmath609 , we obtain a multiflow @xmath1 in @xmath371 . here @xmath1 is a locking in @xmath371 . this follows from the fact ( obtained from uncrossing ) that for @xmath176}}$ ] with @xmath628 ( resp . @xmath604 ) , a minimum @xmath189-cut in @xmath609 ( resp . @xmath610 ) is a minimum @xmath189-cut in @xmath371 . multiflows are kept as node - arc forms . this procedure , called the _ aggregation _ , can be done in @xmath341 time as in ( * ? ? ? * section 2.2 ) . suppose that @xmath629 . suppose that there is an @xmath189-fringe . compute a minimum @xmath189-cut @xmath16 . construct @xmath609 and @xmath610 as above , find locking multiflows @xmath614 in @xmath609 and @xmath615 in @xmath610 , and aggregate @xmath614 and @xmath615 into a locking multiflow @xmath1 in @xmath371 . in @xmath609 , there are two terminals , and a locking multiflow is obtained by a maximum flow . in @xmath610 , there are ( at most ) three terminals but no @xmath189-fringes . after recursing at most three times , we arrive at the situation that there are no fringes . this situation is precisely the same as ( * ? ? ? * section 2.1 ) . then a locking multiflow is obtained in at most three max - flow computations . the time complexity of this algorithm is analyzed in precisely the same way as ( * ? ? ? * section 2.3 ) , sketched as follows . for simplicity of analysis , we use orlin s @xmath341-time algorithm @xcite to find a maximum flow and minimum cut . let @xmath630 denote the time complexity of the algorithm applied to potts @xmath0-submodular functions on graph @xmath430 with @xmath631 , and @xmath632 . suppose that the time complexity of the max - flow algorithm and the aggregation procedure are bounded by @xmath633 and by @xmath634 for constants @xmath20 and @xmath635 , respectively . we show by induction that @xmath636 for a constant @xmath637 ( to be determined later ) . for @xmath638 , it holds @xmath639 . suppose that @xmath640 . then @xmath641 with @xmath642 . by induction , we have @xmath643 where we use @xmath342 and @xmath644 . for @xmath645 , it holds @xmath636 as required . this completes the proof of theorem [ thm : potts ] . the compact representation for @xmath50-closed sets by an elementary pip is kind of a data compression . hence it is natural to consider an efficient way to extract elements of the original @xmath50-closed set . this corresponds to the enumeration of consistent ideals of an elementary pip . as seen in remark [ lem : pip_and_cnf ] , consistent ideals correspond to true assignments of a boolean 2-cnf . thus we can enumerate all consistent ideals in output - polynomial time @xcite ( i.e. , the algorithm stops in time polynomial in the length of the input and output ) . maximal consistent ideals are of special interest , as described in section [ sec : relaxation ] . for a pip @xmath70 , let @xmath646 denote the family of maximal consistent ideals . now we consider the enumeration of @xmath646 . this can also be done in output - polynomial time by using the algorithm of @xcite in @xmath647 time per output . we here develop a considerably faster algorithm for the elementary pip of a potts @xmath0-submodular function @xmath10 . our algorithm utilizes its amalgamated structure by posets ( theorem [ thm : pip_potts ] ) . in fact , the structure of @xmath646 is quite simple , which we now explain . let @xmath570 be the posets , and @xmath571 } } } \sigma_\alpha \times { \left\{{\alpha}\right\}}$ ] the pip defined in the previous section . for distinct @xmath533}}$ ] , let @xmath648 be the subposet of @xmath550 . in particular , @xmath649 is equal to @xmath650 as a set , and the partial order of @xmath649 is the reverse of that of @xmath650 . let @xmath651 } } \setminus { \left\{{\alpha}\right\ } } } \sigma_{\alpha , \beta}$ ] . define the poset @xmath652 by disjoint sum @xmath653 where partial order @xmath396 on @xmath652 is defined as : the relation on @xmath649 is the same as the partial order of @xmath649 and there is no relation between @xmath649 and @xmath654 for @xmath655 . for a subset @xmath656 of @xmath550 , let @xmath657 . for an ideal @xmath658 of @xmath652 , let @xmath659 be defined by @xmath660 [ thm : enumeration_potts ] the map @xmath661 is a bijection from the ideal family of @xmath652 to @xmath646 . let @xmath658 be an ideal of @xmath652 . we first show that @xmath659 is a consistent ideal . consider @xmath662 . if @xmath663 , then @xmath664 since @xmath665 is an ideal by ( ep1 ) . suppose that @xmath666 . then @xmath667 . since @xmath658 is an ideal in @xmath649 , @xmath668 is an ideal in @xmath550 . consequently @xmath669 , and @xmath670 . suppose that @xmath671 . then @xmath672 . observe that @xmath673 is an ideal in @xmath674 . from this , we obtain @xmath670 , as above . since @xmath659 contains exactly one of @xmath675 with @xmath447 , the image @xmath659 is consistent and maximal . let @xmath109 be a maximal consistent ideal of @xmath395 . necessarily @xmath109 contains @xmath665 for all @xmath176}}$ ] . consider @xmath675 with @xmath447 . then @xmath109 contains exactly one of @xmath676 . otherwise , consider the principal ideal @xmath677 of @xmath678 , and the ideal @xmath679 . by maximality , @xmath679 is inconsistent . then there are @xmath680 and @xmath681 with @xmath682 . by ( ep1 ) , ( ep2 ) and lemma [ lem : intersection_of_isolating_cuts ] , there is @xmath683 with @xmath684 . since @xmath109 is an ideal , it holds @xmath685 . necessarily @xmath686 and @xmath687 . this is a contradiction . for distinct @xmath533}}$ ] let @xmath688 . then @xmath649 is the disjoint union of @xmath689 and @xmath690 ( as a set ) . thus , letting @xmath691 , we have @xmath692 . therefore our problem of enumerating all maximal minimizers of @xmath693 is reduced to the enumeration of all ideals of poset @xmath652 . this is a well - studied enumeration problem . one of the current best algorithms is squire s algorithm @xcite that enumerates all ideals of an @xmath3-element poset in amortized @xmath694 time per output . from the elementary pip for the minimizer set of potts @xmath0-submodular function @xmath436 , all maximal minimizers of @xmath10 can be enumerated in amortized @xmath694 time per output . the above poset @xmath652 may be viewed as a `` compact representation '' of maximal minimizers of potts @xmath0-submodular function @xmath10 . in a general elementary pip @xmath70 ( for the minimizer set of a general @xmath0-submodular function ) , such a compact representation is still possible if @xmath70 has a maximal consistent ideal @xmath637 satisfying the following property : * @xmath637 contains exactly one of @xmath291 and @xmath290 for each @xmath695 of the case ( ep2 - 1 ) . in this case , as in @xmath661 , there is a bijection between @xmath696 and @xmath646 . one can see that the pip @xmath697 has a simple structure similar to the above poset @xmath652 ( though it is not elementary ) . we developed an algorithm to enumerate consistent ideals of @xmath697 in @xmath12 time per output , and announced in the conference version of this paper @xcite that such a fast enumeration is possible for maximal consistent ideals of pip @xmath70 . however we found an elementary pip that having no maximal consistent ideal with the property ( p ) ; consider pip @xmath698 with @xmath699 , @xmath700 , @xmath701 , @xmath702 , @xmath703 , and @xmath704 . therefore ( * ? ? ? * theorem 14 ) is not true for such a pip . a _ @xmath0-submodular relaxation _ @xmath359 of a function @xmath705}}^n}{{\overline{\setr}}}$ ] is a @xmath0-submodular function on @xmath33 such that @xmath706 for all @xmath450}}^n { \left({\subseteq { s_k}^n}\right)}$ ] . yoshida @xcite investigated @xmath0-submodular relaxations as a key tool for designing efficient fpt algorithms . kolmogorov @xcite applied @xmath0-submodular relaxations to labeling problems on computer vision , which we describe below . a label assignment is a process of assigning a label to each pixel of a given image . for example , in the object extraction , each pixel is labeled as `` foreground '' or `` background '' . in stereo matching , the disparity of each pixel is computed from the given two photos taken from slightly different positions , and the pixel is labeled according to the estimated depth . we consider the labels to be numbered from 1 to @xmath0 . such a labeling problem is formulated as the problem of minimizing an _ energy function_. a potts energy function is simple but widely used energy function . however the exact minimization of a potts energy function is computationally intractable . gridchyn kolmogorov @xcite applied the @xmath0-submodular relaxation , that is , the energy function is relaxed to a @xmath0-submodular function by allowing some pixels to have 0 ( meaning `` non - labeled '' ) . the following property , called _ persistency _ @xcite , is the reason why they introduced the relaxation . [ thm : persistency ] let @xmath705}}^n}{{\overline{\setr}}}$ ] be a function and @xmath354 a @xmath0-submodular relaxation of @xmath1 . for every minimizer @xmath386 of @xmath359 , there exists a minimizer @xmath707}}^n$ ] of @xmath1 such that @xmath358 implies @xmath173 for each @xmath36}}$ ] . namely , each minimizer of @xmath359 gives us partial information about a minimizer of @xmath1 . an efficient algorithm for minimizing @xmath0-submodular relaxations of potts functions was also proposed in @xcite . hence we can obtain a partial labeling extensible to an optimal labeling , which we call a _ persistent labeling_. in section [ sec : potts ] we gave an efficient algorithm to construct the elementary pip representing all the minimizers of a potts @xmath0-submodular function . since minimizers that contain more nonzero elements have more information , we want to find a minimizer whose support is largest . in fact , such minimizers are precisely maximal minimizers . let @xmath107 be a @xmath50-closed set on @xmath33 . the supports of maximal elements in @xmath107 are the same . let @xmath134 be maximal and @xmath708 . for each @xmath36}}$ ] , it holds @xmath709 if @xmath710 and @xmath711 if @xmath712 . in particular , @xmath713 and @xmath714 hold . since @xmath163 is maximal , we obtain @xmath715 and @xmath716 . by changing the role of @xmath35 and @xmath163 , we also have @xmath717 . thus @xmath209 . from this lemma , it turns out that all maximal minimizers of @xmath359 have the same and largest amount of information about minimizers of @xmath1 . in labeling problem with potts energy , all maximal persistent labelings ( with respect to a @xmath0-submodular relaxation ) can be efficiently generated by the algorithm in section [ sec : enumeration ] . we implemented our algorithm on the stereo matching problem with potts energy function . this has an aspect of the replication of the experiment in @xcite , but we computed not only one of the persistent labelings but also its pip - representation . we used `` tsukuba '' and `` cones '' in the middlebury data @xcite as input images . [ [ problem - setting - and - formulation . ] ] problem setting and formulation . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we are given photo images @xmath56 and @xmath652 taken from left and right positions , respectively . the images @xmath56 and @xmath652 are @xmath718 arrays such that entries @xmath719}}$ ] and @xmath720}}$ ] are rgb vectors @xmath721 of the intensity at pixel @xmath722 , where each pixel is represented by a pair @xmath722 of its horizontal coordinate @xmath723 and vertical coordinate @xmath724 . the goal of the stereo matching problem is to assign to each pixel the `` disparity label '' @xmath725}}$ ] that represents the depth of the object on the pixel . we model this problem as a minimization of a potts energy function on diagonal grid graph @xmath430 , where @xmath364 is the set of pixels , and two pixels @xmath722 and @xmath726 have an edge in @xmath727 if and only if @xmath728 and @xmath729 . the first and the second term of are called `` data term '' and `` smoothness term '' @xcite , respectively . for each pixel @xmath579 , the data term @xmath438 measures how well the estimated disparity of pixel @xmath285 agrees with the pair of given images . we employed the traditional averaged ssd ( sum of squared difference ) costs as in @xcite : @xmath730 } } - { { r}{{\mathopen{}\mathclose\bgroup\left}[{x + d_\alpha , y}{\aftergroup\egroup\right}]}}}\right\rvert}^2 $ } \quad { \left({\alpha \in { { \left[{k}\right]}}}\right)},\end{aligned}\ ] ] where @xmath731 is the @xmath732 window centered at @xmath733 ( i.e. , the set of pixels @xmath726 with @xmath734 and @xmath735 ) , @xmath736 is the 2-norm , @xmath737 is the disparity corresponding to the label @xmath189 . we simply set @xmath738 for each @xmath176}}$ ] . for each pair of adjacent pixels @xmath433 , the smoothness term increases the energy by @xmath434 if @xmath285 and @xmath286 have different labels . we set every @xmath434 to be the same value @xmath739 as in @xcite , and conducted experiments with @xmath740 and 20 to see the effect of @xmath739 . [ [ experimental - results . ] ] experimental results . + + + + + + + + + + + + + + + + + + + + + and `` cones '' ( bottom row ) with @xmath741 and `` cones '' ( bottom row ) with @xmath741 and `` cones '' ( bottom row ) with @xmath741 and `` cones '' ( bottom row ) with @xmath741 and `` cones '' ( bottom row ) with @xmath741 and `` cones '' ( bottom row ) with @xmath741 figure [ fig : result_image ] and table [ tbl : result ] show the results of our experiment . the pixels labeled in the minimum persistent labeling are colored in gray of the brightness corresponding to each label . the blue pixels are unlabeled even in maximal persistent labelings . the red pixels are the difference between the minimum persistent labeling and maximal ones , i.e. , the pixel was labeled in ( any of ) maximal persistent labelings but not in the minimum one . we can observe that there are few red pixels as mentioned in @xcite , and they are mainly located on the boundary of two regions with different labels . a possible reason is the following : consider a simple 1-dimensional case where a pixel @xmath285 is adjacent only to pixels @xmath286 and @xmath742 . let @xmath386 be the minimum persistent labeling and assume @xmath743 . then the increment of the energy is the same ( @xmath744 ) even if @xmath37 is set to any of @xmath745 . therefore the pixel @xmath285 will be red if @xmath746 , and thus we think that this will occur in boundaries more frequently than inside of regions . with regard to the effect of @xmath739 , the larger @xmath739 decreases the percentages of gray and red pixels on both tsukuba and cones , and increases the blue pixels to the contrary . this result agrees with the experiments in @xcite . we consider that this is due to the fact that the value of each @xmath747 is moderately lower in @xmath451 since @xmath747 is the average of the minimum and the second minimum values of @xmath438 as described in section [ sec : potts ] . hence if @xmath739 is large , the energy will be lower just by letting all @xmath748 than by tuning each @xmath37 finely according to the values of the corresponding data term @xmath451 . .experimental result [ cols="<,>,^,^,^,^",options="header " , ] [ tbl : result ] [ [ the - structure - of - the - pip . ] ] the structure of the pip . + + + + + + + + + + + + + + + + + + + + + + + + + ] figure [ fig : result_pip ] shows the pip - representation for the partial labelings on tsukuba with @xmath740 . the pip consists of many small connected pips , which correspond to each of connected red regions in figure [ fig : result_image ] . thus we can easily calculate the number of maximal persistent labelings by multiplying the number of maximal consistent ideals of each connected pip ( notice that an ideal of the pip is maximal and consistent if and only if it contains all elements having no inconsistent pair and one element of each minimally inconsistent pairs ) . the right most column of table [ tbl : result ] shows the number of maximal persistent labelings in each experiment . we discovered the fact that there are plenty of maximal persistent labelings even though the percentages of red pixels are little . we thank kazuo murota for helpful comments . this work was partially supported by jsps kakenhi grant numbers 25280004 , 26330023 , 26280004 , and by jst , erato , kawarabayashi large graph project .
most mass in the universe is in the form of dark matter , and most dark matter is gravitationally clustered in the form of halos . dark matter halos are therefore the building blocks of the universe . knowledge of the formation and evolution history of dark matter halos is essential for an understanding of the halos structure and dynamics , as well as the environment that harbors galaxies and clusters . designs for dark matter search experiments also rely on the predicted spatial and velocity distributions of dark matter particles in our own galaxy . according to the theory of structure formation in current cosmological models , dark matter halos arise from tiny primordial fluctuations seeded in the matter and radiation fields in the early universe at , for example , the end of an inflationary era . these small inhomogeneities grow via gravitational instability , with the slightly overdense regions becoming denser and the slightly underdense regions becoming emptier . at a given time , there exists a spectrum of density fluctuations over a wide range of length and mass scales , whose spatial distribution is characterized by the power density in fourier space @xmath0 , the power spectrum . dark matter halos therefore rarely form and evolve in isolation , growing instead by frequent accretion of smaller - mass halos and occasional major mergers with other halos of comparable mass . the earliest theoretical insight into cosmological halo formation came from the spherical radial infall model , which considers an isolated spherical overdense volume and describes the secondary infall of bound but initially expanding collisionless matter onto this region @xcite . if the initial density perturbation spectrum is scale - free and the velocity distribution follows the hubble law , the infall solution is found to approach a self - similar form in an einstein - de sitter universe . the resulting density profile asymptotically approaches @xmath1 with @xmath2 @xcite . the exact value of @xmath3 depends on the index of the initial perturbation spectrum @xmath4 : it is isothermal ( @xmath5 ) for @xmath6 and for @xmath7 steepens to @xmath8 @xcite or @xmath9 @xcite . the collapse of halos therefore retains memory of the initial conditions in the radial infall model . by contrast , modern cosmological n - body simulations find dark matter halos over a wide range of mass scales to follow a density profile that is nearly independent of the initial power spectrum and is shallower near the halo centers ( with @xmath10 ) than that predicted by the radial infall model @xcite . furthermore , the dark matter in the simulated halos is not entirely smoothly distributed spatially ; instead , at least 10% of the halo mass is in the form of hundreds to thousands of smaller , dense satellite halos of varying mass @xcite . thus even though the analytic similarity solutions derived from one - dimensional infall onto a smooth overdensity have provided valuable insight into the evolution of self - gravitating systems , realistic halo formation is clearly more complicated . numerical simulation has been the tool most frequently used in cosmology to study structure formation . it has provided powerful insight and occasionally surprising results ( e.g. universal density profile ) unexplainable by the simulations themselves . our understanding of structure formation can certainly be enhanced if a complementary tool is developed . complementary approaches are also especially needed at this time to help us better understand and perhaps resolve the `` cuspy halo '' and `` dwarf satellite '' problems of the cold dark matter model . in this paper we offer a new perspective on the subject of dark matter halo formation by proposing an approach different from both the radial infall model and n - body simulations . we focus on the phase - space distribution of dark matter in galaxy halos instead of on the motion of a mass shell ( as in infall models ) or the orbit of a simulation particle ( as in n - body ) . we attempt to understand the evolution of the phase - space distribution with a statistical description based on kinetic theory . our method is to be contrasted with the standard method where the individual dark matter particles are described by the collisionless boltzmann equation ; n - body simulations are in essence a monte carlo method for solving this equation . instead , we consider the phase space of dark matter particles in an average halo and develop a cosmological kinetic theory for halo evolution in the presence of a spectrum of density fluctuations . the essential features of our approach are represented by the fluctuation - dissipation theorem for time - dependent stochastic processes . to elaborate on this concept , we recall that stochastic modelling was first developed to explain the brownian motion of pollen particles suspended in a fluid . ( see [ sec : brownian ] for a more detailed discussion . ) this phenomenon is now understood to arise from a fluctuating force on a brownian particle due to many rapid collisions with the molecules in the surrounding fluid . these fluctuations cause successive small changes in the particle velocity , hence to a random walk . the consequence of many such random walks is diffusion , a form of dissipation . due to the large number of molecules , it is impossible to solve the coupled equations of motion for the trajectories of all molecules and the pollen particle in the fluid fortunately , a stochastic treatment suffices . one mathematical prescription is provided by the langevin equation @xcite , which models the forces arising from the fluctuations as a stochastic term that is added to all other deterministic forces in the equation of motion for a particle . an alternative prescription is given by the fokker - planck equation @xcite , which is an equation of motion for the probability density or distribution function of particles undergoing random walks . in this case , the collective changes in the velocities due to the fluctuations in the system are represented by a drift coefficient and a diffusion coefficient arising from the collision terms in the collisional boltzmann equation . stochastic phenomena occur beyond molecular scales . in astrophysics , the best studied example is the evolution of dense stellar systems such as globular clusters @xcite . a star in a globular cluster experiences a fluctuating force from many gravitational two - body encounters and suffers a succession of small velocity changes . the fluctuations in these systems arise from granularity in the mass distribution at the locations of the stars . the fokker - planck approximation is valid for these weak encounters . we explore this example in greater detail and contrast it with our model of galaxy halo evolution in [ sec : globular ] . the key feature common to all of these systems is the presence of stochastic fluctuations that result in dissipation . in the cosmological setting addressed by this paper , the fluctuations arise from cosmological perturbations produced in the early universe . at early times , these perturbations are gaussian and fully characterized by their power spectrum ; at late times , the substructures in halos seen in recent simulations are the nonlinear counterparts of this spectrum of early ripples . these fluctuations scatter particles . if one considers an ensemble of halos , the masses and locations of the substructures ( or their gaussian predecessors ) are random variables . the substructures thereby introduce `` noise '' into the evolution of dark matter halos . as we show in this paper , the result is dissipation in the form of drift and velocity - space diffusion . recent work has explored the use of the fokker - planck equation in cosmological problems . @xcite , for example , searched for steady - state solutions of the fokker - planck equation assuming the fluctuations arise from two - body encounters . it also assumed a simple isotropic phase space distribution dependent only on the total energy and a radial power law form for the potential and density . under these assumptions , @xcite found that @xmath11 was a steady - state solution under the effects of encounters . there was no dynamics in this work , however : the system was assumed to be in collisionless equilibrium and the nature of the clumps producing the collisions was left arbitrary . it also remains to be determined whether galaxy halos , either dark matter or stellar , actually evolve to @xmath12 cusp . @xcite examined the dynamics of halos due to stochastic fluctuations from three different noise processes for the fokker - planck scattering term : the merging clumps were assumed to fly by with constant velocity , to follow decaying orbits due to dynamical friction , or to orbit inside halos . the results of this study were promising and supported the notion that the density profiles of halos were affected by mergers and substructures . the fluctuations in this work , however , were constructed by hand instead of being self - consistently calculated , and the results about a central cusp were inconclusive . our objective here is to construct a kinetic equation in which the drift and diffusion terms are derived from realistic density fluctuations in current cosmological models . we will also retain the full phase space information by , in effect , using @xmath13 without averaging over the angular momentum or radial velocity . we also do not assume collisionless equilibrium ; instead we base our approach on exact dynamics . the organization of this paper is as follows : in 2 , we derive the kinetic equation relevant for the evolution of an average halo , starting with the collisionless boltzmann equation for individual dark matter particles . this procedure gives rise to a collision term on the right - hand side of the kinetic equation . this term represents a correlated force density and arises from fluctuations in the ensemble averaged gravitational potential due to clustering of the matter distribution . in , we introduce the concepts and physical variables needed to apply this equation to cosmological problems . the key step is in relating the phase space density @xmath14 to the probability densities of two standard variables in cosmology : the mass density @xmath15 and velocity @xmath16 . we then focus on the quasi - linear regime in which the fluctuations about the mean density and hubble flow of an isotropic and homogeneous universe are small . this focus allows us to relate @xmath15 and @xmath16 to the density and displacement fields @xmath17 and @xmath18 , which are gaussian random fields in standard cosmological models . we leave the most mathematical part of the paper to appendix [ sec : pgauss ] , where we describe how to use the statistics of constrained gaussian random fields to express the probability densities of 3 in terms of the ( constrained ) means and covariances of the density and displacement fields . the idea of a constrained field is central to this study because our aim is to construct a kinetic theory for dark matter particles that will later reside in a collapsed halo rather than at any random location in the universe . the relevant perturbations are for a density field constrained to have certain properties , e.g. , a specified height or gradient at some point in space . the spectrum for such constrained fields is related but not identical to the familiar linear power spectrum @xmath19 for unconstrained gaussian density fluctuations . in [ sec : fpeq ] , we put the final expression for the correlated force derived in appendix [ sec : pgauss ] back into the kinetic equation , and show that the resulting equation is of the fokker - planck form where one term represents a drift force and the other term represents diffusion in velocity - space . we work out these diffusion coefficients for the currently - favored cold dark matter plus a cosmological constant ( @xmath20cdm ) model , and provide explicit expressions for these coefficients in terms of the linear matter fluctuation power spectrum . we also present the results for the coefficients computed for this model . [ sec : discuss ] discusses further the physical meaning of the results in [ sec : fpeq ] . we interpret the diffusion terms by taking velocity moments of the kinetic equation . to place our cosmological kinetic equation in a broader perspective , we discuss in more detail the classical examples of brownian motion and globular clusters , and point out the similarities and differences among the three cases . [ sec : conclu ] provides a summary and conclusions . in appendix [ sec : stationary ] we investigate the relaxation to stationary solutions of our cosmological fokker - planck equation . we derive the kinetic equation for galaxy halo evolution following the methods presented by @xcite for the derivation of the bbgky hierarchy . the starting point is the one - particle phase space density for dark matter particles in a single halo , the klimontovich ( or klimontovich - dupree ) density @xcite @xmath21 \delta_{\rm d}[\vec v-\vec v_i(t)]\ ] ] where @xmath22 is the dirac delta function and the subscript k denotes the use of the klimontovich density . we suppose that each halo is made of equal - mass particles of mass @xmath23 , for example , the elementary particles of non - baryonic dark matter models . for convenience , we use velocity rather than momentum and we choose units so that @xmath24 is a mass density . we use proper coordinates except for [ sec : cosmovar ] and appendix [ sec : pgauss ] , where we will switch to the appropriate cosmological ( comoving ) coordinates . the klimontovich density is simply a way of casting the exact trajectories of all @xmath25 particles in a system into the form of a phase space density . the number of particles in @xmath26 about @xmath27 is @xmath28 , which equals either 0 or 1 . the klimontovich density consists of delta functions because , when one is dealing with a single system ( as opposed to an ensemble ) , every particle has a well - defined position and velocity . the reason for introducing this quantity is that it gives a particularly clean way of going from exact n - body dynamics to a statistical description by taking expectation values of @xmath29 or products of more than one @xmath29 over an ensemble of n - body systems . this approach to kinetic theory is different from the usual method that begins with the n - particle distribution for an ensemble . the klimontovich density obeys the klimontovich - dupree equation @xmath30 where @xmath31 for notational convenience we have grouped together all 6 phase space coordinates into @xmath32 . ( [ klim ] ) is exact , as may be verified by substituting eq . ( [ phaseden ] ) and using the equations of motion for individual particles , @xmath33 in practice , the gravitational force may be softened by modifying eq . ( [ gravec ] ) to reduce artificial two - body relaxation , as is done in n - body simulations . ( [ phaseden ] ) and ( [ klim ] ) are entirely equivalent to eqs . ( [ newton ] ) for one system . the klimontovich phase space density retains _ all _ information about a particular halo because it specifies the trajectories of all particles . ( [ phaseden])([newton ] ) are formally equivalent to a perfect n - body simulation . that is not what we want . rather than giving a perfect description of a single halo , we average over halos to obtain a statistical description of halo evolution . while it is impossible to completely describe a single galactic halo of dark matter particles , by using statistical mechanics methods we _ can _ describe the average of infinitely many halos ! before proceeding further , we make one important modification to the usual procedure in kinetic theory . ( [ newton ] ) assume that we work in an inertial frame . this is satisfactory for an isolated halo but not for a halo formed in a cosmological context where neighbors cause it to orbit in the net gravitational field produced by all matter . the center of mass of a halo typically moves much farther in a hubble time than the size of the halo itself , even in an inertial frame in which the halo begins at rest . for example , for the cold dark matter family of models , the displacement of a mass element from its initial position ( in comoving coordinates ) is typically several mpc today . if we statistically average over an ensemble of halos each of which wanders this far in a hubble time , the phase space density will be artificially smeared out , making the resulting average `` halo '' much larger than any realistic halo . we avoid this by doing the same thing that simulators do when they construct halo profiles : we work in the center - of - mass frame of each halo . indeed , at each moment the halo center - of - mass must be at rest in our reference frame . this frame is non - inertial because of the gravitational forces acting on the halo . thus we must modify the kinetic equation to apply to an accelerating frame . fortunately , the accelerations are small and the frame is extended only a few mpc , so we do not have to worry about relativistic effects . we simply include the inertial ( `` fictitious '' ) forces associated with working in an accelerating frame . we choose to define @xmath34 as the center - of - mass of each halo . transforming eqs . ( [ newton ] ) to the center - of - mass frame is very simple : we replace @xmath35 by the _ tidal _ gravitational field relative to the center of each halo , @xmath36 , and then subtract the center - of - mass velocity from the initial velocity of each particle . ( [ klim ] ) is unchanged except that @xmath35 is replaced by @xmath37 the subscript t is a reminder that this is the tidal gravitational field . now we return to the usual procedure for deriving a kinetic equation . the klimontovich density is the phase space density for a single realization of a halo . we might imagine performing many n - body simulations of different halos , each with slightly different initial conditions . if we average the klimontovich density over this ensemble of halos , the result ( at least , in the limit where the size of the ensemble becomes infinite ) is no longer a sum of delta functions but instead is a smooth _ one - particle distribution function _ : @xmath38 where angle brackets denote the ensemble average . recall that for notational convenience we have grouped all six phase space variables into @xmath39 . our goal is to derive a tractable kinetic equation for @xmath40 . the obvious next step is to take the ensemble average of eq . ( [ klim ] ) . to proceed , however , we will need the _ two - particle distribution function _ @xmath41 , which follows from averaging the product of klimontovich densities at two points : @xmath42 the delta - function term @xmath43 corresponds to the case in which the two points lie on top of one particle . two distinct particles give the @xmath44 term , which may always be written as a product term plus a two - particle correlation : @xmath45 this equation defines the two - point correlation function in phase space , @xmath46 . the averaging we perform is similar to the averaging used in defining ordinary clustering correlation functions in cosmology . the differences here are that we are not averaging over arbitrary systems but only over those that lead to formation of a halo at @xmath34 , and that we retain velocity information . had we averaged over all space and integrated over velocities , we would have gotten @xmath47 and @xmath48 where @xmath49 is the cosmic mean mass density and @xmath50 is the usual spatial two - point correlation function . now , by taking the ensemble average of eq . ( [ klim ] ) we arrive at a kinetic equation for the evolution of the average halo phase space density @xcite : @xmath51 where we define the ensemble - averaged tidal field @xmath52 and the correlated force density @xmath53=-g\int d^6w'\,f_{2c}(\vec w , \vec w^{\,\prime},t)\left(\frac{\vec r-\vec r^{\,\prime}}{\vert \vec r-\vec r^{\,\prime}\vert^3}+\frac{\vec r^{\,\prime}}{\vert \vec r^{\,\prime}\vert^3}\right)\ .\ ] ] the covariance of two random variables is denoted @xmath54\equiv \l ( a-\l a \r)(b-\l b \r)\r = \langle ab\rangle-\langle a\rangle\langle b\rangle$ ] . the product of @xmath55 and @xmath29 involves a product of two klimontovich densities , for which we have used eqs . ( [ f2 ] ) and ( [ f2c ] ) . the one - particle discreteness term in eq . ( [ f2 ] ) makes no contribution because the particle has no self - force . ( [ boltzeq ] ) is the first bbgky hierarchy equation @xcite . in cosmological applications this equation is usually derived starting from the liouville equation for the n - particle distribution function . in order to clarify the averaging that is done here , we have started instead from the one - particle klimontovich density . the result , eq . ( [ boltzeq ] ) , is an evolution equation that looks very much like eq . ( [ klim ] ) . by averaging over halos , however , we have introduced a correlation integral term on the right - hand side . the _ average _ halo therefore does not evolve according to the vlasov equation ; instead , it evolves according to a kinetic equation with an effective collision term . ( [ boltzeq])([g2 ] ) are exact . no approximation has been made yet . ( [ boltzeq ] ) is a continuity equation for density in phase space . without the right - hand side , it says that phase space density is conserved along trajectories given by eq . ( [ newton ] ) modified for the tidal force . the right - hand side gives a correction due to the fact that fluctuations about the ensemble average lead to correlated forces . as we will see , the effects of these fluctuations lead to dissipation . because the correction term is a divergence , the mass density @xmath56 is not dissipated but instead is conserved . in most applications of kinetic theory , the right - hand side of eq . ( [ boltzeq ] ) arises from particle collisions . in the usual treatment of a gas of particles with short - range forces , one assumes @xmath57 always except during particle collisions . this is the assumption of molecular chaos that boltzmann introduced to derive his celebrated kinetic equation . our situation , however , is different : we can not neglect @xmath46 , because it can be much larger than the single - particle term in eq . ( [ f2c ] ) due to strong gravitational clustering . heuristically , @xmath46 describes the substructure within a galaxy halo at the two - point level . ( higher order correlation functions would be needed for a complete description . ) current cosmological models and observations imply that the initial density field has fluctuations that cause the formation of many small halos which subsequently merge into present - day halos . the lumpiness of the matter distribution represents a fluctuation about the ensemble average density field . through the correlated force density @xmath58 , these fluctuations cause changes in the energy and angular momentum of individual particle orbits that are important for the evolution of the one - particle distribution function . ( [ boltzeq ] ) is useful only if we find an expression for @xmath46 . one way to proceed would be to derive a kinetic equation for @xmath46 by taking the ensemble average of eq . ( [ klim ] ) multiplied by the klimontovich density . this leads to the second bbgky equation , which depends on the three - point correlation in phase space , which obeys the third bbgky equation , and so on . thus we either have an infinite chain of coupled equations of increasing dimensionality , or we must close the system . @xcite handled this problem in their study of gravitational clustering by writing the three - point function in terms of products of two - point functions . we will proceed differently , by seeking to close the hierarchy at the first level , eq . ( [ boltzeq ] ) . to summarize this section , we have derived a formally exact evolution equation for the average halo . the one - particle phase space density is conserved up to a fluctuating force term arising from the two - point correlation function in phase space . physically this term represents the effects of substructure within and around halos , which scatter particles to new orbits . in the next two sections , we will show that we can in fact express @xmath46 in terms of the one - particle phase space density @xmath40 using a relation that is exact to second order in cosmological perturbation theory . this relation will enable us to close the first bbgky hierarchy . solving our evolution equation in eqs . ( [ boltzeq])([g2 ] ) requires specifying the phase space density @xmath40 at an initial time and the two - particle correlation @xmath46 at all times . in this section we show how to compute these quantities from the probability distributions of mass density and velocity at one and two points in space . we wish to derive expressions that would relate @xmath59 and @xmath60 to probability distributions of mass density and velocity , because these latter quantities are naturally specified by a cosmological model for structure formation . the method is to examine the mass and momentum in a small volume of space and to construct ensemble averages of the klimontovich density using the probability distribution of density and velocity . we use the symbol @xmath61 to refer to the probability density with respect to its arguments , where the joint probability distribution for @xmath62 is @xmath63 . we assume that these probability distributions are always normalized to unit total probability . the variables relevant for this paper are the densities @xmath15 and velocities @xmath64 at two points @xmath65 and at one time @xmath66 : @xmath67 , @xmath68 , @xmath69 , and @xmath70 . initially we assume that the velocity field in a given realization is single - valued in space . we use the following probability distributions : @xmath71 where the different probability densities are related to one another by @xmath72 these probability distributions depend on time but we suppress the @xmath66-dependence for clarity . for a small volume in phase space , @xmath73 equals the mean mass contained in @xmath74 . the same quantity can be calculated by averaging @xmath75 over density at fixed velocity using @xmath76 , i.e. , @xmath77 . equating the results gives @xmath78 we have used the conditional probability density @xmath79 to define the conditional mean @xmath80 . note that the probability densities depend on @xmath81 through @xmath82 and @xmath83 , so eq . ( [ f1p ] ) yields the desired one - particle phase space dependence on @xmath84 . this argument is easily extended to the two - point phase space density , using the joint probability distribution of density and velocity at two points . as eq . ( [ g2 ] ) shows , we need @xmath85 , which depends on @xmath86 but not @xmath87 , i.e. , we need the velocity at only one point . recalling that @xmath88 , we obtain @xmath89\nonumber\\ & = & \int_0^\infty d\rho_1\,\rho_1\int_0^\infty d\rho_2\,\rho_2 \left[p(\rho_1,\rho_2\vert\vec v_1)-p(\rho_1\vert\vec v_1 ) p(\rho_2)\right]\,p(\vec v_1)\nonumber\\ & = & \left[\left\langle\rho_1\rho_2\vert\vec v_1\right\rangle -\left\langle\rho_1\vert\vec v_1\right\rangle\left\langle \rho_2\right\rangle\right]\,p(\vec v_1)\ .\end{aligned}\ ] ] although we have derived eqs . ( [ f1p ] ) and ( [ f2cp ] ) assuming a single - valued velocity field in each realization , they are also valid when there is a distribution of velocities at each point . one simply interprets @xmath90 as the total mass density while @xmath86 is a single particle velocity . therefore these equations are fully general . at high redshift , before dark matter halos formed , the density and velocity fields were only slightly perturbed from a uniformly expanding cosmological model . it is thought that the fluctuations of density and velocity responsible for all cosmic structure were gaussian random fields produced during the inflationary era of the very early universe . these fluctuations provide us with both the initial conditions and the random element leading to a probabilistic description for halo formation . our goal now is to obtain expressions for eqs . ( [ f1p ] ) and ( [ f2cp ] ) at early times while the density fluctuations are small . before proceeding further , we review the description of density and velocity fields at high redshift in an almost uniformly expanding cosmological model . first , the mean expansion is described by the cosmic expansion scale factor @xmath91 which is related to the redshift by @xmath92 . the hubble expansion rate is @xmath93 . the matter contributes a fraction @xmath94 to the closure density . we account for the mean expansion by defining new coordinates conventionally called comoving coordinates : @xmath95 . adopting the standard cosmological model , we assume that the dark matter is cold . this assumption means that the velocity field is single - valued in space at early times before halos form . we can then write the density field @xmath15 and velocity field @xmath16 in terms of perturbations @xmath17 and @xmath18 around a uniformly expanding background : @xmath96\ , \quad \vec v(\vec r , t)=h(t)\left[\vec r+b(t)\vec\psi(\vec x , t)\right]\ .\ ] ] here @xmath97^{0.6}$ ] is the growth rate for density fluctuations with time - dependence @xmath98 . we emphasize that eq . ( [ pertdv ] ) does not assume that the density fluctuations are small ; it only assumes that the velocity field is single - valued . we are implicitly assuming that all of the matter may be described as a single cold non - interacting fluid ( i.e. we assume that baryons move like cold dark matter over the length scales of interest ) . the computation of the functions @xmath91 , @xmath99 , @xmath100 , @xmath101 , and @xmath102 is standard in cosmology and will not be described here . at early times when @xmath103 , mass conservation relates the density and velocity perturbation fields by @xmath104 , or @xmath105 now we make a crucial step by changing variables from @xmath106 used previously to @xmath107 . the reason for doing so is that the description in comoving coordinates @xmath108 and perturbation variables @xmath18 and @xmath17 is much simpler in the small perturbation limit . following the notation in 3.1 , we write @xmath109 , @xmath110 , and @xmath111 . these variables have normalized joint probability densities @xmath112 , @xmath113 , and @xmath114 in analogy with eqs . ( [ probs ] ) and ( [ pdfs ] ) . to change variables from @xmath115 to @xmath116 , we use @xmath117 , @xmath118 , and @xmath119 . ( [ f1p ] ) then becomes @xmath120 p(\vec\psi_1 ) \ , ( hb)^{-3}\ , \ ] ] where @xmath121 and @xmath122 is the conditional distribution of @xmath123 . similarly , eq . ( [ f2cp ] ) becomes @xmath124p(\vec\psi_1)(hb)^{-3}\ .\ ] ] although the derivation of eqs . ( [ f1p1a ] ) and ( [ f2cp1a ] ) assumed that the velocity field is single - valued , they are valid in general provided that one interprets @xmath123 and @xmath125 as total mass density perturbations ( not necessarily small ) while @xmath126 is proportional to the single particle velocity . ( [ f1p1a ] ) and ( [ f2cp1a ] ) are equivalent to eqs . ( [ f1p ] ) and ( [ f2cp ] ) expressed in different variables . their utility becomes evident in appendix [ sec : pgauss ] where we evaluate the probability distributions of density and velocity in cosmological models . all that remains is to calculate the expectation values in eqs . ( [ f1p1a ] ) and ( [ f2cp1a ] ) and to substitute the latter into eq . ( [ g2 ] ) . this calculation is somewhat complicated and is presented in appendix [ sec : pgauss ] . part of the complication arises because of an important detail we have so far neglected : halos form preferentially in regions that are initially overdense . our calculations assume that a halo forms at @xmath127 . consequently we should not evaluate the expectation values for arbitrary points in space but rather for those that satisfy appropriate constraints . strictly speaking it is impossible to specify in advance the regions that will form halos in a cosmological model with random initial fluctuations because the dynamics of gravitational clustering is too complicated . a simple model , however , has been found to give reasonable agreement with numerical simulations : halos form at maxima of the smoothed initial density field @xcite . these constraints are found to modify the expectation values in eqs . ( [ f1p1a ] ) and ( [ f2cp1a ] ) . to emphasize the application of constraints , we rewrite these expectation values with a subscript @xmath128 : @xmath129 p(\vec\psi_1 ) \ , ( hb)^{-3}\ , \ ] ] and @xmath130p(\vec\psi_1)(hb)^{-3}\ .\ ] ] eqs . ( [ f1p1 ] ) and ( [ f2cp1 ] ) are the main results of this section . we will use them to specify the initial conditions on @xmath40 and to determine the correlated force density in eq . ( [ g2 ] ) . the next step is to evaluate the necessary constrained expectation values using the standard cosmological model . we present the details of this calculation in appendix [ sec : pgauss ] , where we choose to err on the side of simplicity by following @xcite and using only the zeroth and first derivatives of the smoothed initial density field @xmath131 as constraints . detailed comparison has shown that this simple model does not predict perfectly all halo formation sites @xcite , and more complicated constraints work better , e.g. @xcite . the procedure of appendix [ sec : pgauss ] can presumably be extended to incorporate more complicated constraints . after having devoted [ sec : dfgauss ] and appendix [ sec : pgauss ] to probability and statistics we are now ready to return to the kinetic equation in [ sec : kinetic ] and write down an explicit expression for the correlated force density @xmath132 on the right - hand - side of the kinetic equation ( [ boltzeq ] ) . the approach we have followed is exact rather than phenomenological our starting point was the first bbgky hierarchy equation , an exact kinetic equation . the only approximation we have made is to assume that the density and velocity fields are gaussian random fields . the resulting two - particle correlation function , eq . ( [ f2cexact ] ) , depends in a simple way on the one - particle distribution @xmath133 . as we will show next , this simple dependence leads to the fokker - planck equation . this is a remarkable and non - trivial result . the fokker - planck equation is a great simplification of the exact and intractable bbgky hierarchy . usually the fokker - planck equation is _ assumed _ to hold a priori , and then its diffusion coefficients are estimated . here , on the other hand , we have _ derived _ the fokker - planck equation . along with it we obtain the diffusion coefficients in the quasi - linear regime , as follows . substituting eq . ( [ f2cexact ] ) into eq . ( [ g2 ] ) using eq . ( [ f2cp ] ) , converting to comoving coordinates , using eq . ( [ linpsi ] ) to relate @xmath18 to @xmath16 , and using eq . ( [ f1lin ] ) , to third order in @xmath17 we get @xmath134 \,f \right . \nonumber\\ & & + \left . ( hb ) \left [ c(\vec\psi,\vec\psi\,)-c(\vec\psi_0,\vec\psi\ , ) -\langle\vec\psi\,\rangle_c\otimes c(\delta,\vec\psi\ , ) \right ] \cdot\frac{\partial f}{\partial\vec v}\right\}+o(\delta^4)\ , \end{aligned}\ ] ] where the covariance matrices @xmath128 of variables @xmath17 and @xmath18 are given in eqs . ( [ covfun ] ) and ( [ c0 ] ) . here , @xmath135 and @xmath136 . note that all the terms with subscript 0 arise from our use of the tidal field rather than the total gravitational acceleration . substituting eq . ( [ g2lin ] ) into eq . ( [ boltzeq ] ) , in proper coordinates our kinetic equation in the quasilinear regime reduces to the fokker - planck equation @xmath137 specified by a vector field @xmath138 and a tensor field @xmath139 . together , @xmath138 and @xmath139 are called diffusion coefficients . heuristically , @xmath138 represents a drift force ( @xmath140 has units of acceleration ) while @xmath139 represents the diffusive effects of fluctuating forces . the flux density in velocity space is @xmath141 . keeping only terms up to second order in the density perturbations , these quantities are given by @xmath142 \nonumber\\ & = & - \hbox{cov}(\delta,\vec g_0\vert\vec\psi\ , ) = \hbox{cov}(\delta,\vec g_{\rm t}\vert\vec v\,)\end{aligned}\ ] ] and @xmath143 in eqs . ( [ driftlin ] ) and ( [ difflin ] ) we have also used the fact that in the quasilinear regime the peculiar gravity and displacement are related by @xmath144 @xcite , and the tidal field is @xmath145 , where @xmath146 . in writing the second line of eq . ( [ driftlin ] ) we have used eq . ( [ covy ] ) with @xmath147 and @xmath148 . thus , @xmath149 should be understood to be subject to the same extremum or other constraints as e.g. the constrained covariance function @xmath150 . the displacement and velocity are related by the second of eqs . ( [ pertdv ] ) . thus , in linear theory @xmath151 so that conditioning on @xmath16 makes the fluctuation in @xmath35 vanish , allowing us to replace @xmath152 by @xmath153 in eq . ( [ driftlin ] ) . in all cases , the covariances are taken subject to the extremum constraints discussed in appendix [ sec : pgauss ] . it is worth noting that had we used the total gravitational acceleration @xmath154 instead of the tidal field @xmath155 , the drift vector @xmath138 would have been zero and @xmath156 would have been larger . we conclude that tidal effects are crucial for a proper description of halo evolution @xcite . this is evident in the final expressions for @xmath138 and @xmath139 . after the lengthy derivations , these expressions are remarkably simple , suggesting that they may have greater validity than the case of small - amplitude gaussian fluctuations that we have analyzed here . in going from eqs . ( [ f1lin ] ) and ( [ g2lin ] ) to eqs . ( [ driftlin ] ) and ( [ difflin ] ) we have dropped the contributions made by @xmath157 and @xmath158 because they do not contribute at second order in @xmath17 . at this order , @xmath138 and @xmath139 are independent of the value of the smoothed density @xmath159 used for the constraint . however , the constraint enters in third order . carrying out the derivation to one higher order in @xmath17 requires using the following result for gaussian random variables : @xmath160 . applying this with @xmath161 , after some algebra we find @xmath162 although these results are correct for gaussian fluctuations to third order in the density perturbations , nonlinear evolution will generate non - vanishing irreducible three - point correlations ( i.e. , non - gaussianity ) that have been neglected here . for example , nonlinear evolution generates corrections to the zeldovich approximation so that velocity and gravity are no longer proportional to each other as we have assumed . for this reason , eqs . ( [ diffuns ] ) should not be regarded as being more accurate than eqs . ( [ driftlin ] ) and ( [ difflin ] ) . despite their limitations , eqs . ( [ diffuns ] ) are useful in showing the qualitative effect of applying an initial constraint . choosing the proto - halo to have @xmath163 means that typically @xmath164 in the proto - halo , which decreases @xmath138 and @xmath165 compared with eqs . ( [ driftlin ] ) and ( [ difflin ] ) . conversely , the diffusion coefficients are enhanced in void regions . note also that the constraint value appears explicitly in the initial conditions for @xmath40 in the linear regime through the term @xmath166 in eq . ( [ f1lin ] ) . ( [ driftlin ] ) and ( [ difflin ] ) imply that , in the quasilinear regime , the drift and diffusivity are independent of the velocity within the halo . ( although @xmath138 is equivalent to a covariance at fixed @xmath16 , it is independent of the particular value of @xmath16 . ) this follows because @xmath46 depends on @xmath167 only through @xmath133 and @xmath168 in eq . ( [ f2cexact ] ) . as noted above , at second order in perturbation theory the drift and diffusivity are also independent of the initial overdensity of the peak @xmath159 ( but they depend on the smoothing scale used to find extrema ) . thus , both the fokker - planck equation and the diffusion coefficients appear to be remarkably robust quantities . to make further use of eqs . ( [ driftlin ] ) and ( [ difflin ] ) , we need to write out the coefficients explicitly . using eqs . ( [ covfun ] ) , ( [ driftlin ] ) , and ( [ difflin ] ) we first factor out the dependence on cosmology by writing [ adscaled ] @xmath169 \,,\end{aligned}\ ] ] where @xmath170 and @xmath171 are scaled radial and tangential diffusion coefficients . the scaled variables are [ coeffs ] @xmath172\right\ } \ , \label{scala}\\ d_r(r , t ) & = & c_r(r ) - \frac{d\gamma(r)}{dr}= \sigma_\psi^2-\frac{d\gamma(r)}{dr}-\left[\frac{r\bar\eta(r ) } { \sigma_0}\right]^2-\left[\frac{\bar\xi(r)-2\bar\eta(r)}{\sigma_1 } \right]^2\ , \label{scaldr}\\ d_t(r , t ) & = & c_t(r ) - \frac{\gamma(r)}{r}= \sigma_\psi^2-\frac{\gamma(r)}{r}-\left[\frac{\bar\eta(r)}{\sigma_1 } \right]^2 \ . \label{scaldt}\end{aligned}\ ] ] here , @xmath173 is the comoving distance from the center of the halo . the time dependence arises only because the functions appearing in eqs . ( [ coeffs ] ) depend on the time - dependent power spectrum of density fluctuations . these functions are summarized here for convenience : [ sumfun ] @xmath174 where we recall that @xmath175 and @xmath176 are the covariances of the constraints @xmath159 and @xmath177 defined in eqs . ( [ varcon ] ) , and @xmath178 is a smoothing window defined in eqs . ( [ smoothdel ] ) and ( [ window ] ) . we use the gaussian window @xmath179 . to gain some insight into the significance of these results , let us consider a proto - halo at high redshift . if we suppose that the gravity field is dominated by its small - scale components , we may replace @xmath180 by @xmath35 . if we further ignore the effects of initial constraints , in linear theory @xmath181 where @xmath182 is the one - dimensional velocity dispersion . in one realization , the forces are deterministic , but taken across the ensemble they are random and fluctuating . a diffusivity @xmath183 implies that these fluctuations cause the typical particle to change its velocity by order itself in about a hubble time . there is also a net drift force @xmath138 with a dissipative timescale of about @xmath184 dynamical times , where @xmath17 is a typical magnitude of the density fluctuations . as a result , friction and diffusion will significantly alter halos within a hubble time of their collapse at high redshift . we conclude that it is not valid to model halos using the spherically symmetric vlasov equation . this conclusion has also been reached previously by @xcite . to summarize , eqs . ( [ fplin])([difflin ] ) , and their more detailed version , eqs . ( [ adscaled])([sumfun ] ) , are the main results of this subsection . ( [ fplin ] ) has the standard form of a fokker - planck equation . the fokker - planck form arises naturally and eqs . ( [ driftlin])([difflin ] ) are exact for gaussian fluctuations up to second order in perturbation theory . the diffusion coefficients @xmath138 and @xmath139 are due to fluctuations in the gravitational field over the ensemble of halos . the drift vector @xmath140 represents a source of gravitational acceleration in addition to the acceleration @xmath180 produced by the mean matter distribution . the diffusion term @xmath185 causes the temperature ( i.e. velocity dispersion ) to increase ( for a positive definite @xmath139 ) . having derived analytic expressions for the diffusion coefficients in the previous subsection , we now evaluate them for the currently favored cosmological model at high redshift . figures [ fig : friction ] and [ fig : diffusion ] show @xmath186 , @xmath170 , and @xmath171 as a function of radius @xmath173 ( measured from the halo center ) for a range of gaussian smoothing lengths @xmath187 in @xmath178 . we obtain these results by numerically integrating eqs . ( [ coeffs ] ) and ( [ sumfun ] ) with the linear power spectrum @xmath19 for a flat @xmath20cdm cosmological model with @xmath188 . the figures show a number of interesting features . first , the diffusion coefficients are negative over some regions of radius for some parameters . as we will show below , @xmath138 is the gravitational acceleration on particles in the halo caused by drift . if the friction were produced by the chandrasekhar mechanism of tidal wakes , one would expect @xmath189 ( see [ sec : globular ] ) . however , here @xmath138 is independent of @xmath16 and the particle mass . the source of friction is not the tidal wake of a flow past a moving massive particle . instead it is the fluctuations of the gravitational field of a gaussian random process , and these fluctuations can lead to a radially outward or inward force . we discuss this further in [ sec : discuss ] . more surprising at first is the fact that the diffusivities @xmath170 and @xmath171 can be negative . negative diffusivity causes the velocity dispersion ( or temperature ) to decrease , indicating an instability . such behavior is possible for gravitational systems . in fact , this result is to be expected in the quasilinear regime of gravitational instability because second - order perturbations enhance gravitational collapse of dark matter halos @xcite . in the strongly nonlinear regime , after halo virialization , we expect the diffusivities to become positive . to elucidate the origin of the negative diffusivity , we show in figures [ fig : fig3 ] and [ fig : fig4 ] that once the extremum constraint @xmath190 is removed ( by setting @xmath191 ) , we obtain @xmath192 and @xmath193 for all radii . the extremum constraint is therefore crucial to generating positive @xmath186 and negative diffusivity seen in figures [ fig : friction ] and [ fig : diffusion ] . this effect is easy to understand : if one stands at a random place in a gaussian field , nonlinear effects can speed up or slow down gravitational collapse . gravitational instability , however , speeds up the evolution in the vicinity of a density extremum . another obvious systematic effect with the diffusion coefficients in figures [ fig : friction ] and [ fig : diffusion ] is their dependence on the smoothing radius @xmath187 used to set the initial constraints . for the power spectrum assumed here , the maximum of @xmath186 occurs at about @xmath194 . in the limit @xmath195 , the variances @xmath175 and @xmath176 diverge , causing the constraint terms in the diffusion coefficients ( i.e. , those terms proportional to @xmath196 and @xmath197 ) to vanish . figure [ fig : diffusion ] shows that the diffusivity is isotropic at the halo center but shows radial anisotropy outside the center . at small radii the tangential diffusivity is larger in absolute value than the radial diffusivity . tangential diffusion gives the infalling matter angular momentum . our treatment therefore provides a new statistical description of the growth of angular momentum by tidal torques @xcite and may enable in principle an explanation for the universal angular momentum profile found recently by @xcite . it is instructive to work out analytically the asymptotic behavior of the functions in eqs . ( [ coeffs ] ) for small and large @xmath173 . for small @xmath173 we find , [ difcosmallr ] @xmath198 \right\ } + o(r^3 ) \\ d_r(r , t ) & = & -\left[\frac{\bar\eta(0)}{\sigma_1}\right]^2 + r^2 \left [ \frac{1}{10}\sigma_\delta^2 + \frac{3}{5}\alpha \bar\eta(0 ) -\left(\frac{\bar\eta(0)}{\sigma_0 } \right)^2 \right ] + o(r^4 ) \ , \\ d_t(r , t ) & = & -\left[\frac{\bar\eta(0)}{\sigma_1}\right]^2 + r^2 \left [ \frac{1}{30}\sigma_\delta^2 + \frac{1}{5}\alpha \bar\eta(0 ) \right ] + o(r^4 ) \ , \end{aligned}\ ] ] where we have defined @xmath199 eqs . ( [ difcosmallr ] ) show that if @xmath200 is finite , then as @xmath201 , @xmath186 drops to zero while @xmath170 and @xmath171 approach a negative constant as discussed above , @xmath202 . this central value is decreased as the smoothing scale is decreased ( i.e. higher @xmath203 and @xmath204 ) as shown in figure [ fig : diffusion ] . as @xmath195 , we have @xmath205 and @xmath206 as long as @xmath207 as @xmath208 . in the limit @xmath206 ( no extremum constraint ) , @xmath170 and @xmath171 are non - negative everywhere ( cf . figs . 2 and 4 ) . thus , negative diffusivity ( instability ) is produced by constraining the slope of the density field . constraints have the opposite effect on @xmath186 : as the comparison of figures 1 and 3 confirms , the contributions to @xmath186 from constraint terms are positive . for large @xmath173 , both @xmath209 and @xmath210 approach @xmath211 , and @xmath212 ( assuming @xmath213 as @xmath214 ) , so @xmath215 figure [ fig : diffusion ] shows that @xmath170 rises slightly above @xmath216 at a few hundred mpc ( for the standard model power spectrum ) before reaching @xmath216 . this is because @xmath217 and @xmath218 is negative for this range of @xmath173 . the drift @xmath219 approaches zero at large @xmath173 . in practice , the large-@xmath173 limit is unimportant because halo infall occurs only from within about 10 mpc . another way to gain insight into the diffusion coefficients is to examine their behavior for scale - free spectra @xmath4 . many of the integrals are analytic in this case . we define a set of functions @xmath220 the integral converges for @xmath221 . using properties of the spherical bessel functions @xmath222 , we obtain the following useful relations : [ glrel ] @xmath223 we also use the following relation , valid for @xmath221 : @xmath224^{-1}\ .\ ] ] for gaussian - tapered power - law spectrum @xmath225 and gaussian smoothing @xmath226 , eqs . ( [ sumfun ] ) give , for @xmath227 and @xmath228 , [ sumfunp ] @xmath229 here , @xmath230 is euler s constant . for a pure power - law spectrum with @xmath231 , @xmath232 and @xmath211 diverge at long wavelength . however , eqs . ( [ dr00 ] ) and ( [ dt00 ] ) are valid for @xmath228 . for @xmath231 , @xmath233 in eq . ( [ scala ] ) . for @xmath234 , @xmath211 converges at long wavelength but diverges at short wavelength with @xmath235 as @xmath227 . the limiting behavior of @xmath236 and @xmath237 are given by [ limbarfun ] @xmath238\ , & r\ll r\ , \\ -ng_1(n+2)\,r^{-(n+3)}\ , & r\gg r\ , \end{cases } \label{barxiap}\\ \bar\eta(r ) & \approx & b\begin{cases } \frac{1}{6}\gamma\left(\frac{n+3}{2}\right)\left(\frac{r } { \sqrt{2}}\right)^{-(n+3)}\left[1-\frac{(n+3)}{10}\frac{r^2 } { r^2}+o(r^4)\right]\ , & r\ll r\ , \\ g_1(n+2)\,r^{-(n+3)}\ , & r\gg r\ . \end{cases } \label{baretaap}\end{aligned}\ ] ] for all @xmath239 , these functions obey the relation @xmath240 . the drift coefficient is plotted in figure [ fig : a - power ] for @xmath241 . as expected from the previous section , constraints make @xmath186 more positive . for power - law spectra , however , @xmath200 diverges so one can not use eqs . ( [ difcosmallr ] ) to obtain the behavior as @xmath242 . the exact results here show that for @xmath243 , @xmath244 as @xmath242 . for @xmath245 , @xmath246 as @xmath247 . these results show that models with lots of small - scale power and substructure ( @xmath248 as @xmath208 ) have a strong inward drift force , while models that are smoother on small scales ( @xmath245 ) have negligible drift force as @xmath242 . this is easy to understand qualitatively in the context of halo formation . density fields with lots of substructure have a nonspherical mass distribution , with mass concentrated into subhalos . for a given halo , the mass between @xmath249 and @xmath173 , for example , is concentrated into overdense substructures whose gravity field on infalling particles is stronger , on average , than if this mass were uniformly distributed over a spherical shell as it is for the average halo . given a density correlation function that rises with decreasing @xmath173 , there is a greater concentration of such clumps interior to @xmath173 than outside of @xmath173 , leading to a net inward drift force . figure [ fig : d - power ] shows the radial and tangential diffusivity for power - law spectra with @xmath231 . for @xmath234 the diffusivity diverges because the velocity dispersion @xmath211 diverges as @xmath208 . for @xmath250 the divergence is logarithmic resulting in a logarithmic dependence of @xmath170 and @xmath171 on radius . figure [ fig : d - power ] shows that the effect of the constraints is to increase the diffusivity at radii within a few smoothing lengths of the peak . as we noted above , constraints on the density gradient make the diffusivities negative as @xmath242 . for @xmath231 the diffusivities continue to rise with increasing @xmath173 because @xmath211 diverges as @xmath214 . for the physical power spectrum used in figures [ fig : friction ] and [ fig : diffusion ] , @xmath211 converges at both small and large @xmath251 , so the asymptotic behavior is different . the use of the fokker - planck equation to describe dark matter clustering is novel . in order to gain insight into what this approach may reveal , in this section we examine the fokker - planck equation from several different perspectives . the fokker - planck equation describes relaxation processes in a weakly collisional fluid . in this regard it is similar to the boltzmann equation , in which the right - hand side of eq . ( [ boltzeq ] ) is replaced by a two - body collision integral that is bilinear in the one - particle distribution function . the present situation is physically very different however `` collisions '' are not due to individual particles but rather to the gravity field of density fluctuations in a gaussian random field . nevertheless , we can borrow one of the techniques used to gain insight into the boltzmann equation , viz . velocity moments , to obtain a clear physical interpretation of the meaning of the fokker - planck equation and the diffusion coefficients @xmath138 and @xmath139 . moment equations are obtained by multiplying eq . ( [ fplin ] ) by powers of @xmath252 and integrating over velocity . the three lowest moments give the mass density @xmath15 , fluid velocity @xmath253 , and velocity dispersion tensor @xmath254 : [ momdef ] @xmath255 here we are using physical ( proper ) spatial coordinates rather than comoving coordinates . the density , velocity , and velocity dispersion tensor obey the following equations : [ momeqs ] @xmath256 the first equation is the usual fluid continuity equation . the second is the usual euler equation , except that the pressure is generalized to a matrix @xmath257 because the spatial stress need not be diagonal for a collisionless fluid . the right - hand side has two force terms : the gravitational tidal field @xmath180 produced by @xmath258 , and the drift acceleration @xmath259 caused by the correlated fluctuations of the mass density field . the third equation introduces a tensor @xmath260 , which is defined exactly like @xmath254 except the integrand in eq . ( [ thetadef ] ) acquires an additional factor @xmath261 . ( [ heat ] ) is a generalized heat equation , which requires a little more discussion . the velocity dispersion tensor @xmath254 generalizes the temperature of an isotropic gas to the case of an anisotropic velocity distribution . for an isotropic gas with @xmath262 ( here @xmath263 has units of @xmath264 ) and @xmath265 , this equation takes the more familiar form @xmath266 where @xmath267 is the heat flux . the velocity gradient terms in eq . ( [ heat ] ) generalize the @xmath268 term ( i.e. , the @xmath269 term ) of eq . ( [ isoheat ] ) to an anisotropic velocity distribution . for a perfect gas , the heat flux term is absent because the maxwellian velocity distribution has vanishing skewness . for an imperfect or collisionless gas , however , the third moment of the velocity distribution generates a generalized heat flux @xmath270 . notice that the heating rate is given not by the heat flux but instead by its divergence . for a weakly imperfect gas , the heat flux can be approximated ( e.g. in the chapman - enskog expansion approach ) by a conductive flux @xmath271 where @xmath272 is the conductivity , but for a collisionless gas , @xmath260 can not be obtained simply from @xmath15 , @xmath253 , and @xmath254 . thus , eqs . ( [ momeqs ] ) do not form a closed system . this is precisely why we must use a phase space description for a collisionless gas instead of trying to modify the fluid equations . ( [ momeqs ] ) and ( [ isoheat ] ) , however , are pedagogically useful in showing us how to interpret @xmath138 and @xmath139 . the velocity diffusivity tensor @xmath273 appears as a new term in the anisotropic heat equation , with no familiar counterpart in the dynamics of a collisional gas . although @xmath273 comes from a _ diffusion _ ( i.e. , gradient ) term in _ phase space _ , it appears as a direct _ local heating _ term in _ real space _ ( i.e. , coordinate space @xmath274 ) . diffusion in velocity space is equivalent to local heating : the temperature increases when particles spread out in velocity at a given point in space . the diffusion occurs in velocity space not in real space . the presence of this term begs the question : where does the energy come from ? from the exact kinetic equation ( [ boltzeq ] ) , it follows that the heating term is given by the symmetrized first velocity moment of the correlated force density @xmath58 produced by fluctuating gravitational fields . from this we see that the heating comes from gravitational energy . for this reason , it is possible for @xmath275 to be positive or negative : gravitational fluctuations can either locally heat or cool a gas . one should not try to draw conclusions about energy conservation from these results because our calculations are performed in the accelerating frame centered on a halo . however , the equations of motion used in deriving our kinetic theory imply local energy conservation . the fokker - planck equation describes the evolution of the probability distribution for particles undergoing random walks . instead of describing the system by a distribution function , we can specify equations of motion for individual particle trajectories using the langevin equation @xcite . although this is reversing the logic we used in deriving the fokker - planck equation , it is useful for providing both an understanding and a monte carlo method for numerically solving the fokker - planck equation . the langevin equation is a stochastic differential equation for the individual particle trajectories . the langevin equation corresponding to eq . ( [ fplin ] ) is given by the pair of equations @xmath276 where @xmath277 is a force due to a zero - mean gaussian random process with covariance @xmath278 the velocity of a particle at time @xmath279 , @xmath280 , is therefore given by a deterministic term @xmath281 dt$ ] , plus a random term drawn from a gaussian distribution with variance @xmath282 . subtle differences exist between the cases where the variance @xmath275 is evaluated at @xmath283 ( `` ito calculus '' ) vs. at @xmath284 ( `` stratonovich calculus '' ) . details can be found in @xcite . note that the langevin equation requires @xmath285 . the interpretation of the langevin equation is given by the following result @xcite : if a sufficiently large sample of particles is drawn at random at time @xmath286 from the phase space distribution @xmath287 , and their orbits are integrated using eqs . ( [ langevin ] ) , then at any later time @xmath66 their positions and velocities are a sample from the phase space density @xmath14 obtained by integrating eq . ( [ fplin ] ) . ( [ langevin ] ) are similar to the exact equations of motion for individual particles given by eqs . ( [ newton ] ) , but there are several important differences . the gravity field @xmath288 of one realization is replaced by the spherically symmetric gravity field of the ensemble - average density profile . the effects of spatially - varying forces in @xmath289 are represented by the drift term @xmath138 and the stochastic acceleration @xmath277 . these additional terms have the effect that , at least in the quasilinear regime ( which we assumed in our derivation of the fokker - planck equation ) , the phase space evolution implied by the langevin dynamics is equivalent to a monte carlo sample from the phase space density . this does not mean we have come full circle . for individual realizations of the ensemble , the density field is clustered and not spherically symmetric and the phase space is six - dimensional . the langevin equations give orbits in a smooth , spherically symmetric matter distribution . the latter is far easier to model numerically , with much higher resolution possible , than the original dynamics . by taking advantage of the symmetries of the average halo we have reduced the phase space from six dimensions to three . the fokker - planck equation was first written almost a century ago to describe brownian motion @xcite . in this case , a small macroscopic particle ( a brownian particle ) undergoes a random walk due to its collisions with individual molecules in a liquid . instead of the phase space density of molecules , the relevant quantity is the probability density @xmath290 for the brownian particle to have velocity @xmath16 at time @xmath66 . assuming a spatially homogeneous and isotropic medium , this probability distribution obeys the fokker - planck equation @xmath291 where @xmath292 is a constant ( given , for a small spherical body of mass @xmath23 and radius @xmath293 immersed in a fluid of viscosity @xmath294 , by the stokes formula @xmath295 ) , @xmath263 is the temperature , and @xmath296 is the boltzmann constant . ( [ fpbrown ] ) has the same form as eq . ( [ fplin ] ) if we identify @xmath297 for the drag term and @xmath298 for the diffusion term . we note that the brownian motion can be equally described by the langevin equations in eqs . ( [ langevin ] ) and ( [ stofor ] ) with @xmath299 , @xmath297 , and @xmath300 . brownian motion describes a very different physical situation than dark matter halo formation . why are both described by a fokker - planck equation ? the reason is that both problems involve random walks . let us examine this more closely . in the case of brownian motion , the random walks arise from impulsive collisions that cause the brownian particle s velocity to change significantly on a timescale @xmath301 . during this time interval the particle moves a distance @xmath302 . after an elapsed time @xmath66 , the particle undergoes @xmath303 steps and therefore moves a distance @xmath304 . in thermal equilibrium , @xmath305 so the particle moves a typical distance @xmath306 where @xmath307 is the velocity diffusivity . a random walk in space leads naturally to a diffusion equation for the probability distribution @xcite . the dark matter case at first appears to be very different . instead of a large particle being buffeted by small ones , we are studying small particles being buffeted by large mass fluctuations ( i.e. large in mass compared with the dark matter particles themselves , whose mass never enters our discussion ) . moreover , our derivation of the fokker - planck equation has assumed that the motion is in the quasilinear regime described by eq . ( [ pertdv ] ) with a universal time - dependence for @xmath308 . therefore each particle has moved in a straight line with ( in appropriate units ) a constant velocity , which is not a random walk in space . however , _ the velocity _ of a given particle is a random walk since @xmath18 is obtained by adding up the gravitational accelerations produced by all the mass fluctuations in a gaussian random field . the net force on a particle is a random walk at fixed time . although cosmic gaussian random fields lead to a fokker - planck equation , the diffusion coefficients are very different from brownian motion . first , the drift acceleration @xmath138 is independent of velocity , in contrast with the viscous drag acceleration @xmath309 for brownian motion . in the quasilinear regime of gravitational clustering , the drift is not a friction at all . instead it is radially directed ( because of spherical symmetry for the average halo ) , either inward or outward depending on the density correlations of fluctuating substructure . second , the velocity diffusivity tensor @xmath273 is not proportional to the temperature ; the dark matter is treated as being completely cold with vanishing temperature . in both cases the diffusivity is proportional to the mean squared displacement . in the cosmological case this depends simply on the power spectrum of density fluctuations and not on the thermal velocity of the particles . in the brownian case @xmath138 and @xmath139 are both proportional to the stokes drag coefficient @xmath292 . the proportionality between @xmath138 and @xmath139 is no accident but is a consequence of the fluctuation - dissipation theorem , which states that these coefficients have a proportionality given by the condition of thermal equilibrium . it is instructive to see this in a generalized case of eq . ( [ fpbrown ] ) . suppose that the brownian particle is moving in a time - independent potential @xmath310 . the fokker - planck equation then becomes @xmath311 the stationary solution ( for suitable boundary conditions ) is given by the boltzmann distribution @xmath312\ , \ \ e=\frac{1}{2}mv^2+m\phi(\vec r\,)\ .\ ] ] where @xmath25 is a normalization constant . with this solution not only does the left - hand side of eq . ( [ fpb ] ) vanish , but so does the velocity flux on the right - hand side . drag and diffusion balance each other in thermal equilibrium . note that the mean squared velocity at fixed @xmath313 equals @xmath314 . for a distribution of masses of brownian particles , thermal equilibrium results in an equipartition of energy with the heaviest particles moving most slowly . this is a natural consequence of having @xmath315 equal the thermal speed . in thermodynamic systems , the fokker - planck equation governs the relaxation to thermal equilibrium . in the cosmological case of quasilinear clustering considered here , it is not thermal equilibrium but rather the statistics of gaussian random fields that relate @xmath138 and @xmath139 . in appendix [ sec : stationary ] we investigate our cosmological fokker - planck equation for the relaxation to stationary solutions . globular cluster evolution is the most studied application of the fokker - planck equation in astrophysics . unlike the brownian case , gravity plays the dominant role here . a star in a globular cluster can be approximated as experiencing two types of gravitational forces : a smoothly varying potential @xmath316 due to the smoothed matter distribution in the system , and a fluctuating force due to many two - body interactions with other stars . the phase space density of the stars obeys the fokker - planck equation @xcite @xmath317 + \frac{1}{2 } \frac{\partial^2}{\partial v_i \partial v_j } \left [ d_{ij}(\vec v\,)f \right ] \,.\ ] ] this looks very similar to the cosmological kinetic equation ( [ fplin ] ) in the quasilinear regime , aside from a factor of two in the diffusivity that is purely a matter of differing conventions , and the velocity dependence of the diffusion coefficients . however , the physics of @xmath138 and @xmath273 is very different here . the relaxation process in globular cluster dynamics is _ two - body relaxation _ @xcite , i.e. the dissipation arising from the fluctuating forces of discrete newtonian gravitating point masses . examining this case will shed light on the relaxation processes that can affect galaxy halo evolution . the calculation of diffusion coefficients for two - body relaxation is described in 8.3 of @xcite . to understand the results it is helpful to consider the process of chandrasekhar dynamical friction acting on a test particle with mass @xmath318 and velocity @xmath319 moving through a spatially homogeneous sea of background particles of mass @xmath320 and velocity distribution @xmath321 with mass density @xmath322 . for an isotropic maxwellian distribution @xmath323 , the test body feels an acceleration @xmath324 \vec v_t\ , \ ] ] where @xmath325 . now treat each star within the globular cluster one at a time as a test particle , with all the other stars serving as background particles . if the velocity distribution is maxwellian and spatial gradients are ignored , dynamical friction will contribute @xmath326 given by eq . ( [ dynfric ] ) with @xmath319 replaced by @xmath16 . more generally , for a spatially homogeneous distribution on scales larger than the interparticle spacing , two - body relaxation gives @xcite [ ad ] @xmath327 where @xmath328 is the coulomb logarithm , and the rosenbluth potentials @xmath329 and @xmath330 are given by @xcite [ rosenbluth ] @xmath331 for the isotropic maxwellian distribution @xmath332 , @xmath333 is a drag force and the needed coefficients are [ ad2 ] @xmath334 \ , \\ d_\parallel & = & 8\pi g^2 \rho_b m_b \frac{\sigma_b^2}{v^3 } \ln \lambda \left [ { \rm erf}(x ) - \frac{2x}{\sqrt{\pi } } e^{-x^2 } \right ] \,,\\ d_\perp & = & 8\pi g^2 \rho_b m_b \frac{1}{v } \ln \lambda \left [ \left ( 1- \frac{1}{2 x^2 } \right ) { \rm erf}(x ) + \frac{1}{\sqrt{\pi}x } e^{-x^2 } \right ] \ , , % d_\parallel&=&\frac{2m_b}{m_t+m_b}\,\sigma_b^2\gamma\ , \\ % d_\perp&=&\frac{8\pi g^2\rho_bm_b\ln\lambda}{v}{\rm erf}(x ) % -d_\parallel\ , \end{aligned}\ ] ] where @xmath335 is defined as before ( with @xmath336 ) , and @xmath273 is decomposed into two components : @xmath337 we note that the drift @xmath138 contains two terms : the first term is proportional to the test particle mass @xmath318 while the second is independent of @xmath318 . the first term represents a `` polarization cloud '' effect , which arises from the fact that a test particle moving relative to a background deflects more upstream than downstream background particles @xcite . this process results in a density gradient and hence a drag force on the test particle , whose amplitude increases linearly with @xmath318 since the gravitational deflection is proportional to @xmath318 . in contrast , the second term in @xmath138 and the diffusivity tensor @xmath139 in eq . ( [ ad2 ] ) are both @xmath338 and independent of @xmath318 . these two terms are due to fluctuations in the gravitational potential caused by granularity in the background particle distribution . the derivation of the fokker - planck diffusion coefficients for two - body relaxation is completely different from our derivation based on quasilinear cosmological density fluctuations . in both cases , fluctuating forces lead to random walks in velocity space , hence to the fokker - planck equation . the nature of the fluctuating forces and their consequences , however , are very different in the two cases . our cosmological fokker - planck derivation is exact ( in second - order cosmological perturbation theory ) ; it yields no drag ; and the resulting drift and diffusion coefficients depend on position but are independent of velocity , as shown in eqs . ( [ adscaled])([sumfun ] ) . in contrast , the two - body relaxation calculation is based a taylor series expansion of the phenomenological ( not exact ) master equation ; it yields no radial drift ; and the drag and diffusion coefficients depend on velocity but are ( in the usual derivation ) independent of position @xcite . furthermore , the two - body relaxation rates are proportional to @xmath339 while the rates we compute are proportional to @xmath340 from eq . ( [ adscaled ] ) . in the cosmology case , the phase space correlations are built into the initial power spectrum of density fluctuations , whose amplitude is given independently of @xmath340 . with two - body relaxation , however , correlations are gravitationally induced by fluctuations instead of being present ab initio , hence the rates pick up another factor of @xmath340 . the differences in the @xmath138 term ( radial drift versus drag ) have an important consequence for equipartition of energy . as shown previously in the discussion of brownian motion , drag and diffusion work together to drive a system to thermal equilibrium in which the mean squared velocity is proportional to @xmath341 . binney and tremaine show ( p. 513 of binney & tremaine 1988 ) show that this also happens with two - body relaxation . because equilibrium systems with only gravitational forces generally have negative specific heat , there is no stable thermal equilibrium . nonetheless , two - body relaxation tends to drive the velocity distribution toward the maxwellian form , with equipartition of different mass species . in the quasilinear cosmological case , the drift and diffusion coefficients are independent of both velocity and particle mass . therefore the equilibrium velocity distribution can not depend on particle mass there is no equipartition of energy . this result is similar to violent relaxation , a process arising during the initial collapse and virialization of a dark matter halo when the gravitational potential is rapidly varying in time @xcite . our relaxation process , however , arises in the quasilinear regime and is present even if @xmath342 as it is ( at fixed comoving position ) if @xmath343 . moreover , violent relaxation ( and its partner , phase mixing ) are generally understood as arising from the collisionless dynamics represented by the left - hand side of eq . ( [ fp_globular ] ) , while both two - body relaxation and our cosmological relaxation process are represented by the `` collision '' terms on the right - hand side . why did our cosmological calculation not yield a drag term ( and a corresponding diffusion term linked to it by the fluctuation - dissipation theorem ) ? we suspect that it is because our calculation was limited to small - amplitude perturbations about a homogeneous and isotropic expanding cosmological model . the induced effects of substructure would only come in at higher order in perturbation theory . in the nonlinear regime we expect two types of drift terms to be present : @xmath344 , where @xmath3 is the radial drift and @xmath292 is the drag coefficient . the chandrasekhar calculation suggests that the drag and its accompanying diffusivity will depend on both position ( through @xmath345 ) and velocity . the radial drift is absent in the chandrasekhar calculation because the background there is assumed to be spatially homogeneous ; with an inhomogeneous distribution in a virialized halo there should be a radial drift ( and corresponding diffusivity ) that could depend on both position and velocity . in the nonlinear regime drag should arise from dynamical friction just as with the globular clusters , except that the `` background '' particles whose discreteness causes the @xmath320 terms in eqs . ( [ ad2 ] ) are now the subhalos and substructures that rain upon a halo . because we have not yet extended our derivation to the fully nonlinear regime , these expectations , while plausible , have yet to be demonstrated . in this paper we have developed a cosmological kinetic theory , valid to second - order in perturbation theory , to describe the evolution of the phase - space distribution of dark matter particles in galaxy halos . this theory introduces a new way to model the early phases of galaxy halo formation , which has traditionally been studied by analytic infall models or numerical n - body methods . the key physical ingredients behind our kinetic description are stochastic fluctuations and dissipation caused by substructures arising from a spectrum of cosmological density perturbations . the kinetic equation that we have obtained at the end of our derivation , eqs . ( [ fplin])([difflin ] ) , has the standard form of a fokker - planck equation . the diffusion coefficients @xmath346 and * d * represent acceleration due to a drift force and velocity - space diffusion , respectively . to second order in perturbation theory , these coefficients are related to various covariance matrices of the cosmological density and velocity fields given by eqs . ( [ driftlin])([difflin ] ) . these expressions can in turn be written in terms of the familiar linear power spectrum @xmath19 of matter fluctuations , as shown in eqs . ( [ coeffs ] ) and ( [ sumfun ] ) . the results for @xmath346 and * d * from second - order perturbation theory are shown in figures 14 for the currently favored @xmath20cdm model , and in figures 5 and 6 for various scale - free models with power - law @xmath4 . our results indicate that dissipative processes are important during the initial collapse of a dark matter halo , and it is not valid to model halos using the spherically symmetric vlasov ( collisionless boltzmann ) equation . furthermore , we find that the diffusivity is initially negative close to the center of a halo , indicating a thermodynamic instability . the source of this instability is gravity : perturbations enhance gravitational collapse of dark matter halos . we emphasize that our derivation leading to the fokker - planck equation given by eqs . ( [ fplin])([difflin ] ) is exact to second order in cosmological perturbation theory . we do not follow the frequently used approach ( e.g. binney & tremaine 1988 ) that begins by expanding the collisional terms of the non - exact master equation in an infinite series of changes in phase - space coordinates ( i.e. @xmath347 ) , and arrives at the fokker - planck equation by assuming weak encounters ( i.e. small @xmath348 ) and truncating the series after the second - order terms . instead , our starting point is eq . ( [ boltzeq ] ) , the first bbgky hierarchy equation for a smooth one - particle distribution function . this is an exact kinetic equation ; solving it requires specifying the one - particle distribution at an initial time and the two - particle correlation function at all times . we have shown in [ sec : dfgauss ] and appendix [ sec : pgauss ] that if the cosmological density field @xmath17 and displacement ( or velocity ) field @xmath349 are gaussian random fields , the resulting two - particle correlation function depends on the one - particle distribution in a simple way given by eq . ( [ f2cexact ] ) . ( [ f2cexact ] ) is exact to second order in perturbation theory and provides the closure relation for the bbgky hierarchy in eq . ( [ boltzeq ] ) . combining eq . ( [ f2cexact ] ) with eq . ( [ boltzeq ] ) results in eq . ( [ fplin ] ) , which is a fokker - planck equation . the cosmological dissipative processes identified in this paper are quite different from the standard two - body relaxation , which also leads to a fokker - planck equation . we find that there is no dynamical friction ( i.e. terms proportional to @xmath350 in the diffusion coefficients ) in second - order cosmological perturbation theory . instead , there is a radial drift toward ( or away from ) the center of a halo due to the clustering of matter within the halo . the usual treatment of two - body relaxation has no such drift term because it assumes the medium to be homogeneous . thus we have identified a new relaxation process that must affect the phase - space structure of dark matter halos . although our derivation is exact to second order in the density fluctuations , it is valid only when the fluctuations are small . we are therefore describing only the early stages of dark matter halo formation . the expressions we obtain for the diffusion coefficients are plainly invalid in the nonlinear regime , where we expect dynamical friction to be present as well as radial drift . in a later paper we will investigate the nonlinear generalizations of the exact second - order results derived in this paper . the fokker - planck description should still apply in the nonlinear regime provided that the fluctuating force on a particle can be modelled as a markov process consisting of a random walk of many steps . we conjecture that this description will be approximately valid when the matter distribution is modelled as a set of clumps ( i.e. , the halo model ) which scatter individual dark matter particles away from the orbits they would have in a smooth , spherical potential . the fokker - planck description should apply not only to the initial collapse and virialization of a dark matter halo containing substructure , but also to the subsequent evolution under minor mergers . our inability to solve exactly the nonlinear dynamics , however , means that the diffusion coefficients will have to be calibrated using the nonlinear halo model or numerical n - body simulations . calibrating the diffusion coefficients is not equivalent to reproducing the n - body simulations . n - body simulations provide a monte carlo solution of the bbgky hierarchy , which is much more general than the fokker - planck equation . if the simulation results are describable by fokker - planck evolution , this already represents a significant new result . moreover , even if functional forms for the radial dependence of the diffusion coefficients have to be calibrated with simulations , the solution of the fokker - planck equation contains far more information because it gives the complete phase - space density distribution , not simply radial profiles . another important step will therefore be to actually solve the fokker - planck equation . in deriving it we have identified and quantified the relaxation processes affecting halo formation and evolution , but this is only the first step . in appendix [ sec : stationary ] we presented solutions in unrealistically simple cases just to highlight the roles played by radial drift , drag , and diffusion . an outstanding question is whether these processes actually are rapid enough to erase the memory of cosmological initial conditions sufficiently so that dark matter halos relax to a universal profile . once we have a nonlinear model for the diffusion coefficients , we will address this important question . the work presented in this paper has a number of implications . first , we have highlighted the importance of the phase space density for understanding dark matter dynamics , and have developed a method for calculating its evolution during the early phases of structure formation . second , we have identified a dissipative process that may erase memories of initial conditions during the early phases of galaxy halo evolution . this erasure is necessary for universal halo density profiles , although further investigation is needed to quantify how this dissipative process affects the density profiles . third , the statistical description of the potential fluctuations caused by substructure that we have developed can also be added to investigate related problems such as the heating of galactic disks @xcite and the merging of central black holes when galaxy halos merge @xcite . we thank jon arons , avishai dekel , and martin weinberg for useful conversations . the aspen center for physics provided a stimulating environment where a portion of the research was carried out . m is partially supported by an alfred p. sloan fellowship , a cottrell scholars award from the research corporation , and nasa grant nag5 - 12173 . in this appendix we derive expressions for the probability distributions of displacement and density perturbation , @xmath133 and @xmath351 , that are needed in eqs . ( [ f1p1 ] ) and ( [ f2cp1 ] ) . we then use them to evaluate eqs . ( [ f1p1 ] ) and ( [ f2cp1 ] ) at high redshift when the matter distribution was described by a gaussian random field of small - amplitude density perturbations . we use comoving spatial coordinates @xmath352 . in accordance with the standard cosmological model , we assume that @xmath353 is a gaussian random field with power spectrum @xmath0 , and that the displacement and density are related by eq . ( [ linpsi ] ) . these are the appropriate assumptions at high redshift in standard models of structure formation . strictly speaking , at high redshift , when linear theory applies , the evolution of the distribution function is trivial because the density and velocity fields evolve by spatially - homogeneous amplification . here we use the statistics of gaussian random fields to compute quantities driving the lowest - order corrections to linear evolution . we will find that the fluctuating force is formally second order in perturbation theory . a gaussian random field is described most simply by its fourier transform , which we define by @xmath354 we suppress the time dependence for convenience . before we apply a constraint , both @xmath355 and @xmath356 have zero mean . the covariance of @xmath357 gives the power spectrum , @xmath358 where @xmath22 is the dirac delta function . in the linear regime , @xmath359 is a linear functional of @xmath355 . the multivariate distribution of any linear functional of a gaussian random field is itself gaussian ( i.e. multivariate normal ) and is therefore described completely by its mean and covariance matrix . this remains true even for a constrained gaussian random field @xcite . in our case , we wish to constrain the initial density field to represent a proto - halo . the constraints are applied to the smoothed density field @xmath360 where @xmath361 is a smoothing window with unit volume integral , @xmath362 . the subscript @xmath187 indicates the smoothing length . we assume that the smoothing window is spherically symmetric so that @xmath178 is real and depends only on the magnitude of the wavevector . throughout the paper we will use the gaussian window @xmath363 and @xmath226 . unfortunately there is no analytic theory to tell us where halos will form . however , there are a variety of plausible approaches . for example , following @xcite we might constrain @xmath364 to have a maximum of specified height , orientation , and shape at @xmath365 . this peak constraint involves ten variables ( @xmath364 and the components of its first and second derivatives with respect to each of the coordinates ) , presenting a formidable challenge to evaluating the covariance matrix of the four variables @xmath366 . the peak constraint is appealing but it is difficult to calculate and in practice it has not been found to predict well the actual formation sites of dark matter halos @xcite . instead of trying to impose complicated constraints for halo formation ( e.g. monaco et al . 2002 ) , we err on the side of simplicity by choosing to use only the zeroth and first derivatives of the smoothed initial density field @xmath367 . we require that @xmath368 be an _ extremum _ of the smoothed density field by requiring @xmath369 . of course , an extremum may be a maximum , a minimum , or a saddle point . we therefore keep track also of the smoothed density @xmath370 . provided that @xmath159 is greater than 2 standard deviations , the extrema of @xmath371 are close to the peaks @xcite . how will we know if our results depend strongly on the choice of constraint ? we will do this in two ways . first , we will vary the height of the extremum , @xmath159 . second , we will drop the gradient constraint and use the single constraint @xmath159 . while these tests will not prove that we have a good model for the sites of halo formation , they do provide control cases for us to assess the sensitivity of our results to the details of the initial constraints . now we must calculate the mean and covariances of @xmath366 subject to constraints @xmath372 where @xmath373 the method for this calculation is based on the theorem presented in appendix d of @xcite , which states that if @xmath374 and @xmath375 are zero - mean gaussian variables ( more generally , vectors of any length ) , then the conditional distribution for @xmath375 given @xmath374 , @xmath376 , is gaussian with mean @xmath377 and covariance matrix @xmath378 where @xmath379 . the symbol @xmath380 denotes a tensor product and angle brackets denote mean values . thus , @xmath381 is the covariance matrix of the constraints while @xmath382 is the covariance matrix of @xmath375 subject to the constraints . ( the symbol @xmath128 , used either as a function or as a subscript , implies that a constraint is applied . ) the juxtaposition of two matrices implies matrix multiplication . to apply these results , first we compute the covariance matrix of constraints @xmath383 . using eqs . ( [ fourier])([window ] ) , we get the covariances [ varcon ] @xmath384 where @xmath385 is the unit tensor . next we consider the covariance matrix of the variables @xmath386 . since @xmath387 , the displacement field @xmath18 has an additional factor @xmath388 in the integrand of eq . ( [ fourier ] ) . for the off - diagonal elements of the covariance matrix of @xmath375 we obtain [ covyb ] @xmath389 where the spherical bessel functions are @xmath390 and @xmath391 . the diagonal variances are [ covybb ] @xmath392 note that in eqs . ( [ covyb ] ) ( and in the following ) we are using @xmath313 as a comoving displacement vector and not as the radius vector in proper coordinates . proper coordinates are not used in this appendix . finally , we have the cross - covariances between our primary variables and the constraints , i.e. the components of @xmath393 . these follow from eqs . ( [ fourier])([smoothdel ] ) : [ covyab ] @xmath394 to arrive at the last expression above , we have used the identity @xmath395\hat r\otimes\hat r+\frac{j_1(kr)}{kr}\,{\bf i}\ , \ ] ] where @xmath396 , @xmath397 and @xmath385 is the unit tensor . using eqs . ( [ meany ] ) and ( [ varcon])([covyab ] ) , we get the mean values subject to the extremum constraints @xmath398 , [ meanext ] @xmath399 the subscript @xmath128 denotes `` subject to constraints . '' ( [ meanext ] ) would be unchanged had we dropped the extremum condition and imposed instead the single constraint @xmath400 ; the constraint @xmath401 has no effect on the mean values . this follows because @xmath402 so that the only way that @xmath177 can enter the mean values is through terms proportional to @xmath403 which vanish when the extremum constraint is imposed . we now compute the covariance matrix of our variables @xmath386 subject to the extremum constraints @xmath383 . the covariances follow from eqs . ( [ covy])([covyab ] ) . recalling the definition @xmath404 , we obtain [ covfun ] @xmath405\ , \label{c2i}\\ c(\delta_1,\vec\psi_1 ) & = & \frac{\bar\xi_1\bar\eta_1\vec x_1}{\sigma_0 ^ 2 } + \frac{1}{\sigma_1 ^ 2}\frac{d\bar\xi}{dr_1}(\bar\xi_1 - 2\bar\eta_1 ) \hat x_1\ , \label{c1i}\\ c(\vec\psi_1,\vec\psi_1 ) & = & c_r\hat r_1\otimes\hat r_1 + c_t ( { \bf i}- \hat r_1\otimes\hat r_1 ) \ , \nonumber\\ c_r(r_1 ) & \equiv & \sigma_\psi^2-\left(\frac{r_1\bar\eta_1 } { \sigma_0}\right)^2 - \left(\frac{\bar\xi_1 - 2\bar\eta_1}{\sigma_1 } \right)^2\ , \label{cpar}\\ c_t(r_1)&\equiv&\sigma_\psi^2-\left(\frac{\bar\eta_1}{\sigma_1 } \right)^2\ , \label{cper}\end{aligned}\ ] ] where @xmath406 and @xmath407 , and we have used the relation @xmath408 . note that @xmath409 is a vector and @xmath410 is a second - rank tensor . later we will need the tensor @xmath411 , which is the matrix inverse of @xmath410 . the role of constraints is easy to see in eqs . ( [ covfun ] ) . the constraint on @xmath159 is responsible for the terms proportional to @xmath196 while the constraint on @xmath177 gives rise to the terms proportional to @xmath197 . therefore , the constraints are easily removed by setting @xmath412 ( to eliminate the constraint on @xmath177 ) or @xmath196 ( to eliminate the constraint on @xmath159 ) . as expected , in the absence of any constraints , the covariance matrix recovers the same expressions as the unconstrained covariance in eqs . ( [ covyb ] ) and ( [ covybb ] ) . we will need one more covariance , @xmath413 subject to the extremum constraints where @xmath414 . using eqs . ( [ covy ] ) , ( [ covyab ] ) , and ( [ nntens ] ) , we get @xmath415 now we are ready to calculate the probability distribution functions and means of various quantities needed in [ sec : dfgauss ] . specifically , we need @xmath133 , @xmath416 , @xmath417 , @xmath418 , and @xmath419 that appear in the key eqs . ( [ f1p1 ] ) and ( [ f2cp1 ] ) for @xmath420 and @xmath421 . ( [ m2c ] ) already gives @xmath422 but we need to calculate the other quantities . first , the probability density distribution for the displacement @xmath133 ( subject to the extremum constraints ) is a three - dimensional gaussian whose mean and covariance are given above , implying @xmath423\ .\ ] ] note that the argument of the exponential is a scalar the dot denotes a contraction ( dot product ) . the joint distribution of displacement and density ( subject to the constraints ) is a little more complicated to work out since this requires inverting the @xmath424 covariance matrix of @xmath366 . the distribution factors into conditional and marginal distributions , @xmath425 . straightforward algebra gives the conditional distribution , @xmath426 ^ 2\right\}\ , \ ] ] where @xmath427 note that the second term in @xmath428 ( and the similar term in eq . [ f2cond ] ) is a scalar . writing it out with indices and using the summation convention , @xmath429 . ( [ f2cond ] ) and ( [ vardel2 ] ) could also have been derived by noting that @xmath430 for @xmath431 and @xmath432 . then using eq . ( [ meany ] ) and relations @xmath433 and @xmath434 , we obtain @xmath435 which is identical to the mean in eq . ( [ f2cond ] ) . the constrained mean @xmath436 follows simply by replacing @xmath422 with @xmath437 and @xmath438 with @xmath439 . the only remaining constrained mean that we need to evaluate is @xmath419 . direct evaluation of the conditional distribution would require inverting a @xmath440 matrix . it is much simpler to use eqs . ( [ meany ] ) and ( [ covy ] ) with @xmath441 and @xmath442 . from the off - diagonal element of @xmath443 , we get @xmath444 at last we are ready to evaluate the one- and two - particle distribution functions for small - amplitude gaussian fluctuations . substituting @xmath445 into eq . [ f1p1 ] ) , to second order in @xmath17 we obtain @xmath446 where @xmath166 is given by eq . ( [ m1c ] ) with @xmath447 and @xmath150 is a vector equal to the covariance of the constrained @xmath17 and @xmath18 given in eq . ( [ c1i ] ) . without an initial constraint we would have @xmath448 and @xmath449 . an initial constraint on the proto - halo shifts the mean density and causes the density and velocity to be correlated hence it changes the phase space density . note that eq . ( [ f1lin ] ) implies that in linear theory the spatial and velocity distributions decouple . as expected , the mean spatial density is @xmath450 . the velocity distribution at each point in the halo is simply the underlying gaussian of eq . ( [ fpsi1 ] ) with the mean shifted by @xmath150 . note that the value of the proto - halo smoothed density constraint @xmath159 affects the density profile , @xmath451 , but it does not affect the velocity distribution up to second order in perturbation theory . the imposition of the extremum constraint @xmath401 affects the phase space density only through the term in @xmath452 . setting @xmath412 everywhere removes the constraint on @xmath177 . the two - particle correlation is found by substituting eqs . ( [ d2psi1 ] ) and ( [ d1d2psi1 ] ) into eq . ( [ f2cp1 ] ) . the result is @xmath453 \cdot\frac{\partial}{\partial\vpsia } \log p(\vpsia ) \right\ } p(\vpsia)(hb)^{-3}\ .\end{aligned}\ ] ] this equation is the main result of this appendix . it is exact for gaussian random fields . stationary solutions of the fokker - planck equation are relevant for systems that relax on a timescale faster than external changes take place . the stationary solution for classical brownian motion was given in [ sec : brownian ] . stationary solutions may also exist for cosmological systems where the constant - temperature heat bath is replaced by small - scale structure that rains upon dark matter halos . stationary solutions have @xmath454 and @xmath455 so that the flux vanishes in velocity space . we consider spherically symmetric solutions as appropriate for the average halo . from jeans theorem , the distribution function is a function of the integrals of motion @xcite . for spherical systems it is often assumed that stationary solutions have the form @xmath456 where @xmath457 ( the particle masses are irrelevant , so we use the specific energy and angular momentum . ) the collisionless boltzmann equation gives no constraint on @xmath458 . the fokker - planck equation , however , constrains @xmath40 because stationarity requires that @xmath459 . assuming spherical symmetry , we write @xmath460 introducing the radial drift coefficient @xmath3 and drag coefficient @xmath292 . we assume isotropic diffusivity @xmath275 . for the globular cluster case with chandrasekhar s dynamical friction , @xmath461 . for the case of quasilinear cosmological fluctuations arising from gaussian random fields , we have seen that @xmath462 . ( the cosmological case also allows for different diffusivities in the radial and tangential directions , but that is an unnecessary complication here . ) requiring the phase space flux to vanish now gives the following constraints on the distribution function @xmath458 : @xmath463 it is interesting to see what conditions lead to thermal equilibrium with a maxwellian distribution of velocities , i.e. eq . ( [ equilsol ] ) with @xmath464 . ( [ station0 ] ) shows that the conditions are @xmath461 and @xmath465 . thus , the drift must be velocity dependent and behave like a drag , @xmath466 . friction converts shear into heat in such a way as to drive the system towards thermal equilibrium . chandrasekhar dynamical friction satisfies these conditions ; therefore , two - body relaxation drives the velocity distribution towards the maxwell - boltzmann form . quasilinear cosmological fluctuations , however , are quite different , having @xmath462 and @xmath467 , implying @xmath468 . while these equations do not have simple general solutions , it is clear that cosmological fluctuations do not drive a system towards thermal equilibrium with equipartition of energies . our cosmological relaxation process is similar to violent relaxation in this respect @xcite . although the general cosmological case of relaxation in a dark matter halo with @xmath467 is complicated and can not be fully solved here , there is a special case worthy of closer examination , namely , the evolution of an unconstrained cosmological gaussian random field , where fluctuations generate peculiar velocities . that is , we drop the constraint of having a halo centered at @xmath469 and instead consider the velocity distribution of dark matter particles at a randomly selected point in space . to consider the cosmological case , we should use the appropriate comoving variables to factor out the hubble expansion . in place of @xmath313 and @xmath16 , we use the following variables : @xmath470 where @xmath91 is the cosmic scale factor and @xmath471 . with this change of variables , the fokker - planck equation ( [ fplin ] ) becomes @xmath472 if the velocities are measured in the comoving frame , @xmath473 is replaced by @xmath35 . when using comoving coordinates the gravitational field is reduced by the hubble acceleration , @xmath474 . if the mean density field is homogeneous , then all particles move with the unperturbed hubble expansion , @xmath475 , implying @xmath476 . thus , both the velocity and acceleration terms on the left - hand side vanish when @xmath477 . if we apply no initial constraint on the density field , then the one - point distribution function @xmath40 is a function only of time and peculiar velocity @xmath478 . we use the variable @xmath253 for the velocity because @xmath478 decreases in proportion to @xmath479 for a freely - falling body in an unperturbed homogeneous and isotropic expanding universe . if not for the right - hand side , the solution of eq . ( [ fpcomov ] ) for a homogeneous and isotropic model would be @xmath480 , i.e. the the initial peculiar velocity distribution would simply redshift with cosmic expansion and remain spatially homogeneous . the fokker - planck equation does not , however , describe an unperturbed homogeneous and isotropic universe . rather it describes the statistical average of an ensemble of universes with fluctuations . in the case of unconstrained initial conditions , @xmath481 and eq . ( [ fpcomov ] ) reduces to @xmath482\ , \ ] ] where @xmath483 and @xmath484 in general . we have used statistical isotropy to require @xmath485 where @xmath486 is a unit vector in the direction of @xmath253 . in second order perturbation theory ( [ sec : fpdiffco ] ) , @xmath487 ( no drag ) and @xmath488 ( assuming that we measure velocities in the comoving frame . ) the fokker - planck equation then takes the form of a diffusion equation with @xmath275 independent of @xmath489 , and the general solution is @xmath490 the effect of fluctuations is simply to spread the initial velocity distribution ( if @xmath285 ) while retaining the gaussian form . physically , the gravitational fluctuations of substructure cause local heating . contrary to eq . ( [ diffsol ] ) , cosmological simulations show that the distribution of peculiar velocities in the nonlinear regime is closer to exponential rather than gaussian , at least in the tails @xcite . it is interesting to ask whether the fokker - planck equation can explain such a result . with the drift term retained , eq . ( [ fpcomhom ] ) has the stationary solution @xmath491 if @xmath492 is positive and independent of @xmath489 , the result is indeed an exponential distribution of peculiar velocities @xmath489 . in second - order perturbation theory we found that @xmath487 and @xmath275 is independent of @xmath489 , however chandrasekhar s dynamical friction gives @xmath493 . evidently , a third case is needed to explain the exponential velocity distribution there must be a drag such that @xmath492 is independent of @xmath489 , quite unlike dynamical friction from two - body relaxation . determining whether such a drag term arises from nonlinear gravitational clustering is a subject for further work . as one final example , we consider the effects of both drag and radial drift in a simplified model illustrating the effects of collisions on a system for which the relaxation time is much longer than the dynamical time ( e.g. the crossing - time for a particle in a halo ) . under these conditions , the system at all times is nearly in collisionless equilibrium with the phase space density being a function of the actions alone ( and not the angles , using action - angle variables ) . because the actions are conserved in the absence of collisions , for quasi - static evolution eq . ( [ fplin ] ) simplifies to @xmath494\ , \ ] ] where we have generalized the radial drift to an arbitrary vector field @xmath495 which we assume to be independent of @xmath16 . we can solve this equation exactly for the simple case where @xmath495 , @xmath292 , and @xmath139 are all constants . the general solution is the green s function solution @xmath496 with green s function @xmath497-\frac{1}{2\gamma}\vec s\cdot{\bf d}\cdot\vec s \left(1-e^{-2\gamma t}\right)\right\}\ .\ ] ] the parameter @xmath301 is a relaxation time . ( [ fpqsgreen ] ) simplifies to a normalized gaussian in two limits , @xmath498 where @xmath499 is a multivariate normal with mean @xmath253 and covariance matrix @xmath500 . the solution for @xmath501 was given previously in eq . ( [ diffsol ] ) . now we have a more general solution showing that the equilibrium ( for constant diffusion coefficients ) is a maxwellian with velocity dispersion tensor @xmath502 just in eq . ( [ fpstat ] ) if @xmath503 , but now the drift vector @xmath495 produces an offset of the mean velocity @xmath504 . thus , velocity - independent drift @xmath3 and velocity - dependent drag @xmath309 play completely different roles in collisional relaxation . in later papers we will investigate these roles for more realistic models of nonlinear halo evolution . plotted as a function of halo radius @xmath173 for gaussian smoothing lengths @xmath505 0.5 , 0.25 , 0.1 , and 0.05 @xmath506 mpc . it is computed from eq . ( [ scala ] ) for the @xmath20cdm model with @xmath507 . for comparison , the dashed curve shows @xmath186 for an unconstrained field . the vertical axis shows the dimensionless @xmath508 , i.e. the drift normalized by the rms of the peculiar gravitational acceleration . ] in the radial ( @xmath173 ) and tangential ( @xmath66 ) directions plotted as a function of halo radius @xmath173 for gaussian smoothing lengths @xmath509 , 0.5 , and 0.05 @xmath506 mpc . it is computed from eqs . ( [ scaldr ] ) and ( [ scaldt ] ) for the same cosmological model as fig . 1 . the dashed curve in each panel is for an unconstrained field . the vertical axis shows the dimensionless @xmath510 , i.e. the diffusivity normalized by the mean squared peculiar gravitational acceleration . ] for power - law spectra , @xmath4 , plotted as a function of halo radius @xmath173 scaled to the gaussian smoothing radius @xmath187 . results are shown for four spectral indices , @xmath511 , and 1 , and for constrained and unconstrained fields . the amplitude of @xmath186 corresponds to @xmath512 normalized to @xmath513 . ] in the radial ( @xmath173 ) and tangential ( @xmath66 ) directions for power - law spectra , @xmath4 , plotted as a function of halo radius @xmath173 scaled to the gaussian smoothing radius @xmath187 . results are shown for two spectral indices , @xmath250 and @xmath514 , and for constrained and unconstrained fields . for @xmath515 , constraints cause the diffusivity to approach a negative constant for @xmath516 . the amplitude of @xmath275 corresponds to @xmath512 normalized to @xmath513 . ] antonuccio - delogu , v. & colafrancesco , s. 1994 , , 427 , 72 bardeen , j. m. , bond , j. r. , kaiser , n. , & szalay , a. s. 1986 , , 304 , 15 benson , a. j. , lacey , c. g. , frenk , c. s. , baugh , c. m. , & cole , s. f 2003 , astro - ph/0307298 bertschinger , e. 1985 , , 58 , 39 bertschinger , e. 1993 , in statistical description of transport in plasma , astro- , and nuclear physics , ed . j. misguich , g. pelletier , & p. schuck , ( commack , ny : nova science ) , 193 bertschinger , e. 1996 , in cosmology and large scale structure , proc . les houches summer school , session lx , ed . r. schaeffer , j. silk , m. spiro , and j. zinn - justin ( amsterdam : elsevier science ) , 273 binney , j. j. & tremaine , s. 1988 , galactic dynamics ( princeton : princeton university press ) bullock , j. s. , dekel , a. , kolatt , t. s. , kravtsov , a. v. klypin , a. a. porciani , c. , & primack , j. r. 2001 , , 555 , 240 chandrasekhar , s. 1943 , , 97 , 255 davis , m. & peebles , p. j. e. , 34 , 425 dekel , a. , devor , j. , & hetzroni , g. 2003 , , 341 , 326 dupree , t. h. 1967 , phys . fluids 10 , 1049 evans , n. w. & collett , j. l. 1997 , , 480 , l103 fillmore , j. a. & goldreich , p. 1984 , , 281 , 1 fokker , a. d. 1914 , ann . physik 43 , 810 gardiner , c. w. 2002 , handbook of stochastic methods ( berlin : springer ) ghigna , s. , moore , b. , governato , f. , lake , g. , quinn , t. , stadel , j. 1998 , mnras , 300 , 146 gilbert , i. h. 1968 , , 152 , 1043 gunn , j. e. & gott , j. r. 1972 , , 176 , 1 haehnelt , m. , & kauffmann , g. 2002 , , 336 , l61 hoffman , y. , & shaham , j. 1985 , , 297 , 16 hughes , s. a. , & blandford , r. d. 2003 , 585 , l101 ichimaru , s. 1973 , basic principles of plasma physics : a statistical approach ( reading : addison - wesley ) jain , b. , & bertschinger , e. 1994 , , 431 , 495 kauffmann , g. , & haehnelt , m. 2000 , , 311 , 576 katz , n. , quinn , t. , & gelb , j. 1993 , , 265 , 689 klimontovich , yu . l. 1967 , the statistical theory of non - equilibrium processes in a plasma ( cambridge , mass . : mit press ) klypin a. , kravtsov a. v. , valenzuela 0 . , prada f. 1999 , apj , 522 , 82 langevin , p. 1908 , comptes rendus , 146 , 530 lynden - bell , d. 1967 , , 136 , 101 monaco , p. , theuns , t. , taffoni , g. , governato , f. , quinn , t. , & stadel , j. 2002 , , 564 , 8 moore , b. , governato , f. , quinn , t. , stadel , j. , & lake , g. 1998 , , 499 , l5 moore , b. , ghigna , s. , governato , f. , lake , g. , quinn , t. , stadel , j. , tozzi , p. 1999 , apj , 524 , 19l mulder , w. a. 1983 , a&a , 117 , 9 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1996 , , 462 , 563 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1997 , , 490 , 493 nusser , a. & sheth , r. k. 1999 , , 303 , 685 peebles , p. j. e. 1980 , the large scale structure of the universe ( princeton : princeton university press ) planck , m. 1917 , sitzber . p. 324 risken , h. 1989 , the fokker - planck equation ( berlin : springer ) rosenbluth , m. n. , macdonald , w. m. , & judd , d. l. 1957 , phys . 107 , 1 sheth , r. k. , & diaferio , a. 2001 , , 322 , 901 spitzer , l. 1987 , dynamical evolution of globular clusters ( princeton : princeton university press ) syer , d. , & white , s. d. m. 1998 , , 293 , 337 weinberg , m. d. 2001 , , 328 , 311 ; 328 , 321 white , s. d. m. 1984 , , 286 , 38 zeldovich , ya . b. 1970 , a & a , 5 , 84
we present a numerical study of exact breather dynamics in two - dimensional nonlinear schrdinger lattices . this issue fits well with the enduring scientific interests of serge aubry in the many faceted subject of localization and transport in nonlinear macroscopic discrete systems , where serge s outstanding contributions are widely recognized . many ideas and lines of study followed in the investigation reported here have found a source of inspiration in early works of aubry and collaborators , among which the doctoral works of th . cretegny and j.l . marn deserve special mention . the existence and properties of localized solutions in extended discrete systems have atracted interest in a broad range of physical fields @xcite . discrete breathers , sometimes referred to as intrinsic localized modes , are spatially localized and time periodic solutions . these solutions arise in the context of nonlinear discrete systems and are of fundamental interest for varied physical applications such as pulse propagation in nonlinear optics , energy storing and transport in biomolecules , plasma physics , etc ... the existence of discrete breathers in these systems has been proved rigorously @xcite for a number of equations with physical relevance and , contrary to continuous nonlinear equations , their existence can be regarded as a generic feature of these systems . one of the most important class of equations are the so - called _ discrete nonlinear schrdinger lattices _ the existence of discrete breathers has been proven for a wide range of systems belonging to this class of nonlinear difference - differential equations . in particular , the most important example of wide aplicability is the standard nonlinear schrdinger equation . for instance , this equation was employed in @xcite for describing the propagation of localized beams in an array of nonlinear ( kerr type ) waveguides , having experimental validation subsequently reported in @xcite . the study of two - dimensional nonlinear schrdinger lattices has atracted much atention @xcite in recent years due to the new phenomena emerging when the dimensionality of the lattice is increased . some examples of these new features are the existence of vortex - breathers @xcite which supports energy flux , the appearance of an energy threshold for the creation of discrete breathers @xcite and the ubiquity of an instability ( the quasi - collapse ) of some discrete breather solutions leading to a highly localized pulson state @xcite . these theoretical efforts have their counterpart in recent advances in the field of nonlinear optics . the studies of two - dimensional arrays of coupled nonlinear waveguides allow the experimental observation of those effects studied theoretically . specially relevant is the recent experimental breakthrough ( theoretically designed in @xcite ) by fleisher _ et al _ @xcite , where a two - dimensional array of nonlinear waveguides is induced in a photosensitive material . this technique provides a clear experimental verification of the two - dimensional discrete breather existence in this system . in particular , besides the observation of standard discrete breathers , these works reported the first observations of staggered discrete breathers . our study here will focus on the computation of numerically exact dicrete breathers in two - dimensional anisotropic nonlinear schrdinger lattices , _ i.e. _ where the couplings in the two spatial directions are different . the use of shooting methods allow us to find these solutions and analyze their structural and stability properties . both pinned and mobile discrete breathers are studied . in the latter case we will study only the ones whose motion is along one axis of the lattice . the analysis of the numerically exact solutions shed light on some features of the properties and stability of localized solutions reported in previous works . the plan of the paper is as follows : in section [ sec : numerical ] we provide the technical background and details needed for self - contained purposes . first we summarize in [ subsec:1d ] the main conclusions on the dynamics of 1d schrdinger discrete breathers reported in @xcite . a detailed account of the numerical methods ( svd - regularized newton continuation of operator fixed points ) that we have used can be found in that reference . also in [ subsec:1d ] , we discuss briefly the most relevant formal differences with respect to alternative approaches to ( one - dimensional ) exact mobility of discrete breathers , _ e.g. _ those in refs @xcite and @xcite . in subsection [ sec : numerics ] we introduce the two - dimensional anisotropic salerno lattice and provide explanations on the implementation of the numerical procedures used to study the dynamics of 2d discrete breathers . the analysis of the results of our numerics on pinned discrete breathers for anisotropic nonlinear schrdinger lattices is reported in section [ sec : pinned ] . we present the numerical computations of the fixed point norm , as a function of three parameters : breather frequency , transversal coupling , and nonlinearity ( see below ) . they show , as anticipated , the so - called _ quasi - collapse _ transition , associated with the ( well - known ) existence of thresholds for the breather norm in two - dimensional lattices . we present numerically computed sectors of the bifurcation surface . we conclude this section with a brief look at the nonlinear dynamics on the unstable manifold , whose typical trajectories have been called _ pulson _ states . early numerical work on the 2d quasi - collapse phenomena in isotropic lattices was reported in @xcite and @xcite . a two - year - old account of the `` state of knowledge '' on 2d schrdinger lattices can be found in section six of @xcite . in section [ secc : mobile ] we show results on a type of mobile breather , namely those moving along the direction of stronger lattice coupling constant . the structure of each of these mobile exact discrete breathers is that of a localized moving _ core _ superimpossed on a specific extended state of resonant small amplitude radiation , the _ background_. we present here the results of an extensive floquet stability analysis of this type of solutions in two sectors of the three - dimensional parameter space , which clearly show the existence of two different transitions . the tangent space eigenvectors associated to each of the transitions are presented , and the relation of the unstable manifold trajectories to pulson states is analyzed afterwards . we conclude with section [ secc : conclusions ] , where we briefly review the results obtained and illustrate their possible implications for mobility of 2d discrete breathers . the use of numerical tools for the continuation of discrete breather solutions has been widely employed since their existence proof was reported ( see _ e.g _ @xcite ) . in particular , the design of numerical techniques for finding exact mobile breathers based on those employed for the pinned ones has been explored in the recent years @xcite and paves the way to resolving the still open question about discrete breather mobility . here , after reviewing the most important features of 1d mobile breathers in nonlinear schrdinger lattices , we briefly explain how the continuation method is implemented in our system . as was introduced in section [ sec : introduction ] , exact mobility of dicrete breathers in 1d schrdinger nonlinear lattices were numerically studied by the authors in previous works @xcite . in particular , a numerical continuation from the integrable _ ablowitz - ladik lattice _ , a - l , ( where exact mobile breathers can be calculated analytically @xcite ) , @xmath1 to the standard _ discrete nonlinear schrdinger equation _ , dnls , @xmath2 was performed within the so - called _ salerno model _ @xcite , @xmath3 where @xmath4 , @xmath5 and @xmath6 are the parameters accounting for the strength of the nonlinear terms . the above equation includes the former two relevant equations ( [ eq : al ] ) and ( [ eq : dnls ] ) , for ( @xmath7 , @xmath8 ) and ( @xmath9 , @xmath10 ) respectively , providing the desired interpolation needed to develop the continuation scheme . the integrability of equation ( [ eq : al ] ) provides for @xmath11 a two - parametric family of discrete breathers @xmath12\;{\mbox{exp}}[{\mbox{i}}(\alpha(n - x(t))-\omega(t))]\;,\ ] ] where the two parameters @xmath13 $ ] and @xmath14 describe the velocity and internal frequency of the solution @xmath15 the salerno lattice ( [ eq : salerno ] ) , possess two dynamical invariants , namely the hamiltonian , @xmath16 where @xmath17 denotes the complex conjugate of @xmath18 , and the norm @xmath19 in order to find exact mobile discrete breathers in the salerno model we define a ( @xmath20)-resonant solution @xmath21 referred to some time scale @xmath22 such that @xmath23 within the above definition , a mobile breather that translates @xmath24 sites after @xmath25 periods of the internal oscillation will satisfy equation ( [ eq : p - q_resonant ] ) when @xmath26 . in this sense , the continuation focuses on the families of ( @xmath20)-resonant discrete breathers , that is breather solutions with the two characteristic time scales ( corresponding to the breather velocity , @xmath27 , and the internal frequency , @xmath28 ) being commensurate . in all the computations we have used finite lattices with periodic boundary conditions ( pbc ) so that @xmath29 and @xmath30 ( with @xmath31 being the lattice size ) . in the previous works @xcite the authors start from those a - l solutions which are ( @xmath20)-resonant , we discretize ( fine grid ) the path @xmath32 ( with @xmath5 and @xmath6 being positive ) along the salerno model , and go through it computing the corresponding ( @xmath20)-resonant solutions for the pairs @xmath33 . we can choose the path without loss of generality because of the scaling property of equation ( [ eq : salerno ] ) . then , each solution is numerically computed as a fixed point of the map @xmath34 where @xmath35 is the lattice translation operator @xmath36 , and @xmath37 is the @xmath38-evolution map ( @xmath39 ) following the dynamics dictated by equation ( [ eq : salerno ] ) for the corresponding value of @xmath6 ( @xmath40 ) ; _ i.e. _ , @xmath41=\{\phi_{n}(t_{0}+t_b)\}$ ] . the continuation was then performed for a fine grid of frequencies belonging to the family of ( @xmath42)-resonant discrete breathers . the most important conclusion about discrete breathers in these nonlinear schrdinger lattices is that mobility in the non - integrable regime ( @xmath43 ) demands the presence of an extended _ background _ to which the fixed point solution is spatially asymptotic ( @xmath44 ) , _ i.e. _ the solution is exactly written as @xmath45 this expression defines the purely localized component @xmath46 of the solution . the background is a finite linear combination of nonlinear plane waves , @xmath47 $ ] . these plane waves are exact solutions of the salerno model ( [ eq : salerno ] ) being the nonlinear dispersion relation @xmath48\cos k_{j } - 2\nu |a_{j}|^{2}$ ] . the results concerning the characterization of the background are discussed in detail in @xcite , here we briefly summarize the most important features : * the set of `` @xmath49 '' plane waves which take part in the background of a @xmath50resonant discrete breather with internal frequency @xmath51 is derived by the simple selection rule for the wave - numbers @xmath52 @xmath53 _ i.e. _ only the plane waves which are ( @xmath20)-resonant with the internal period of the breather can be components of @xmath54 . the number of solutions of ( [ resonant ] ) fixes `` @xmath49 '' . * the amplitudes @xmath55 of the nonlinear plane waves differ by orders of magnitude . * there exist a strong positive correlation between the amplitude of the background and the strength of the peierls - nabarro barrier arising from the periodic lattice . this correlation is particularly clear when symmetry breaking transitions occur for the also studied case of @xmath56 and @xmath57 ( see reference @xcite ) , and reflects the link between non - integrability and the existence of the background dressing of the mobile core . * finally , the interpretation of the correlation described in _ ( iii ) _ is reinforced from a study of the energy evolution of the mobile core : _ there is an energy balance brought by the background when the core moves along the lattice_. in particular , it can be observed how the core energy oscillates periodically so that it takes the maximum energy value when the core visits the inter - site configuration . this extra energy periodically obtained by the core is provided by the background , with the energy maximum clearly related to the background amplitude . before concluding this subsection , it is worth commenting on some of the differences between the newton continuation of fixed points that we use in this paper , and other important recent approaches to breather numerics . the work by ablowitz _ et al _ @xcite uses discrete fourier analysis to obtain a nonlinear nonlocal integral equation , from where the `` ... soliton is thus viewed as a fixed point of a nonlinear functional '' ( sic ) in the fourier transformed space of functions . following these authors , their results seem to differ from those of early pioneering work @xcite ( nowadays textbook material @xcite ) `` in which a _ continuous _ travelling solitary waves were reported using fourier series expansions with finite period @xmath58 while assuming convergence as @xmath59 '' ( sic ) . et al _ term _ continuous _ a solution that can be defined off the lattice points , which they see as `` necessary when discussing travelling waves in lattices '' ( sic ) , and disagree with some conclusions reported in the earlier works . the ( `` orthodoxy matters '' ) discussion above helps us to clarify how differently our numerical approaches `` sees '' the discrete schrdinger breather problem : the very concept of a variable defined off the lattice points is intrinsically alien to our discrete approach , which neither needs of it nor excludes its eventual consideration . in contrast to those views ( but not at all in logical opposition ) , we consistently view the thermodynamical limit ( @xmath60 ) in lattice space , much in the sense used _ e.g. _ by serge aubry in his celebrated work on the frenkel - kontorova ground state problem @xcite : the infinite size limit is built up from a subsequence of pbc ( finite ) lattices for which the limit is well defined . this will make the fourier - transformed @xmath61-space continuum . closer to our approach in some respects , though technically different in many others , is the formal approach purposed recently by james and collaborators @xcite . this approach , which uses recent central manifold theorems , was brought to our attention very recently and unfortunately , we have to defer to a future publication a comparison of our 1d computations with theirs , this is beyond the scope of this paper , namely , the anisotropic 2d schrodinger lattices , which we introduce in the next subsection . motivated by the method described above , we can extend the continuation scheme for calculating exact discrete breathers in higher dimensional systems . in particular we focus on the two - dimensional nonlinear schrdinger lattice @xmath62 this lattice can be viewed as the _ two - dimensional salerno model_. the two coupling parameters @xmath63 and @xmath64 provide a technical advantage for numerics ( see below ) , but they are also introduced for theoretical and experimental interest . the possibility of controlling the ratio between the two linear couplings of the two transversal directions has been studied in various works as a way of analysing how the intrinsic 2d phenomena ( such as the quasi - collapse ) emerge . in fact , for @xmath65 , @xmath9 and @xmath10 equation ( [ eq:2d - dnls ] ) describes a set of weakly coupled nonlinear wavequide arrays and can be considered as a case of `` _ intermediate dimensionality _ '' . this extreme has been studied experimentally in @xcite and using perturbative methods in @xcite . on the other hand , this equation incorporates , as two particular limits , the physically relevant standard two - dimensional dnls equation ( @xmath9 , @xmath10 ) and the two - dimensional a - l lattice ( @xmath7 , @xmath8 ) . the continuation between these two limits provides a useful tool for studying the interplay between the on - site and inter - site nonlinearity in the 2d anisotropic lattice . moreover , the anisotropy ( or freedom in the values of the coupling parameters @xmath66 and @xmath67 ) allows to include an integrable model among the members of the family of nonlinear lattices described by ( [ eq:2d - dnls ] ) . that is , for @xmath8 , @xmath68 and @xmath69 one obtains a set of integrable a - l chains . in this way , every 2d model included in ( [ eq:2d - dnls ] ) is connected with an integrable model where analytic discrete breathers are available . in what follows we set @xmath70 and @xmath71 with @xmath72 and @xmath57 . the dynamics ( [ eq:2d - dnls ] ) can be derived from the salerno poisson structure @xmath73 with the hamiltonian @xmath74 as in the 1d salerno model there is also a second conserved quantity , the norm , due to the phase invariance of the equations of motion ( [ eq:2d - dnls ] ) @xmath75 in the same manner as in the 1d case we will focus on a special set of 2d discrete breathers . for this , we have to generalize the definition ( [ eq : p - q_resonant ] ) of a resonant solution in the two - dimensional case . in this context discrete breathers solutions are characterized by three time scales . namely , one associated with the internal oscillation @xmath28 and the other two derived from the translation of the localization center , _ i.e. _ its velocity @xmath76 . the subset of @xmath77-tuples ( @xmath28,@xmath78 ) that fulfil the ( @xmath79)-resonance condition @xmath80 ( where @xmath81 , @xmath82 and @xmath25 are integers ) denote the breather solution that can be obtained with our continuation method . these solutions are those that after @xmath25 periods of the internal frequency , @xmath83 , translates @xmath81 and @xmath82 lattice sites in the @xmath84 and @xmath85 direction of the square lattice , respectively , _ i.e. _ @xmath86 where again pbc are applied @xmath87 , @xmath88 , @xmath89 and @xmath90 ( with @xmath91 and @xmath92 being the lattice size in the @xmath84 and @xmath85 direction respectively ) . consequently , a ( @xmath79)-resonant state will be a solution of the following set of equations @xmath93 & = & { l}_{y}^{p_{y}}{l}_{x}^{p_{x } } { t}_{(\omega_{b } , \nu , c_{1 } , c_{2})}^q[\{\hat{\phi}_{n , m}(t_{0})\}]= \nonumber \\ & = & \{\hat{\phi}_{n , m}(t)\}\ ; , \label{eq:2d - map}\end{aligned}\ ] ] where the operators @xmath94 are the lattice translation in the @xmath95-direction , @xmath96&=&\{\phi_{n+1,m}(t_{0})\}\ ; , \\ l_{y}[\{\phi_{n , m}(t_{0})\}]&=&\{\phi_{n , m+1}(t_{0})\}\;.\end{aligned}\ ] ] besides , @xmath97 is the time evolution operator given by equation ( [ eq:2d - dnls ] ) over one period @xmath98 , @xmath99= \{\phi_{n , m}(t_{0}+t_{b})\}\;.\ ] ] in order to illustrate the 2d time scales resonance we consider the plane wave solutions of equation ( [ eq:2d - dnls ] ) : @xmath100 $ ] . these solutions possess the following nonlinear dispersion relation @xmath101 hence , we can obtain the subset of plane waves which are ( @xmath79)-resonant with some time scale @xmath22 ( _ i.e. _ after a time @xmath102 they have translated @xmath81 and @xmath82 sites in the @xmath84 and @xmath85 direction , respectively ) . each member of these subsets will be labelled by the pair @xmath103(@xmath104 , @xmath105 ) and from the condition ( [ eq:2d - map ] ) it follows that the corresponding set of values of @xmath106 for each family will satisfy the relation @xmath107 where @xmath108 is an integer and @xmath109(@xmath81 , @xmath82 ) . in figure [ fig : resonances ] the corresponding values of @xmath106 are represented for two resonances of the type @xmath110 and @xmath111 . cc the method used for solving equation ( [ eq:2d - map ] ) for each resonant @xmath77-tuple ( @xmath28,@xmath78 ) is the same as in the 1d case , which is extensively described in @xcite . the implicit function theorem assures that a fixed point solution of a map ( [ eq:2d - map ] ) given by @xmath112(@xmath81 , @xmath82 , @xmath25 , @xmath28 , @xmath6 , @xmath63 , @xmath64 ) can be obtained provided that _ ( i ) _ the jacobian of the operator @xmath113-{i}$ ] is invertible and _ ( ii ) _ we know a fixed point of a map corresponding to an infinitesimally close set of parameters , @xmath114 ( @xmath81 , @xmath82 , @xmath25 , @xmath115 , @xmath116 , @xmath117 , @xmath118 ) . the first demand can be satisfied using a _ singular value decomposition _ ( svd ) @xcite of the jacobian in order to obtain the pseudo - inverse operator . on the other hand , when the second condition is fulfilled convergence of the newton - raphson iterative scheme is guaranteed . for this , we start with a sufficiently good trial solution , @xmath119 and solve the equation @xmath120^{-1}\cdot f_{\vec{\xi}}[\{\phi^{0}_{n , m}(t_{0})\}]\ ; , \label{eq : iteration}\ ] ] in order to obtain @xmath121 . we iterate this calculations to the desired convergence , and then the solution @xmath122 , is obtained . in our numerics this is the case when @xmath123<n\cdot10^{-16}\;,\ ] ] ( where @xmath31 is the total number of sites in the square lattice ) is fulfilled . once the solution is found we use it as the following trial solution , @xmath119 , for solving the map ( [ eq:2d - map ] ) corresponding to the next set of parameters @xmath124 . as an additional benefit , this method provides the linear stability of the computed solution . for this , we only have to compute the eigenvalues of the jacobian of the operator around the solution @xmath125 $ ] . in fact , this jacobian is the extended floquet matrix of the solution . then , a solution is linearly stable if all the eigenvalues of the corresponding floquet matrix are inside the unit circle . moreover , the symplectic character of the dynamics ( [ eq:2d - dnls ] ) implies that all the eigenvalues appear in quadruplets ( @xmath126 , @xmath127 , @xmath128 , @xmath129 ) and thus for a linearly stable solutions all the eigenvalues of its extended floquet matrix lie on the unit circle . there are two possible paths for developing the continuation method depending on the choice of the starting point of the continuation . one possibility is to start from the full anti - continuum limit , @xmath130 , where a pinned breather solution of frequency @xmath28 is written as @xmath131 starting from the above solution , we can perform the continuation increasing the parameters @xmath63 and @xmath64 as usual , and so obtain the whole family of ( @xmath132)-resonant discrete breathers . an alternative path starts from the one - dimensional limit , @xmath133 . the choice of this second limit ( which implies taking as the very initial trial solution of the continuation the whole set of 1d solutions described in section [ subsec:1d ] ) is justified when seeking mobile solutions . as stated above , this limit offers the possibility of studying strongly anisotropic lattices as a controlled interpolating situation between one and two dimensions . on the other hand , employing this strategy we can only obtain those solutions which are ( @xmath134)-resonant , _ i.e. _ the two - dimensional continuation of those one - dimensional ( @xmath135)-resonant discrete breathers . hence , the solution from which we start is @xmath136 where @xmath137 is the corresponding ( @xmath135)-resonant one - dimensional solution . in what follows we will employ both continuation paths when we study the case of pinned breathers ( section [ sec : pinned ] ) , and we will show that the results obtained are the same when approaching the same limit ( the standard two - dimensional dnls ) . as we have discussed , we can choose two different starting points for the continuation of @xmath138-resonant fixed points ( pinned breathers ) of equation ( [ eq:2d - map ] ) : _ ( i ) _ the full anti - continuum ( ac ) limit ( @xmath139 ) , or _ ( ii ) _ the ( one - dimensional , 1d ) limit of uncoupled chains ( @xmath140 ) , where they can be obtained from continuation along increasing values of the parameter @xmath6 from the 1d a - l lattice ( [ eq : al ] ) . as a test for our codes , we have checked that both paths arrive to the same solution . in fact , unique continuations can proceed along any path on the plane of parameters ( @xmath141 ) that we have explored . early works @xcite on the isotropic two - dimensional standard dnls equation analyzed the so - called _ quasi - collapse _ instability of pinned discrete breathers , _ i.e. _ the condensation of all the energy into a few modes in discrete nonlinear systems , which corresponds to the onset of a singularity ( wave collapse ) @xcite in multidimensional continuum models . subsequent numerical works @xcite extended these studies to the isotropic 2d salerno lattice and addressed the question of how the instability is affected by the presence of impurity lattice sites . as expected , our results further corroborate the existence of quasi - collapse instabilities in the anisotropic case : the phase diagram in parameter space ( @xmath51 , @xmath67 , @xmath6 ) consists of two regions ( stable and unstable ) separated by the surface of transition . as we perform the continuation of breather solutions across the parameter space we scan the floquet stability of the computed solution . in figure [ fig : ctresholds ] we present the two stability transition curves in the plane ( @xmath28 , @xmath64 , @xmath142 ) , _ i.e. _ the function @xmath143 , corresponding to the two different continuation starts . the continuation from the ac limit is made through the path @xmath144 and the one from the 1d limit is made at @xmath145 . the convergence of the two paths at @xmath146 is clearly seen . cc the criterion for stability of the pinned discrete breather solution derived in @xcite , @xmath147 is of a very general character and our numerics illustrate it clearly . on the other hand , the floquet stability analysis detects the dimensionality ( and a basis in tangent space ) of the unstable linear manifold associated with the quasi - collapse instability that these exact discrete breathers experience for some parameter values . we have computed numerically , for a fine grid of @xmath51 values and a coarser grid of @xmath67 and @xmath6 , the function @xmath148 , from which we show some sectors in figures [ fig : norma - c2 ] and [ fig : norma - nu ] . in figure [ fig : norma - c2 ] we show the numerically computed norm ( [ eq:2dnorm ] ) as a function of the breather frequency @xmath149 , for three different values of the transversal coupling @xmath67 , and a fixed value of @xmath150 ( anisotropic dnls limit ) . we observe the existence of a minimum value , @xmath151 , which is thus seen as an _ excitation threshold _ for the creation of these solutions . the position of the minimum @xmath152 , which _ naturally _ increases with @xmath67 , separates the stable and unstable branches of pinned breathers . breathers corresponding to values of @xmath51 where @xmath149 has a negative slope are unstable : this is shown in the insets , where the floquet spectra of two representative examples of pinned discrete schrdinger breathers are plotted in the complex plane . note that the high accuracy of the numerical solution allows an unprecedented detailed floquet analysis of the instability , paving the way to rigorous analytical characterizations of the quasi - collapse unstable manifold . this is a one - dimensional manifold , as our numerical results unambiguously confirm . then , in the regime of small time scales , the unstable manifold is fully characterized by a single floquet eigenvector . cc figure [ fig : norma - nu ] shows the ( surface ) function @xmath153 for the volume sector of constant @xmath67(@xmath154 ) . most noticeably , the critical ( threshold ) line of bifurcation points ( @xmath155 ) , as seen in the inset , does not define a monotone function @xmath156 . in fact , in the whole interval of @xmath157 values , the range of values of @xmath158 is quite short , indicating the insensitibity of the gross features of the quasi - collapse transition to the value of @xmath6 . however , considering finer details , one sees that the threshold curve @xmath156 smoothly reaches its slightly larger values around midway between the dnls and the a - l limits . in other words , intermediate values of the interpolating ( salerno ) parameter @xmath6 somewhat favour the enhancement of the quasi - collapse unstable region . these conclusions are in contrast with the stated conclusion ( for isotropic lattices ) in @xcite that the ablowitz - ladik term increases the stability regime . cc when instability is allowed to develop beyond the fixed point tangent space into the nonlinear realm of perturbations , the trajectory obtained by direct integration of the equations of motion invariably ends after a transient ( of time scale given by the real floquet exponent larger than @xmath159 ) in a localized solution with complex dynamics , the _ pulson states_. these states were characterized in @xcite in the following terms `` ... where the peak intensity @xmath160 oscillates between the central site and its four nearest neighbours ... it is not known whether these pulson states represent true quasiperiodic solutions to the dnls equation '' . what makes these trajectories on the unstable nonlinear quasi - collapse manifold of much practical relevance and interest is their ubiquity : they appear as persistent localized states in the hamiltonian dynamical evolution from a wide variety of initial conditions . their description requires at least two frequencies , namely the internal ( genuine breather - like frequency ) and the frequency of the oscillations of the _ breather width _ around a mean width value , which turns out to be less than the width of the unstable exact discrete breather . second and outer shells of neighbours ( in both lattice axes ) also participate in the width oscillations . cc though a more detailed characterization of the pulson states will be presented elsewhere , it is illustrative to consider ( figure [ fig : qc - fft ] ) the power spectrum of the field at the central site of a typical trajectory on the unstable nonlinear manifold of a quasi - collapsing pinned discrete breather . this shows peaks at the combinations @xmath161 ( @xmath162 ) , where @xmath163 is the frequency of the width oscillations characterizing the pulson state , while @xmath164 is a frequency higher than the ( initial condition ) fixed point frequency @xmath51 . the new frequency @xmath165 turns out to be very close to the breather frequency of the same ( initial ) norm on the stable branch . in other words , the instability drives a shift of breathing frequency towards the stable branch , while the excess energy is transferred to the oscillatory motion of the observable width . this behavior seems to be the essence of the physical characterization of the nonlinear quasi - collapse manifold dynamics . the numerical observation of a two - frequency power spectrum for a typical pulson state points towards an eventual positive answer to the question ( on true quasiperiodicity ) arised in @xcite , though , for more details we have to refer to the phd dissertation @xcite . this point serves to illustrate how the high accuracy of the fixed point numerical solution provides detailed clues on many still unsolved ( from a mathematical and physical point of view ) questions on two - dimensional schrdinger localization , which are of prospective experimental interest in nonlinear ( photonic , josephson , ... ) physics technologies . early and current attempts to explore straightaway discrete breather mobility in isotropic 2d schrdinger lattices seem to agree @xcite that `` kicking '' procedures meet huge difficulties in delivering good mobile solutions , contrary to the numerical experiences in 1d lattices . we note here that the formal basis for those methods @xcite takes advantage of the floquet spectra analysis of exact pinned breathers , where the so - called _ depinning _ ( symmetry - breaking ) mode is identified . this allows , _ provided peierls - nabarro barriers are small enough _ , to obtain nice numerical 1d mobile discrete breathers , by computing trajectories from perturbations of the exact pinned breather along the tangent space direction specified by the depinning eigenvector . the presence of symmetry - breaking instabilities leading to exchange of stability between one - site and two - site centered pinned breathers @xcite and the associated lowering of the peierls - nabarro barriers to breather displacements , hugely facilitating the success of these procedures when applied to ( both hamiltonian and dissipative ) one - dimensional lattices . @xcite in contrast , our `` anisotropic lattice '' continuation approach takes advantage of the availability of exact 1d mobile solutions by monitoring the parameter @xmath67 of transversal coupling , and then does not rely on how easily one promotes clean mobility from pinned localization . in this way we obtain accurate numerical ( @xmath166 ) fixed points , that is schrdinger discrete breathers moving along the strong coupling direction . in a forthcoming paper we will address the ( much more difficult ) question for arbitrary direction of motion . in figure [ fig : mob - structure ] we visualize the instantaneous real and imaginary components of the 2d discrete field profile of a typical ( 1,0,1 ) schrdinger breather . its structure can be seen as the natural extension to two - dimensional lattices of the structure of mobile schrdinger breathers analyzed in @xcite . the numerical solution is spatially asymptotic to a _ finely tuned _ small - amplitude extended ( delocalized ) radiation state @xmath167 when @xmath168 . the fixed point solution can be thus decomposed as @xmath169 which defines @xmath170 , the spatially _ localized _ component of the solution . it turns out that the spatially _ delocalized _ component is a highly localized state in the ( continuum , in the thermodynamic limit ) @xmath61-space of wavevectors . more precisely , @xmath171 is a finite linear combination of ( 1,0,1 ) -resonant nonlinear ( _ i.e. _ amplitude - dependent frequency @xmath172 ) 2d planewaves . it can be said that , as might be expected , 1d schrdinger breather mobility smoothly persists when ( strong @xmath66-coupling ) 1d chains are coupled transversally . importantly , the numerical continuation for increasing values of the transversal coupling @xmath67 proceeds far from the weak coupling regime into where the genuine two - dimensional effects start to be manifest , as we will see below . cc most noticeable , the svd - regularized newton procedure invariably selects the values @xmath173 for all values of @xmath67 and @xmath6 , and thus the values of @xmath174 for the 2d resonant planewave are independent of @xmath67 ( so it remains equal to the @xmath61 values of the 1d ( @xmath175 ) fixed point for the uncoupled chain ) . the appearance of an extended background modulation in the transversal direction of @xmath176 appears naturally as the _ best _ choice to take advantage of approximately 1d breather propagation along strong coupling direction , for it keeps the value of @xmath174 favoured by the strong coupling @xmath66 value : any other value of @xmath177 would entail a different @xmath174 value . note however that this provides only a _ plausibility argument _ for the interpretation of the numerical observation ( @xmath176 ) . the high accuracy of the computed solutions allows a detailed analysis of many issues concerning 2d schrdinger breather exact mobility along the strong coupling direction . we leave aside in this paper many of them , and focus here on how the existence of quasi - collapse instabilities of pinned schrdinger breathers , for increasing @xmath67-coupling values , influences the stability properties of moving ( @xmath178 ) breathers . in other words , we search here for genuine 2d effects on these `` strong - coupling -direction '' ( quasi-1d ) moving breathers . we have performed an exhaustive exploration of two sectors of the parameter space ( @xmath179 ) , corresponding to the breather frequency values @xmath180 , and @xmath181 , by computing the continued ( @xmath178 ) fixed point . these values of @xmath51 were chosen low enough to allow the analysis of pinned breather quasi - collapse effects on mobility , which occurs at relatively low values of @xmath182 for these values of @xmath172 . cc + the floquet analysis of the computed solutions provides the stability diagrams represented in figures [ fig : mob - stability ] . both show no qualitative differences : there are two regions in the ( @xmath183 ) plane where the ( @xmath178 ) mobile breather is linearly unstable . the figures are not `` schematic '' : every point of the plane in a fine grid of values of @xmath67 and @xmath6 has been analyzed , _ i.e. _ the floquet spectrum of the computed ( @xmath178 ) fixed point is scrutinized , as shown in figures [ fig : mob - stability].b and [ fig : mob - stability].d , where the modulus of the floquet eigenvalues is shown as a function of either @xmath6 ( figure [ fig : mob - stability].b ) or @xmath67 ( figure [ fig : mob - stability].d ) . the first unstable region appears at low values of @xmath67 and intermediate to high values of the salerno parameter @xmath6 , _ i.e. _ it does not occur close to the a - l limit . this unstable region is also bounded above in the direction of @xmath67 : the variation of the modulus of the unstable floquet eigenvalue versus the transversal coupling parameter @xmath67 shows that the mobile breather becomes stable again at larger values of @xmath67 , before the second instability at even higher coupling takes place . an important observation is that the pinned discrete breather of the same frequency is _ linearly stable _ at the points in this unstable region for ( 1,0,1 ) mobile breathers . thus this instability can not be ascribed to pinned quasi - collapse effects . the second transition occurs for values of @xmath67 close to , but slightly higher than , the values @xmath184 of the quasi - collapse of the pinned breather of the same frequency . we had already seen in the previous section that the quasi - collapse transition @xmath182 is only very weakly dependent on the value of @xmath6 , and note that the same is true for this mobile breather bifurcation . these results suggest that this second transition is related to quasi - collapsing phenomena . significantly , the stability of the ( @xmath178 ) mobile breather persists for a small interval of coupling values above the pinned breather quasi - collapse . this should be regarded as natural , for the mobile breather is a different solution . note in figure [ fig : mob - stability].b that the modulus of the unstable floquet eigenvalue , in the interior of the unstable region , reaches much higher values than those typical for the first type of instability , and decreases for larger values of @xmath67 , before the breather solution ceases to exist and only plane wave solutions are obtained by our numerical method . note that this behaviour of the unstable floquet eigenvalue also fits well to the main features of the pinned quasi - collapse instability strength , as described by the slope @xmath185 . from now on we will refer to this instability of mobile breathers as the quasi - collapse instability . in the next subsection we characterize both generic types of instability , by looking at the details of the unstable manifold associated with each type . as we will see , pulson states turn out to play a role in the description of typical trajectories on the unstable nonlinear manifolds . first , we analyze the quasi - collapse instability of ( @xmath178 ) mobile breathers . the unstable linear subspace in the tangent space of the fixed point is one - dimensional . the typical instantaneous profile of the ( modulus ) unstable floquet eigenvector driving the instability is shown in figure [ fig : unstabledyn].d . it is an exponentially localized 2d profile which decays asymptotically to zero as @xmath186 , _ i.e. _ it does not excite radiation . these characteristics are shared by the quasi - collapse unstable eigenvector of the pinned breathers , which further reinforce the previous considerations leading us to consider this instability as the mobile counterpart of the pinned quasi - collapse transition . in figure [ fig : unstabledyn].c we have visualized the time evolution of the field modulus contour plot for a typical trajectory on the unstable manifold . this is obtained by direct numerical integration of the equations of motion , from an initial condition in which a small perturbation along the quasi - collapse eigenvector has been added to the unstable fixed point solution . one sees that the breather translational motion slows down , and the energy is transferred to width oscillations . these oscillations turn out to be more irregular , see figure [ fig : qc - evolution ] , than those observed in section [ sec : pinned ] when we inspected typical trajectories on the unstable nonlinear manifold of pinned breathers . 1 cc + the difference in the character of the width oscillations in both ( pinned and mobile ) cases may be ascribed to the presence of an extended background component in the mobile breather solution , which naturally enters into the energy transfer taking place during temporal evolution . the slowing down of the translational motion continues and eventually the breather pins into a convulsive pulson state surrounded by the remaining radiation . now we pay attention to the `` low @xmath67 '' instability of ( @xmath178 ) mobile breathers . the modulus profile of the unstable floquet eigenvector that drives this instability is m - shaped ( bimodal ) , as shown in figure [ fig : unstabledyn].b , and is asymptotic to an extended planewave - like profile as @xmath168 , _ i.e. _ it is not a purely localized perturbation . it is indeed rather different from the quasi - collapse unstable eigenvector analyzed above , which is consistent with the fact that the pinned breather of the same frequency is linearly stable in this region of parameter space . as argued above , this instability is not related to quasi - collapse phenomena , and it does not appear in the region of small values of the salerno parameter @xmath6 , close to the a - l limit . cc a typical trajectory on the unstable manifold associated with this instability is shown in figure [ fig : unstabledyn].a , where we have plotted the time evolution of the field modulus contour plot . we can see there that the mobile breather pins quickly while a small pulse moving backwards is ejected , which spreads and finally mixes with the remaining delocalized background . however some energy is transferred to width oscillations of the pinned breather so that also in this case we observe the formation of pulson states surrounded by the remaining radiation . as the main difference of this behaviour , with respect to the evolution observed on the quasi - collapse unstable manifold , is the ejection of the small moving pulse , we refer to this instability as _ fission_. by increasing the strength of the initial perturbation along the direction of the unstable eigenvector , one observes that the size of the ejected pulse increases . this observation is consistent with the results reported in @xcite , where the evolution of initial moving gaussian pulses in isotropic 2d schrdinger lattices was studied . these numerical experiences lead the authors to conclude that `` the characteristic feature of the discrete quasi - collapse of a moving pulse is the splitting of the initially moving broad pulse into a track of the standing narrow structures ... '' ( sic ) . however , we see from our study of the stability of exact moving discrete breathers that the fission and the quasi - collapse instabilities have different origins and they appear in different regions of parameter space . on the other hand , the ubiquitous phenomenon of width oscillations of pinned localized structures ( pulson states ) can not be ascribed to quasi - collapse . they also appear as the preferred way to allocate excess of ( localization ) energy in regions of parameter space far from the quasi - collapse unstable region . we have studied here the dynamics of exact numerical discrete breathers , both pinned and mobile , in two - dimensional anisotropic nonlinear schrdinger lattices . these solutions are computed using a svd - regularized newton method by continuation from a set of uncoupled 1d chains into increasing non - zero values of the coupling in the transversal direction . we have performed an extensive exploration in the parameter space ( @xmath28 , @xmath64 , @xmath6 ) of breather frequency , transversal coupling and salerno parameter , by computing the floquet spectra of the numerical solutions . we have also computed the breather norm @xmath187 and further corroborate the general validity of the criterion found in @xcite , namely that the partial derivative @xmath188 is positive for stable pinned breathers . furthermore , we have analyzed the dynamics on the quasi - collapse unstable manifold , where the unstable breather experiences a shift in frequency towards the ( higher ) value of the stable breather with the same norm . the excess of energy is coherently transferred to oscillations of the breather width , so that the resulting pulson state is characterized by two frequencies . we have studied discrete breathers moving along the strong coupling direction . these solutions are composed of an exponentially localized core on top of an extended background which is itself the finite sum of a finite set of nonlinear 2d plane waves . the time scales asociated with these plane waves are resonant with the core internal frequency as happens in the 1d case . in particular , the background chooses a finite set of plane waves from a continuous family of resonant solutions . the floquet analysis of these mobile discrete breathers reveals the existence of two distinct types of instability . one is the counterpart , for mobile breathers , of the quasi - collapse experienced by pinned breathers . the other instability occurs in a region of parameter space where pinned breathers are linearly stable . the analysis of the dynamics on the unstable manifold show that the excess of energy is partly transferred to a small moving pulse , ejected from the center of localization , which justifies the designation of a fission instability . however , part of the energy excess is also transferred to width oscillations . the appearance of pulson states far from the quasi - collapse regime indicates that the tendency to allocate energy in the form of width oscillations is a general 2d feature , not exclusively associated to quasi - collapse instabilities . in a future work we will focus on mobility of 2d discrete breathers in an arbitrary lattice direction . the results obtained here shed light about how this mobility can be obtained . in fact , our experiences show that mobility of pinned breathers can be induced based on the existence of the extended background in the numerically exact mobile solution . on the other hand , the results obtained here and the aforementioned future work may help to design and better understand recent numerical experiments reported in @xcite , concerning the interaction between high amplitude pinned breathers and mobile ones . these experiments provides a possible way for routing and blocking mobile discrete breathers via the interaction with the high amplitude pinned ones , resulting in a plausible implementation of logical functions . the authors acknowledge f. falo , yu . kivshar , r.s . mackay and m. peyrard for sharing thoughts , and pointing out some important references to us . jg - g and lmf are grateful to m. johansson and b. malomed for discussions on some issues regarding the `` travelling wave '' ( orthodox ) perspective on discrete breathers . financial support came from mcyt ( project no . bfm2002 00113 ) , dga and bifi . jg - g akcnowledges financial support from the mecyd through a fpu grant . work at los alamos performed under the auspices of the us doe . j. c. eilbeck and m. johansson , in proceedings of _ localization and energy transfer in nonlinear systems _ , edited by l. vzquez , r. s. mackay , and m. p. zorzano ( world scientific , singapore , 2003),p.44 . t. peschel and f. lederer , proceedings of the confererence on nonlinear coherent structures in physics and biology , d.b . duncan and j.c . eilbeck eds . , published on www : http://www.ma.hw.ac.uk/solitons/procs/
recently , localization systems have found widespread application , since they allow to develop a number of new services in both outdoor and indoor environments @xcite . conventional localization systems acquire positional information from a set of observations ; these are usually extracted from noisy wireless signals propagating in harsh environments . unavoidably , this limits the accuracy that can be achieved by such systems . localization accuracy can certainly benefit from the availability of any form of _ a priori _ knowledge . in the technical literature , the sources of prior knowledge commonly exploited are represented by the positions of some specific nodes ( called _ anchors _ ) or by some characteristic of the communication channel ; for instance in @xcite the impact of different factors ( like non - line - of - sight , nlos , propagation and network synchronization ) on localization accuracy have been thoroughly analysed employing cramer - rao bounds ( crbs ) and , when a priori knowledge is available , bayesian cramer - rao bounds ( bcrbs ) . in @xcite multipath propagation in ultra wide band ( uwb ) systems and its effects on time - of - arrival ( toa ) estimation have been investigated by means of crbs or bcrbs . in @xcite crbs and bcrbs for the analysis of cooperative localization techniques have been derived . in recent times , however , some attention has been paid to the possibility of improving accuracy by endowing localization systems with _ map - awareness _ , i.e. , with the knowledge of the _ map _ of the environment in which they operate . specific examples of map - aware algorithms have been developed in @xcite , which propose the adoption of non - linear filtering techniques ( namely , particle filtering and extended kalman filtering ) to embed map - based a priori information in navigation systems . this work evidences that this kind of information plays an important role in improving localization and navigation accuracy ; however , the impact of map - awareness on the performance limits of localization systems is still an open problem . in fact , as far as we know , the technical literature dealing with the fundamental limits of localization accuracy @xcite has not considered this issue yet . in this paper , novel accuracy bounds for map - aware localization systems are developed and their applications to specific environments are analysed . specifically , the bcrb , the extended zik - zakai bound ( ezzb ) and the weiss - weinstein bound ( wwb ) for the above mentioned systems are derived . these bounds , which provide several novel insights into map - aware localization , have the following features : 1 . they are characterized by different tightness / analytical complexity ; 2 . they can be evaluated for any map geometry and , in the cases of the bcrb and the ezzb , admit a closed form for rectangular maps ; 3 . they allow to a ) identify map features ( e.g. , shape , area , etc . ) influencing localization accuracy and b ) quantify such an influence . then , such bounds are evaluated and compared for simple representative environments . our results allow to assess the importance of map - awareness in localization systems in the presence of noisy observations . in particular , they evidence that : 1 . maps should be expected to play a significant role at low signal - to - noise ratios ( snrs ) and in the presence of obstructions ; 2 . in some cases simplified map models can be adopted for localization purposes , since map details have a negligible impact on estimation accuracy . the manuscript is organized as follows . in section [ sec : scenario ] our reference scenario is illustrated and the map model is defined . performance bounds for map - aware localization are derived and evaluated in specific scenarios in sections [ sec : perf_bounds ] and [ sec : num_res ] , respectively . finally , some conclusions are offered in section [ sec : conc ] . _ notations _ : matrices are denoted by upper - case bold letters , vectors by lower - case bold letters and scalar quantities by italic letters . the notation @xmath0 denotes expectation with respect to the random vector @xmath1 ; @xmath2 denotes the trace of a square matrix ; @xmath3 denotes a square matrix with the arguments on its main diagonal and zeros elsewhere ; @xmath4_{i , l}$ ] is the element on the @xmath5-th row and @xmath6-th column of its argument . @xmath7 denotes the @xmath8 identity matrix ; @xmath9 means that the matrix @xmath10 is positive semi - definite . the probability density function ( pdf ) of the random vector @xmath11 evaluated at the point @xmath1 is denoted as @xmath12 , whereas @xmath13 denotes the pdf of a gaussian random vector having mean @xmath14 and covariance matrix @xmath15 , evaluated at the point @xmath1 . @xmath16 denotes the so called indicator function for the set @xmath17 ( it is equal to 1 when @xmath18 and zero otherwise ) . in the following we focus on the problem of localizing a single device , called _ agent _ , in a 2-d environment , i.e. , of estimating its position @xmath19^{t}$ ] , in the presence of the following information : a ) a map representing the a priori knowledge about the environment in which the localization system operates ; b ) an observation vector @xmath20 related to the true position @xmath21 of the agent . in a bayesian framework the map of a given environment can be modelled for localization purposes as a pdf @xmath22 of the random variable @xmath21 , representing the agent position . such a pdf is characterized by a 2-d support @xmath23 having a finite size and consisting of the set of points in the environment not occupied by obstructions ( e.g. , walls , buildings , etc ) . in the absence of other prior knowledge , a natural choice for @xmath22 is a simple uniform model , i.e. , @xmath24 where @xmath25 denotes the area of @xmath23 . this model is referred to as _ uniform map _ in the following and is preferable to other , more detailed , prior models since it can be adopted when only basic knowledge of the environment is available ( e.g. , the floor plan of a building in indoor environments or satellite photos in outdoor environments ) . an alternative statistical model for maps , better suited to mathematical analysis in the bcrb case , is a `` smoothed '' version of ( [ eq : uniform_map ] ) ; this is obtained modifying the uniform pdf in a narrow area around the edges of @xmath23 in order to introduce a smooth transition to zero along the boundaries of its support . note that the adoption of a smoothed uniform model leads to analytical results similar to those found with the uniform model if the transition region is small with respect to the size of @xmath23 . as it will become clearer in the following , for a generic map , a detailed description of the structure of its support @xmath23 should be provided to ease the formulation of the accuracy bounds referring to the map itself . to begin , let us define , for each @xmath26 , the sets @xmath27 and @xmath28 representing the intersection between @xmath23 and an horizontal ( _ h _ ) and vertical ( _ v _ ) line identified by the abscissa @xmath29 and ordinate @xmath30 , respectively . then , let : * @xmath31 and @xmath32 denote the number of connected components of @xmath33 and @xmath34 , respectively ; * @xmath35 and @xmath36 denote the set of the _ odd _ values @xmath37 and @xmath38 @xmath39 , respectively ; * @xmath40 and @xmath41 , with @xmath42 and @xmath43 , denote the length of the connected components of @xmath33 and @xmath34 respectively ; * @xmath44 and @xmath45 denote the set of the _ even _ values @xmath46 and @xmath47 @xmath48 , respectively . * @xmath49 and @xmath50 , with @xmath51 and @xmath52 , denote the length of the connected components of @xmath53 and @xmath54 respectively ( note that @xmath53 and @xmath54 represent the complementary sets of @xmath33 and @xmath34 with respect to @xmath55 ) ; in other words , they denote the horizontal and vertical size of map obstructions , respectively ; * @xmath56 and @xmath57 denote the projection of @xmath23 on the @xmath29 and @xmath30 axis , respectively ; the graphical meaning of these parameters for a specific uniform map is exemplified by fig . [ fig : non_convex_generic_shape ] . in the following , we assume that the localization system is able to acquire a set of noisy observations , collected in a vector @xmath20 and related to the agent position @xmath21 according to a specific statistical model . note that , on the one hand , the results illustrated in sec . [ sub : bcrb ] hold for a general statistical model relating @xmath20 to @xmath21 ; on the other hand , in sec . [ sub : zzb ] and [ sub : wwb ] , it is assumed that @xmath58 for mathematical convenience , where @xmath59 and @xmath60 ( i.e. , @xmath20 is affected by uncorrelated noise ) . it is important to point out that : 1 . the model ( [ eq : obs_model_gaussian ] ) is purposely abstract and does not explicitly refer to a specific localization technique ( e.g. , toa or received signal strength ( rss ) ) or to particular propagation conditions . generally speaking , it is suitable to describe the position estimate @xmath20 generated by a map - unaware and unbiased localization algorithm ; in addition , it leads to useful bounds which unveil the impact of _ map _ modelling ( instead of that of _ observation _ modelling ) on estimation performance ; readers interested in an in - depth analysis on the impact of observation modelling can refer to @xcite . 2 . in principle , the bounds derived in this manuscript can be extended to the case of correlated noise , accounting for the presence of an off - diagonal term @xmath61 in @xmath15 . however , the presence of noise correlation is neglected in the following derivations since : a ) the magnitude of @xmath62 in a real world system depends on several parameters ( the type of measurements processed by the system , the multi - lateration or angulation technique employed , the properties of the propagation channel , etc ) and can not be easily assessed ; b ) our research work evidenced that the presence of noise correlation does not provide significant additional insights on the impact of map - awareness on performance bounds . in this section various bounds about the accuracy of map - aware position estimation are derived and their implications are analysed . given an estimator @xmath63 of @xmath21 based on the observation vector @xmath20 , the bcrb establishes that the bayesian mean square error ( bmse ) matrix @xmath64 satisfies @xmath65 @xcite , where @xmath66 is the bayesian fisher information matrix ( bfim ) associated to the statistical models employed by @xmath63 . this result entails that , for 2-d localization : @xmath67 where @xmath68_{1,1}$ ] , @xmath69_{2,2}$ ] , @xmath70_{1,1}$ ] and @xmath71_{2,2}$ ] . it is not difficult to show that the bfim @xmath66 can be put in the form ( e.g. , see ( * ? ? ? ( 75 ) ) ) @xmath72 where @xmath73^{t}\right\ } \label{eq : bfim_cond}\ ] ] and @xmath74^{t}\right\ } \label{eq : bfim_apriori}\ ] ] represent the contributions originating from the observation vector ( i.e. , the so called _ conditional _ information ) and that due to _ a priori _ information , respectively . the last decomposition allows us to analyse the two above mentioned sources of information in a separate fashion . therefore , in the following , we mainly focus on the evaluation of the matrix @xmath75 ( [ eq : bfim_apriori ] ) , since this term unveils the contribution of map information in the estimation of agent position . as already mentioned in sec . [ sub : scenario_map ] a smoothed uniform map model is adopted in the derivation of @xmath75 ( analytical details are provided in appendix [ apd : proofs_bfim ] ) so that its pdf @xmath22 , unlike its ( discontinuous ) uniform counterpart ( [ eq : uniform_map ] ) , satisfies the regularity condition@xcite @xmath76 required for the evaluation of the bayesian fisher information . moreover it is assumed that : 1 . the smoothing of @xmath22 is described by a smoothing function @xmath77 , which is required to be a continuous and differentiable pdf for which the a priori fisher information ( fi ) @xmath78 can be evaluated ; 2 . the smoothing of @xmath22 along the @xmath29 axis is independent from that along the @xmath30 axis . in appendix [ apd : proofs_bfim ] it is proved that , given the assumptions illustrated above , @xmath75 ( [ eq : bfim_apriori ] ) can be approximated as @xmath79 note that such an expression is independent from the observation model and depends from the smoothing function through its fi @xmath80 only the value to be assigned to @xmath80 is derived for the case of rectangular maps . ] . given @xmath81 , the computation of the bcrb requires the knowledge of @xmath82 ( [ eq : bfim_cond ] ) ( see ( [ eq : j_somma ] ) ) ; for its evaluation , instead of considering a specific observation model , we just assume that the cross - information about @xmath29 and @xmath30 is negligible , so that such a matrix can be put in the form if the specific observation model ( [ eq : obs_model_gaussian ] ) is adopted ; this result will be exploited later for comparison with other bounds derived for the model ( [ eq : obs_model_gaussian ] ) . ] @xmath83 then , the bcrb associated with the pdf @xmath22 can be put in the form ( see ( [ eq : bfim_map_generic_mult_intervals ] ) , ( [ eq : jzp ] ) and ( [ eq : j_somma ] ) ) @xmath84 it can be shown that the last two expressions are exact , and can be put in a closed form ( not involving any integral ) , if @xmath23 is the union of a set of disjoint rectangles ( all having parallel sides ) . from ( [ eq : bfim_map_generic_mult_intervals ] ) and ( [ eq : bcrb_generic ] ) it can be inferred that : 1 . map - aware estimators should be expected to be always more accurate than their map - unaware counterparts ; in fact @xmath81 ( [ eq : bfim_map_generic_mult_intervals ] ) is positive definite , so that the trace of the matrix @xmath85 ( [ eq : bcrb_generic ] ) ( which bounds the accuracy of map - aware estimators ) is smaller than the trace of @xmath86 ( see ( [ eq : jzp ] ) ) which represents the cramer - rao bound ( crb ) in the considered scenario ( and bounds the accuracy of map - unaware estimators ) . note that this is a well - known result ( the contribution of a priori information is positive and lowers the bcrb ) , but may be concealed by the complex analytical form of ( [ eq : bcrb_generic ] ) . 2 . the bcrb ( [ eq : bcrb_generic ] ) exhibits a complicated dependence on the map properties and , in particular , a complex non linear dependence on the map area @xmath25 ; in fact , this parameter can be related to the functions @xmath87 and @xmath88 appearing in the right - hand side ( rhs ) of ( [ eq : bcrb_generic ] ) as @xmath89 however , the following asymptotic result suggests that the bcrb is expected to increase ( and thus the accuracy of map - aware estimation to worsen ) as @xmath25 gets larger : if @xmath90 the bcrb tends to the crb ( since the map support @xmath91 and the contribution coming from prior information vanishes the prior expressed by ( [ eq : uniform_map ] ) becomes `` improper '' ( see ( * ? ? ? 4.2 ) ) . ] ) which always takes on a bigger value ( see point 1 ) . only the smallest elements in the set @xmath88 ( @xmath87 ) significantly contribute to the sum @xmath92 ( @xmath93 ) appearing in ( [ eq : bfim_map_generic_mult_intervals ] ) and ( [ eq : bcrb_generic ] ) and , consequently , may appreciably influence estimation accuracy ; in the limit , if @xmath94 ( @xmath95 ) for some @xmath96 and @xmath30 ( @xmath97 and @xmath29 ) , then the information about @xmath29 ( @xmath30 ) coordinate becomes infinite because the agent is constrained to lie , for those values of @xmath96 and @xmath30 ( @xmath97 and @xmath29 ) , in a single point of the map support . 4 . the matrices @xmath81 ( [ eq : bfim_map_generic_mult_intervals ] ) and @xmath85 ( [ eq : bcrb_generic ] ) can be easily put in a closed form when @xmath87 and @xmath88 are step functions ( i.e. , the map is made by union of rectangles ) . further insights can be obtained taking a specific class of maps into consideration . in particular , here we focus on the case of a map without obstructions , i.e. , characterized by a support @xmath23 such that @xmath98 and @xmath99 @xmath100 ( note that convex maps belong to this class ) ; under these assumptions , it is easy to show that @xmath75 ( [ eq : bfim_map_generic_mult_intervals ] ) can be simplified as @xmath101 from the rhs of the last expression it can be easily inferred that : @xmath102dy,\right.\nonumber \\ & \left.\frac{\sigma_{y}^{3}}{2\mathcal{a_{r}}}\int_{\mathcal{x}}\left[\sum_{m\in\mathcal{n}_{v}^{o}(x)}\zeta\left(\frac{h_{m}(x)}{\sigma_{y}}\right)+\sum_{m\in\mathcal{n}_{v}^{e}(x)}\zeta_{ov}\left(\frac{\triangle h_{m}(x)}{\sigma_{y}},\frac{h_{m-1}(x)}{\sigma_{y}},\frac{h_{m+1}(x)}{\sigma_{y}}\right)\right]dx\right\ } \label{eq : zzb_generic}\end{aligned}\ ] ] 1 . even in this case the dependence of @xmath81 from @xmath25 is complicated . however , analysing ( [ eq : bfim_maps_nogap ] ) , it is easy to note that if @xmath91 ( so that @xmath90 ) the elements of @xmath81 tend to zero , so that the impact of prior information vanishes , as already mentioned above . 2 . the amount of information concerning the agent position along the @xmath29 ( @xmath30 ) axis direction is proportional to @xmath103 ( @xmath104 ) , where @xmath105 ( @xmath106 ) ; the last integral can be interpreted as the area of a `` virtual map '' characterized by the same projection @xmath57 ( @xmath56 ) as the real map , but whose width ( height ) at each point @xmath107 ( @xmath108 ) is the reciprocal of the width ( height ) of the real map . when prior information dominates over conditional information , the map support @xmath23 minimizing the trace of the bcrb under the constraint of a constant area @xmath25 has a square shape . in fact , if @xmath23 is a rectangle whose sides have lengths @xmath109 and @xmath110 , it is easy to show that @xmath111 ; then , if we set @xmath112 , @xmath113 , where @xmath114 and @xmath115 denote two real positive parameters , the trace of the inverse of ( [ eq : bfim_maps_nogap ] ) can be put in the form @xmath116 and is minimized for @xmath117 ( i.e. , for @xmath118 ) . 4 . in ( [ eq : bfim_maps_nogap ] ) the only contribution coming from the smoothing function @xmath77 is represented by the factor @xmath80 . further comments about the bcrb ( [ eq : bcrb_generic ] ) will be provided in section [ sec : num_res ] , where this bound is computed for some specific maps . finally , it is important to point out that : a ) in localization problems bcrb s often provide useful insights @xcite , but these bounds are usually loose for low snr conditions @xcite , i.e. , when a priori information ( the map in this context ) plays a critical role due to the poor quality of observations ; b ) the bcrb analysis requires the adoption of the smoothed uniform pdf model for prior information ( see section [ sub : scenario_map ] ) . these considerations motivate the search for other bounds and , in particular , for the ezzb , which is usually tighter than the bcrb at low snrs @xcite and does not require ( [ eq : regularity_condition ] ) to hold ( so that the uniform model ( [ eq : uniform_map ] ) can be employed as it is ) ; this is further discussed in the following section . similarly to ( [ eq : bcrb ] ) , the ezzb for 2d localization can be expressed as @xmath119 where ( see appendix [ apd : proofs_zzb ] ) @xmath120p_{\text{\tiny{min}}}^{\mathbf{z}}\left(\boldsymbol{\rho},\boldsymbol{\rho}+h\mathbf{e}_{\nu}\right)h\ , d\boldsymbol{\rho}dh\label{eq : zzb_def}\ ] ] @xmath121 , @xmath122^{t}$ ] , @xmath123^{t}$ ] , @xmath124 represents the error probability referring to the likelihood ratio test in a binary detection problem involving the hypotheses @xmath125 @xmath126 and the observation @xmath20 ; finally , the integration domain is @xmath127 the ezzb ( [ eq : zzb])-([eq : zzb_def ] ) deserves the following comments : 1 . its formulation is obtained from the standard formulation ( see ( * ? ? ? * eq . ( 32 ) ) ) removing the so - called valley - filling function and the maximization operator , as shown in appendix [ apd : proofs_zzb ] ; these modifications simplify the derivation of the bound at the price of a small reduction in its tightness ; 2 . it exhibits a complicated dependence on the observation model , which comes into play in the evaluation of the probability @xmath128 ; 3 . it can not be put in a form similar to ( [ eq : j_somma ] ) , so that , generally speaking , the contribution coming from conditional information ( i.e. , observations ) can not be easily separated from that associated with a priori information ( map awareness in this case ) . as shown in appendix [ apd : proofs_zzb ] , for a uniform map and the observation model ( [ eq : obs_model_gaussian ] ) the ezzb ( [ eq : zzb_def ] ) can be put in the form ( [ eq : zzb_generic ] ) shown at the top of this page where @xmath129 @xmath130 an accurate analysis of the ezzb ( [ eq : zzb_generic ] ) and a comparison with the bcrb ( [ eq : bcrb_generic ] ) lead to the following conclusions : 1 . the ezzb ( [ eq : zzb_generic ] ) exhibits a similar dependence as the bcrb ( [ eq : bcrb_generic ] ) in terms of the quantities @xmath131 , @xmath132 . in fact , if @xmath131 and/or @xmath133 increases , the trace of matrix @xmath134 in ( [ eq : zzb_generic ] ) gets larger , since @xmath135 ( [ eq : f_rho_def ] ) is a monotonically increasing function . 2 . in the same environment the ezzb ( [ eq : zzb_generic ] ) takes on larger values than the bcrb ( [ eq : bcrb_generic ] ) and results to be a bound tighter to the performance of optimal estimators . this is due to the fact that the terms @xmath135 appearing in the rhs of ( [ eq : zzb_generic ] ) give the ezzb approximately the same behaviour of the bcrb ; however , the ezzb also contains terms based on @xmath136 ( [ eq : f_gap_def ] ) , which is a positive valued function . the ezzb ( [ eq : zzb_generic ] ) , unlike the bcrb , is influenced by the shape of map obstructions ( through the @xmath136 function and the @xmath137 , @xmath138 parameters ) . the two terms of ( [ eq : zzb_generic ] ) depending on the function @xmath136 ( [ eq : f_gap_def ] ) are influenced by the spatial extensions @xmath137 and @xmath138 of the obstructions separating _ consecutive _ map segments ( which are characterized by the lengths @xmath139 and @xmath140 , respectively ) ; in particular , the impact of obstructions is significant when @xmath49 ( @xmath141 ) is greater than one of @xmath139 ( @xmath140 ) . 5 . the integrals in the rhs of ( [ eq : zzb_generic ] ) can be easily put in a closed form when @xmath87 and @xmath88 are step functions ( i.e. , the map is made by union of rectangles ) . similarly to the bcrb , further insights can be obtained considering the class of maps without obstructions . in this case the ezzb ( [ eq : zzb_generic ] ) simplifies as @xmath142^{2}}{2\mathcal{\mathcal{a_{r}}}\left[\lambda_{x}\left(h_{x},\mathcal{r}\right)-\exp\left(-\frac{h_{x}^{2}}{2\sigma_{x}^{2}}\right)\gamma_{x}\left(h_{x},\mathcal{r}\right)\right]},\sup_{h_{y}\in\mathbb{r}}\frac{\left[h_{y}\exp\left(-\frac{h_{y}^{2}}{8\sigma_{y}^{2}}\right)\lambda_{y}\left(h_{y},\mathcal{r}\right)\right]^{2}}{2\mathcal{\mathcal{a_{r}}}\left[\lambda_{y}\left(h_{y},\mathcal{r}\right)-\exp\left(-\frac{h_{y}^{2}}{2\sigma_{y}^{2}}\right)\gamma_{y}\left(h_{y},\mathcal{r}\right)\right]}\right\ } \label{eq : wwb_generic}\end{aligned}\ ] ] @xmath143 it is also worth mentioning that , if the map support @xmath23 can be expressed as the union of a set of disjoint rectangles ( all having parallel sides ) , the ezzb ( [ eq : zzb_maps_nogap ] ) can be simplified further , obtaining a closed - form bound . the result ( [ eq : zzb_maps_nogap ] ) deserves the following comments : 1 . the structure of its rhs is similar to that of the equivalent bfim ( [ eq : bfim_maps_nogap ] ) . despite the differences between the integrand functions , our numerical results show that in this case both the ezzb and the bcrb are tight and that the ezzb does not provide additional hints on localization accuracy with respect to the bcrb ( see section [ sec : num_res ] ) . 2 . the ezzb predicts the correct variance of map - aware estimation when the snr is very low whereas a specific assumption about the value of @xmath80 is required to obtain the same result in the bcrb case . for instance , let us consider a uniform map whose support @xmath23 is a rectangle whose sides have lengths @xmath109 and @xmath110 ; with model ( [ eq : obs_model_gaussian ] ) and for @xmath144 and @xmath145 ( i.e. , for a snr approaching zero ) , the bmse of the minimum mean square error ( mmse ) estimator tends to @xmath146 . the ezzb provides this exact result since the rhs of ( [ eq : zzb_maps_nogap ] ) can be put in the closed form @xmath147 and @xmath148 for a snr @xmath149 approaching zero around @xmath150 . ] . this result can be compared with its bcrb counterpart ; in this case , from ( [ eq : bfim_maps_nogap ] ) it can be inferred that @xmath151 , so that @xmath152 should be selected to match the correct variance . the ezzb ( [ eq : zzb_maps_nogap ] ) confirms that map - aware estimators should be expected to be more accurate than their map - unaware counterparts , at least for maps without obstructions : for instance , if a rectangular map of sides @xmath109 , @xmath110 is considered , then ( [ eq : zzb_rectangle_map ] ) is easily proved to be always smaller than @xmath153 ( because @xmath154 ) which is the trivial ezzb ) after proving that @xmath155 for @xmath156 . ] for the observation model ( [ eq : obs_model_gaussian ] ) when the map is unavailable ( i.e. , when no prior information is available ) . this derives from the fact that a priori information lowers the ezzb ( similarly to what occurs in the case of the crb and the bcrb ) ; this is a well - known result , but may not be evident in ( [ eq : zzb_maps_nogap ] ) . 4 . similarly to the bcrb , the dependence on the map area is complicated ; however , it should be expected that the elements of @xmath134 get larger when @xmath157 increases as suggested by the following asymptotic result : when @xmath90 the contribution coming from a priori information vanishes and the ezzb ( [ eq : zzb_rectangle_map ] ) tends to @xmath153 , i.e. , to the map - unaware ezzb , which is larger ( see point 3 ) . as already mentioned above , the ezzb allows us to understand the role played by some map features , which do not have any impact on the bcrb . however , our numerical results evidence that , for complex maps and , in particular , at very low snrs , this bound turns out to be somewhat far from the performance offered by a mmse estimator . this has motivated the derivation of the wwb @xcite , which can be tighter than the ezzb at very low - snrs . similarly to the bcrb and the ezzb ( see ( [ eq : bcrb ] ) and ( [ eq : zzb ] ) , respectively ) , the wwb , for 2-d localization , can be expressed as @xmath158 where @xcite @xmath159^{2}\right\ } } \label{eq : wwb_def}\ ] ] and @xmath160 @xmath161 are functions of the @xmath162 parameter ; moreover @xmath163 is a likelihood ratio and @xmath121 . note that ( [ eq : wwb_def ] ) can be obtained from ( * ? ? ? * eqs . ( 7)-(8 ) ) setting @xmath164 , @xmath165 for the wwb optimization parameters and @xmath166 , since we are interested in a bound on the variance ( not on the co - variance ) of the estimates of @xmath21 . as shown in appendix [ apd : proofs_wwb ] , for a uniform map and the observation model ( [ eq : obs_model_gaussian ] ) the wwb ( [ eq : wwb_def ] ) can be put in the form ( [ eq : wwb_generic ] ) shown at the top of the next page , where @xmath167 , @xmath168 denote the parameters to be optimized to make the bound tighter , @xmath169 and @xmath170 are functions of the parameters @xmath167 , @xmath168 and of the geometrical features of the map support @xmath23 ( dual expressions hold for @xmath171 and @xmath172 ) ; finally @xmath173}(h_{x})\label{eq : omega_def}\ ] ] @xmath174\\ w_{2 } & h\in[\triangle w+w_{2};\triangle w+w_{1}]\\ w_{tot}-h & h\in[\triangle w+w_{1};w_{tot } ] \end{cases}\label{eq : omega_ov_def}\end{gathered}\ ] ] @xmath175 note that the structure of * @xmath176 * ( [ eq : lambda_def ] ) is similar to that of some terms contained in the rhs of the ezzb ( [ eq : zzb_generic ] ) ; however , in the wwb the noise variances appear in some exponential functions ( [ eq : wwb_generic ] ) only . unfortunately , it is hard to infer from ( [ eq : wwb_generic ] ) any conclusion about the dependence of the wwb on the features of a given map ; this is mainly due to the presence of the @xmath177 operators and to the discontinuous behaviour of the functions @xmath178 and @xmath179 . despite this , the wwb may play an important role , since , as shown in section [ sec : num_res ] for specific maps , it is tighter than the bcrb and the ezzb at low - snrs . in the following section the general bounds provided above are evaluated for specific maps , so that some additional insights about the ultimate accuracy achieved by map - aware estimators can be obtained . of a `` l - shaped '' map ( map # 2).[fig : map_2d_lshape],width=288 ] of a map modelling the floor of a big building ( map # 3).[fig : map_2d_indoor_building],width=288 ] the bounds illustrated in the previous section have been evaluated for the following 3 specific maps : 1 . map # 1 : a unidimensional map whose support @xmath23 consists of @xmath180 disjoint segments ( denoted @xmath181 and @xmath182 in the following ) having the same length @xmath183 and separated by @xmath184 meters ; 2 . map # 2 : the `` l - shaped '' 2-d map shown in fig . [ fig : map_2d_lshape ] and fully described by the set @xmath185 @xmath186 of geometrical parameters ; 3 . map # 3 : the 2-d map shown in fig . [ fig : map_2d_indoor_building ] ; note that such a map models the floor of a large building and is quite complex , so that its geometrical parameters are not given and analytical bounds are not evaluated for it ; instead , some numerical results generated by means of computer simulations will be illustrated . it is important to point out that maps # 2 and # 3 are examples of 2-d _ rectangular _ uniform maps , i.e. , each of them is characterised by a support @xmath23 that can be represented as the union of non - overlapping rectangles , all having parallel sides ; such non - overlapping rectangles form the so called _ rectangular covering _ of the map support @xmath23 @xcite . note that any map characterised by a support of _ arbitrary _ shape can be always approximated , up to an arbitrary precision , by a rectangular map if a proper covering , containing a sufficiently large number of rectangles , is selected . for this reason , maps # 2 and # 3 are relevant and practical examples . map # 1 can be also deemed relevant from a practical point of view since it can be thought of as a transversal slice of a 2-d map ( consisting , for instance , of two road lanes or two rooms facing each other ) ; in addition , such a simple scenario allows to acquire more insights than maps # 2 and # 3 , as shown later . in all cases the observation model ( [ eq : obs_model_gaussian ] ) has been adopted ( with @xmath187 and @xmath188 in the 1-d and 2-d scenarios , respectively ) and the tightness of the bounds has been assessed comparing them with the root mean square error ( rmse ) performance of optimal estimators evaluated by means of extensive computer simulations . in addition , in accordance with the theoretical arguments of paragraph [ sub : zzb ] ( see the comments to ( [ eq : zzb_maps_nogap ] ) ) , @xmath152 has been selected in the evaluation of all the bcrb results illustrated in this section . finally , in our simulations @xmath189 realizations of the random parameter @xmath21 have been generated according to a uniform distribution over the map support @xmath23 ; then gaussian noise samples ( characterized by @xmath190 if not explicitly stated ) have been added to generate the noisy observations , according to ( [ eq : obs_model_gaussian ] ) . * bounds - * the bcrb , the ezzb and wwb for map # 1 are ( see ( [ eq : bcrb_generic ] ) , ( [ eq : zzb_generic ] ) , and ( [ eq : wwb_generic ] ) , respectively ) @xmath191 @xmath192\label{eq : map1_zzb}\ ] ] and @xmath193 , respectively , where @xmath194}\label{eq : map1_wwb}\end{aligned}\ ] ] @xmath195 is the snr , @xmath196 , @xmath197 and @xmath198 . it is important to point out that that no approximations have been adopted in the derivations of the bcrb , the ezzb and the wwb for the considered 1-d scenario , so that ( [ eq : map1_bcrb])-([eq : map1_wwb ] ) represent _ exact _ bounds ( of course , a smoothed uniform map has been assumed for the bcrb only ) . + * estimators - * map - aware localization can be accomplished exploiting the mmse estimator @xmath199 where @xmath200 is a normalization factor , or the maximum - a - posteriori ( map ) estimator @xmath201 on the contrary , if the map is unknown , the ( map - unaware ) maximum likelihood ( ml ) estimator @xmath202 can be used ( obviously , the resulting estimate has the same variance @xmath203 as the observation noise ) . note that the rmse performance of these estimators is influenced by the geometrical parameters @xmath183 and @xmath184 and by the noise standard deviation @xmath204 . + * numerical results * * - * figs . [ fig : map1_vs_noise ] and [ fig : map1_vs_deltax ] compare the square roots of the bcrb , ezzb and wwb bounds @xmath205 , @xmath206 , @xmath207 and the rmse of @xmath208 , @xmath209 and @xmath210 versus @xmath149 ( with @xmath183 and @xmath184 fixed ) and @xmath184 ( with @xmath183 and @xmath204 fixed ) , respectively . for the mmse ( [ eq : mmse_estimator_map_1 ] ) , map ( [ eq : map_estimator_map_1 ] ) and ml estimators . map # 1 , @xmath211 and @xmath212 are assumed . the bcrb ( [ eq : map1_bcrb ] ) , ezzb ( [ eq : map1_zzb ] ) and wwb ( [ eq : map1_wwb ] ) are shown for comparison.[fig : map1_vs_noise],width=336 ] for the mmse ( [ eq : mmse_estimator_map_1 ] ) , map ( [ eq : map_estimator_map_1 ] ) and ml estimators . map # 1 , @xmath211 and @xmath190 are assumed . the bcrb ( [ eq : map1_bcrb ] ) , ezzb ( [ eq : map1_zzb ] ) and wwb ( [ eq : map1_wwb ] ) are shown for comparison.[fig : map1_vs_deltax],width=336 ] these results evidence that : 1 . the bcrb is tight at high snrs only ; in such conditions all the bounds and the rmse of both map - aware and map - unaware estimators tend to @xmath204 ( see fig . [ fig : map1_vs_noise ] ) : conditional knowledge dominates . 2 . unlike the bcrb , the ezzb and , in particular , the wwb exhibit a dependence on @xmath184 and are much tighter ( see fig . [ fig : map1_vs_deltax ] ) to the performance of optimal estimators and , for this reason , represent more useful tools for predicting system accuracy . 3 . the role played by map knowledge in estimation accuracy becomes significant as @xmath149 decreases , i.e. , as the quality of the observations gets worse ; indeed map knowledge prevents the bounds and the rmse of the map - aware estimators from diverging ( note that in fig . [ fig : map1_vs_noise ] the map - unaware rmse diverges for @xmath213 ) . the performance gap among the mmse , the map and especially the ml estimators is significant for low snrs ( see fig . [ fig : map1_vs_noise ] ) . if the gap @xmath184 is small ( i.e. , if the segments @xmath181 and @xmath182 are close ) , the rmse of the considered estimators is low . on the contrary , if @xmath184 increases , the rmse gets larger , reaches a maximum and then decreases , tending to @xmath214 ( see fig . [ fig : map1_vs_deltax ] ) . this behaviour can be explained as follows : if the noise level is large with respect to @xmath184 , the estimate of the agent position may belong to the wrong segment , thus resulting in a large error . however , a further increase of @xmath184 entails a reduction of the rmse , since , for a large spacing between @xmath181 and @xmath182 , it is unlikely that the wrong segment is selected in map - aware estimation . and @xmath183 for the mmse estimator ( [ eq : mmse_estimator_map_1 ] ) . map # 1 and @xmath190 are assumed . the data has been smoothed to remove simulation noise and improve readability.[fig : mmse_map1_surface],width=288 ] further insights are provided by fig . [ fig : mmse_map1_surface ] , which shows the rmse versus @xmath183 and @xmath184 for the mmse estimator ( note that the mmse data shown in fig . [ fig : map1_vs_deltax ] can be extracted from fig . [ fig : mmse_map1_surface ] setting @xmath211 ) . in fact , the numerical results shown in this figure evidence that : 1 . for @xmath215 the rmse decreases only mildly as @xmath184 gets larger and the main geometric feature of the map affecting estimation accuracy is the area of @xmath23 ; in this case a map - aware estimator still provides an advantage over map - unaware counterparts ( the maximum rmse for the first estimator is @xmath216 , as shown in fig . [ fig : mmse_map1_surface ] , and is smaller than @xmath190 , which represents the rmse for the ml estimator , as shown in fig . [ fig : map1_vs_deltax ] ) . 2 . for @xmath217 the rmse reaches its maximum for @xmath218 ; for this reason , any combination of the geometric parameters and noise level satisfying such equality should be avoided . based on the numerical results shown in fig . [ fig : mmse_map1_surface ] , we can draw some observations about an indoor localization system operating in two adjacent rooms , each having width @xmath219 and separated by a wall having thickness @xmath220 , and processing noisy observations characterized by @xmath190 ( realistic value for rss systems , see @xcite ) . from fig . [ fig : mmse_map1_surface ] we can infer that , even if @xmath221 , since @xmath222 is appreciably smaller than @xmath223 , the advantage in terms of rmse provided by a map modelling wall obstructions is small with respect to that offered by a map modelling the environment as a single connected room having width @xmath224 . in fact , in fig . [ fig : mmse_map1_surface ] , if the presence of the wall is accounted for , the map - aware estimator performance is associated with the point @xmath225 , whereas , if it is not , with the point @xmath226 . unless the room size @xmath219 is very small ( i.e. , @xmath227 ) the two `` operating points '' will be very close ( due to the small value of @xmath222 ) and the resulting rmse values will be similar so that , in this case , a map modelling wall obstructions is not very useful . note also that , if the quality of observations improves substantially ( i.e. , @xmath228 ) , map information plays a less significant role than the conditional information provided by such observations , so that a detailed a priori map modelling walls is not useful in this case , too . on the contrary , maps may play an important role in localization in outdoor environments , where the region characterized by @xmath217 and @xmath229 in the @xmath230 plane is easier to reach , since noisy observations and large obstructions should be expected ; for this reason , in this case the presence of the obstructions in map modelling should not be neglected . @xmath231},\right.\left.\sup_{h_{y}\in\mathbb{r}}\frac{h_{y}^{2}\exp\left(-\frac{h_{y}^{2}}{4\sigma^{2}}\right)\lambda_{y}^{2}\left(h_{y},\mathcal{r}\right)}{2\mathcal{a_{r}}\left[\lambda_{y}\left(h_{y},\mathcal{r}\right)-\exp\left(-\frac{h_{y}^{2}}{2\sigma^{2}}\right)\gamma_{y}\left(h_{y},\mathcal{r}\right)\right]}\right\ } \label{eq : map2_wwb}\end{aligned}\ ] ] * bounds - * let us now focus on map # 2 . the bcrb and the ezzb in this case are ( see ( [ eq : bcrb_generic ] ) and ( [ eq : zzb_generic ] ) , respectively ) @xmath232 @xmath233,\right.\notag\\ & \quad\left.\frac{\sigma^{2}}{2\rho_{\mathcal{a}}}\left[w_{1}\zeta\left(\frac{h_{1}+h_{2}}{\sigma}\right)+w_{2}\zeta\left(\frac{h_{1}}{\sigma}\right)\right]\right\ } \label{eq : map2_zzb}\end{aligned}\ ] ] respectively , where @xmath234 . the wwb derived from ( [ eq : wwb_generic ] ) is shown in ( [ eq : map2_wwb ] ) at the top of the next page ; in ( [ eq : map2_wwb ] ) the @xmath176 and @xmath235 functions have the expressions @xmath236 @xmath237 respectively ( dual expressions hold for @xmath171 and @xmath172 ) . note that @xmath238 and @xmath239 ( @xmath240 and @xmath241 ) play a symmetric role in ( [ eq : map2_bcrb ] ) , ( [ eq : map2_zzb ] ) and ( [ eq : map2_wwb ] ) ; this suggests that a change in the geometrical features improving @xmath242 will reduce @xmath243 and vice versa , when the overall area @xmath25 of the map is kept constant . + * estimators - * map - aware localization can be accomplished exploiting the mmse estimator @xmath244=(1/c(\mathbf{z}))\iint_{\mathcal{r}}\mathbf{p}\,\mathcal{n}\left(\mathbf{z};\mathbf{p},\boldsymbol{\sigma}\right)d\mathbf{p}\label{eq : mmse_estimator_map_2}\ ] ] where @xmath245 . on the contrary , the ( map - unaware ) ml estimate is given by @xmath246 . + * numerical results - * fig . [ fig : mmse_map2 ] shows the square roots of the bcrb components @xmath205 , @xmath247 , its trace @xmath248 and the rmse of the estimators @xmath249 , @xmath250 , @xmath251 and @xmath252 ( the ezzb ( [ eq : map2_zzb ] ) and wwb ( [ eq : map2_wwb ] ) are not shown to ease the reading ) ; @xmath190 , @xmath253 and @xmath254 have been assumed in this case . in addition , when increasing @xmath241 , @xmath238 is reduced according to the law @xmath255 so that the equality @xmath256 always holds . from these results it is inferred that : 1 . as already noted for map # 1 , map - unaware estimation , which is unaffected by a change in the geometrical features of the map , provides a worse accuracy than its map - aware counterpart ( the rmse of ml estimator is @xmath257 larger than that of its mmse counterpart ) . 2 . increasing @xmath241 and decreasing @xmath238 changes the aspect ratio of a portion of the map ( see fig . [ fig : map_2d_lshape ] ) ; this results in an appreciable reduction in the rmse of the parameter @xmath30 , but also in a mild increase of the rmse of the parameter @xmath29 . therefore , the improvement along one direction is compensated for by the worsening in the orthogonal direction , so that no significant advantage for the rmse of @xmath251 is found . the last result suggests that when the accuracy in the estimation of a given element of @xmath21 is improved maintaining @xmath25 constant , a reduction in the estimation accuracy of the other element should be expected , so that that the overall accuracy does not change much . for the mmse estimator ( [ eq : mmse_estimator_map_2 ] ) . map # 2 ( fig . [ fig : map_2d_lshape ] ) , @xmath190 and @xmath256 are assumed . the bcrb components @xmath205 , @xmath247 , @xmath248 ( see ( [ eq : map2_bcrb ] ) ) and the rmse for the ml estimator are also shown for comparison.[fig : mmse_map2],width=336 ] * observation model * * - * finally , let us analyse the rmse performance of localization systems operating in the environment described by map # 3 in two cases : a ) the map - aware estimator knows map # 3 exactly ; b ) the map - aware estimator assumes that the localization environment is described by the bounding box of map # 3 . note that the last case is significant from a practical point of view , since , in the absence of detailed information about the floor of a building , the bounding box represents the best a priori model which can be adopted to describe the floor itself . + in addition , in this paragraph model ( [ eq : obs_model_gaussian ] ) is compared against a more specific counterpart , called `` ranging observation model '' in the following ; for such modelling we assume that 1 ) the observations are acquired by 4 anchors placed in the map corners ; 2 ) the observation provided by the @xmath5-th anchor can be expressed as @xmath258 with @xmath259 , where @xmath260 represents gaussian noise with variance @xmath203 and @xmath261 represents the @xmath5-th anchor position . the scenario considered in this example is helpful to assess 1 ) to what extent a detailed knowledge of the geometrical features of a map can improve localization accuracy with respect to the availability of a simplified map model ( like a bounding box model ) and 2 ) if a localization system based on the gaussian model ( [ eq : obs_model_gaussian ] ) can exhibit a similar behaviour as that of a localization system based on ( [ eq : obs_model_ranging ] ) , at least in the considered cases . * * + * estimators * * - * given the model ( [ eq : obs_model_ranging ] ) , the map estimator can be expressed as @xmath262 where @xmath263^{t}$ ] and @xmath264 . if the observation model ( [ eq : obs_model_gaussian ] ) is adopted in place of ( [ eq : obs_model_ranging ] ) , the map estimator @xmath265 is obtained , where @xmath266 . + * numerical results * * - * fig . [ fig : mmse_map3 ] compares the rmse performance achieved by the map estimators ( [ eq : map_estimator_map_3_toa ] ) and ( [ eq : map_estimator_map_3_gaussian ] ) versus @xmath204 in cases a and b. these results show that : 1 . surprisingly the selection of the map model ( map # 3 in case a or its bounding box in case b ) does not affect the mmse estimation performance , whatever the model for the observations . this result can be motivated as follows . for each value of @xmath204 , there is some portion of the map where the combination of noise level and geometrical features ( see the analysis of map # 1 ) entails a large rmse and where a detailed map ( map # 3 ) does not provide a substantial advantage over simpler maps ( e.g. , the bounding box ) ; since the bayesian rmse results from an average of the agent position over the entire map , it is biased by such portions of the map for each value of @xmath204 . such an effect influences also any global bayesian bound , like the bcrb , the ezzb and the wwb : their values result from an average over the entire parameter space ( in this case @xmath23 ) and can be strongly influenced by the parameter values ( in this case , the values of @xmath21 ) characterized by the largest errors @xcite . note also that local bounds ( e.g. , crb ) , which model the parameters to be estimated as deterministic variables , do not suffer from this problem , but can not exploit a priori information and represent performance limits for unbiased estimators only @xcite . in our case study a detailed knowledge of map # 3 ( case a ) provides , in various subsets of @xmath23 , a substantial improvement with respect to the knowledge of the bounding box ( case b ) , but the effects of such subsets are then `` averaged out '' when evaluating the bayesian rmse , as shown in fig . [ fig : mmse_map3 ] . 2 . the adoption of the gaussian observation model ( [ eq : obs_model_gaussian ] ) entails a worse accuracy than that provided by the ranging observation model ( [ eq : obs_model_ranging ] ) . this can be related to the placement of the anchors and to their number ; in fact , range observations coming from anchors displaced with a constant angular spacing allow to average out the measurement errors better than the positional observation of the gaussian model ( [ eq : obs_model_gaussian ] ) when the same noise variance @xmath203 is assumed in both models and one about @xmath30 ) are considered since 2-d localization is performed . ] . if the offset between the curves referring to the gaussian and the ranging model is neglected , our results show that both curves exhibit the same behaviour , when @xmath204 is varied ; this is an hint to the fact that the insights obtained from the bounds derived for the model ( [ eq : obs_model_gaussian ] ) may be useful also in the analysis of systems employing different observation models ( like ( [ eq : obs_model_ranging ] ) ) . finally , comment 1 ) suggests that , when assessing the improvement in localization accuracy coming from map awareness , the evaluation of the overall rmse does not provide sufficient information about the importance of a given map . in fact , if the effects of different portions of the maps are analysed , it is found that some of them play a more relevant role than other ones ; in this perspective , the insights and the intuition acquired from the results referring to map # 1 are very useful , given that map # 1 can be thought of as a transversal slice of a subset of any 2-d map . for the map estimators ( [ eq : map_estimator_map_3_toa ] ) and ( [ eq : map_estimator_map_3_gaussian ] ) , using either the gaussian observation model ( see ( [ eq : obs_model_gaussian ] ) ) or the ranging observation model ( see ( [ eq : obs_model_ranging ] ) ) . the a priori map is map # 3 ( case a ) or its bounding box ( case b ) . the rmse of ml estimators is also shown for comparison . [ fig : mmse_map3],width=336 ] in this manuscript the impact of map awareness on localization performance has been investigated from a theoretical perspective , evaluating different accuracy bounds . such bounds provide some general indications about the role and importance of this form of a priori information . our study has evidenced that , unluckily , the tighter is an accuracy bound , the more complicated is its dependence on the geometric parameters of maps . this has motivated the analysis of the developed bounds in specific environments ; in particular , three different maps have been considered and , for each of them , the accuracy bounds have been evaluated and compared with the performance offered by map - aware and map - unaware estimators . our analytical and numerical results have evidenced that : a ) map - aware estimation accuracy can be related to some features of the map ( e.g. , its shape and area ) even though , in general , the relation is complicated ; b ) maps are really useful in the presence of some combination of low snrs and specific geometrical features of the map ( e.g. , the size of obstructions ) ; c ) for some combinations of snrs and geometrical features , there is no need of refined maps since knowledge of map details provides only a negligible performance gain ; d ) in a given environment the ezzb and , in particular , the wwb are usually much tighter than the bcrb ; e ) in the cases of maps containing many features , the importance of map awareness is better captured by the analysis of subsets of those map . future improvements include the development of map - aware bounds for specific observation models ; in particular models including bias due to the non - line of sight ( nlos ) propagation may unveil important insights about map - aware localization systems operating in nlos conditions . finally , it is important to point out that all the bounds derived in this manuscript can be also exploited for other estimation problems ( provided that the vector of random parameters to be estimated is characterized by a uniform a priori pdf ) and can be easily generalized to the case of estimation of an @xmath267-dimensional parameter vector , with @xmath268 . in this appendix the derivation of ( [ eq : bfim_map_generic_mult_intervals ] ) and ( [ eq : bfim_maps_nogap ] ) is sketched . to begin , we consider a real function @xmath77 having the following properties : aaa it is continuous and differentiable ; @xmath269 ; @xmath270 ; @xmath271 ; its support is the interval is the same as that adopted for @xmath21 and @xmath20 ( see sec . [ sec : scenario ] ) . ] @xmath272 $ ] ; @xmath273 . these properties have the following implications : a ) @xmath77 is a pdf function ( see p.1-p.3 ) ; b ) an a - priori fi @xmath80 can be computed for it ( the regularity condition p.4 ensures the fi existence ) @xcite ; c ) @xmath274 can be used to model bounded statistical distributions ( i.e. , maps ) , eventually after some scaling and translation ( see p.5 , p.6 ) . as far as the last point is concerned , it is worth mentioning that @xmath275 with @xmath276 , is still a pdf function and its fi is @xmath277 . any function @xmath274 sharing the properties p.1-p.6 represents a `` smoothing function '' ; specific examples of smoothing functions can be found in @xcite . any smoothing function @xmath77 can be used to define the analytical model of a smoothed uniform map . to show how this can be done , we start considering a unidimensional ( 1-d ) scenario first , where the agent position to estimate is the scalar @xmath29 . the support @xmath278 of a 1-d uniform map can be always represented as the union of @xmath279 disjoint segments spaced by @xmath280 segments where the map pdf @xmath281 is equal to 0 . let the @xmath279 segments of the support @xmath23 be indexed by the odd numbers of the set @xmath282 and let @xmath283 ( @xmath284 ) denote the centre ( width ) of the @xmath96-th segment , with @xmath285 . then the map pdf @xmath281 associated with this scenario can be expressed as @xmath286 where @xmath287 . the a priori fi associated with @xmath281 is given by @xmath288 let us extend these results to 2-d maps . exploiting some definitions given in section [ sec : scenario ] and assuming that the smoothing along @xmath29 and that along @xmath30 are independent , the pdf for an arbitrary smoothed uniform 2-d map can be put in the form . ] @xmath289 where @xmath290 ( @xmath291 ) and @xmath40 ( @xmath41 ) denote the centre and the length , respectively of the @xmath96-th ( @xmath97-th ) segment . let us now compute the elements of the bfim @xmath75 associated with this pdf ( see ( @xmath292 ) ) . the analysis of ( [ eq : pdf_2d_map ] ) can be connected to that of ( [ eq : pdf_1d_multirect ] ) exploiting the formula ( iterated expectation ) @xmath293_{1,1 } & \triangleq\mathbb{e}_{\mathbf{p}}\left\ { \left(\frac{\partial\ln f(\mathbf{p})}{\partial x}\right)^{2}\right\ } \nonumber \\ & = \mathbb{e}_{y}\left\ { \mathbb{e}_{x|y}\left\ { \left(\frac{\partial\ln f(\mathbf{p})}{\partial x}\right)^{2}\right\ } \right\ } \label{eq : bfim_2d_generic_map_x}\end{aligned}\ ] ] and adopting the approximation @xmath294 where @xmath295 . it is worth noting that in ( [ eq : pdf_2d_map_approx ] ) the smoothing along the @xmath30 coordinate is ignored . this is a reasonable approximation whenever the smoothing only modifies a narrow area of the edges of @xmath23 compared to the area of the flat regions of ( [ eq : pdf_2d_map ] ) ; the selection of a proper smoothing function ensures that such a condition holds ( see sec . [ sub : scenario_map ] and @xcite ) . + thanks to ( [ eq : pdf_2d_map_approx ] ) , the evaluation of the inner expectation in ( [ eq : bfim_2d_generic_map_x ] ) is equivalent to that of the fi referring to a 1-d map which consists of @xmath296 segments having widths @xmath88 and centred around the points @xmath297 ; for this reason , from ( [ eq : pdf_1d_multirect])-(@xmath298 ) it is easily inferred that @xmath299 substituting the last result in the rhs of ( [ eq : bfim_2d_generic_map_x ] ) produces @xmath293_{1,1 } & \simeq\frac{\mathrm{j}_{s}}{\mathcal{\mathcal{a_{r}}}}\int_{\mathcal{y}}\sum_{n\in\mathcal{n}_{h}^{o}(y)}\frac{1}{w_{n}(y)}f(y)dy\nonumber \\ & \simeq\frac{\mathrm{j}_{s}}{\mathcal{\mathcal{a_{r}}}}\int_{\mathcal{y}}\sum_{n\in\mathcal{n}_{h}^{o}(y)}\frac{dy}{w_{n}(y)}\label{eq : bfim_2d_generic_map_xbis}\end{aligned}\ ] ] where @xmath300 is the pdf of @xmath30 only and it has been ignored in the integral since the smoothing only affects a small portion of the integration domain ( see section [ sec : scenario ] ) . a similar approach can be adopted in the evaluation of @xmath301_{2,2}$ ] ( ignoring , in this case , the smoothing along @xmath29 ) ; this leads to @xmath302_{2,2}\simeq\frac{\mathrm{j}_{s}}{\mathcal{\mathcal{a_{r}}}}\int_{\mathcal{x}}\sum_{m\in\mathcal{n}_{v}^{o}(x)}\frac{1}{h_{m}(x)}dx\label{eq : bfim_2d_generic_map_y}\ ] ] note also that the approximated results ( [ eq : bfim_2d_generic_map_xbis])-([eq : bfim_2d_generic_map_y ] ) become exact if @xmath23 is a rectangle having its sides parallel to the reference axes , since in this case the smoothing along @xmath29 is truly independent from the smoothing along @xmath30 . for the same reason the cross terms @xmath301_{2,1}$ ] and @xmath301_{1,2}$ ] are exactly equal to zero for a rectangle ; however since 2-d maps usually exhibit a weak correlation between the variables @xmath29 and @xmath30 , these terms can be neglected ; this leads to ( [ eq : bfim_map_generic_mult_intervals ] ) , from which ( [ eq : bfim_maps_nogap ] ) is easily obtained assuming that @xmath98 and @xmath99 @xmath100 . in this appendix the ezzb ( @xmath303 ) is derived following a procedure conceptually similar to the one provided in ( * ? ? ? we start from the identity ( see @xcite ) @xmath304 where @xmath305_{\nu}$ ] . following @xcite , the probability appearing in ( [ eq : zzb_1 ] ) can be put in the form @xmath306 where @xmath307 denotes the pdf of @xmath21 . let @xmath308 and @xmath309 , where @xmath310 is some function of the integration variable of ( [ eq : zzb_1 ] ) . if the integration domain of ( [ eq : zzb_2 ] ) is constrained to be a subset @xmath311 of the map support @xmath23 , such that @xmath312 and @xmath313 , then , dividing and multiplying the rhs of ( [ eq : zzb_2 ] ) by @xmath314 , produces the expression ( * ? ? ? ( 22)-(30 ) ) @xmath315p_{\text{\tiny{min}}}^{\mathbf{z}}(\boldsymbol{\rho},\boldsymbol{\rho}+\boldsymbol{\delta}(h))d\boldsymbol{\rho}\label{eq : zzb_3}\end{gathered}\ ] ] ( @xmath316 is defined in sec . [ sub : zzb ] ) which holds for any @xmath310 satisfying the equality @xmath317 , where @xmath318 . note that ( [ eq : zzb_3 ] ) generalizes ( * ? ? ? * eq . ( 30 ) ) to the case where the pdf @xmath307 of the parameter to estimate has a bounded support . here @xmath319 is selected so that @xmath320 . then substituting ( [ eq : zzb_3 ] ) in ( [ eq : zzb_1 ] ) produces ( @xmath303 ) , where the integration domains of @xmath321 and @xmath322 have been merged in the set @xmath323 . like in the previous appendix , let us evaluate now ( @xmath303 ) for a 1-d uniform map whose support @xmath278 consists of the union of @xmath279 disjoint segments . in the following such segments are indexed by the odd numbers of the set @xmath282 and the lower ( upper ) limit of the segment @xmath285 is denoted @xmath324 ( @xmath325 ) , where @xmath283 ( @xmath284 ) represents the centre ( length ) of the segment itself . moreover , it is assumed , without any loss of generality , that @xmath326 . in this 1-d scenario ( @xmath303 ) simplifies as @xmath327p_{\text{\tiny{min}}}^{z}(\tau,\tau+h)h\ , d\tau\ : dh\label{eq : zzb_1d_1}\ ] ] where @xmath328 , @xmath329 and @xmath330 represents the minimum error probability of a binary detector which , on the basis of a noisy datum @xmath331 and a likelihood ratio test , has to select one of the following two hypotheses @xmath332 and @xmath333 . note that , for any @xmath334 , we have that @xmath335 ( with @xmath336 ) , because of the uniformity of the considered map . if the gaussian observation model ( [ eq : obs_model_gaussian ] ) adopted to the 1-d case is assumed , the average error probability of the above mentioned detector is easily shown to be @xmath337 ( where @xmath204 denotes the standard deviation of the noise observation model ) , which is independent of @xmath338 ; then , ( [ eq : zzb_1d_1 ] ) simplifies as @xmath339 where @xmath340 , @xmath341 and @xmath342 can be shown to be 2-d domain consisting of the union of @xmath279 triangles and @xmath343 parallelograms in the plane @xmath344 , as exemplified by fig . [ fig : appendix ] , which refers to the case @xmath345 . in particular , the contribution to the rhs of ( [ eq : zzb_1d_3 ] ) from the @xmath5-th triangle @xmath346 can be shown to be @xmath347 , where @xmath348 , @xmath349 , the function @xmath350 is defined in ( [ eq : f_rho_def ] ) and @xmath351 ( see @xcite for more details ) . as far as the parallelograms are concerned , we retain only the contributions coming from @xmath352 parallelograms @xmath353 with @xmath354 ( see fig . [ fig : appendix ] ) ; such contributions are given by @xmath355 , where @xmath356 , @xmath357 and @xmath358 is defined in ( [ eq : f_gap_def ] ) for @xmath359 ( it can be shown that @xmath360 ) . note that the contributions to ( [ eq : zzb_1d_3 ] ) coming from the discarded parallelograms are always positive since @xmath361 for @xmath362 . then , substituting the above mentioned contributions in ( [ eq : zzb_1d_3 ] ) produces the lower bound @xmath363\label{eq : zzb_1d}\ ] ] for the ezzb . the approach developed for a 1-d uniform map can be extended to a 2-d scenario keeping into account that a ) the term @xmath124 of ( @xmath303 ) takes on a similar form as @xmath330 of ( [ eq : zzb_1d_1 ] ) , b ) the integral on @xmath322 can be decomposed in a couple of nested integrals , one over @xmath56 and the other one over @xmath57 , and c ) eq . ( [ eq : zzb_1d ] ) can be exploited for the inner integral . further details can be found in @xcite . in this appendix we first derive the wwb for a @xmath364-d uniform map characterized by a support @xmath23 , assuming the observation model ( [ eq : obs_model_gaussian ] ) . note that this derivation can be easily generalised to a @xmath267-d scenario . to begin , we note that , since @xmath365 ( see ( [ eq : obs_model_gaussian ] ) ) , the likelihood ratio @xmath366 ( [ eq : likelihood_ratio ] ) is given by @xmath367 @xmath368 . then , it is not difficult to prove that the expectation appearing in the numerator of ( [ eq : wwb_def ] ) is given by @xmath369 where @xmath370 is a slice of the integration domain @xmath371 involved in the evaluation of the ezzb @xcite and @xmath121 . the expectation appearing in the denominator of ( [ eq : wwb ] ) can be expressed as @xmath372^{2}\right\ } \\ \shoveleft=\mathbb{e}_{\mathbf{z},\mathbf{p}}\left\ { l\left(\mathbf{z};\mathbf{p}+h_{\nu}\mathbf{e}_{\nu},\mathbf{p}\right)\right\ } + \mathbb{e}_{\mathbf{z},\mathbf{p}}\left\ { l\left(\mathbf{z};\mathbf{p}-h_{\nu}\mathbf{e}_{\nu},\mathbf{p}\right)\right\ } \\ -2\cdot\mathbb{e}_{\mathbf{z},\mathbf{p}}\left\ { l^{\frac{1}{2}}\left(\mathbf{z};\mathbf{p}+h_{\nu}\mathbf{e}_{\nu},\mathbf{p}\right)l^{\frac{1}{2}}\left(\mathbf{z};\mathbf{p}-h_{\nu}\mathbf{e}_{\nu},\mathbf{p}\right)\right\ } \end{gathered}\ ] ] i.e. , as the sum of three terms which are denoted @xmath373 , @xmath374 and @xmath375 , respectively , in the following . it is important to note that @xmath376 and @xmath377 where @xmath378 and the functions @xmath379 and @xmath380 are defined in ( [ eq : lambda_def ] ) and ( [ eq : gamma_def ] ) . then , substituting ( [ eq : wwb_num ] ) , ( [ eq : wwb_a_b ] ) , ( [ eq : wwb_c ] ) in ( [ eq : wwb_def ] ) produces ( [ eq : wwb_generic ] ) . finally , it is worth pointing out that @xmath379 ( [ eq : lambda_def ] ) and @xmath380 ( [ eq : gamma_def ] ) can be explicitly related to the functions @xmath381 , @xmath382 , @xmath383 , @xmath384 introduced in section [ sec : scenario ] . to clarify this point , let us focus on @xmath176 and @xmath235 which can be derived , for a given ordinate @xmath30 , analysing the integration domain @xmath385 if the variables @xmath341 and @xmath386 are defined , this domain can be put in the form @xmath387 which describes a slice of the @xmath344 plane of fig . [ fig : appendix ] . the contribution of the @xmath5-th triangle @xmath388 to @xmath389 ( @xmath390 ) is given by @xmath391 ( @xmath392 ) , where @xmath393 is defined in ( [ eq : omega_def ] ) , @xmath394 is defined in ( [ eq : omega_ov_def ] ) and @xmath395 is the slice of the map support @xmath23 at the ordinate @xmath30 . similarly , the contribution from the @xmath5-th parallelogram @xmath396 to @xmath397 is given by @xmath392 . therefore , the overall contribution to @xmath389 can be expressed as @xmath398 whereas that to @xmath390 as @xmath399 finally , integrating ( [ eq : sum1 ] ) and ( [ eq : sum2 ] ) produces ( [ eq : lambda_def ] ) and ( [ eq : gamma_def ] ) , respectively . and of its subsets @xmath400 and @xmath401 in the @xmath344 plane ( @xmath345 is assumed ) . note that the contribution from the `` discarded '' parallelogram is neglected in the evaluation of both ( [ eq : gamma_def])-([eq : lambda_def ] ) and ( [ eq : zzb_1d_3 ] ) . [ fig : appendix],width=336 ] f. gustafsson and f. gunnarsson , `` mobile positioning using wireless networks : possibilities and fundamental limitations based on available wireless network measurements , '' _ ieee signal process . mag . _ , vol . 22 , no . 4 , pp . 41 53 , jul . 2005 . s. gezici , z. tian , g. giannakis , h. kobayashi , a. molisch , h. poor , and z. sahinoglu , `` localization via ultra - wideband radios : a look at positioning aspects for future sensor networks , '' _ ieee signal process . mag . _ , vol . 22 , no . 4 , pp . 7084 , jul . 2005 . n. patwari , j. ash , s. kyperountas , a. o. hero , iii , r. moses , and n. correal , `` locating the nodes : cooperative localization in wireless sensor networks , '' _ ieee signal process . mag . _ , vol . 22 , no . 4 , pp . 5469 , jul . 2005 . s. beauregard , widyawan , and m. klepal , `` indoor pdr performance enhancement using minimal map information and particle filters , '' in _ ieee / ion position , location and navigation symp . _ , monterey , ca , 2008 , pp . 141147 . p. davidson , j. collin , and j. takala , `` application of particle filters for indoor positioning using floor plans , '' in _ proc . ubiquitous positioning indoor navigation and location based service conf . _ , kirkkonummi , 2010 , pp . 14 . m. anisetti , c. a. ardagna , v. bellandi , e. damiani , and s. reale , `` map - based location and tracking in multipath outdoor mobile networks , '' _ ieee trans . on wireless commun . _ , vol . 10 , no . 3 , pp . 814824 , mar . 2011 . d. nicholson and a. vecchio , `` bayesian bounds on parameter estimation accuracy for compact coalescing binary gravitational wave signals , '' _ physical review d _ , vol . 57 , no . 8 , pp . 45884599 , apr . f. montorsi and g. m. vitetta , `` on the performance limits of pilot - based estimation of bandlimited frequency - selective communication channels , '' _ ieee trans . on commun . _ , vol . 59 , no . 11 , pp . 29642969 , nov . 2011 . w. liou , j .- m . tan , and r. lee , `` minimum rectangular partition problem for simple rectilinear polygons , '' _ ieee trans . comput .- aided des . circuits syst . _ , vol . 9 , no . 7 , pp . 720733 , jul . 1990 . s. mazuelas , a. bahillo , r. m. lorenzo , p. fernandez , f. a. lago , e. garcia , j. blas , and e. j. abril , `` robust indoor positioning provided by real - time rssi values in unmodified wlan networks , '' _ ieee j. sel . topics signal process . _ , vol . 3 , no . 5 , pp . 821831 , oct . 2009 . t. routtenberg and j. tabrikian , `` a general class of outage error probability lower bounds in bayesian parameter estimation , '' _ ieee trans . on signal process . _ , vol . 60 , no . 5 , pp . 21522166 , may 2012 . f. montorsi and g. m. vitetta , `` technical report : detailed proofs of bounds for map - aware localization , '' university of modena and reggio emilia , tech . , 2012 , available online at http://frm.users.sourceforge.net/publications.html . francesco montorsi ( s06 ) received both the laurea degree ( _ cum laude _ ) and the laurea specialistica degree ( _ cum laude _ ) in electronic engineering from the university of modena and reggio emilia , italy , in 2007 and 2009 , respectively . he received the ph.d degree in information and communications technologies ( ict ) from the university of modena and reggio emilia in 2013 . he is employed as embedded system engineer for an ict company since 2013 . in 2011 he was a visiting phd student at the wireless communications and network science laboratory of massachusetts institute of technology ( mit ) . his research interests are in the area of localization and navigation systems , with emphasis on statistical signal processing ( linear and non - linear filtering , detection and estimation problems ) , model - based design and model - based performance assessment . dr . montorsi is a member of ieee communications society and served as a reviewer for the ieee transactions on wireless communications , ieee transactions on signal processing , ieee wireless communications letters and several ieee conferences . since 2009 , he has been a postdoctoral fellow in the wireless communication and network sciences laboratory at mit . he previously worked from 2006 to 2009 as a researcher and project manager in a spanish technological center , as well as a junior lecturer in the university of valladolid . his research interests are the application of mathematical and statistical theories to communications , localization , and navigation networks . mazuelas has served as a member of the technical program committee ( tpc ) for the ieee global communications conference ( globecom ) in 2010 , 2011 , and 2012 , the ieee international conference on ultra - wideband ( icuwb ) in 2011 , the 75th ieee vehicular technology conference 2012 , and the ieee wireless telecommunications symposium ( wts ) in 2010 and 2011 . he has received the 2012 ieee communications society fred w. ellersick prize , the best paper award from the ieee globecom in 2011 , the best paper award from the ieee icuwb in 2011 , and the young scientists prize for the best communication at the union radio - scientifique internationale ( ursi ) xxii symposium in 2007 . giorgio m. vitetta ( s89-m91-sm99 ) was born in reggio calabria , italy , in april 1966 . he received the dr . ing . degree in electronic engineering ( _ cum laude _ ) in 1990 , and the ph.d . degree in 1994 , both from the university of pisa , pisa , italy . from 1995 to 1998 , he was a research fellow with the department of information engineering , university of pisa . from 1998 to 2001 , he was an associate professor with the university of modena and reggio emilia , modena , italy , where he is currently a full professor . his main research interests lie in the broad area of communication theory , with particular emphasis on detection / equalization / synchronization algorithms for wireless communications , statistical modelling of wireless and powerline channels , ultrawideband communication techniques and applications of game theory to wireless communications . moe z. win ( s85-m87-sm97-f04 ) received both the ph.d . in electrical engineering and m.s . in applied mathematics as a presidential fellow at the university of southern california ( usc ) in 1998 . he received an m.s . in electrical engineering from usc in 1989 , and a b.s . ( _ magna cum laude _ ) in electrical engineering from texas a&m university in 1987 . he is a professor at the massachusetts institute of technology ( mit ) . prior to joining mit , he was at at&t research laboratories for five years and at the jet propulsion laboratory for seven years . his research encompasses fundamental theories , algorithm design , and experimentation for a broad range of real - world problems . his current research topics include network localization and navigation , network interference exploitation , intrinsic wireless secrecy , adaptive diversity techniques , and ultra - wide bandwidth systems . professor win is an elected fellow of the aaas , the ieee , and the iet , and was an ieee distinguished lecturer . he was honored with two ieee technical field awards : the ieee kiyo tomiyasu award ( 2011 ) and the ieee eric e. sumner award ( 2006 , jointly with r. a. scholtz ) . together with students and colleagues , his papers have received numerous awards including the ieee communications society s stephen o. rice prize ( 2012 ) , the ieee aerospace and electronic systems society s m. barry carlton award ( 2011 ) , the ieee communications society s guglielmo marconi prize paper award ( 2008 ) , and the ieee antennas and propagation society s sergei a. schelkunoff transactions prize paper award ( 2003 ) . highlights of his international scholarly initiatives are the copernicus fellowship ( 2011 ) , the royal academy of engineering distinguished visiting fellowship ( 2009 ) , and the fulbright fellowship ( 2004 ) . other recognitions include the laurea honoris causa from the university of ferrara ( 2008 ) , the technical recognition award of the ieee comsoc radio communications committee ( 2008 ) , and the u.s . presidential early career award for scientists and engineers ( 2004 ) . win is an elected member - at - large on the ieee communications society board of governors ( 20112013 ) . he was the chair ( 20042006 ) and secretary ( 20022004 ) for the radio communications committee of the ieee communications society . over the last decade , he has organized and chaired numerous international conferences . he is currently an editor - at - large for the ieee wireless communications letters , and serving on the editorial advisory board for the ieee transactions on wireless communications . he served as editor ( 20062012 ) for the ieee transactions on wireless communications , and served as area editor ( 20032006 ) and editor ( 19982006 ) for the ieee transactions on communications . he was guest - editor for the proceedings of the ieee ( 2009 ) and ieee journal on selected areas in communications ( 2002 ) .
the accelerated expansion of the universe has now been confirmed by several independent observations including those of high redshift type ia supernovae , and the cosmic microwave background ( cmb ) combined with the large scale structure of the universe @xcite . another way of presenting this kinematic property of the universe is to postulate the existence of a new entity dark energy ( de ) . this latter statement is dynamical in nature and therefore requires some assumptions to be made about the form of gravitational field equations governing the evolution of the ( observed part of the ) universe . although observationally well established , no single theoretical model provides an entirely compelling framework within which cosmic acceleration or de can be understood . indeed , the very many models of de existing in the literature illustrate that its nature is still very much an enigma . at present , all existing observational data are in agreement with the simplest possibility of de being a cosmological constant @xmath2 with @xmath3gev@xmath4 ( inside @xmath5 error bars in the worst case ) . is used throughout the paper . ] this case is internally self - consistent and non - contradictory . the extreme smallness of the cosmological constant expressed in either planck , or even atomic units means only that its origin is _ not _ related to strong , electromagnetic and weak interactions ( in particular , to the problem of the energy density of their vacuum fluctuations ) . where @xmath6 is some characteristic neutrino rest - mass ( the lightest one ? ) suggests that vacuum energy of the interaction responsible for non - zero neutrino rest - masses may be relevant for a non - zero @xmath2 . ] although in this case de reduces to only a single fundamental constant we still have no derivation from any underlying quantum field theory for its small value . within this context it is but natural that other possibilities admitting a ( slightly ) variable dark energy have also been actively studied by the scientific community in recent years . moreover , it is interesting that properties of the currently observed ` late de ' are _ qualitatively _ similar to those of an ` early de ' which is believed to have given rise to accelerated expansion ( inflation ) in the early universe . however in the case of the latter , there are sufficient grounds to support the view that ` early de ' was unstable and , thus , more complicated than a cosmological constant . so , it is natural to conjecture by analogy that the same might also be true of ` late de ' . de models proposed to account for the present cosmic acceleration include : + ( i ) quiessence with @xmath7 , the @xmath2 ( @xmath8 ) is a special member of this class . + ( ii ) quintessence models which are inspired by the simplest class of inflationary models of the early universe and employ a scalar field rolling down a potential @xmath9 to achieve late - time acceleration . quintessence potentials with @xmath10 have the attractive property that approaches a common evolutionary ` tracker path ' from a wide range of initial conditions . + ( iii ) the chaplygin gas model ( cg ) has the equation of state @xmath11 and evolves as @xmath12 where @xmath13 is the redshift , @xmath14 . it therefore behaves like dark matter at early times ( @xmath15 ) and like the at late times . cg appears to be the simplest model attempting to unify de and non - baryonic cold dark matter . + ( iv ) ` phantom ' de ( @xmath16 ) . + ( v ) oscillating de . + ( vi ) models with interactions between de and dark matter . + ( vii ) scalar - tensor de models . + ( viii ) modified gravity de models in which the gravitational lagrangian is changed from @xmath17 to @xmath18 where @xmath17 is the scalar curvature and @xmath19 is an arbitrary function . + ( ix ) dark energy driven by quantum effects . + ( x ) higher dimensional ` braneworld ' models in which acceleration is caused by the leakage of gravity into extra dimensions . + ( xi ) holographic dark energy , etc . + see the reviews @xcite for an exhaustive list of models and references . however , none of these models leads to the reduction of the number of fundamental constants ( parameters in the microscopic lagrangian ) as compared to standard @xmath2cdm . in other words , at the current state - of - the - art , de requires at least one new parameter whose value is set from observations . models with variable de can be broadly divided into two main classes : , in these models de is the energy density of some new , very weakly interacting physical field . ( otherwise dubbed modified gravity models ) . in these models the gravity equations do not coincide with those of einsteinian general relativity . however , it is usually possible to re - write the new equations in the conventional _ einsteinian _ form by transferring all additional terms from the l.h.s . into the r.h.s . of the einstein equations and referring to them as an effective energy - momentum tensor of de ( see sec . 2 below ) . ( precisely this happened to the cosmological constant which originally appeared in the l.h.s . of einstein s field equations but is now felt by many to constitute an effective matter term such as vacuum energy , etc . ) another category , which is even more important from the observational point of view , arises in response to the question whether or not the description of de requires a new field degree of freedom (= a new kind of matter ) . if the answer is in the affirmative then de may be considered as being ` induced ' by other kinds of matter . all physical de models and many geometrical ones belong to this category but there do exist geometrical de models which do not ( for instance the @xmath18 model with the palatini variation of its action ) . faced with the increasing proliferation of de models each with its own physical motivations and assumptions , a concerned cosmologist can proceed in either of two ways : \(i ) test _ every single _ model against observations . \(ii ) try and ascertain properties of dark energy in a _ model independent manner_. in this article , we proceed along route ( ii ) and attempt to review both the successes as well as difficulties faced by methods attempting to reconstruct the properties of dark energy directly from observations in a model independent manner . before attempting to determine its properties , we first need to provide a definition of _ dark energy_. a traditional approach is to use the einstein form of the gravitational field equations [ einst ] r _ - 12g_r = 8 g ( _ a t_^(a ) + t_^de ) as providing a _ definition _ of the effective energy - momentum tensor @xmath20 of de . here , the summation over @xmath21 in the r.h.s . includes all types of matter known from laboratory experiments ( protons , neutrons , photons , neutrinos , etc . ) as well as non - relativistic non - baryonic cold dark matter ( whose energy - momentum tensor is dust - like in the first approximation @xmath22 ) . @xmath23 is the present value of newton s gravitational constant . this definition has a number of advantages : + ( i ) it is simple , well - defined and self - consistent ; + ( ii ) it treats physical and geometrical de on an equal footing ; + ( iii ) in the absence of direct physical interaction between de with known forms of matter or with cold dark matter , the de energy - momentum tensor is conserved : @xmath24 . + one should stress here that using ( [ einst ] ) we automatically ascribe terms describing ( possible ) gravitational interactions between de and non - relativistic matter , as well as the matter energy - momentum tensor multiplied by a change in the effective gravitational constant , to the de energy - momentum tensor . the latter possibility arises , for instance , in scalar - tensor de models . when applied to a homogeneous and isotropic friedmann - robertson - walker ( frw ) cosmological model , eqs . ( [ einst ] ) reduce to two algebraically independent equations : h^2 & = & ( _ a _ a + _ de ) - , [ eq : acc ] + & = & -(_a ( _ a + 3p_a ) + _ de + 3p_de ) [ eq : acc1 ] where @xmath25 is the frw scale factor and the hubble parameter @xmath26 . at late times when radiation may be neglected , one gets [ eq : energy ] _ de = ( 1 - _ m ) where we have omitted the contribution from the curvature term in ( [ eq : acc ] ) for simplicity . @xmath27 is the total density of non - relativistic matter in terms of its critical value . similarly , the expression for the deceleration parameter q - a / ah^2 = x - 1 , x = 1 + z [ eq : decel1 ] ( prime implies differentiation with respect to @xmath28 ) takes the form : p_de = ( q - ) . [ eq : pressure ] dividing ( [ eq : pressure ] ) by ( [ eq : energy ] ) we get the following expression for the _ effective _ equation of state ( eos ) of dark energy ( @xmath29 ) : [ eq : state ] w(x ) = 2 q(x ) - 1 3 ( 1 - _ m(x ) ) . for physical de , the eos makes physical sense . however , this is not so for geometrical de for which the acceleration of the universe is caused by the fact that the field equations describing gravity are not einsteinian . as a specific example of geometrical de consider the widely studied dgp braneworld @xcite for which [ eq : dgp ] h = + , where @xmath30 is a new length scale and @xmath31 and @xmath32 refer respectively to the four and five dimensional planck mass . the acceleration of the universe in this model arises not because of the presence of de but due to the fact that gravity becomes five dimensional on length scales @xmath33 . the contrast between ( [ eq : acc ] ) and ( [ eq : dgp ] ) makes it abundantly clear that for models such as dgp the eos in ( [ eq : state ] ) is an _ effective _ quantity ( @xmath34 ) , which may still be useful for descriptive purposes but which no longer represents any fundamental physical property of an accelerating universe . indeed , instances are known when @xmath35 even when matter itself satisfies the weak energy condition @xmath36 @xcite . it may be instructive to note that , for geometrical de , @xmath1 may show pathological behaviour in certain cases , such as the presence of poles at finite values of redshift , @xmath37 , even though the underlying cosmological model is completely well behaved ( see for instance @xcite ) . for such models , the deceleration parameter @xmath38 and other geometrical parameters prove to be more robust for determining de properties than the eos . we shall return to these important issues in section [ sec : statefinder ] . observational tests of de rely on an accurate measurement of at least one of the following quantities : 1 . the luminosity distance [ eq : lumdis ] = _ 0^z . the angular size distance [ eq : angdis ] d_a(z ) = _ 0^z . the coordinate distance [ eq : coordis ] r(z ) = _ 0^z . here , we have assumed that the universe is spatially flat for simplicity . note that the distance duality relation @xmath39 which follows from ( [ eq : lumdis ] ) and ( [ eq : angdis ] ) is valid only for metric theories of gravity @xcite . its violation ( if observed ) could , therefore , be used to probe alternative theories of gravity . existing data , however , appear to support it ( within observational errors ) @xcite . in all of the above expressions the value of the hubble parameter can be ` reconstructed ' through a relation similar to the one given below for the luminosity distance @xcite : [ eq : h ] h(z)=^-1 . differentiating a second time allows one to reconstruct the equation of state of de ( [ eq : state ] ) . equations ( [ eq : state ] ) and ( [ eq : h ] ) immediately inform us that @xmath1 will be a noisier quantity than @xmath0 since two successive differentiations are needed for the reconstruction @xmath40 while a single differentiation suffices for @xmath41 . this has led several people to suggest @xmath0 ( or @xmath42 ) as being better suited for providing a model independent description of properties of de . another important difference between @xmath0 defined in ( [ eq : h ] ) and @xmath1 in ( [ eq : state ] ) is that the former is independent of the value of the matter density parameter @xmath27 while the latter is not . as a result , uncertainties in the current value of @xmath27 affect the reconstruction of the eos far more profoundly than they do @xmath0 . we shall return to this issue in section [ sec : obstacle ] . knowing @xmath0 ( either through ( [ eq : h ] ) or using corresponding relations for other observational tests ) , allows us to extend cosmological reconstruction to other important physical properties of the universe including : * its age [ eq : age ] t(z ) = _ z^ . * the deceleration parameter ( [ eq : decel1 ] ) and the equation of state ( [ eq : state ] ) ; * the electron - scattering optical depth to a redshift @xmath43 ( z_reion ) = c_0^z_reion , [ eq : reion ] where @xmath44 is the electron density and @xmath45 is the thomson cross - section describing scattering between electrons and cmb photons . * the product @xmath46 , which plays a key role in the alcock paczynski anisotropy test @xcite . * the product @xmath47 , which is used in the volume - redshift test @xcite . * the parameter @xmath48 associated with the determination of the baryon acoustic peak @xcite a = _ 0^z_1 ^2/3 = 0.469 0.017 , where @xmath49 and @xmath50 is the redshift at which the acoustic scale has been measured in the redshift sample . * the ` shift ' parameter @xmath17 associated with the cmb @xcite + r = _ 0^z_lss = 1.7 0.03 . for quintessence , it is additionally possible to reconstruct its potential , since the einstein equations @xmath51 can be rewritten as ( see , for instance , @xcite ) : @xmath52 integrating ( [ eqn : phidot ] ) , one determines @xmath53 to within an additive constant . the inversion @xmath54 followed by substitution into ( [ eqn : vzed ] ) allows us to reconstruct @xmath9 , since @xmath0 and its first derivative can be determined from observations using ( [ eq : h ] ) . the presence of @xmath55 in ( [ eqn : vzed ] ) and ( [ eqn : phidot ] ) implies that the value of the matter density must be known rather precisely for an accurate reconstruction of @xmath9 . in several important cases ( [ eqn : vzed ] ) and ( [ eqn : phidot ] ) have led to a closed form expression for @xmath9 , for instance : * de with a constant equation of state @xmath56 is described by @xcite + v()=3h_0 ^ 2(1-w)(1-_m0)^1/|w| 16g_m0^ ^-2(|w| ( -_0+_1 ) ) , [ quequi ] where = , _ 0=(t_0 ) , _ 1= . consequently , a universe consisting of such a scalar field will have expansion properties which closely mimic a different kind of de , e.g. , a tangled network of cosmic strings ( @xmath57 ) or domain walls ( @xmath58 ) . * the chaplygin gas model which unifies dark matter and de and has @xmath59 can be described by the scalar field potential @xcite ( in the absence of additional cold dark matter and neglecting baryons and photons ) : + v ( ) = ( ( 2 ) + ) . note that the behaviour of the chaplygin gas may also be modelled completely differently using a scalar field with the born - infeld kinetic term @xcite ( a specific type of k - essence ) . this illustrates once more that , even for physical de , the equation of state @xmath1 does not uniquely define an underlying field - theoretical model is quite different from the quintessence lagrangian @xmath60 . ] ( see also @xcite ) . from the fact that the left - hand side of eq . ( [ eqn : phidot ] ) is always non - negative , follows an important restriction on the expansion law for the universe which must be satisfied before attempts to reconstruct the potential are made , namely 3h_0 ^ 2(1+z)^2 . [ ineq ] equation ( [ ineq ] ) is simply a restatement of the weak energy condition @xmath61 . integrating ( [ ineq ] ) , we get the relation h^2(z)h_0 ^ 2(1+(1+z)^3 ) [ int - ineq ] which is easier to verify observationally . in the case of physical de , with its implied minimal coupling to gravity , the expansion history @xmath0 can also be reconstructed from the growth rate of inhomogeneous density perturbations in the non - relativistic matter component on scales significantly less than the hubble radius @xmath62 , ( provided perturbations in the de component can be neglected and the effective gravitational constant does not change with time ) @xcite . indeed , the linearized perturbation equation and other quantities referring to the evolution of a frw background that , in turn , can be used to determine @xmath0 @xcite . ] + 2h - 4g_m= 0 [ eq : recon3 ] can easily be inverted , with the result : h(z)= h(0 ) - 3_0^z dz^. [ eq : recon7 ] furthermore , an interesting relationship follows between the current value of the matter density @xmath55 and @xmath63 : = ^2(0)(3_0^dz)^-1 , [ eq : recon6 ] which could provide a consistency check on direct observational determinations of @xmath55 . at sufficiently low redshifts @xmath64 , the advent of deep redshift surveys probing large scale structure may help in reconstructing @xmath0 using ( [ eq : recon7 ] ) . although it is unlikely that the value of @xmath63 for @xmath65 will be reliably known in the near future , the fact that the universe is expected to become matter dominated at fairly low redshifts , @xmath66 at @xmath67 , allows us to use the spatially flat matter dominated solution @xmath68 to extrapolate to higher redshifts and thereby evaluate ( [ eq : recon6 ] ) . as a result , for physical de we now have two independent methods to reconstruct the same quantity @xmath69 : ( i ) through quantities like @xmath70 referring to an unperturbed frw background , and ( ii ) through @xmath63 describing the growth of perturbations on small scales . it therefore follows that one should be able to reconstruct the density perturbation @xmath63 from observations of the luminosity distance . consider the quantity @xmath71 which can also be written as the difference of conformal times : @xmath72 . for our purpose it will be useful to invert the quantity @xmath73 ( determined from observations ) to @xmath74 . next , we again assume that de does not cluster and rewrite ( [ eq : recon3 ] ) in the following integral form : ( e(z ) ) = ( 0 ) + (0)_0^e ( 1+z(e))de + _ 0^e ( 1+z(e_1))de_1_0^e_1 ( e_2)de_2 . [ eq : recondelta ] here @xmath75 since @xmath76 for @xmath77 . clearly ( [ eq : recondelta ] ) can be solved iteratively if the value of @xmath78 is known . the value of @xmath79 does not really matter since we are only interested in the ratio @xmath80 . the requirement @xmath81 leads to the integral condition : (0)= - ( 3/2)_m_0^e_lss(e)de , and , as in the case of ( [ eq : recon6 ] ) , one needs to postulate some reasonable interpolating behaviour for @xmath82 at large redshifts for which no supernova data exist at present . since @xmath83 , the value of this quantity should be possible to determine from the cmb . both ( [ eq : recon7 ] ) and ( [ eq : recondelta ] ) can be used as consistency checks for physical de which , by assumption , is minimally coupled to gravity . for geometrical de on the other hand , the linearized perturbation equation is usually modified from its conventional form ( [ eq : recon3 ] ) , see for instance @xcite for scalar - tensor gravity or @xcite for the dgp model . is expected to be different from the reconstructed expression ( [ eq : recondelta ] ) provides an interesting possibility to distinguish between physical and geometrical models of de . ] as a result , de reconstruction becomes more complicated . in particular , a generic scalar - tensor model of de depends on two arbitrary functions @xmath9 and @xmath84 the scalar field potential and its coupling to gravity respectively . in this case one needs information from _ both _ @xmath70 and @xmath63 in order to reconstruct the de model unambiguously @xcite . to summarize , we have shown in this section that the reconstruction of the expansion rate @xmath0 and other de properties from observational data is well defined and unambiguous from the mathematical point of view . in practice , however , the situation is much more difficult since all observational functions such as @xmath70 and @xmath63 are noisy and known only at discrete values of the redshift @xmath85 ( associated with the redshifts of @xmath86 supernovae in the case of @xmath70 ) . thus , it is impossible to directly differentiate them with respect to redshift as formulae ( [ eq : h ] ) and ( [ eq : recon7 ] ) require . add to this the fact that the dispersion in the luminosity distance is not expected to get significantly better than @xcite @xmath87 , and one is confronted with a noisy quantity @xmath88 sampled at a set of discrete intervals @xmath89 . therefore , to convert from @xmath88 to the function @xmath0 defined at _ all _ redshift values within the interval ( say ) @xmath90 using ( [ eq : h ] ) clearly requires a crucial additional step : some sort of _ smoothing _ procedure . this is usually accomplished using either parametric or non - parametric reconstruction which we turn to next . this approach is based on the assumption that the quantities @xmath91 vary ` sufficiently slowly ' with redshift and can therefore be approximated by a fitting formula ( ansatz ) which relies on a small number of free parameters @xmath92 . the ansatz for @xmath93 is compared against observations , and the values of the free parameters @xmath94 are determined using a minimization procedure ( usually maximum likelyhood ) . quite clearly a successful fitting function should , in principle , be able to faithfully reproduce the properties of an entire class of de models . different parameterizations have been suggested for : @xmath95 @xcite , @xmath0 @xcite , @xmath1 @xcite and @xmath96 @xcite . we attempt to summarize some of these approaches below . 1 . * fitting functions to the luminosity distance @xmath95 * + an approximation to a function can easily be generated by expanding it in a taylor series about some redshift @xmath97 . applied to the luminosity distance ( with @xmath98 ) , this method gives the following fitting function @xcite = _ i=1^n a_iz^i . [ eq : poly ] unfortunately , it appears that in order to accurately determine quantities of interest such as @xmath0 and @xmath1 one must take a large number of terms in ( [ eq : poly ] ) which greatly increases the errors of reconstruction @xcite and reduces the reliability of this ansatz . + a more versatile pad - type ansatz was suggested in @xcite = 2 , x = 1+z , [ eqn : star ] which is able to _ exactly _ reproduce the results both for cdm ( @xmath99 ) and the steady state model ( @xmath100 ) . + another accurate fit with a greater number of free parameters is @xcite & & = ( 1)-(y ) , + & & ( y)=2 ^ -1/8 , y = 1/ . [ eq : fit_chiba ] 2 . * fitting functions to the de density * + the dark energy density can be written as a truncated taylor series polynomial in @xmath101 , @xmath102 . this leads to the following ansatz for the hubble parameter @xcite [ eq : poly_fit ] h(x ) = h_0x^3 + a_1 + a_2x + a_3 x^2^ , [ eq : taylor ] which , when substituted in the expression for the luminosity distance ( [ eq : lumdis ] ) , yields = _ 1 ^ 1+z . [ eq : taylor1 ] this fitting function gives exact results for the cosmological constant @xmath8 ( @xmath103 ) as well as for quiessence with @xmath104 ( @xmath105 ) and @xmath106 ( @xmath107 ) . the presence of the term @xmath108 in ( [ eq : poly_fit ] ) ensures that the ansatz correctly reproduces the matter dominated epoch at early times ( @xmath67 ) . for quintessence models as well as the chaplygin gas , the luminosity distance @xmath70 in ( [ eq : taylor1 ] ) can be determined to an accuracy of better than @xmath109 if @xmath110 @xcite . * fitting functions to the equation of state * + in this approach one assumes that the de equation of state @xmath1 is an unknown variable whose behaviour is ` guessed ' by means of a suitable fitting function @xmath111 . since h^2(z ) & = & h_0 ^ 2 ( 1+z)^3 + ^2 , + + & = & ( 1- ) , [ eq : hubble_recon ] the values of @xmath94 can be determined by comparing the luminosity distance ( [ eq : lumdis ] ) or @xmath112 , against observations . + several possible fits for @xmath1 have been suggested in the literature . perhaps the simplest is the taylor expansion @xcite w(z ) = _ i=0^n w_iz^i , [ eq : wpol ] which , for @xmath113 , gives results significantly better than the taylor expansion for the luminosity distance ( [ eq : poly ] ) . the simple two parameter representation @xmath114 is however of limited utility since it is only valid for @xmath115 . a considerably more versatile four parameter ansatz has been suggested in @xcite . + a popular two parameter fit was suggested by chevallier and polarski @xcite and by linder @xcite w(a ) = w_0+w_1(1-a ) = w_0 + w_1 , [ eq : cpl ] the associated luminosity distance can be obtained by substitution into ( [ eq : hubble_recon ] ) and ( [ eq : lumdis ] ) . a more general form for this fit is w(a ) = w_p + ( a_p - a)w_a , where @xmath116 is the value of the scale factor at the ` sweet spot ' where the equation of state @xmath117 is most tightly constrained . the value of @xmath116 usually depends upon the data set being used @xcite . @xmath118 \epsfxsize=2.7 in \epsffile{h_exp_w_g.epsi } & \epsfxsize=3.3 in \epsffile{huterer3.ps } \\ \end{array}$ ] as mentioned earlier , for an ansatz to be regarded as being successful it should embrace within its fold the behaviour of a reasonably wide class of de models . most of the ansatz s discussed above successfully accommodate de whose equation of state evolves moderately with redshift . it is quite clear that these simple fits can not be used to rule out ( using observations ) models with rapidly evolving @xmath1 . one reason for this is that applying an ansatz such as ( [ eq : taylor ] ) or ( [ eq : cpl ] ) to sne data is equivalent to smoothing the evolution of the universe over a redshift interval @xmath119 where @xmath86 is the number of free parameters in the ansatz ( the concrete form of an ansatz defines the form of a smoothing redshift function ) . clearly , an implicit smoothing such as this will cause rapid transitions in dark energy to disappear not because of disagreement with data but simply because of the manner in which the data have been ` massaged ' @xcite . ( an example of the disastrous results of applying the prior @xmath120 constant to models with an evolving eos is discussed in section [ sec : obstacle ] . ) models with a fast phase transition in dark energy ( @xmath121 over a narrow range of redshift @xmath122 ) have been discussed in @xcite . to accommodate such models the following ansatz was suggested @xcite w(z)=w_i+ , [ eq : bass ] where @xmath123 is the initial equation of state at high redshifts , @xmath124 is a transition redshift at which the equation of state falls to @xmath125 and @xmath126 describes the rate of change of @xmath1 . substitution into ( [ eq : hubble_recon ] ) and ( [ eq : lumdis ] ) gives the luminosity distance . finally , models with oscillating de have also been discussed in the literature @xcite . to accommodate an oscillating eos the following ansatz has been suggested @xcite w ( ) = w_0 + w_1where @xmath127 are free parameters whose values must be obtained by fitting to observations . it is encouraging to note that , in spite of some ambiguity in the form of the different fits , when applied to the same supernova data set most of them give consistent results in the range @xmath128 ( where there is sufficient data ) @xcite . one should also note that although increasing the number of parameters usually increases the accuracy of reconstruction of the ` best fit ' , this is often accompanied by severe degeneracies which limit the utility of introducing a large number of free parameters . in addition , extra free parameters are quite severely penalized by information criteria such the akaike information criterion @xcite and the bayesian information criterion @xcite , see also @xcite . however , there exists a subtlety when applying these information criteria to fitting functions of @xmath129 etc . as opposed to concrete theoretical models of de . in the case of the former , one should keep in mind the possibility that a fit with a larger number of parameters might in fact be describing the behaviour of a fundamental de model ( not yet known ) containing a smaller number of truly free parameters . the reverse is also true , a primitive fit ( hence not penalizable ) may , if the information criteria are applied , detract our attention from a more complicated but also more fundamental explanation of a phenomenon . let us illustrate this with an example . the theory of gravitational instability informs us that the linear density contrast @xmath130 and the peculiar velocity field @xmath131 are related as @xcite = - [ eq : velocity ] where the function @xmath132 can be approximated as @xmath133 . an exact calculation however reveals @xmath134 to be a more complicated function involving elliptical integrals , etc . therefore a naive application of information criteria to ( [ eq : velocity ] ) would needlessly penalize the latter , which is correct but complicated , in favour of the simpler @xmath133 . another popular approach the so - called principal component analysis is based on expanding @xmath1 in terms of a basis of orthogonal functions . the form of these functions depends upon the kind of data used and its constraining capabilities @xcite . we end this section by noting that the reconstruction approach has also been extended to scalar - tensor gravity @xcite , string inspired cosmology with the gauss - bonnet term coupled to a scalar field @xcite and dark energy with non - canonical kinetic energy terms @xcite . other applications of cosmological reconstruction may be found in @xcite . non - parametric smoothing usually involves directly smoothing either @xmath95 , or some other quantity appropriately binned in redshift space with some characteristic smoothing scale . different ways of implementing this approach have been discussed in @xcite . all reconstruction methods must deal with the fact that the data sample is usually sparse and the coverage of redshift space uneven . consider for instance the comoving coordinate distance to a single supernova within a larger sample r_i = r(z_i ) + n_i . here , @xmath135 is the noise term @xmath136 , and @xmath137 is related to the distance modulus @xmath138 of the supernova by @xcite @xmath139 . direct differentiation of @xmath140 to give @xmath141 will obviously result in a very noisy quantity . one way to tackle this is to fit @xmath142 piece - wise using a set of basis functions . following this approach , daly and djorgovsky @xcite reconstructed the deceleration parameter from the dimensionless coordinate distance @xmath143 by : ( i ) fitting @xmath144 locally within a small redshift bin by means of a second order polynomial requiring that at least 10 data points lie within each bin ; ( ii ) determining the first and second derivatives of @xmath144 from the fit coefficients ( for each bin ) and reconstructing the deceleration parameter from @xmath144 by means of the relation -q(z ) = 1 + ( 1+z ) . applying their method to a set of radio galaxies in addition to type ia sne , daly and djorgovsky @xcite found that although the cosmological constant provided a very good fit to the data , modest evolution in the dark energy density was also perfectly acceptable . the transition from deceleration to acceleration occured at @xmath145 , with a best fit value of @xmath146 . these results are in broad agreement with those obtained using parametric approaches @xcite . a different approach to non - parametric reconstruction is discussed by shafieloo @xcite who generalize a smoothing ansatz widely used in the analysis of large scale structure . according to this method , a smoothed quantity @xmath147 is constructed from a fluctuating ` raw ' quantity @xmath148 using a low pass filter @xmath19 having a smoothing scale @xmath126 d^s(*x * , ) = d(*x * ) f(|*x*-*x*| ; ) d*x * . in large scale structure studies , @xmath149 is the density field @xcite whereas for cosmological reconstruction @xmath149 could be either of @xmath150 . commonly used filters include : ( i ) the ` top - hat ' filter , which has a sharp cutoff @xmath151 where @xmath152 is the heaviside step function ( @xmath153 for @xmath154 , @xmath155 for @xmath156 ) and ( ii ) the gaussian filter f_g . when applied to snap - quality data using a gaussian filter , this method reconstructs the hubble parameter to an accuracy of @xmath157 within the redshift interval @xmath158 . the look - back time t(z)=t(z=0)-t(z)=h_0 ^ -1_0^z is reconstructed to an even better accuracy of @xmath159 at @xmath160 . despite the considerable success of these approaches , neither they , nor the parametric methods discussed earlier provide independent measurements of either @xmath0 or @xmath38 within a given redshift range . one approach towards an uncorrelated determination of cosmological quantities is to bin the data in redshift bins of size @xmath126 with a weight ( or window ) function assigned to each bin @xcite . each window function vanishes outside its redshift bin and the two supernovae falling on either side of a bin boundary are discarded . this method ensures that , since two adjacent window functions have no common supernovae , the measurement of quantities within these bins will have uncorrelated error bars . this method is quite successful and complements other reconstruction approaches in many respects . one weakness , which practitioners of this approach have to guard against is that , since some information at the bin boundary is lost , discontinuities in the fit values at bin edges may be present @xcite . these , in turn , could lead to unphysically large derivatives of derived quantities such as @xmath0 and @xmath1 . results obtained by applying a variant of this method to recent cmb+sne+lss data are shown in figure [ fig : wangm ] and appear to agree with the results of the parametric approach in figure [ fig : eos ] . other promising methods for obtaining uncorrelated estimates of cosmological evolution are discussed in @xcite . in all of the above approaches , it is important to choose the value of the smoothing scale @xmath126 ( or bin size ) optimally , so that ( i ) a sufficient number of data points is accommodated within each bin to reduce the effects of shot noise , ( ii ) the bin size is not too large to cause excessive smoothing . the following useful formula giving the relative error bars on @xmath0 comes to our aid @xcite [ eq : tegmark ] , where n is the total number of supernovae ( assuming an approximately uniform distribution ) and @xmath161 is the noise of the data . clearly , a reduction in the value of @xmath126 by 3 should be compensated for by having 27 many times more supernovae if one wants to keep @xmath162 unchanged . the situation is far worse for @xmath1 which has a @xmath163 scaling , since the data are being differentiated twice to get @xmath1 . with respect to cosmological reconstruction , considerable caution must be exercised when setting priors on the values of cosmological parameters . this is true both for the density parameter @xmath27 whose value is currently known to about @xmath164 accuracy and the de equation of state . maor @xcite gave a particularly insightful example of the dangers of incorrect reconstruction by assuming a fiducial de model with @xmath165 and an evolving eos @xmath166 . in their reconstruction , the results of which are shown in figure [ fig : maor ] , it was assumed that @xmath167 was a constant and , additionally , that @xmath168 . these two ( incorrect ) priors led to a gross underestimation of @xmath167 and an overestimation of @xmath27 as shown in figure [ fig : maor ] . the above example shows that the eos of dark energy can be badly reconstructed if we assume a misleading prior for @xmath169 . since @xmath1 depends upon the value of @xmath27 through ( [ eq : state ] ) , it follows that an incorrect assumption about the value of the matter density has the potential to affect cosmological reconstruction quite significantly . this is demonstrated in figure [ fig : arman ] in which the de eos is reconstructed for a fiducial @xmath2cdm model with @xmath170 and @xmath165 . cosmological reconstruction is based on the ansatz ( [ eq : poly_fit ] ) in which the _ incorrect _ value @xmath171 is assumed . this results in the reconstructed eos evolving with redshift when , in fact , no such evolution is present in the fiducial model ! as we have seen , the eos has certain blemishes it is obtained from @xmath95 after differentiating _ twice _ and it is sensitive to the _ precise _ value of @xmath55 . both these features hamper its reconstruction which has led several authors to propose the hubble parameter or the de density as being more appropriate for cosmological reconstruction @xcite . carrying these ideas further , the possibility of extracting information about the equation of state from the reconstructed hubble parameter by constructing a _ weighted average _ of the equation of state was explored in @xcite . this quantity , dubbed the @xmath169-probe , gleans information about the equation of state from the _ first derivative _ of the luminosity distance . therefore , it is less noisy and better determined than @xmath1 . the weighted average of the equation of state is defined as @xcite [ eq : weight1 ] 1+|w= ( 1+w(z ) ) . an important feature of the @xmath169-probe ( @xmath172 ) is that it can be expressed in terms of the difference in dark energy density over a given redshift range [ eq : weight2 ] 1+|w(z_1,z_2 ) & = & + & & ( ) . + here @xmath130 denotes the total change of a variable between integration limits and @xmath173 ( @xmath174 ) . thus @xmath172 is easy to determine if the hubble parameter has been accurately reconstructed . an important property of @xmath172 is that it is less sensitive to uncertainties in the value of @xmath55 than @xmath1 . this is demonstrated in figure [ fig : margin ] where the results for @xmath172 are shown after marginalizing over the matter density . remarkably , the value of @xmath172 for the fiducial @xmath2cdm model remains close to @xmath175 , while @xmath172 for the evolving de model shows a clear signature of evolution . thus , small uncertainties in the value of the matter density do not appear to adversely affect the accuracy of the reconstructed @xmath169-probe . furthermore , several excellent methods for determining @xmath176 and @xmath0 have been suggested in the literature @xcite , any of which could be used to determine @xmath172 using ( [ eq : weight2 ] ) . cosmological observations made during the past two decades have brought about both qualitative and quantitative changes in our perception of the universe . observational support has come for many theories ( such as inflation ) which earlier were regarded as being at the level of hypothesis . on the other hand , fundamentally new properties of our universe , such its current state of acceleration , have also been unravelled . the remarkable qualitative similarity between current acceleration ( fueled by de ) and inflationary acceleration ( supported by what may be called primordial dark energy ) may help us understand in which direction cosmology may be headed . the discovery by cobe of an approximately scale - invariant spectrum of primordial fluctuations ( @xmath177 ) was regarded by many as providing tacit support for the inflationary scenario . however , virtually all models of inflation predict small departures from the scale invariance @xmath178 , where @xmath179 is a model dependent quantity . strict scale invariance ( @xmath180 ) arises for a specific 2-parameter family of inflaton potentials only @xcite , which reduces to @xmath181 in the slow - roll approximation . in this context , the recent 3-year data release of the wmap experiment suggests @xmath182 , which is in excellent agreement with predictions made by the simplest inflationary models such as chaotic inflation with @xmath183 . the present situation concerning de resembles , in some respects , the status of inflation just after the release of the cobe data in 1992 . the cosmological constant is in excellent agreement with data and departures ( if any ) from @xmath184 are believed to be quite small . nevertheless , since at present no fundamental theory predicting the ( small ) value of the @xmath2-term exists , it is of utmost importance that departures from @xmath184 be probed to accuracies of at least @xmath109 by future generations of experiments . since virtually all de models ( other than @xmath2 ) have either @xmath185 or @xmath186 , the need for probing both the equation of state as well as its first derivative become crucial if any deep insight is to be gained into the nature of de . it is useful to recall that several de models which agree well with the current data arise because of modifications to the gravitational sector of the theory @xcite . for these _ geometrical de _ models , the eos no longer plays the role of a fundamental physical quantity and it would be very useful if we could supplement it with a diagnostic which could unambiguously probe the properties of all classes of de models . the @xmath169-probe discussed earlier , provides one such method , since it is based on the expansion history @xmath187 . since the expansion factor @xmath25 is an essential feature of all metric theories of gravity , it is worthwhile trying to explore the properties of de by considering the generic form [ eq : taylor0 ] a(t ) = a(t_0 ) + a_0 ( t - t_0 ) + ( t - t_0)^2 + ( t - t_0)^3 + ... . cosmic acceleration appears to be a fairly recent phenomenon @xcite , so we can confine our attention to small values of @xmath188 in ( [ eq : taylor0 ] ) . the second and third terms in the rhs of ( [ eq : taylor0 ] ) have played a crucial role in the development of cosmology . indeed , for a long time the large errors in @xmath189 and @xmath190 impaired a precise picture of cosmic expansion and only recently has the sign of @xmath191 been determined to sufficient accuracy for us to make the statement that the universe is accelerating . links the ricci scalar @xmath17 to @xmath191 . therefore , an accelerating ( @xmath192 ) spatially flat universe corresponds to @xmath193 . ] keeping in mind the significant progress expected in observational cosmology over the next decade , we feel the time to be ripe to supplement @xmath194 and @xmath191 with @xmath195 , which is the next logical step in the hierarchy of cosmological parameters . it is remarkable that @xmath196 for the spatially flat @xmath2cdm model which is the ` standard ' cosmological model at present , as well as for the cdm model . supplementing @xmath197 ( the first statefinder ) with the second statefinder @xmath198 permits us to break this degeneracy and to characterize different de models in a very informative manner . was also dubbed the cosmic jerk in @xcite . ] the statefinder pair @xmath199 is a ` geometrical ' diagnostic since it depends upon @xmath25 and hence upon the space - time geometry . an important property of the statefinder is that spatially flat lcdm corresponds to the fixed point _ lcdm = 1,0 . [ eq : fixedpt ] so the basic question of whether de is the @xmath2-term or ` something else ' can be rephrased into whether the equality ( [ eq : fixedpt ] ) is satisfied for de . indeed , the departure of a de model from @xmath200 provides a good way of establishing the ` distance ' of this model from lcdm @xcite . the fact that different classes of de models show distinctly different behaviour when plotted in the @xmath199-plane adds to the practical utility of this diagnostic . see figure [ fig : statefinder ] and @xcite for some applications of the statefinder . finally , the pair @xmath199 can also be expressed in terms of @xmath201 quite simply [ eq : statefinder1 ] r & = & 1 + ( 1+w ) - , + s & = & 1+w- . it is instructive to note that the second statefinder can be rewritten as @xcite s=(+ p)p [ eq : statefinder2 ] where @xmath202 is the _ total _ pressure including that of de . from ( [ eq : statefinder2 ] ) we find that @xmath203 is very sensitive to the epoch when the total pressure in the universe vanishes . for @xmath2cdm , this takes place at the redshift @xmath204 , when the positive radiation pressure cancels the negative pressure in @xmath2 . consequently , @xmath205 as @xmath206 . in general , the redshift @xmath207 at which the total pressure in the universe becomes zero is quite sensitive to the nature of de . -18pt lll level & geometrical parameter & physical parameter + + 1 & @xmath208 & @xmath209 , + & & @xmath210 + + + 2 & @xmath211 & @xmath212 + & @xmath213 & @xmath214 + + + 3 & @xmath215 & @xmath216 , @xmath217 + + & @xmath218 & + + [ table : dark ] finally , returning to our analogy between an inflaton and dark energy , an interesting correspondence exists between the @xmath219 pair describing de and the _ slow roll parameters _ @xmath220 , @xmath221 , describing inflation q + 1 & = & + r - 1 & = & ( 2 - 3 ) . we end this section by noting that our current description of cosmology relies on parameters which are either _ geometrical _ or _ physical _ in nature . in the table above , we have divided cosmological parameters according to their _ level _ which is related to the number of differentiations of the expansion factor needed to construct that parameter . note that all relative energy densities in the table @xmath222 are defined using the present value @xmath223 of the newton gravitational constant in agreement with the definition ( [ einst ] ) . each higher level requires an additional differentiation of observational data @xmath224 and , therefore , demands a higher level of accuracy for the latter . for current data , we broadly have about 10% accuracy for quantities belonging to the first level , 50% for the second level , while determination of third level parameters lies in the future . by smoothing , we lower the level of a parameter by unity . relations between geometrical and physical parameters are the following : w(z ) = 2 q(z ) - 1 3 ( 1 - _ m(z ) ) ; + _ v(z ) = - _ m0(1+z)^3 ; + _ t(z ) = - _ mo(1+z)^3 ; + _ ( z ) = ( r - 3q - 4 + _ m0 ( 1+z)^3 ) . [ relation ] the nature of dark energy is clearly one of the outstanding physical and cosmological puzzles of this century . although many distinct theoretical models have been advanced to explain the cosmic acceleration , an alternative approach to study and understand de is to determine de properties from observational data in a model independent fashion , thus , _ reconstructing _ de from observations . in this review , we have attempted to briefly summarize some important approaches of cosmological reconstruction . these approaches are briefly classifiable into parametric and non - parametric methods . both methods appear to give consistent results when applied to current data . future directions for model independent reconstruction of dark energy have also been briefly discussed . in the first approximation , within current observational errors , the reconstructed de behaves like a cosmological constant . but future , much more precise data , may well show some deviations from this behaviour . however small , these could have a profound influence on the whole of physics . _ acknowledgments _ : we are grateful to dragan huterer , irith maor , yun wang and jun - qing xia for agreeing to provide figures for this review . vs acknowledges a useful conversation with ranjeev misra . as was partially supported by the russian foundation for basic research , grant 05 - 02 - 17450 , and by the research program astronomy " of the russian academy of sciences . he also thanks the centre emile borel , institut henri poincare , paris for hospitality in the period when this paper was finished . a. g. riess , 1998 , , 1009 [ astro - ph/9805201 ] ; s. j. perlmutter , 1999 , , 565 [ astro - ph/9812133 ] ; r. a. knop , 2003 , , 102(k ) [ astro - ph/0309368 ] . a. g. riess , 2004 , , 665(r ) [ astro - ph/0402512 ] . p. astier , 2005 , astron . astroph . * 447 * , 31 [ astro - ph/0510447 ] . s m. carroll , 2001 , living rev.rel . * 4 * , 1 [ astro - ph/0004075 ] ; p. j e. peebles and b. ratra , 2003 , rev.mod.phys . * 75 * , 559 [ astro - ph/0207347 ] ; t. padmanabhan , 2003 , phys . rep . * 380 * , 235 [ hep - th/0212290 ] ; v. sahni , 2004 , lect . notes phys . * 653 * , 141 [ astro - ph/0403324 ] ; 2005 , astro - ph/0502032 ; s. nojiri and s. d. odintsov , 2006 , hep - th/0601213 ; r. p. woodard , 2006 , astro - ph/0601672 ; e. j. copeland , m. sami and s. tsujikawa , 2006 , hep - th/0603057 ; s. bludman , 2006 , astro - ph/0605198 . a. lue , r. scoccimarro and g. d. starkman , 2004 , , 124015 [ astro - ph/0401515 ] ; k. koyama and r. maartens , 2006 , jcap * 0601 * , 016 [ astro - ph/0511634 ] ; g. deffayet , g. gabadadze and a. inglesias , 2006 , jcap * 0608 * , 012 [ hep - th/0607099 ] . v. sahni and l. m. wang , 2000 , 103517 [ astro - ph/9910097 ] ; s. dodelson , m. kaplinghat and e. stewart , 2000 , , 5276 [ astro - ph/0002360 ] ; c. rubano , p. scudellaro , e. piedipalumbo and s. capozziello , 2003 , , 123501 [ astro - ph/0311535 ] ; g. barenboim , o. mena , c. quigg , 2005 , , 063533 [ astro - ph/0412010 ] ; s. nojiri and s. d. odintsov , 2006 , , 139 [ hep - th/0603062 ] . j. m. f. maia and j. a. s. lima , 2002 , , 083513 [ astro - ph/0112091 ] ; m. szydlowski and w. czaja , 2004 , , 083507 [ astro - ph/0309191 ] ; 2004 , , 083518 [ gr - qc/0305033 ] ; n. j. nunes and j. e. lidsey , 2004 , , 123511 [ astro - ph/0310882 ] ; u. alam , v. sahni , t. d. saini and a. a. starobinsky , 2004 , astro - ph/0406672 ; v. b. johri and p. k. rath , 2005 , astro - ph/0510017 ; m. sahln , a. r. liddle and d. parkinson , 2005 , , 083511 ; s. tsujikawa , 2005 , , 083512 ; d. polarski and a. ranquet , 2005 , , 1 ; y. wang and k. freese , 2006 , , 449 ; f. simpson and s. bridle , 2006 , , 083001 ; a. a. sen , 2006 , jcap * 0603 * , 010 [ astro - ph/0512406 ] ; x. zhang , 2006 , astro - ph/0604484 ; astro - ph/0609699 . v. gorini , a. kamenshchik and u. moschella , 2003 , , 063509 [ astro - ph/0209395 ] ; w. zimdahl and d. pavon , 2004 , gen . . grav . * 36 * , 1483 [ gr - qc/0311067 ] ; a. k. d. evans , i. k. wehus , o. gron and o. elgaroy , 2005 , astron . astrophys . * 430 * , 399 [ astro - ph/0406407 ] ; u. debnath , a. banerjee and s. chakraborty , 2004 , class . quant . grav . * 21 * , 5609 [ gr - qc/0411015 ] ; x. zhang , 2005 , , 1 [ astro - ph/0503075 ] ; x. zhang , 2005 , int.j.mod.phys . * d14 * , 1597 , [ astro - ph/0504586 ] ; x. zhang , feng - quan wu , j. zhang , 2006 , jcap * 0601 * , 003 [ astro - ph/0411221 ] ; p. wu and h. yu , 2005 , int d * 14 * , 1873 [ gr - qc/05090365 ] ; e. e. o. ishida , 2006 , astro - ph/0609614 ; d. rapetti , s. w. allen , m. a. amin , r. d. blandford , astro - ph/0605683 .
stochastic partial differential equations ( spdes ) , or partial differential equations ( pdes ) with uncertainties , play an important role in many areas of engineering and applied sciences such as turbulence , flows in random media , solid mechanics , filtering , and finance . numerical simulations of spdes are typically based on the monte carlo ( mc ) or the polynomial chaos ( pc ) methods @xcite . in both cases , the long - time simulation of spdes proves to be quite expensive @xcite . in this paper , we extend the pc - based method developed for stochastic differential equations in @xcite to the setting of spdes . the pc method originates from the works @xcite and enables us to expand ( square integrable ) functionals of brownian motion in a basis of hermite polynomials . applications of such a framework to random flows and turbulence theory are examined in @xcite . more recently , works in @xcite combined the pc method with the karhunen loeve ( kl ) expansion @xcite to study structural mechanics problems . the generalized polynomial chaos ( gpc ) developed in @xcite next extends the hermite pc to a set of non - gaussian random parameters . the main advantage of the pc method is that it allows us to propagate stochasticity by providing expansions of quantities of interest in terms of appropriate uncertainties while in effect replacing the stochastic equations by systems of deterministic equations . once these deterministic equations are solved , the statistical properties of the solution can be readily inferred from the coefficients of the expansion , which facilitates uncertainty quantification . in some cases , the pc method can propagate uncertainties with a substantially lower cost than mc methods ; especially for low dimensional uncertainties @xcite ; see also @xcite . however , in cases of high dimensional random parameters , the efficiency of the pc method is reduced because of the large number of terms that appear in the expansion . the presence of a temporal random forcing is a serious challenge to the pc method as the number of stochastic variables increases linearly with time , which hinders the capability of the pc method for long - time computations @xcite . moreover , the optimality of the initial basis typically degrades as time evolves due to the presence of nonlinearities in the equation . these drawbacks were addressed in @xcite . in our previous work @xcite , we proposed the _ dynamical generalized polynomial chaos _ ( dgpc ) method to solve low dimensional stochastic differential equations ( sdes ) driven by white noise . the novelty of the dgpc algorithm is that it utilizes a restart procedure and constructs polynomial chaos expansions ( pces ) dynamically in time by using orthogonal polynomials of the projections of the solution at the current steps and the future random forcing @xcite . this allows the algorithm to keep reasonably sparse random bases , and to mitigate the aforementioned curse of dimensionality and loss of optimality . numerical experiments in that reference illustrate that such a restart procedure has the ability to accurately estimate long - time solutions of sdes ; see the relevant computational and theoretical details in @xcite . although the algorithm performs well for low dimensional sdes , extension to larger systems is challenging and requires serious modifications . in this manuscript , we extend dgpc to the framework of spdes driven by white noise . while solutions to spdes are , in general , high dimensional random fields , they may lend themselves in some cases to low dimensional representations @xcite . armed with this fact , we propose at each restart to use the kl expansion to compress the solution into a representation involving a finite number of random modes , i.e. , a projection onto a lower dimensional manifold . in cases where the modeling equations contain non - forcing random inputs other than brownian forcing , such as a random viscosity , the kl expansion is applied to the solution and the random parameters together so that the algorithm automatically selects the intrinsic variables , which have the largest influence on the solution . since kl expansions are known to be optimal in the mean square sense , at each restart point in time , only a few dominating , most energetic , random modes are chosen and incorporated into pce to represent the future solution . however , the combination of the random kl modes and the random forcing variables brings about high dimensionality . the computational challenges then become : ( i ) computing orthogonal polynomials of arbitrary multivariate distributions ; ( ii ) keeping the number of terms in the expansion as small as possible . the construction of orthogonal polynomials of evolving multivariate distributions is possible by estimating their statistical moments @xcite , which is , in general , a computationally intensive procedure . in this work , we estimate the moments of the solution using its pce through a sampling approach to greatly reduce their computational cost ; see @xcite for a similar sampling methodology . we also show that the method is robust with respect to re - sampling . the second challenge ( ii ) is a major problem for all pc - based methods . the kl decomposition is computationally expensive as it requires solving a large eigenvalue problem . for problems of moderate size , we find that the eigenvalue problem is solved efficiently by using a krylov subspace method . for larger problems , in order to mitigate both memory requirements and computational cost of the kl expansion , we find low - rank approximations to the covariance matrices based on their pc representations and without assembling them . the algorithm leverages randomized projection methods as introduced in @xcite , and uses appropriate random matrices to obtain fast and accurate solutions of large eigenvalue problems . after selecting the dominating modes in the kl expansion , we make use of the sparse truncation technique proposed in @xcite to further reduce the number of terms in a pce . for long - time computations , we also develop an adaptive restart scheme , which adapts the time lag between restarts based on the acuity of the nonlinear effects . the use of compression techniques to exploit intrinsic lower dimensional structures of solutions of spdes is not new and is in fact necessary in many contexts ; see @xcite . the novelty of our approach is that a lower dimensional representation of the solution is learned online and integrated into a pce to integrate future random forcing and represent future solutions . this procedure is computationally viable and the combination of the aforementioned ideas allows us to attain a reasonable accuracy in the long - time evolutions of spdes for a reasonable computational cost . as we are interested in the long - time evolution of spdes , we restrict ourselves here to dynamics with a dissipation mechanism . equilibrium statistics and asymptotic properties of the solutions are relevant in many applications and have been extensively studied in the literature @xcite . based on these motivations , we provide numerical experiments for a 1d randomly forced burgers equation and a 2d stochastic navier stokes ( sns ) system . all equations are driven by white noise and satisfy periodic ( spatial ) boundary conditions . to demonstrate the efficiency of our algorithm , we present both short- and long - time computations . in some cases , we model the viscosity as a random process . statistical moments obtained by the algorithm are compared to the standard mc methods for short times to assess the accuracy of the algorithm . results show promising speed - ups over standard mc methods . we exhibit convergence behavior in terms of the degree of the expansion , the number of random modes retained in the kl expansion , and the ( adaptive ) restart time . in some cases , we provide a purely pce - based numerical verification of the convergence of the process to its invariant measure . the outline of the paper is as follows . section [ sec : kl_pc ] introduces the basic setting for the pc and kl expansions . in the following section , the main algorithm and its implementation details are discussed . numerical experiments are conducted in section [ sec : numerical ] using several settings for the stochastic burgers equation and navier stokes system . some concluding remarks are offered in section [ sec : summary ] . throughout the manuscript , we consider the following time - dependent stochastic partial differential equation ( spde ) driven by the white noise @xmath0 : @xmath1 , \\ & u(x,0,\omega)= u_0(x,\omega ) , \quad x \in g , \ , \omega \in \omega , \end{aligned } \right.\end{aligned}\ ] ] where , for concreteness , @xmath2 is the @xmath3-dimensional torus so that @xmath4 and its derivatives are periodic functions in the variable @xmath5 . the dgpc algorithm may easily be extended to more general boundary conditions and geometries . above , @xmath6 is a possibly non - linear differential operator in the spatial variables . the solution takes values in @xmath7 . in the numerical simulations , the parameters @xmath3 and @xmath8 are set to either @xmath9 or @xmath10 . given a probability space @xmath11 , where @xmath12 is a sample space equipped with the sigma - algebra @xmath13 and the probability measure @xmath14 , we denote by @xmath15 the hilbert space of square integrable random fields on @xmath2 . for a random field @xmath16 , we define the expectation @xmath17 = \int_{\omega } u(x,\omega ) { \mathbb{p}}(d\omega),\ ] ] and the covariance @xmath18 , \quad x , y \in g,\end{aligned}\ ] ] where @xmath19 denotes the transpose . the kl expansion of a time - independent random field @xmath16 with a continuous covariance is as follows : @xmath20 where the eigenvalues @xmath21 are nonnegative , and the eigenfunctions @xmath22 and the mean zero random variables @xmath23 are given by @xmath24 \ , , \phi_l \rangle_{l^2(g)}.\end{aligned}\ ] ] here , @xmath25 denotes the spatial inner product on @xmath2 . the spatial and random modes satisfy bi - orthogonality , i.e. @xmath26 = \delta_{lk}$ ] and @xmath27 , where @xmath28 denotes the kronecker delta function ; see @xcite . the major feature of the kl decomposition is that the truncation after a finite number , denoted by @xmath29 hereafter , of terms is optimal in the @xmath30 sense . how @xmath29 should be chosen obviously depends on the spectrum of the covariance operator . when the process shows a high degree of correlation , then typically @xmath29 may be chosen relatively small due to the rapid decay of the eigenvalues @xcite . this makes the kl expansion a useful dimensionality reduction technique in many applications and will play a crucial role in our algorithm to compress the dimensionality of stochastic solutions of . we assume that the random solution @xmath31 of is a second - order stochastic process for all @xmath32 $ ] , and for simplicity , that the initial condition is deterministic . the pc method for is constructed as follows . given a complete countable orthonormal system @xmath33 $ ] , @xmath34 , we project the brownian motion @xmath35 onto @xmath36 $ ] by defining @xmath37 . then , @xmath38 consists of a countable number of independent and identically distributed standard gaussian random variables , and the convergence @xmath39 ^ 2 \rightarrow 0 , \quad k \rightarrow \infty,\end{aligned}\ ] ] holds for all @xmath40 . basic examples of orthonormal systems are trigonometric functions and wavelets ; see @xcite . the spde solution is a nonlinear functional of brownian motion on the interval @xmath41 $ ] . therefore , the projection onto @xmath42 enables us to consider the solution depending on a countable number of random variables , i.e. , @xmath43 . then , the polynomial chaos expansion ( pce ) of the solution @xmath44 at @xmath32 $ ] is given by @xmath45 . \end{aligned}\ ] ] note that the expansion is a sum of orthogonal projections onto orthogonal subspaces spanned by the wick polynomials @xmath46 , where @xmath47 is the @xmath48th degree normalized one - dimensional hermite polynomial and @xmath49 belongs to set of multi - indices with finite number of nonzero components @xmath50 the completeness of the orthonormal polynomials @xmath51 in @xmath30 , i.e. convergence of the expansion , is established in @xcite and the result is known as the cameron - martin theorem . in the following , we will use the term hermite pce to emphasize that the expansion utilizes gaussian random variables . all statistical information of the random field is contained in the deterministic coefficients @xmath52 of the expansion . specifically , the first two moments are given by @xmath53 = u_{\boldsymbol{0}}(x , t ) , \qquad { \mathbb{e}}[u^2 ] = \sum_{{\boldsymbol{\alpha}}\in \mathcal{j } } u_{\alpha}^2(x , t).\end{aligned}\ ] ] higher order moments can be obtained using the triple products @xmath54 $ ] or the hermite product formula for @xmath42 ; see @xcite . in practice , we truncate the expansion using polynomials of maximum degree @xmath55 and @xmath56 number of random variables @xmath57 , which results in the following approximation @xmath58 this simple truncation gives rise to @xmath59 terms in the approximation . two major problems of the pce in this context are as follows . to maintain a prescribed accuracy as time evolves , the degree of freedom @xmath56 should grow in accordance with the error behavior of , which brings about high dimensional randomness that needs to be incorporated in the expansion . therefore , the computational cost increases rapidly with dimension , which decreases the efficiency of the pc method . the second related problem of pc is that initial predetermined bases may lose their optimality and even fail to converge for long - time evolutions @xcite . to alleviate the aforementioned bottlenecks of the pc method , namely high dimensionality and long - term integration , several approaches have been proposed @xcite . in the following , we briefly summarize the approach called _ dynamical generalized polynomial chaos _ ( dgpc ) , which was developed in @xcite to solve ( reasonably low dimensional ) stochastic differential equations ( sdes ) driven by brownian motion over long times . the solution @xmath60 of a @xmath3-dimensional sde system can be seen as a nonlinear functional of the initial condition @xmath61 and the countable number of variables @xmath42 . the approach in @xcite hinges upon two important observations : ( i ) for sufficiently small time lags @xmath62 , the solution @xmath63 , can be efficiently captured by low - order polynomials in @xmath64 ; and ( ii ) the solution @xmath63 depends on brownian motion only on the interval @xmath65 $ ] by the markov property . these two remarks allowed us to employ evolving pces based on polynomials of the random forcing and projections of the solution onto underlying chaos spaces . in this way , the algorithm can forget " about the past and pc representations remain reasonably sparse in the evolving basis while the accuracy is retained for long times . to be more specific , and given a set of increasing restart times @xmath66 on @xmath67 , @xmath68 and @xmath69 , the algorithm projects the random variable @xmath70 onto a finite chaos space at each @xmath66 . this projection will be denoted by @xmath71 in the following . a major challenge in the implementation is to construct orthogonal polynomials of the varying distribution of the random variable @xmath71 . the pce gives rise to natural formulas to compute statistical moments of the random variable . the algorithm constructs orthogonal polynomials of @xmath71 based on the knowledge of its moments via a gram schmidt procedure . to obtain evolution equations for the expansion coefficients , the algorithm then computes necessary triple products for @xmath71 and performs galerkin projection onto subspaces spanned by orthogonal polynomials in @xmath71 and @xmath42 . convergence of the algorithm is related to the unique solvability of the moment problem of the measures associated to variables @xmath71 ; see @xcite . related theoretical aspects and the estimates of computational costs can be found in detail in @xcite . in @xcite , the algorithm is applied to low dimensional nonlinear sde systems and provides accurate long - time simulations with a small cost compared to standard hermite pce . in particular , we computed invariant measures of sdes in some cases , which standard hermite pce typically fails to achieve . in the subsequent sections , we propose an extension of the dgpc algorithm to spdes , which requires several key modifications . in this section , we present the algorithmic details of the dgpc method applied to the general spde . the main ideas are summarized as follows . let a decomposition @xmath72 be given . following our discussion in section [ sec : dgpc ] and using the markov property , the solution @xmath73 , @xmath74 , can be represented in a pc expansion in terms of @xmath75 and @xmath76 , where @xmath77 denotes the gaussian random variables required for brownian forcing on the interval @xmath78 $ ] . let @xmath79 denote the projection of the solution @xmath80 onto the polynomial chaos space . we will also use the shorthand notation @xmath71 to denote this projection . to construct a pce in terms of polynomials of @xmath71 , we separate the spatial dependence and randomness via the kl expansion : @xmath81 let @xmath82 denote the random modes . since the solution @xmath31 , @xmath83 $ ] , is a functional of the random forcing @xmath76 and modes @xmath84 , the next step pce @xmath85 is given by @xmath86,\end{aligned}\ ] ] with the notation @xmath87 , where @xmath88 denotes an orthonormal basis in its arguments . the expansion dynamically needs a pc basis depending on the random forcing and evolving random kl modes of the solution . the coefficients @xmath89 satisfy a pde system obtained by galerkin projection of the spde onto the space spanned by @xmath90 . statistical properties can be retrieved after solving the induced pde system . here are the computational bottlenecks of this approach : ( i ) the simple truncation yields a large number of terms in the expansion , which leads to long computational times to solve the deterministic evolution equations ; ( ii ) estimating the terms appearing in the kl expansion is a major computational bottleneck , especially in higher spatial dimensions ; and ( iii ) computation of the orthogonal polynomials @xmath91 may also require intensive amount of computation . in the following , we address these issues in turn . the number of terms in the simple truncation in the hermite pce increases rapidly with respect to @xmath55 and @xmath56 . given sufficient regularity of the solution , the expansion coefficients decay both in the number of gaussian variables @xmath56 and the degree of polynomials @xmath55 . this observation led the authors @xcite to introduce a sparse truncation of the multi - index set and retain a truncated random basis , which keeps lower ( higher ) order polynomials in @xmath92 with larger ( smaller ) subscripts . this truncation can be quantified using an estimate for the decay rate of the coefficients ; see @xcite . following @xcite , we introduce a sparse index @xmath93 and define the corresponding sparse multi - index set @xmath94 basically , the index @xmath95 keeps track of how much degree we want in each variable @xmath92 . using , one can define the corresponding version of the pc expansion which might have drastically reduced number of terms . this is possible by a suitable choice of the index @xmath95 so that ineffective cross terms in high degree polynomials are eliminated . at each restart @xmath66 , we employ the kl expansion for the projected random field @xmath71 , and the expansion is truncated after a finite number of @xmath29 terms . the decomposition yields the eigenvalues @xmath96 and the eigenfunctions @xmath97 for @xmath98 . therefore , in addition to the gaussian variables @xmath76 , we have the random modes @xmath84 in the pce so that the total number of random variables becomes @xmath99 . to accommodate @xmath84 , we extend the multi - index set to @xmath100 . this can also be done by the tensorization @xmath101 of the multi - index sets since @xmath76 and @xmath84 are independent . however , since tensorization yields higher number of terms in the pce for most values of @xmath56 and @xmath29 , it is not considered in the following . assuming that the orthonormal basis @xmath102 is constructed in the previous step , the solution @xmath80 is approximated using the truncated pce @xmath103 where the time dependence of the coefficients @xmath104 is omitted for brevity . to avoid confusion , we note that both the projection and its truncation will be denoted by @xmath71 . armed with this approximation , using the orthogonality of random bases , the covariance of @xmath71 is easily estimated by @xmath105 where @xmath106 . in practice , we consider a discretization of the spatial domain @xmath2 with an even mesh parameter @xmath107 and grid points @xmath108 , @xmath109 . denote by @xmath110 the resulting covariance matrix . in general , we can use a spectral method , e.g. , fourier series in the case of periodic functions , to approximate the coefficients @xmath111 as @xmath112 where @xmath113 are orthogonal global basis functions on @xmath2 and @xmath114 ; see @xcite . thus , @xmath104 is approximated by a vector @xmath115 on the grid . therefore , the dimension of the covariance matrix @xmath110 becomes of order @xmath116 . after computing the covariance matrix , the corresponding eigenvalue problem can be solved for the first @xmath29 largest eigenvalues . the random modes @xmath84 are given as @xmath117 ) \ , , \phi_{j , l } \rangle_{l^2(g ) } = \frac{1}{\sqrt{\lambda_{j , l } } } \sum_{{\boldsymbol{\alpha}}>0 } \langle u_{j,{\boldsymbol{\alpha } } } , \phi_{j , l } \rangle_{l^2 } \ , t_{{\boldsymbol{\alpha}}}({\boldsymbol{\xi}}_{j-1 } , { \boldsymbol{\eta}}_{j-1 } ) , \quad l=1,\ldots , d . \end{aligned}\ ] ] this representation yields the random modes @xmath84 as a function of @xmath118 and the previous modes @xmath119 . here , we assume that the integrals @xmath120 can be computed by an accurate quadrature method . when the solution @xmath4 is more than one - dimensional , several implementations of the kl expansion can be considered . for instance , we may apply the kl expansion to each component of the solution @xmath4 separately and incorporate the resulting individual random modes into one pce . although this approach certainly makes the kl step faster , we found that it yielded inaccurate results in dgpc and needed a large number of variables in the pce to represent the solution accurately since cross covariance structures between the components of the solution are lost . therefore , in the following , we implement the kl expansion directly to the multi - dimensional solution and produce one set of random modes @xmath121 which represents all of its components . depending on the resolution of the discretization of the domain @xmath2 and the dimension @xmath3 , assembling the covariance matrix and solving the corresponding eigenvalue problem may prove dauntingly expensive . several methods have been devised to reduce computational costs , such as fast fourier techniques ( e.g. , in the case of stationary covariance kernels ) or sparse matrix approximations together with krylov subspace solvers @xcite . here , the assembly of the covariance matrix is performed at each restart via the summation formula . in our one dimensional simulations , with @xmath122 , this assembly can be carried out reasonably fast . since the solution of the eigenvalue problem is required only for the number @xmath123 of eigenvalues and eigenfunctions , krylov subspace methods @xcite perform well . we also utilize the implicitly restarted arnoldi method to efficiently find the few largest eigenvalues and corresponding eigenfunctions ; @xcite . in the higher dimensional simulations , when @xmath124 , computing and storing such large covariance matrices become challenging . covariance matrices are , in general , not sparse and require @xmath125 units of storage in the memory . moreover , assembling a large matrix at every restart is computationally very expensive for long - time simulations . the problem of computing and storing a large covariance matrix resulting from the kl expansion is not new and was addressed before in @xcite . it was noted that although the covariance matrices were dense , they were usually of low - rank ; @xcite . based on this observation , we next introduce an approximation approach , which leverages low - rank structures and avoids assembling large matrices . a low - rank approximation @xmath126 tries to capture most of the action of the matrix @xmath110 by a product of two low - rank matrices @xmath127 and @xmath128 . several efficient algorithms , e.g. , the fast multipole method and @xmath129-matrices , depend on low - rank approximations @xcite . we approximate eigenvalues and eigenvectors of the correlation matrix @xmath110 by using low - rank approximations as follows . given a low rank approximation @xmath130 of the symmetric covariance matrix @xmath110 , where the matrix @xmath131 is of size @xmath132 with @xmath133 orthonormal columns , the eigenvalue problem of @xmath110 can be approximated efficiently by applying @xmath134 or @xmath135 algorithm to the much smaller matrix @xmath136 . in the dgpc setting , this amounts to computing @xmath137 the explicit assembly of the covariance matrix @xmath110 is avoided by computing only the matrix - vector product @xmath138 . an approximate eigenvalue decomposition @xmath139 is deduced from the eigenvalue decomposition of the smaller matrix @xmath140 by setting @xmath141 . the crucial step of the computation is the construction of a low - rank matrix @xmath131 with @xmath142 orthonormal columns that accurately describes @xmath110 . we tried an approach based on the discrete unitary fourier transform to map the coefficients @xmath143 to the frequency space and retain only the lowest frequencies . although this approach allowed us to obtain reasonable compressions of the covariance matrix and enabled faster computations , the following approach consistently yielded much better results . following @xcite , we construct the matrix @xmath131 using random projections . algorithm 4.1 in @xcite draws an @xmath144 gaussian random matrix @xmath145 and forms the matrix @xmath146 . the matrix @xmath131 with @xmath147 orthonormal columns is then obtained by the @xmath134 factorization @xmath148 . note again that we do not assemble the matrix @xmath110 . rather , the matrix - matrix product @xmath149 is computed as in equation @xmath150 . since we require the largest @xmath29 eigenvalues , the target low - rank becomes @xmath29 . as indicated in @xcite , we use an oversampling parameter @xmath8 by setting @xmath151 . typical values of @xmath8 are @xmath152 or @xmath153 . since eigenvalues decay rapidly in our applications , we found @xmath154 to be accurate . with overwhelming probability , the spectral error @xmath155 is bounded within a small polynomial perturbation of the theoretical minimum , the @xmath156 singular value of @xmath110 ; for relevant theoretical details , see ( * ? ? ? * section 10 ) . in practice , we found this randomized approach to be highly accurate in our computational simulations . moreover , since assembly of large covariance matrices are avoided , running times and memory requirements for the kl expansion in @xmath157 are reduced drastically compared to the previously described methods ; see section [ sec : sns ] . in this subsection , we consider the case in which the differential operator @xmath6 in contains additional random input parameters , i.e. , @xmath158 . the random inputs will be denoted by the process @xmath159 of a dimension @xmath160 . a typical case is that of a random viscosity , e.g. , depending on a set of uniformly distributed random variables . we assume that the process @xmath161 is independent of time and that the corresponding orthogonal polynomials are available ; for instance in the askey family @xcite . we first observe that the solution @xmath31 is now a functional of brownian motion @xmath162 and the random process @xmath161 . therefore , assuming @xmath30 integrability , it can be written as a pce in terms of the associated orthogonal polynomials of @xmath162 and @xmath161 . at the restart @xmath66 , there are two options to carry out the kl expansion : ( i ) apply the kl expansion to only the solution @xmath71 and keep the basis variables for @xmath161 in addition to @xmath84 in the next pce ; and ( ii ) compress the combined random variable @xmath163 using the kl expansion and denote by @xmath84 the combined random modes which represent both @xmath71 and @xmath161 . the first approach will yield pces which provide functional representations of the solution in terms of @xmath162 and @xmath161 for each time @xmath66 . in the second approach , the functional dependence of the solution in terms of @xmath161 is lost in the first kl expansion step . however , the moments of the solution can still be computed through the combined random kl modes . in many uq settings , rather than a functional dependence , it is statistical information of the underlying solution , e.g. , moments of the invariant measure in the long - time , that we are after . moreover , the second approach can be seen as a dimensionality reduction technique , which compresses @xmath71 and the process @xmath161 together , thereby further reducing the number of terms in pce . when additional random parameters appear in the equation , we found it reasonable to implement the second approach to reduce cost while the first approach may be used as a reference computation to assess accuracy . it is useful to note that by combining the random fields @xmath71 and @xmath161 , the algorithm automatically chooses the important part of the random process @xmath161 that influences the solution while keeping the moments of the solution accurate ; see section [ sec : numerical ] . after obtaining the random modes @xmath84 , @xmath164 , we need to construct the following orthonormal basis : @xmath165 notice that since @xmath76 is gaussian and identically distributed for each @xmath166 , the corresponding orthonormal polynomials of @xmath76 are known to be the hermite polynomials for each @xmath166 . however , the probability distribution of @xmath84 is arbitrary and changes at each restart . therefore , the computation of orthonormal polynomials is computationally intensive and can be performed using the gram schmidt method as follows . we note that the set can be computed based on the knowledge of moments of the variables @xmath76 and @xmath84 . following @xcite , we assemble the gram matrix @xmath167 with the entries @xmath168 , \quad { \boldsymbol{\alpha}}_k,{\boldsymbol{\alpha}}_l \in \mathcal{j}^r_{k+d , n}. \end{aligned}\ ] ] the matrix @xmath167 is a @xmath169-dimensional , square and symmetric matrix . for theoretical reasons , we assume that the moments up to @xmath170 exist and the measure of @xmath171 is non - degenerate . then , the cholesky factorization is employed to @xmath167 and the polynomials @xmath172 is found by inverting the resulting upper triangular matrix as @xmath173 where @xmath174 are real coefficients . the kl expansion yields uncorrelated random variables @xmath84 . if the underlying process is gaussian , it is known that these variables are also independent . however , in general , marginals of @xmath84 are dependent variables . multi - index operations can still be used to construct the polynomial set with respect to the joint distribution , although the estimation of multivariate moments of @xmath84 becomes necessary because of such a dependency . in this case , it is known that orthogonal polynomials are not unique and depend on the ordering imposed on the multi - index set ; see @xcite . in all computations , we use the graded lexicographic ordering for multi - indices . the completeness of the orthogonal polynomials @xmath175 is closely related to the moment problem of the random variables @xmath76 and @xmath84 . in particular , if the moment problem is uniquely solvable , i.e. , the measure is determinate , then the orthogonal polynomials are dense in @xmath30 @xcite . some basic conditions that guarantee determinacy of the measure of a continuous random variable on a finite dimensional space are compact support and exponential integrability . gaussian measures are determinate and the hermite pce converges by the cameron - martin theorem . however , in general , whether the distribution of @xmath84 is determinate or not is unknown . this problem is addressed in our previous work in the case of finite dimensional sde systems ; see @xcite . theoretical results are applied in the setting where the solutions are approximated by compactly supported distributions under appropriate assumptions . in the following , we assume that the measures associated to @xmath84 are determinate so that convergence is guaranteed , which is consistent with our numerical simulations ; see section [ sec : numerical ] . based on the above discussion , the orthonormal basis requires the computation of the moments . the exact moments of the gaussian variables @xmath76 are computed by analytical formulas and then stored during the offline stage . however , the distribution of @xmath84 is varying with @xmath166 . therefore , the computation of moments should be carried out based on information provided by the pce . several methods are available to compute moments of probability distributions in the pc context such as , e.g. , quadrature methods , monte carlo sampling , or a pure pce approach . this procedure is notoriously expensive and ill - posed @xcite . the pure pce approach computes the moments of @xmath84 by repeatedly multiplying its pce and taking expectation ; see @xcite . this approach is discussed in detail in our earlier manuscript @xcite and works reasonably well for a low dimensional sde systems . however , it becomes prohibitively expensive if the dimension of the random variables in the pc basis is even moderate ; see the computational complexity section in @xcite . therefore , in this work , we consider an alternative approach using monte carlo sampling , which drastically reduces the computational cost for computing moments compared to the pure pce approach . we assume that independent samples of the initial condition ( therefore , the samples of @xmath176 ) are provided so that the algorithm can be initialized . to construct the set , based on , we need to compute the first @xmath170 moments of the joint random variable @xmath171 . moreover , since the triple products will be required for the evolution of the pce coefficients , the first @xmath177 moments need to be computed as well ; see section [ sec : galerkin_proj ] . using the same ordering of the multi - index set @xmath178 , we require the computation of the following moments : @xmath179 = { \mathbb{e}}[{\boldsymbol{\xi}}_j^{(\alpha_1,\ldots,\alpha_k ) } ] \,{\mathbb{e}}[{\boldsymbol{\eta}}_j^{(\alpha_{k+1},\ldots,\alpha_{k+d } ) } ] , \quad { \boldsymbol{\alpha}}\in \mathcal{j}^r_{k+d,3n},\end{aligned}\ ] ] where we used the independence of @xmath76 and @xmath84 . let @xmath180 denote independent samples of the random modes for @xmath181 , where @xmath182 . then , provided the samples @xmath183 are given , the moments of @xmath184 can be approximated by @xmath185 \approx \frac{1}{s } \sum_{i=1}^s ( { \boldsymbol{\eta}}_j(\omega_i))^{(\alpha_{k+1},\ldots,\alpha_{k+d } ) } = \frac{1}{s } \sum_{i=1}^s \prod_{l=1}^d ( \eta_{j , l}(\omega_i ) ) ^{\alpha_{k+l}},\ ] ] where we used the usual multi - index notation for powers . therefore , multivariate moments are computed by a combination of the analytical formulas for @xmath76 and a sampling approximation for @xmath84 . note that in applications , we use small values of @xmath55 with a sufficiently large number of samples @xmath186 to guarantee accuracy . although we discussed computing moments based on samples , we have not explained how the samples of @xmath84 are acquired except for @xmath176 . the distribution of @xmath84 , @xmath187 , is evolving in time . however , owing to the pce of the each component @xmath188 , we can write @xmath189 this representation gives a natural way to sample from the distribution of @xmath84 by the recursion @xmath190 assuming that we obtained samples of @xmath191 at the previous restart @xmath192 . independent samples @xmath193 are obtained through sampling a multivariate gaussian distribution . therefore , on the subinterval @xmath194 $ ] , pce acts like a transport map which maps previously obtained samples of @xmath119 and new samples of @xmath118 to the samples of the new random modes @xmath84 . note that @xmath84 is a function of the variables @xmath118 and @xmath119 . therefore , the number of samples of @xmath84 should be ideally @xmath195 provided the same of number of samples @xmath186 is used for each @xmath118 and @xmath119 . however , in practice , this is not feasible in our method as the number of samples grows in time . instead , the equation keeps only the diagonal terms in the sample space . in the numerical simulations , we use large values of @xmath186 so that the loss of accuracy incurred from discarding some samples would be minimal . moreover , it is important to notice that entails samples from the joint distribution of @xmath84 so that monte carlo method is used to approximate the expectations while preserving the dependence structure of marginals . [ rmk : samples ] the method blends monte carlo sampling into a pc approach to exploit the virtues of the both methods , namely , rapid computation of expectations and spectral accuracy provided by the mc and pc methods , respectively . although the method is utilizing samples for the computation of moments , samples are not used in the evolution stage . the algorithm essentially propagates moments of the measures between successive times , where moments are computed using a sampling technique . to test the robustness of the method with respect to sampling , imagine that the algorithm starts with an infinite supply of independent samples of @xmath196 . we discard the first sample after using it to construct the corresponding orthogonal polynomials at the end of the first time interval and propagate the remaining samples with the pce map to construct ( an infinite supply of ) samples of @xmath197 . the algorithm iteratively estimates the distribution , hence the moments , of each @xmath84 while samples are discarded at each restart . we tested this idea by starting with a set of @xmath48 independent samples of @xmath176 and propagated them by pce for a maximum of @xmath48 restarts . we found that the accuracy of the calculations was not affected by such a re - sampling tool . this comparison showed the robustness of the algorithm under changes of samples . since in practice , such re - sampling increases the computational costs compared to , it is not considered in the numerical experiments presented in the next sections . we also emphasize that the sampling approach readily returns samples of the approximated solution at the endpoint @xmath198 through its kl expansion without a further sampling procedure . these samples can also be useful in uncertainty quantification to estimate further statistical properties of the solution such as probabilities on prescribed sets or probability density functions . once an orthonormal basis is obtained , the algorithm performs a galerkin projection onto the space spanned by the basis , and this requires the computation of triple products . using the representation , the required triple products can be written as @xmath199 = \sum _ { { \boldsymbol{\alpha}}_{k ' } \leq { \boldsymbol{\alpha}}_k}\sum _ { { \boldsymbol{\alpha}}_{l ' } \leq { \boldsymbol{\alpha}}_l } \sum _ { { \boldsymbol{\alpha}}_{m ' } \leq { \boldsymbol{\alpha}}_m } a_{k'k } \ , a_{l'l } \ , a_{m'm } \ , { \mathbb{e}}[({\boldsymbol{\xi}}_j,{\boldsymbol{\eta}}_j)^{{\boldsymbol{\alpha}}_{k ' } + { \boldsymbol{\alpha}}_{l'}+ { \boldsymbol{\alpha}}_{m ' } } ] , \end{aligned}\ ] ] where all multi - indices belong to the set @xmath100 . thus this formula can be computed by the knowledge of moments of order up to @xmath177 . depending on the choice of the sparse index @xmath95 , the multi - index @xmath200 might not be an element of @xmath178 . therefore , once we fix the multi - index set @xmath100 in the offline stage , we also compute and store the indices that are elements of @xmath178 . finally , we perform galerkin projection of the spde and obtain the following pde system for the coefficients @xmath89 of , @xmath83 $ ] : @xmath201 + \sigma(x ) \ , { \mathbb{e}}[t_{{\boldsymbol{\alpha}}}({\boldsymbol{\xi}}_j , { \boldsymbol{\eta}}_j ) \dot{w}(t)].\end{aligned}\ ] ] the first expectation in the above line is computed with the aid of the triple products and the second using the representation . note that the initial conditions @xmath202 can be obtained by noticing that the representation of @xmath71 is nothing but a sum involving linear polynomials in @xmath188 . it can therefore be rewritten in the basis @xmath203 with the help of galerkin projections . hence , the only coefficients that survive in at @xmath204 are the mean and the ones which correspond to the first degree polynomials in @xmath84 . then , the pde system can be solved in time using a time - integration method combined with the aforementioned spectral method . if the initial condition @xmath61 of is deterministic , we employ the hermite pce on the first interval @xmath205 $ ] , which does not necessitate the computation of the kl decomposition . so far , the method uses a predetermined restart time @xmath206 . for long - term simulations , an adaptive restart scheme that sets the restarts online depending on the properties of the solution can reduce the computational cost . we propose to adapt the restart time based on the following two criteria : ( i ) preserve the accuracy of the representation of the random forcing ; and ( ii ) mitigate the effect the nonlinearities in the accuracy of the polynomial expansions . for a prescribed number of dimensions to describe the random forcing , the algorithm can not take too large steps to preserve accuracy . also , nonlinearities force the solution to be less accurately described by low - degree polynomials in the initial condition as time increases . in both cases , we wish @xmath206 to be as large as possible for a given accuracy in mind . to this end , let @xmath207 denote the adaptive time - step starting from time @xmath66 . to ensure an accurate representation of the forcing term , we set a maximum value @xmath208 for @xmath209 for all @xmath166 , i.e. @xmath210 . in practice , @xmath208 is based on the error analysis of random forcing by a finite dimensional approximation ; see for instance @xcite . to address nonlinearities , we consider the following ratio for the pc coefficients @xmath111 : @xmath211.\end{aligned}\ ] ] the condition measures the norm ratio of the nonlinear terms in the variance to the norm of the variance . in applications , the ratio is computed at each time integration point . similar other conditions were used in different settings in @xcite . consider a threshold value @xmath212 . we propose the following conditions for adaptive time - steps using @xmath213 $ ] : a. @xmath214 , [ item : less3eps ] b. @xmath215 , [ item : bigger3eps ] where @xmath216 is a polynomial approximation to @xmath217 , which can be found by fitting a @xmath8-degree polynomial to @xmath217 on the interval @xmath218 $ ] . this approximation is only required for time values @xmath219 satisfying @xmath220 . the time - steps @xmath209 , @xmath221 are set adaptively . for short time - steps , we do not expect dynamics to change drastically between successive intervals . therefore , condition [ item : less3eps ] verifies whether the ratio is smaller than @xmath222 on the current interval , and then sets the adaptive time - step for the next interval . when @xmath223 , then the algorithm selects a bigger time - step by finding the root of @xmath224 . note that the current evolution on the interval @xmath225 $ ] is not prematurely stopped at the end point . although pces converge at any point inside the interval , errors , however , are known to wildly oscillate inside the interval and become spectrally accurate only at the end point ; see @xcite . ( this is because @xmath226 in is spectrally accurate at @xmath227 and much less so on @xmath67 . ) condition [ item : bigger3eps ] essentially verifies whether the ratio becomes too large ( i.e. @xmath228 ) , and when this happens , forces the evolution to restart from the current initial point @xmath192 . this control ensures that the algorithm does not take too large steps . our procedure is summarized in algorithm [ alg : dgpc ] , where , for simplicity , we only present the version which uses a predetermined number of restarts . + decompose the time domain @xmath41 = [ 0,t_1 ] \cup \ldots \cup [ t_{n-1},t]$ ] + initialize the degrees of freedom @xmath229 + choose the sparse index @xmath230 + compute the indices used in the triple - product formula + compute moments of @xmath231 apply the kl expansion to @xmath71 and obtain @xmath232 compute the moments @xmath233 $ ] construct orthogonal polynomials @xmath175 compute the associated triple products perform galerkin projection onto @xmath234 set up initial conditions for @xmath235 evolve the pce coefficients of @xmath235 we now present numerical simulations for the burgers equation in one spatial dimension and a navier stokes system in two spatial dimensions both driven by white noise . we consider these equations for two reasons . first , the statistical behavior of solutions of these equations is of importance in statistical mechanics and turbulence theory ; see , e.g. , @xcite . second , they serve as challenging test beds for the pce methodology . we illustrate convergence results in terms of the degrees of freedoms @xmath29 , @xmath55 and the time - step @xmath206 , and consider both short - time and long - time evolutions . the convergence of the method in terms of @xmath56 is treated in detail in @xcite . as a general comment , we do not recommend using large values of @xmath55 since the computation of orthogonal polynomials is quite ill - posed . in the following , we use polynomials of degree up to @xmath236 . the algorithm mitigates the ill - posedness by choosing frequent restarts and small degree ; i.e. small @xmath206 and @xmath55 . the settings we consider here closely follow those addressed in the manuscripts @xcite . we consider the following one dimensional viscous stochastic burgers equation for @xmath31 : @xmath237 \times [ 0,t ] , \end{cases}\end{aligned}\ ] ] where @xmath226 is a brownian motion in time , @xmath238 is the viscosity ( which will be either deterministic or random ) , the initial condition @xmath61 is deterministic , and the solution itself is periodic in the spatial domain @xmath239 $ ] . following @xcite , we choose cosine functions as an orthonormal basis @xmath240 , for @xmath241 $ ] . employing the equation for each subinterval @xmath242 $ ] and using galerkin projection , we derive the governing equations for the pc coefficients @xmath89 of @xmath235 : @xmath243,\ ] ] where @xmath244 . since the initial condition is deterministic , we employ hermite pce in the subinterval @xmath245 $ ] . the pc coefficients @xmath246 of the nonlinearity @xmath247 are computed by multiplying the corresponding pces with the help of pre - computed triple products . since the coefficients @xmath89 are periodic in the physical space , we utilize a truncated fourier series : @xmath248,\ ] ] with the even number of frequencies @xmath249 to be chosen . further , using the equidistant partition for the spatial domain @xmath239 $ ] and fast fourier transform ( fft ) , we can compute the fourier coefficients @xmath250 with a reasonable computational cost . this procedure gives rise to an ode system which is then integrated in time using a second order predictor - corrector method combined with an exponential integrator for the stiff term @xcite . in the implementation of the burgers equation , the algorithm assembles the covariance matrix at each restart as discussed in section [ sec : klexpansion ] . a large memory is not required for such a one dimensional spatial problem . we found that using the random projection technique described in section [ sec : klexpansion ] did not result in significantly shorter total computational times because the computation of the eigenvalue problem is already efficient in this case by means of krylov subspace methods . [ ex : burgers1 ] for this numerical simulation , we choose the spatial part of the random forcing as @xmath251 , the initial condition as @xmath252 and set the viscosity @xmath253 . under these sets of parameters , it has been proved that there exists a unique invariant measure , which is the long - time limit of the dynamics ( * ? ? ? * theorem 2 ) . thus , the main aim of the following simulations is to demonstrate the efficiency of the algorithm in the long - time setting . for the parameters of dgpc , we take @xmath254 , @xmath255 , and vary the number of the kl modes @xmath29 . final time is @xmath256 and the interval divided into @xmath257 pieces by taking @xmath258 . the spatial mesh size @xmath249 is set to be @xmath259 and the number of samples @xmath186 to compute moments is taken as @xmath260 . finally , sparse indices @xmath95 are chosen as follows : a. if @xmath261 , then @xmath262 , and if @xmath263 , we set @xmath264 resulting in @xmath265 terms in the expansion . b. if @xmath266 , then @xmath267 , and if @xmath263 , we set @xmath268 resulting in @xmath269 terms in the expansion . c. if @xmath270 , then @xmath271 , and if @xmath263 , we set @xmath272 resulting in @xmath273 terms in the expansion . note that the first @xmath56 indices in @xmath95 correspond to degrees of polynomials in @xmath42 and the remaining to @xmath121 . choosing @xmath267 means using only first degree polynomials is @xmath274 and @xmath275 . also , setting @xmath268 for @xmath276 eliminates the cross terms involving first degree polynomials of @xmath275 in the second order terms . note that using a sparse index not only reduces the number of terms in pce but also alleviates the computation of moments . to compare our algorithm , we use a second order weak runge - kutta scheme since the exact solution is not available . to make a fair comparison , both algorithms ( monte carlo & dgpc ) use the same time - step @xmath277 and the same mesh size @xmath278 . the number of samples used in mc algorithms are @xmath279 and @xmath260 and the corresponding algorithms will be denoted by mc1 , mc2 and mc3 , respectively . the exact solution is taken as the result of mc3 and the relative @xmath30 error @xmath280 - { \mathbb{e}}[u_{mc } ] ||_{2}/|| { \mathbb{e}}[u_{mc}]||_2 $ ] of the mean is computed . errors for higher order centered moments are computed similarly . from figure [ fig : ex1_1 ] , we observe that all errors grow with time in the initial stages and in particular , the degree of freedom @xmath261 is the least accurate , which is expected as the dynamics change rapidly during initial stages . increasing the number of kl modes entails more accurate expansion up to some order . it can be observed that all the error levels stabilize for moderate times while @xmath270 is the most accurate . this phenomenon is explained by the convergence to a stationary measure such that statistics do not change considerably after some time . we now increase the final time to @xmath281 . dgpc algorithms use the following parameters : @xmath282 , @xmath255 , @xmath283 , @xmath284 and @xmath258 . the corresponding total number of terms for each subinterval becomes @xmath285 , @xmath286 and @xmath287 . the mesh size is taken as @xmath288 , which offers better spatial resolution . table [ table : burgdet ] summarizes the relative @xmath30 errors of the dgpc algorithms with different degrees of freedom and the mc methods . all errors are computed by taking mc3 as the exact solution . the time ratio column is computed as the total time required by the each algorithm divided by the elapsed time of mc3 with @xmath289 and @xmath290 . the parameters for mc1 and mc2 algorithms remain the same as above ; the algorithms are executed a few times and the resulting errors are averaged . we also include the elapsed times for the offline computation in dgpc algorithms . in practice , the required data for the offline step can be computed once and stored for further executions of the algorithm to speed - up the running time . .relative errors for the centered moments by dgpc and mc methods at t=6 . each time ratio is computed by comparing to mc3 . [ cols=">,^,^,^,^,^",options="header " , ] table [ table : rel_err_vor_rand_vis ] exhibits the relative errors of dgpc for the vorticity using the random matrix approach for the kl expansion . comparing table [ table : rel_err_vor_rand_vis ] and [ table : rel_err_fourier_truncation ] , we see that relative elapsed times of dgpc with respect to mc3 are further improved . additional randomness for mc means an extra dimension to sample from , whereas for dgpc , it means an extra variable that needs to be compressed into the modes @xmath121 . since the dynamics crucially depend on the behavior of the viscosity , using few realizations for viscosity sampling in mc is not recommended . iin this setting , we found that our mc simulations demanded a high cpu time compared to dgpc . note , however , that viscosity sampling could clearly be performed in parallel in a mc framework something that is not as easily feasible in the pce setting . the preceding simulations were concerned with short time evolutions of sns and comparisons of the proposed algorithm with a monte carlo method . numerical results for reasonably short time computations indicated that our algorithm achieved a similar accuracy compared to mc typically for a smaller computational cost . we are now interested in long time simulations and convergence to steady states . since there is no random forcing acting upon the temperature equation in , the ( uncoupled ) temperature diffuses to zero quickly . therefore , we only solve the vorticity equation in the system . the following numerical experiment considers the vorticity equation with a deterministic viscosity @xmath291 and a spatial forcing as described in example [ ex : sns_short1 ] . the parameters of the simulation are @xmath292 , @xmath293 , @xmath294 and @xmath295 . pc expansions with thirty number of terms are employed on each subinterval . the four - step adams predictor - corrector method is used for the time integration . figure [ fig : ex7_1 ] shows three different initial conditions for the vorticity . the first layer is supported around @xmath296 while the others are aligned horizontally . widths of all layers are widened and different sinusoidal perturbations are considered . in figure [ fig : ex7_2 ] , we show the @xmath30-norm of the successive differences of the first two moments in time . each column represents one of the initial conditions presented in the corresponding column in figure [ fig : ex7_1 ] . after a ( very ) long time , the norms of the successive differences drop below @xmath297 , which ( numerically ) indicates that statistical moments no longer change significantly in time . in all cases , we found that the dynamics converged to the same state , which is shown in figure [ fig : ex7_3 ] . notice that the invariant measure is a non - gaussian random field and the moments have oscillations in the @xmath5 variable . we also see that high variance regions correspond to where the mean fields display peaks . based on these findings for this scenario , we assert that the dynamics converge to an invariant measure which is numerically captured in the long - time by the dgpc algorithm . + in long time computations , the mc method usually requires the propagation of many realizations in time , which renders the method hardly affordable in some cases . however , if the dynamical system possesses a unique ergodic invariant measure , the mc method may be carried out to sample such a measure by considering a single , very long time mc realization , which repeatedly visits the whole state space . while carrying out such a sampling is also computationally expensive , it is likely to compare favorably to our dgpc algorithm in this case . in general , ergodicity or uniqueness of an invariant measure may not be known or not hold for complicated physical dynamics ( e.g. , invariant measures parametrized by a random parameter as in example [ ex : randomvis ] ) . in such cases , our algorithm offers a viable alternative to the mc method to capture the long - term dynamics by providing statistical information resulting from the expansion coefficients . we have presented a pc - based algorithm , called dynamical generalized polynomial chaos , to tackle long - time integration and high dimensional randomness in the context of pdes with markovian forcing . to deal with these challenges , dgpc uses a restart procedure , which constructs pces dynamically based on the polynomials of the projections of the solution , the random forcing and the random parameters found in the equation . the karhunen loeve expansion is employed at each restart to find a representation of the solutions which correspond to low - dimensional dynamics of the underlying physical processes . the relevant modes are then incorporated in a pce for the future evolution . using sparse multi - index sets and frequent restarts , the algorithm provides an efficient way to capture the solutions in a fairly sparse random bases in terms of orthogonal polynomials of dynamically evolving measures . the main computational bottlenecks of the algorithm are the simulation of the deterministic evolution equation , the kl expansion , and the computation of moments . the cost of the deterministic evolution is dictated by the complicated nature of the spdes . the kl expansion is an expensive dimensionality reduction technique . since the algorithm constructs pces online , kl expansions ( or other dimensionality reductions methods ) are unavoidable at each restart . we found that for large covariance matrices , the kl cost was drastically reduced when the covariance matrix was estimated by a low - rank approximation obtained by random projections . the estimation of the orthogonal polynomials and corresponding triple products of evolving arbitrary measures is also a costly step as in most pc - based methods . using a 1d randomly forced burgers equation and a 2d stochastic navier stokes system , we provided several numerical simulations for both short- and long - time solutions . we compared the accuracy and computational time of the algorithm to the standard monte carlo method and found that the proposed algorithm achieved similar error levels for a ( generally significant ) lower computational cost in most cases . the substantial speed - up of dgpc is especially promising when the equation contains additional , time - independent , random contributions , which is one of the main reasons to use pc based methods in general . to demonstrate the efficiency of the algorithm for long time simulations , we computed invariant measures for both equations , which is not a trivial task for two dimensional navier stokes systems . other methods such as the recent multilevel monte carlo techniques offer improvements over the standard mc methods @xcite . the restarting step of our algorithm remains expensive computationally , especially for problems in two ( or three ) spatial dimensions . however , the restart method provides a viable means to keep the number of random variables to reasonable levels . its ability to compute statistical properties of long - time evolutions for fairly complicated equations is quite promising . the authors would like to thank the reviewers for their critical reading of the manuscript , and for multiple remarks and suggestions that improved the presentation of our method . this work was partially funded by nsf grant dms-1408867 and onr grant n00014 - 15 - 1 - 2679 . h. c. ozen , g. bal , http://dx.doi.org/10.1137/15m1019167[dynamical polynomial chaos expansions and long time evolution of differential equations with random forcing ] , siam / asa j. uncertain . quantif . 4 ( 1 ) ( 2016 ) 609635 . r. ghanem , j. red - horse , propagation of probabilistic uncertainty in complex physical systems using a stochastic finite element approach , phys . d 133 ( 1 - 4 ) ( 1999 ) 137144 , predictability : quantifying uncertainty in models of complex phenomena ( los alamos , nm , 1998 ) . g. bal , http://dx.doi.org/10.1007/978-3-319-11259-6_9-1[propagation of stochasticity in heterogeneous media and applications to uncertainty quantification ] , in : r. ghanem , d. higdon , h. owhadi ( eds . ) , handbook of uncertainty quantification , springer international publishing , cham , 2016 , pp . 124 . g. potte , d. lucor , http://dx.doi.org/10.1016/j.jcp.2011.12.038[non intrusive iterative stochastic spectral representation with application to compressible gas dynamics ] , j. comput . phys . 231 ( 9 ) ( 2012 ) 35873609 . g. berkooz , p. holmes , j. l. lumley , the proper orthogonal decomposition in the analysis of turbulent flows , in : annual review of fluid mechanics , vol . 25 , annual reviews , palo alto , ca , 1993 , pp . 539575 . m. jardak , c .- h . su , g. e. karniadakis , http://dx.doi.org/10.1023/a:1015125304044[spectral polynomial chaos solutions of the stochastic advection equation ] , in : proceedings of the fifth international conference on spectral and high order methods ( icosahom-01 ) ( uppsala ) , vol . 17 , 2002 , pp . 319338 . m. arnst , r. ghanem , e. phipps , j. red - horse , measure transformation and efficient quadrature in reduced - dimensional stochastic modeling of coupled problems , internat . methods engrg . 92 ( 12 ) ( 2012 ) 10441080 . n. halko , p. g. martinsson , j. a. tropp , http://dx.doi.org/10.1137/090771806[finding structure with randomness : probabilistic algorithms for constructing approximate matrix decompositions ] , siam rev . 53 ( 2 ) ( 2011 ) 217288 . j. xu , j. li , sparse wiener chaos approximations of zakai equation for nonlinear filtering , in proceedings of the 21st annual ieee international conference on chinese control and decision conference ( ccdc09 ) ( 2009 , ) pp . 910913 . j. s. hesthaven , s. gottlieb , d. gottlieb , spectral methods for time - dependent problems , vol . 21 of cambridge monographs on applied and computational mathematics , cambridge university press , cambridge , 2007 . r. b. lehoucq , d. c. sorensen , c. yang , arpack users guide : solution of large - scale eigenvalue problems with implicitly restarted arnoldi methods , vol . 6 of software , environments , and tools , society for industrial and applied mathematics ( siam ) , philadelphia , pa , 1998 . b. j. debusschere , h. n. najm , p. p. pbay , o. m. knio , r. g. ghanem , o. p. le matre , numerical challenges in the use of polynomial chaos representations for stochastic processes , siam j. sci . comput . 26 ( 2 ) ( 2004 ) 698719 . h. n. najm , uncertainty quantification and polynomial chaos techniques in computational fluid dynamics , in : annual review of fluid mechanics . 41 , vol . 41 of annu . fluid mech . , annual reviews , palo alto , ca , 2009 , pp .
by far the study of neutron stars has been mainly focused on the relationship between the equation of state ( eos ) of nuclear matter and the observed maximum mass . the connection has been achieved by solving the hydrostatic equilibrium equations based on general relativity . the first generation observed masses exhibited an average value around 1.5 solar masses . this value requires a soft eos that can be easily obtained by introducing new degrees of freedom like hyperons , kaons or quarks accompanied or not by a phase transition . sometimes the softening was so large that the neutron star is predicted to collapse into a black hole , as for the sn1987a @xcite . in the new generation of observations the masses are distributed within a large range , up to 2 solar masses , that requires a stiff eos , i.e. , hadronic matter without new degrees of freedom . since it is hard to imagine pure hadronic matter to sustain the high pressure predicted in the inner core , new scenarios have to be advanced to explain the coexistence , in the phenomenology of neutron stars , of low and high mass spectra . recently @xcite it has been argued that the two observed classes of neutron stars might correspond to two different evolutionary scenarios of neutron stars . in one case , the hot and dense remnant of the supernova explosion rapidly evolves into a hybrid star , where the transition to a quark phase softens the nuclear matter so that @xmath0 ; in the other case a slow evolution could lead the neutron star to a large mass via a mass accreting from the coupling with a white dwarf . from this point of view the destiny of the remnant is strongly affected by the initial conditions , i.e. density , temperature , leptonization degree etc . for instance , if the mass of the remnant is below the mass threshold for quark nucleation the transition to the quark phase is forbidden @xcite . if the mass is slowly accreting the transition is allowed . the role of temperature or other parameters defining the initial state of a new born neutron star has not yet been studied . to investigate the possible phase transition to quark matter in neutron stars , we need also to know the eos of quark matter . although we have in hand the fundamental theory of strong interactions , i.e. , quantum chromodynamics ( qcd ) , we still do not know the true ground state . it is now generally expected that quark matter is in the color - flavor locked phase ( cfl ) @xcite at extremely high densities when the finite current mass of strange quarks becomes unimportant . in the density range from nuclear saturation to cfl , there may exist a rich and varied landscape of phases , e.g. , the 2sc , g2sc , gcfl etc . presently , however , these phases suffer from the so - called chromomagnetic instability problem for both the two- @xcite and three - flavor @xcite cases . on the other hand , experiments show that quarks become asymptotically free rather slowly @xcite . therefore , in the present study we are dealing with the ordinary strange quark matter ( sqm ) @xcite . the special problem in studying the eos of ordinary quark matter is to treat quark confinement in a proper way . in the conventional standard approach , an extra constant term , the famous bag constant @xmath1 , is added to the energy density of the system , which provides a negative pressure to confine quarks within a finite volume , usually called a ` bag ' . the quark mass is infinitely large outside the bag , and a finite constant within the bag . a vast quantity of investigation have been performed within the framework of the bag model @xcite . as is well known , however , particle masses vary with environment . such masses are usually called effective masses . effective masses of hadrons and quarks have been extensively discussed , e.g. , within the nambu - jona - lasinio model @xcite and within a quasi - particle model @xcite . in principle , the density dependence of quark masses should be connected to the in - medium chiral condensates @xcite . taking advantage of the density dependence , one can describe quark confinement without using the bag constant . instead , the quark confinement is achieved by the density dependence of the quark masses derived from in - medium chiral condensates @xcite . the two most important aspects in this model are the quark mass scaling @xcite and the thermodynamic treatment @xcite . both aspects will be reviewed in this paper . in the present contribution , the transition from hadron phase ( hp ) to strange quark phase ( sqp ) in the inner core of a neutron star is investigated within the fully consistent nuclear and quark models . in the hadron sector we adopt the equation of state from brueckner - bethe - goldstone ( bbg ) approach with three - body forces ( tbf ) @xcite . this theory , being a completely microscopic approach , can easily incorporate degrees of freedom such as nucleon resonances [ @xmath2(1232 ) or @xmath3(1440 ) ] , which are expected to appear at higher hadron densities . it is found that the mixed hadron - quark phase can occur , for reasonable values of the confinement parameter , a little above the normal saturation density , and can undergo the transition to pure quark matter at about 5 - 6 times the saturation . this result is quite different from the previous results from nambu - jona - lasinio ( njl ) model in which the mixed quark phase can not appear at neutron - star densities @xcite . afterwards , the influence of the mixed and quark phases on the structure of compact stars is discussed by solving the tolman - oppenheimer - volkov ( tov ) equation and extracting the mass - radius plots for neutron stars . finally , it is shown that the transition to the deconfined phase turns out to be incompatible with the onset of kaon condensation . sqm has been one of the hot topics in nuclear physics since witten s famous stability conjecture @xcite . in many studies , the quark confinement was treated adopting the bag mechanism @xcite . an alternative approach to obtain confinement is based on the density dependence of quark masses @xcite . this mechanism has been extensively applied to investigating the properties of sqm @xcite . in this section , we first give a short review of the two most important aspects , with the main inconsistency of the original model . then we present a fully self - consistent thermodynamic treatment . the properties of sqm will be given in the new treatment . in the present paper , however , the main application of the new approach is the study of the phase transition in compact stars . as mentioned above , the quark confinement in this model is achieved by the density dependence of quark masses . therefore , the first important question is how to determine the quark mass scaling which can reasonably produce confinement . originally , the interaction part of the quark masses was assumed to be inversely proportional to the density @xcite . this linear scaling has been extensively applied to studying the properties of sqm . there are also other mass scalings @xcite . their main drawback is that they are pure parametrizations without any convincing derivation . therefore , a cubic scaling was derived based on the in - medium chiral condensates and linear confinement at both zero @xcite and finite temperature @xcite . this new scaling has been applied to investigate the viscosity of sqm and the damping time scale that is due to the coupling of the viscosity and r mode @xcite , the quark - diquark equation of state and compact star structure @xcite , the properties of strangelets versus the electric charge and strangeness @xcite , and the new solutions for cfl slets @xcite . in the present paper , we use the chirally determined quark mass scaling @xcite to study the phase transition in neutron stars . for this we need a completely self - consistent thermodynamic treatment of the eos of quark matter . the thermodynamic treatment of the system with confinement via the density dependent quark masses has been a long story . originally , the thermodynamic formulism was regarded the same as the constant - mass case @xcite . in this first treatment , the internal pressure can not be zero , and the properties of sqm were rather different from those in the bag model . but it was later pointed out that the difference was caused by the incorrect thermodynamic treatment @xcite . it was found that an additional term is to be added to the pressure and energy expressions @xcite . this second treatment makes it possible sqm to be self - bound . however , two serious problems came out : one is the unreasonable vacuum limits , the other one is the discrepancy between the energy minimum and zero pressure . it was shown that the added term in the pressure , due to the density dependence of quark masses , should not be appended to the energy . after discarding this term in the energy , while keeping it in the pressure , the two inconsistencies mentioned above were immediately removed @xcite . this third treatment has recently been extended to finite temperature @xcite . the thermodynamic formulism in ref . @xcite was also adopted in ref . @xcite , though a different quark mass scaling was used there . a common feature of the last two thermodynamic treatments @xcite , as well as other recent references using this model @xcite , is that they all regard the thermodynamic potential the same as in the fermi gas model . because of the additional term , the pressure becomes obviously not equal to the minus thermodynamic potential density , contradicting the thermodynamic equality @xmath4 for a homogeneous system . one can also easily check that the fundamental differentiation equality @xmath5 for homogeneous systems at zero temperature was not fulfilled in the mentioned references . in the rest of this section , we will present a fully self - consistent thermodynamic treatment of the confinement by density - dependent mass model ( cddm ) . let us consider a quark model with three flavors . denoting the fermi momentum in the phase space by @xmath6 , the particle number densities can then be expressed as @xmath7 and the energy density as @xmath8 equations ( [ nimod ] ) and ( [ emod ] ) are familiar expressions , where the summation index goes over all considered particle types . to let the model be valid for both particles and anti - particles , the particle number density , or accordingly , the fermi momentum , are formally assumed to be negative for anti - particles . therefore , in the upper limit of the integration , the absolute value has to be taken . if the particle masses @xmath9 are constant , the relation between the fermi momenta @xmath6 and the chemical potentials @xmath10 is @xmath11 as is well - known , however , the quark mass depends on density and temperature . in principle , the quark mass scaling should be determined from qcd , which is obviously impossible presently . based on the in - medium chiral condensates , a cubic scaling law was derived at zero temperature @xcite , and it has been recently extended to finite temperature @xcite . at zero temperature , we have the simple cubic scaling @xmath12 where @xmath13 is the quark current mass , @xmath14 is the total baryon number density , the exponent of density is @xmath15 @xcite , and the constant @xmath16 is to be discussed a bit later . in the following , we show that the density - dependence of particle masses will modify the fermi momentum , i.e.the relation in eq . ( [ nuimuifree ] ) for free - particle systems should be modified to include interactions . in fact for the quark flavor @xmath17 we have @xmath18 since the quark masses are density dependent , the derivatives generate an additional term with respect to the free fermi gas model . we get @xmath19 where @xmath20.\ ] ] the pressure is then given by @xmath21 with @xmath22 being the free - particle contribution : @xmath23.\end{aligned}\ ] ] due to the additional term in the chemical potential , the pressure also has an extra term . the inclusion of such a term guarantees that the hughenoltz - van hove theorem is fulfilled in the calculations . in the quasi - particle model @xcite , one also has an extra term . because the quark masses there depend on chemical potentials , and the extra term is not used in the relation between the fermi momenta and chemical potentials , an effective bag constant has to be added to the energy expression to consider confinement @xcite . in cddm quark model , however , we do not need a bag constant any more . quark confinement is achieved automatically by the density dependence of quark masses , or by the strong interaction between quarks . in fact , the exponent @xmath24 in eq . ( [ mqt0 ] ) is derived from the linear confinement interaction @xcite . in the present model , the parameters are : the electron mass @xmath25 mev , the quark current masses @xmath26 , @xmath27 , @xmath28 , and the confinement parameter @xmath16 . although the light - quark masses are not without controversy and remain under active investigations , they are anyway very small , and so we simply take @xmath29 the current mass of strange quarks is @xmath30 mev according to the latest version of the particle data group @xcite . conventionally , the stability of strange quark matter ( sqm ) is judged by the minimum energy per baryon @xcite . if it is less than 930 mev ( the mass of @xmath31fe divided by 56 ) , then sqm is absolutely stable . if it is bigger than 930 mev but less than 939 mev ( the mass of nucleons ) , then it is metastable . otherwise , if it is larger than 939 mev , sqm is un - stable . in case of two - flavor quark matter , it should be no less than 930 mev , in order not to contradict standard nuclear physics . this is the witten - bodmer hypothesis @xcite . in fig . [ dms0 ] , we show the different regimes in the @xmath32-@xmath28 plane . the area below the full line is forbidden where the energy per baryon of two - flavor quark matter is less than 930 mev . above the dotted line , the energy per baryon of sqm is more than 939 mev , and thus sqm is unstable . the area bounded by the dotted and dashed lines is the metastable region where the energy per baryon is between 930 mev and 939 mev . only when the @xmath33 pair is in the range between the full and dashed lines , sqm can be absolutely stable , i.e. , its energy per baryon is less than 930 mev . therefore , the range of @xmath16 values is very narrow for a chosen @xmath28 value , if the witten - bodmer hypothesis is correct . if we take the modest value @xmath34 mev , for example , then @xmath35 is in the range of 158160 mev . the lower bound 158 mev is obtained by taking @xmath36 . if @xmath26 and @xmath27 are given a small finite value , the lower bound can then be a little bit smaller , e.g. 156 mev @xcite . unfortunately we presently do not have a definite conclusion on the stability of sqm , so we treat @xmath16 as a free parameter . however , the first condition , i.e. , @xmath16 greater than about @xmath37 , should always be satisfied . on the other hand , we can connect @xmath16 to the pion mass @xmath38 , pion decay constant @xmath39 , pion - nucleon sigma term @xmath40 , string tension @xmath41 , and the vacuum chiral condensate @xmath42 by @xcite @xmath43 from the known range of the vacuum condensate , we can have an upper bound @xmath44 . therefore , @xmath35 should not be out of the range ( 156 , 270 ) mev . as usually done , we consider sqm as a mixture of @xmath45 , @xmath46 , @xmath47 quarks , and electrons . the relevant chemical potentials @xmath48 , @xmath49 , @xmath50 , and @xmath51 satisfy the weak - equilibrium condition @xmath52 because all particle masses do not depend on the density of electrons , i.e. , @xmath53 , eq . ( [ mui ] ) gives @xmath54 with @xmath55 for @xmath56 quarks , and @xmath57 for electrons . in eq . ( [ muivsni ] ) , @xmath58 is the second term on the right hand side of eq . ( [ mqt0 ] ) , so we have @xmath59 the pressure is then obtained from eq . ( [ pressure ] ) as @xmath60 substituting these expressions into eqs . ( [ weak1 ] ) and ( [ weak2 ] ) , we have @xmath61 and @xmath62 we also have the baryon number density @xmath63 and the charge density @xmath64 the charge - neutrality condition requires @xmath65 . for a given total baryon number density @xmath14 , we can obtain the respective @xmath66 , @xmath67 , @xmath68 , and @xmath69 by solving the four equations ( [ qmeq1 ] ) , ( [ qmeq2 ] ) , ( [ qmeq3 ] ) , ( [ qmeq4 ] ) . the chemical potentials @xmath48 , @xmath49 , @xmath50 , and @xmath51 can then be calculated by eqs . ( [ muivsni ] ) and ( [ muevsne ] ) . therefore , the energy density of the quark matter is a function of the baryon number density @xmath14 and the charge density @xmath70 , i.e. , @xmath71 or @xmath72 where the two partial derivatives , @xmath73 and @xmath74 are called the baryon chemical potential and charge chemical potential , respectively . it can be easily shown that they are connected to the quark chemical potentials by @xmath75 in fact , according to the fundamental differentiation equality of thermodynamics , we have @xmath76 on the other hand , we have @xmath77 , @xmath78 , @xmath79 , and @xmath80 from eqs . ( [ weak1 ] ) , ( [ weak2 ] ) , ( [ qmeq3 ] ) , and ( [ qmeq4 ] ) . substituting these four equalities into eq . ( [ deqthermo ] ) leads to @xmath81 . comparison of this with eq . ( [ deqbaryonchem ] ) immediately gives eq . ( [ baryonchemmuud ] ) . mev and @xmath82 mev . , width=309 ] in fig . [ nvsnb ] , the quark fractions , i.e. , @xmath83 , @xmath84 , @xmath85 , and the @xmath86 times the electron number divided by the total quark number , @xmath87 , have been shown versus the baryon number density for @xmath88 mev and @xmath82 mev . it is seen that the fraction of up quarks is nearly always one third . the fraction of down quarks increases rapidly with decreasing densities , while the fraction of strange quarks approaches to zero when the density decreases to a certain lower density . . , width=309 ] in order to compare the relation between the fermi momentum and chemical potential , we plot , in fig . [ nui - mu ] , the fermi momentum of up ( solid line ) , down ( dotted line ) , and strange ( dashed line ) quarks , respectively , as a function of the corresponding quark chemical potential , in both the present model ( lines with a solid circle ) and the previous model ( lines with an open circle ) . it is very obvious that the difference is very large , especially at comparatively lower densities . in both models , the fermi momentum of up or down quarks is higher than that of strange quarks due to the fact that strange quarks are heavier than up or down quarks . for the same chemical potential , however , the fermi momentum in the present model is generally bigger than that in the previous model , due to the quark mass - density - dependence which reflects the strong interaction between quarks . in mev for the solid , dashed , dotted , and dash - dotted cures are ( 200,150 ) , ( 180,120 ) , ( 170,95 ) , and ( 160,80 ) , respectively . the dash - dot - dot line is for the two - flavor quark matter at @xmath88 mev . it is very obviously shown that the energy minimum , marked with a full triangle on each line , is located exactly at the same point of the zero pressure indicated by an open circle . , width=309 ] figure [ enb ] shows the energy per baryon of quark matter for different parameter sets in the present model . each line has a minimum , corresponding to the lowest energy state ( marked with a solid triangle ) . one can see that the pressure at this minimum is exactly zero . so this special point is marked with an open circle as well . at the same time , we also display the energy per baryon for the two - flavor quark matter by a dash - dot - dot line . we see that the two - flavor quark matter is less stable than sqm . even for a smaller @xmath16 value , e.g. , @xmath89 , its energy will finally exceed that of sqm for a bigger @xmath16 with increasing densities . mev , @xmath90 mev ) and in the bag model ( @xmath91 mev ) . the former has a minimum at a lower density depending on the value of @xmath16 and @xmath28 , while the latter is always a monotonic function of density . the velocity of sound in both models has also been given on the right axis . , width=309 ] in principle , the cddm quark model contains more physics than the simple bag model . to demonstrate this we plot , in fig . [ comp ] , the baryon chemical potential in cddm with @xmath92 mev and @xmath90 mev , and in the bag model with @xmath91 mev . in cddm , the baryon chemical potential decreases with decreasing density to a certain value depending on @xmath16 and @xmath28 , then it increases very rapidly , i.e. , it saturates at a definite density marked with a bullet . when the density is lower than the bullet , the derivative @xmath93 becomes negative , and so quark matter is unstable against phase separation and falls apart at lower densities . in the bag model , however , the baryon chemical potential is always a monotonic function of density , which means that quark matter does not fall apart at any lower densities . the velocity of sound has also been plotted on the right axis . we observe that it is , in the bag model , nearly the same as that of a non - interacting fermi gas . but in cddm , it decreases to zero with decreasing densities . at high densities , it becomes asymptotically identical to the ultrarelativistic limit , as expected . since the single particle energies depend on density and temperature via the quark masses , the thermal properties should be founded on the canonical ensemble , but , as is well known , the partition function is not easy to calculate . therefore a different statistical procedure is usually adopted , which is based on the quasiparticle assumption . according to that the energy density is written as @xmath94 where the fermi distribution function is @xmath95/t}}. \label{ass3}\ ] ] if antiparticles are included , the sum must be extended to antiparticles for which @xmath10 must be replaced by @xmath96 . from the landau definition of the single - particle energy extended to finite temperature , we have @xmath97 where @xmath98 is the dispersion relation of free particles . the extra term @xmath99 can be added to the chemical potential , so defining @xmath100 accordingly , the net density of the particle type @xmath17 is @xmath101,$]or , explicitly , we have @xmath102/t}}\right . \left.-\frac{1}{1+e^{[\varepsilon_i(p)+\mu^*_i]/t}}\right\ } \frac{p^2\mbox{d}p}{2\pi^2}.\ ] ] inverting this equation , one determines @xmath103 as a function of @xmath104 so that the free energy density @xmath105 \label{fttot}\ ] ] with @xmath106 \nonumber\\ & & \phantom{g_i\int_0^{\infty}\bigg\ { } \pm\frac{\mu_i^*}{1+e^{(\sqrt{p^2+m_i^2}\mp\mu_i^*)/t } } \bigg\ } \frac{p^2\mbox{d}p}{2\pi^2 } \label{fitmum}\end{aligned}\ ] ] will be a function of respective particle densities , instead of chemical potentials . one can then determine the real chemical potentials and pressure , according to the well - known relations @xmath107 these quantities will completely describe the thermal equilibrium of pure quark phase and the transition to the quark - hadron mixed phase . a more detailed analysis , where the thermodynamic formulism is developed to much more extent , is reported in appendix [ appea ] . in ref . @xcite , the @xmath10 should be implicitly understood as @xmath108 . mev and @xmath90 mev . , width=309 ] in fig . [ enbt ] , we give the energy per baryon of sqm as a function of both density and temperature . the parameters used for this three - dimensional plot are @xmath109 mev and @xmath90 mev . the brueckner - bethe - goldstone ( bbg ) theory is among the most advanced microscopic theories of nuclear matter . in the recent years it was recognized that the three - body forces , which are expected to have a dominant role at high nuclear density , also affect the saturation point and , in fact , after including the three - body forces in the brueckner theory , the empirical saturation properties are reproduced quite well @xcite . in the neutron star interior , where high baryonic density values are reached , processes like the excitation of nucleon - antinucleon pairs ( @xmath110-diagrams ) and nucleonic resonances ( together with the production of other hadrons ) sizeably influence the two - body nuclear interaction . the former process involves the virtual excitation of negative energy states , which is absent from the standard brueckner theory , and thus it represents a pure relativistic effect . from the comparison with the dirac - brueckner theory it turns out that it is by far the most important relativistic effect @xcite . this process together with nucleon resonances can be incorporated in the interaction as medium virtual excitations in tbfs . one could guess that many body ( more that three ) forces are also important at high density as large as @xmath111 , but it is hard to imagine that pure baryon matter can exist at so high density . at lower density two and three - body forces are dominant since the hole line expansion is , roughly speaking , an expansion in density powers . the global effect of tbfs at high density is strongly repulsive , leading to a remarkable increase of the maximum mass in the study of the neutron star structure . but also a correct estimate of the saturation point is important , since , as we will see below , in strongly asymmetric nuclear matter the threshold for the transition to mixed nucleon - quark phase can appear close to the saturation density . therefore the corresponding eos could be used as input for transport - model simulations of heavy ion collisions , where strongly isospin - asymmetric systems are formed in central events . the brueckner theory extended to tbf is described elsewhere @xcite . here we simply give a brief review of the brueckner - hartree - fock ( bhf ) approximation at finite temperature @xmath112 @xcite . the starting point is the reaction @xmath113-matrix , which satisfies the bbg equation , @xmath114 where @xmath115 , denotes the single particle momentum , the @xmath116-component of spin and isospin , respectively , and @xmath117 is the starting energy . the @xmath113-matrix , the pauli operator @xmath118 and the single particle energies @xmath119 depend on the neutron and proton densities and temperature . the interaction @xmath120 given by @xmath121 where @xmath122 is the bare two body force ( 2bf ) and @xmath123 is an effective 2bf derived by the average of the bare tbf on the third particle as follows @xmath124\displaystyle \frac{1}{4 } { \rm tr}\sum_n \int { \rm d } { \vec r_3 } { \rm d } { \vec r_3^{\ \prime}}\phi^*_n(\vec r_3^{\ \prime } ) [ 1-\eta(r_{13}',t ) ] [ 1-\eta(r_{23}',t ) ] \\[6 mm ] \times \displaystyle w_3(\vec r_1^{\ \prime}\vec r_2^{\ \prime } \vec r_3^{\ \prime}|\vec r_1 \vec r_2 \vec r_3 ) \phi_n(r_3 ) [ 1-\eta(r_{13},t)][1-\eta(r_{23},t ) ] . \end{array } \label{eq : tbf}\ ] ] since the defect function @xmath125 is directly determined by the solution of the bbg equation @xcite , @xmath123 must be calculated self - consistently with the @xmath113 matrix and the s.p . potential @xmath126 on the basis of bbg equation . it is clear from eq . ( [ eq : tbf ] ) that the effective force rising from the tbf in nuclear medium is density and temperature dependent through the defect function . a detailed description and justification of the method can be found in ref . @xcite , including a discussion on the averaging procedure . the validity of such a procedure has been numerically tested in the comparison between the bhf eos plus z - diagrams with @xmath127 meson exchange and the dirac - bhf eos , which are expected to be equal . the calculation @xcite gives an impressive agreement between the two eos s , although the tbf due to the @xmath110-diagrams is averaged according to eq . ( [ eq : tbf ] ) . for @xmath122 we adopt the argonne @xmath128 two - body interaction @xcite . the tbf is constructed from the meson - exchange current approach @xcite and contains virtual particle ( @xmath2 and @xmath129 ) excitations and , in addition , relativistic effects induced by the excitations of particle - antiparticle pairs . this description of the interaction is not completely consistent since , in principle , the two and three body forces should be derived from the same meson parameters , but a recent calculation @xcite replacing argonne potential with bonn potential @xcite and the tbf built up with the bonn meson parameters substantially leads to the same results . let us start with symmetric nuclear matter . in the bhf approximation , the thermodynamic potential can be written @xmath130 , \\ w_l&=&-\frac{1}{2}\sum_k f_k(t ) u_k(t),\end{aligned}\ ] ] where @xmath131 is the thermodynamic potential for a system of independent particle with single particle spectrum @xmath132 , and the @xmath133 the sum of all linked cluster diagrams to the lowest order in the hole line expansion@xcite . @xmath134 is the selfconsistent mean field , @xmath135 the finite temperature bhf approximation suffers from the same difficulty as any strongly interacting fermi system . the difficulty is the same as in the cddm quark model , since in both cases the s.p . spectrum is density and temperature dependent . whereas at @xmath136 the density ( or fermi momentum ) can be fixed , at @xmath137 the role of density is taken by chemical potential @xmath138 ( grand canonical ensemble ) , and one is forced to fix the chemical potential at each iteration due to the presence of the fermi distribution @xmath139/t)\}^{-1}$ ] . since this procedure does not converge @xcite , one should fix the density and invert the equation relating density and chemical potential , @xmath140 which is not a viable task . the usual approximation is to drop out the derivative in the previous equation , which corresponds to the quasiparticle approximation above discussed within the cddm quark model . so doing , the resulting @xmath141 looses its meaning of chemical potential . in this approximations the energy and entropy densities are given by : @xmath142 , \\ s_n & = & -\frac{1}{v}\sum_k\big\{f_k(t)\ln f_k(t ) \nonumber\\ & & \phantom{-\sum_k\big\ { } + \big[1-f_k(t)\big ] \ln\big[1-f_k(t)\big]\big\}.\end{aligned}\ ] ] after one calculates @xmath141 in terms of @xmath143 , the thermodynamics is developed from the free energy density @xmath144 . the free energy per particle , calculated from bhf approximation , is depicted in fig . [ eoslast ] . due to the difficulty of extending the bhf code to very high temperature an extrapolation from the real numerical results to high @xmath112 has been performed adopting the so called frozen approximation based on @xmath112-independent single particle spectrum , i.e. the latter is frozen at @xmath136 . this turns out to be a good approximation up to @xmath145 mev @xcite . the relevant thermodynamical quantities , i.e. chemical potentials and pressure , are derived from free energy as follows @xmath146 let us consider asymmetric nuclear matter with baryon density @xmath147 and asymmetry parameter @xmath148 , where @xmath149 ( @xmath150 ) is the neutron ( proton ) density . the baryon chemical potentials can be expressed as @xmath151 where @xmath152 is the free energy density . assuming the parabolic law for the latter , we get the simple expression for the chemical momentum isotopic shift @xmath153 where @xmath154 is the symmetry free energy density . the parabolic law is well satisfied at low density , but at high density additional terms of the @xmath155-expansion must be considered . in neutron star inner core nuclear matter is supposed to be in @xmath155-equilibrium under the condition of charge neutrality . assuming that only electrons are present ( the muon contribution is negligible ) , the two preceding conditions require @xmath156 @xmath157 is the net charge density of nuclear matter . it is zero for pure neutral nuclear matter . for a given set of @xmath158 , we can solve the chemical potentials @xmath159 , @xmath160 , and @xmath51 from the above equations . then all other quantities can be obtained for a fixed temperature . in other words , all thermodynamic quantities can be regarded as a function of the nucleon density @xmath143 , charge density @xmath157 , and temperature @xmath112 . at zero temperature , for example , the energy density can be regarded as a function of @xmath143 and @xmath157 , i.e. , @xmath161 . with a similar approach as in the preceding section to obtain eq . ( [ baryonchemmuud ] ) , we can easily show that the baryon chemical and charge chemical potentials of nuclear matter can be expressed as @xmath162 the system turns out to be in a strongly isospin asymmetric state . the isotherms of free energy and pressure of nuclear matter in @xmath155-equilibrium are shown in fig . [ beos ] . three values of the temperature are considered.,width=309 ] let us study the nuclear matter , consisting of nucleons and electrons , in equilibrium with a gas of @xmath45 , @xmath46 , @xmath47 quarks and electrons . according to glendenning @xcite , we assume the total charge conservation , in addition to total baryon and energy conservation . now we first consider the case of zero temperature , and then extend to finite temperature . the conservation laws can be imposed by introducing the quark fraction @xmath163 defined as @xmath164 where @xmath165 is the total volume , @xmath166 is the volume occupied by quarks . then the total baryon density is @xmath167 the total electric charge is @xmath168 and the total energy density is @xmath169 where @xmath143 , @xmath157 , and @xmath170 are , respectively , the baryon number density , electric charge density , and energy density of nuclear matter , while @xmath14 , @xmath70 , and @xmath171 are the corresponding quantities of quark matter . @xmath170 is a function of @xmath143 and @xmath157 , @xmath171 is a function of @xmath14 and @xmath70 , i.e. , @xmath161 , @xmath71 . differentiating eq . ( [ etot ] ) , one obtains @xmath172 on the other hand , differentiating eqs . ( [ rhotot ] ) and ( [ qtot ] ) at a given pair of @xmath173 and @xmath174 , we have @xmath175 to minimize @xmath176 , we substitute eqs . ( [ drhoh ] ) and ( [ dqh ] ) into eq . ( [ de01 ] ) . then setting @xmath177 , we find @xmath178 where @xmath179 the conditions in eq . ( [ gibbscond ] ) are nothing but the gibbs ones , i.e. , the baryon chemical potential , the charge chemical potential , and the pressure in nuclear and quark matter should be equal to each other to minimize the total energy of the mixed phase . in the previous two sections , we have linked the baryon chemical potential and charge chemical potential to the respective constituent particle chemical potentials in eqs . ( [ baryonchemmuud ] ) and ( [ baryonchemnm ] ) . as application of these equalities , we immediately see that the first two equations in ( [ gibbscond ] ) are equivalent to @xmath180 in general , all other chemical potentials in quark sector can be related to @xmath48 and @xmath49 , e.g. , @xmath78 , @xmath77 . similarly , all chemical potentials in nuclear sector can be linked to @xmath159 and @xmath160 , e.g. , @xmath181 . therefore , eq . ( [ munpud ] ) means that we can choose either ( @xmath48,@xmath49 ) or ( @xmath159,@xmath160 ) as the two independent chemical potentials . the latter can then be determined by solving the charge neutrality equation and the pressure balance equation for a given total baryon number or a given quark fraction . at finite temperature , we similarly have the phase equilibrium condition @xmath182 the condition in eq . ( [ phasetcon03 ] ) only tells us that the temperature in nuclear and quark sectors are equal , so we have a common temperature @xmath112 . the chemical equilibrium condition in eq . ( [ phasetcon02 ] ) is equivalent to that in eq . ( [ munpud ] ) . therefore , we still have only two independent chemical potentials . for a given total density @xmath173 at a fixed temperature @xmath112 , the two independent chemical potentials and the quark fraction @xmath163 can then be determined by solving the three equations in eqs . ( [ rhotot ] ) , ( [ qtot ] ) with @xmath183 , and ( [ phasetcon01 ] ) . similar to the case at zero temperature , the lower critical density @xmath184 , which separates the nuclear and mixed phases , is defined by @xmath185 , while the critical density @xmath186 between the mixed and quark phases is determined by @xmath187 . , width=309 ] in fig . [ mixe17095 ] , we display , for the parameter @xmath32 = 170 mev , the density dependence of the energy per baryon in pure nuclear matter , in pure quark matter , and in the mixed phase . the quark fraction has also been depicted on the right y - axis . in fig . [ mixp17095 ] the corresponding pressure is reported . it is seen that the nuclear matter is the most favorite phase at lower densities , and the quark matter is the most stable phase at higher densities , while at intermediate densities , the mixed phase has the lowest energy . ( cures with an open circle ) and 20 mev ( cures with an solid circle ) . the nuclear density @xmath143 and the quark density @xmath14 in the mixed phase are also shown on the right axis . , width=309 ] the quark baryon number density @xmath14 and nuclear density @xmath143 are also plotted on the right axis of fig . [ mixp17095 ] . we see that the quark density is always higher than the nuclear density . the transition from hadron phase to mixed phase occurs at the density a bit less than @xmath188 @xmath189 , well below the saturation density . but it is hard to observe in terrestrial laboratories , since the nuclear matter so far realized in exotic nuclei or heavy - ion collisions is much less neutron rich . the transition from mixed phase to pure quark phase occurs at the total density @xmath190 @xmath189 , where the nuclear density is only @xmath191 @xmath189 . the density range of mixed phase is only slightly depending on the temperature , at least in temperature range of interest in neutron stars . the critical densities depend on the parameter @xmath16 . in the left panel of fig . [ rhocd ] , both nuclear critical density ( the solid line , separating the pure nuclear phase and the mixed phase ) and the quark critical density ( the dashed line , delimiting the pure quark phase ) are displayed as a function of @xmath35 . if @xmath192 mev , the two critical densities approach zero , and accordingly sqm is absolutely stable . when 161.6 mev @xmath193 162.5 mev , mixed phase can exist at any lower densities . only when @xmath194 mev , nuclear matter is more stable at lower densities . if we have only two flavor quarks in the quark sector , the critical densities are usually higher . in the same panel we also plot the lower critical density for the two - flavor case . because we know in our real world the two - flavor quark matter does not exist below the saturation density , @xmath35 should be on the right of the first full dot at @xmath195 mev ( the intersection of the dotted and dot - dashed lines ) in fig . [ rhocd ] . on the other hand , to let sqm have a chance to appear below the saturation density , @xmath35 should be on the left of the second full dot where @xmath196 mev . in plotting fig . [ mixe17095 ] and [ mixp17095 ] , we adopt @xmath197 mev . the temperature dependence of the lower and higher critical densities is also plotted in fig . [ rhocd ] ( right panel ) . the two lines at fixed @xmath16 mark the boundaries of the three phases . . the horizontal line is the nuclear saturation density . the dotted line is the quark critical density in the case of two - flavor quark system . right panel : phase diagram for three values of the confinement parameter and @xmath198 mev . , width=309 ] with the equation of state that has the mixed and/or quark phase derived in sec . [ phase ] , we are ready to study the structure of hybrid stars by solving the tolman - oppenheimer - volkov equation @xmath199 where @xmath200 mev@xmath201 is the gravitational constant , @xmath202 is the distance from the center of the star , @xmath203 and @xmath204 are the energy density and pressure at the radius @xmath202 , respectively . the subsidiary condition is @xmath205 with @xmath206 being the mass within the radius @xmath202 . at variance with pure nuclear or quark stars , a hybrid star contains pure quark matter in the core , pure nuclear matter near the outer part , and , in between , a mixed phase of the quark and nuclear matter . in this case , therefore , we must use the eos in whole density range . mev for four values of the confinement parameter @xmath35 of 170 , 180 , 190 , 200 mev . , width=309 ] the resulting gravitational mass for the hybrid star is plotted in fig . [ mass ] , as a function of both radius and central density , for four values of the confinement parameter @xmath35 of 170 , 180 , 190 , 200 mev . the main effect of the phase transition is , as expected , a large reduction of the maximum mass due to softening of the eos . concerning the confinement parameter @xmath16 , we observe a slight decrease of the maximum mass when the @xmath16 value becomes smaller , from 1.74@xmath207 at @xmath35 = 200 mev down to 1.6 , 1.55 , 1.52 @xmath207 corresponding to @xmath35 = 190 , 180 , 170 mev . this can be easily understood in this way : since quark phase is occurring at lower density for smaller values of @xmath16 , sometimes even less than the nuclear saturation density ( for example only 0.15 @xmath189 for @xmath35 = 170 mev ) , then the quark population of the star would be more numerous , since the stronger softening of the eos may only support less gravitational mass . in particular , the most favorable case of @xmath35 = 180 and 190 mev , when the quark confinement appears at around 2@xmath208 consistent with the heavy ion experiment , the predicted maximum mass can match very well the s branch of neutron stars mentioned by haensel et al.with an error less than @xmath209 @xcite . in order to study the temperature effect on the mass of hybrid stars , we plot the star mass , vs the star radius or central density , at @xmath136 and @xmath210 mev in fig . it is seen that the temperature influence on the maximum mass is very limited . otherwise the effect is a quite strong increase of the ns radius for a fixed amount of gravitational mass . but , larger is the mass smaller is the radius variation , similarly to a few other calculations @xcite . the condensation of @xmath211 mesons in neutron stars is widely discussed in the literature ( see refs . @xcite and references therein quoted ) . in dense matter , the condensation of @xmath211 mesons is originated by the reaction @xmath212 if the effective mass of the @xmath211 drops below the chemical potential of the electron , this reaction becomes possible in dense matter , indicating the presence of kaon condensation . since almost all the studies of kaon medium properties @xcite suggest the consistent picture that the attraction from nuclear matter would bring the @xmath211 mass down , so the threshold condition for the onset of @xmath211 condensation @xmath213 , which follows from eq . ( [ eq : kaon.1 ] ) , could be fulfilled in the center of neutron stars @xcite at @xmath214 . however , the deconfinement phase transition from the hadronic phase to the quark phase occurs also at rather low density , which leads to the onset of mixed phase in neutron stars . then it is very interesting to explore how the quark deconfinement affects the @xmath211 condensation threshold . mev for three selected values of the confinement parameter @xmath35 of 180 , 200 , 225 mev . , width=309 ] we take the antikaon dispersion relation constrained by the heavy - ion data as empirical indication of an attractive antikaon potential in dense matter @xcite , and combine this with the bhf model of nuclear matter together with the above quark model without bag constant , to calculate the effective kaon mass in hybrid stars . the result is illustrated in fig . [ kaon ] , where the effective kaon mass and the electron chemical potential are shown in normal nuclear matter and especially in mixed phase for three selected values of the confinement parameter @xmath16 . one sees that in normal nuclear matter , @xmath211 medium mass decreases with increasing density and meets the electron chemical potential at @xmath215 0.6 @xmath189 ( the solid bullet in fig . 14 ) , so the kaon condensation would be present for sure in this case ; however , once the quark phase sets in at the total density of 0.26 , 0.4 , 0.53 @xmath189 respectively for @xmath35 = 180 , 200 , 225 mev , the decreasing speed of the kaon effective mass in the matter slows down a little bit . in addition , more conclusively , the electron population begin to decrease instead , for the reason that the electric charge neutrality can be achieved more efficiently through the charged quarks themselves . as a result the threshold condition of @xmath211 condensation is much more difficult to satisfy , unless the confinement parameter @xmath16 is chosen to be extremely high ( at least @xmath216 mev ) when the presence of quark phase is pushed to very high densities , then one may expect that @xmath217 can finally equal to @xmath51 ( which is indicated with a circle ) . those high values of @xmath16 may even not be realistic , therefore we would like to conclude that the inclusion of quark phase may make the kaon condensation impossible in neutron stars , or at least hinder it very strongly . in this paper the neutron star inner structure has been studied . the lack of strong observational constraints demands for sophisticated models of the ns composition and interaction mechanisms . in this study we only included baryons and quarks in equilibrium with leptons and kaons . since in the ns interior high baryon densities are reached , n@xmath218 and nucleonic excitations are expected to play a major role . their effects can be incorporated in a three body force . the baryon eos in weak coupling equilibrium with electrons was derived within the bbg theory suitably extended so to include the three body force . in the quark sector , we adopted the semiphenomenological cddm quark model , which exhibits a confinement mechanism alternative to the crude mit bag model . furthermore , in contrast with the extension of the mit model , where the density dependence is introduced artificially @xcite , cddm quark model shapes its density dependence in agreement with chiral requirements @xcite . hadrons and quarks in @xmath155-equilibrium were considered at zero and finite temperature . since typical temperatures of protoneutron stars are as high as @xmath219 mev and beyond , rigorous thermodynamic potentials have to be derived . this was possible with cddm quark model , due to its simplicity , whereas some approximations are needed to calculate the hadron phase in the brueckner theory . the latter is still a main drawback of the finite temperature microscopic theory of strongly interacting fermi systems . the transition from the low density hadron phase to high density quark phase in beta equilibrium was studied under the glendenning hypothesis of total charge neutrality . the gibbs construction enabled to follow the evolution of the mixed hadron - to - quark phase , varying the temperature @xmath112 and the confinement parameter @xmath16 . the eos , in terms of pressure , energy density and chemical composition , was constructed as a function of @xmath112 and @xmath16 . moreover the phase diagram @xmath112-@xmath16 was also depicted . the transition density from hadron to mixed phase is strongly dependent on the confinement parameter and , for d small enough , it becomes lower than the nuclear saturation density . however , at that point nuclear matter is so strongly neutron rich that it hard to imagine that it can be observed in terrestrial laboratory experiments . at high temperature the transition density is comparatively smaller . the tov equations have been solved in the above discussed model of neutron stars . the transition to quark matter produces a strong reduction of the maximum mass , from @xmath220 to @xmath221 for the lowest value of the confinement parameter @xmath16 , but the corresponding radius is not changed , being in both cases about @xmath222 km . the other @xmath223-@xmath224 configurations ly in a range @xmath225 km depending on the value of @xmath16 . with increasing @xmath16 , the maximum mass increases from @xmath226 to @xmath227 solar masses . other quark models have been adopted to describe the ns transition to a deconfined phase ( see , for instance , refs . @xcite ) . except for the nambu - jona lasinio model which exhibits instability of neutron stars @xcite , the other models point to a softening of the eos of nuclear matter . the mit model with density - dependent bag constant predicts a beta - stable quark phase quite similar to our model @xcite , but the comparison on the ns structure turns out to be difficult , since in that calculation the transition to quark phase is built on top of hyperonized nuclear matter . the effect of temperature on the maximum mass is negligible up to @xmath210 mev , but as expected the heated system can support much less gravitational mass in the same volume for all @xmath223-@xmath224 configurations . thus , an evolutionary study of isolated neutron stars would show a strong compression from the newborn phase to the long era phase as fast as the star cools down from @xmath219 mev to approximately zero . beyond the confinement parameter , the transition to quark phase is interrelated to other properties , in particular the possible onset of kaon condensation . in our analysis we concluded that the quark phase could not be compatible with the kaon condensation in neutron stars , or at least could hinder it very strongly . the model developed in this paper misses some important aspects . the first one is the effect of neutrinos . neutrinos play the major role in new born neutron stars , when they are still trapped in the interior . in this case the neutronization process is still hindered and the eos is that of symmetric nuclear matter . moreover , the onset of kaon condensation is shifted to higher densities @xcite and probably kaons have no more any chance to compete with quarks . the second one is the inclusion of hyperons and their competition with other mechanisms such as kaon condensation . in refs . @xcite the hybrid stars are studied by including hyperons in the hadron phase , which makes the hadron eos very soft and brings the maximum mass to @xmath228 . the introduction of the quark phase rises up again the maximum mass to @xmath229 , a value which is consistent with our prediction without hyperons . on the other hand , the appearance of hyperons much depends on the threshold of kaon production . therefore the interplay between hyperons and kaons turns out to be quite important and deserves additional investigation , as soon as more reliable empirical inputs will be available , especially on the hyperon - nucleon and hyperon - hyperon interaction . one of us ( u.l . ) is indebted to dr m. baldo for several discussions on statistical thermodynamics . g.x . would like to thank support from nsfc ( 10675137 , 10375074 , 90203004 ) and kjcx2-yw - n2 , and also acknowledge warm hospitality at the dipartimento di fisica e astronomia ( universit di catania ) , the laboratori nazionali del sud ( infn , catania ) , and the mit center for theoretical physics ( mit - ctp ) .. let s start from the fundamental thermodynamic differentiation relation @xmath230 where @xmath231 is the entropy density , and ( anti)particles are assumed to be uniformly distributed in a volume @xmath165 . by using the free energy density @xmath232 , it becomes @xmath233or , equivalently , @xmath234 because of the uniformity , the free energy density has nothing to do with the volume . we thus have @xmath235 at finite temperature , we should consider both particles and anti - particles , and particles / anti - particles are not always located bellow the fermi energy . therefore , the net particle number densities can be expressed as @xmath236 where the superscripts @xmath237 indicates particles , and @xmath238 signifies antiparticles : @xmath239 in eqs . ( [ nitmuspm ] ) and ( [ nitmum ] ) , @xmath108 are effective chemical potentials of respective particles . if quark densities are not density and/or temperature dependent , @xmath108 are nothing but the actual chemical potentials . in our present case , however , quark masses depend on both density and temperature , in order to include the strong interaction between quarks . therefore , the real chemical potentials should be derived according to fundamental thermodynamic laws . to determine the thermodynamic properties , we need to give a characteristic function . at zero temperature , we use the energy density in eq . ( [ emod ] ) due to zero entropy . now the temperature @xmath112 and the densities @xmath104 are chosen as the independent system variables , we should , therefore , choose the free energy in eq . ( [ fttot ] ) as the characteristic function . defining @xmath247\ ] ] with @xmath248 p^2\mbox{d}p,\ ] ] then we can write @xmath249 . substituting this into eq . ( [ stmus ] ) and ( [ muimus ] ) , then applying eq . ( [ dnidmui ] ) and @xmath250 , we have @xmath251 and @xmath252 obviously , the free energy density can be given as @xmath253 the energy density is obtained by @xmath254 , giving @xmath255 and the pressure is obtained by substituting eq . ( [ muitmusexp ] ) into eq . ( [ ptexpr ] ) : @xmath256 eqs . ( [ stmusexp])-([ptmusexp ] ) here are in complete accordance with the eqs . ( 58)-(61 ) in @xcite if one regards the @xmath10 there as @xmath108 and @xmath257 as @xmath22 which were not explicitly stated . the expression in eq . ( [ muitmusexp ] ) is special in the present paper because we need the real chemical potential to investigate the mixed phase . in the above derivation , we choose volume @xmath165 , the temperature@xmath112 , and the particle number densities @xmath104 as the independent system variables . in this case , the free energy is the characteristic function from which we get the complete set of thermodynamic functions . for this purpose we have defined the intermediate variables @xmath108 in eqs . ( [ nitmuspm ] ) and ( [ fitmum ] ) . because the quark matter we are considering is a strongly interacting system , the relations between the chemical potentials and the densities are , in principle , not the same as those of a free fermi gas . however , with the ` effective ' chemical potentials @xmath108 , the densities and the free energy are really of the same form as those of a non - interacting fermi gas . this is what the ` effective ' means . the actual chemical potentials of each type of particles are determined from the fundamental thermodynamic equality ( [ dfexpr ] ) which results in eq . ( [ muitmusexp ] ) . peng , nucl . a _ 747 _ , 75 ( 2005 ) ; g.x . peng , m. loewe , u. lombardo , and x.j . wen , nucl . b ( proc . suppl . ) * 133 * , 259 ( 2004 ) ; g.x . peng , h.c . chiang , p.z . ning , u. lombardo , and m. loewe , int . a * 18 * , 3151 ( 2003 ) ; g.x . peng , u. lombardo , m. loewe , and h.c . chiang , phys . b * 548 * , 189 ( 2002 ) . y. zhang , r.k . su , phys . c * 65 * , 035202 ( 2002 ) ; y. zhang , r. k. su , s. q. ying , and p. wang , europhys . lett . * 56 * , 361 ( 2001 ) ; y. zhang and r. k. su , phys . rev . c * 67 * , 015202 ( 2003 ) . p. wang , phys . c * 62 * , 015204 ( 2000 ) . wiringa , v.g.j . stoks , and r. schiavilla , phys . c * 51 * , 38 ( 1995 ) . r. machleidt , adv . nucl . phys . * 19 * , 189 ( 1989 ) ; r. brockmann and r. machleidt , phys . c * 42 * , 1965 ( 1990 ) ; r. machleidt , phys . rev . c * 63 * , 024001 ( 2001 ) . brown , c .- h . lee , and m. rho , phys . rept . * 462 * , 1 ( 2008 ) . g.q . lee , and g.e . brown , phys . lett . * 79 * , 5214 ( 1997 ) ; f. weber , prog . . phys . * 54 * , 193 - 288 ( 2005 ) ; and quoted therein .
as very few gamma - ray burst ( grb ) sources have astronomical counterparts at other wavebands , empirical studies of grbs have been largely restricted to the analysis of their gamma ray properties : bulk properties such as fluence and spectral hardness , and evolution of these properties within a burst event ( fishman & meegan 1995 ) . while bursts exhibit a vast range of complex temporal behaviors , their bulk properties appear simpler and amenable to straightforward statistical analyses . studies fall into two categories : examination of whether grb bulk properties comprise a homogeneous population or are divided into distinct classes ; and search for relationships between bulk properties . both types of study may lead to astrophysical insight , just as the distinction between main sequence stars and red giants and the measurement of a luminosity - mass relation along the main sequence assisted the development of stellar astrophysics early in the century . the most widely accepted taxonomy of grbs is the division between short - hard and long - soft bursts proposed by dezalay et al . ( 1992 ) and kouveliotou et al . ( 1993 , henceforth k93 ) . k93 noticed a bimodality in the burst duration variable @xmath1 ( time within which 90% of the flux arrived ) , suggesting the presence of two distinct types of bursts separated at @xmath2 sec . the short bursts have systematically harder gamma - ray spectra than longer bursts . the two groups seemed indistinguishable in most other bulk properties , although the larger group of long - soft bursts may have a subclass with a different fluence distribution ( i.e. , different @xmath3 ; katz & canel 1996 ) and the groups may have different galactic latitude distributions ( belli 1997 ) . other researchers point to small groups of bursts with distinctive properties such as the soft - gamma repeaters ( norris et al . 1991 ) , two possible classes with differing short - timescale variability ( lamb , graziani & smith 1993 ) , fast - rise exponential - decay bursts ( bhat et al . 1994 ) , and two types of bursts with different ratios of total fluence and @xmath4300 kev fluence ( pendleton et al . 1997 ) . a variety of relationships between burst properties have also been reported . norris et al . ( 1995 ) find an anti - correlation between @xmath1 ( calculated after wavelet thresholding ) and peak intensity , consistent with a cosmological time dilation . however , a positive correlation between @xmath1 and total fluence is also seen which does not agree with the simplest cosmological interpretation ( lee & petrosian 1997 ) . additional reported relationships include : @xmath1 correlated with peak heights ( lestrade 1994 ) , peak energy correlated with peak flux ( mallozzi et al . 1995 ) , and peak duration anticorrelated with gamma - ray energy ( fenimore et al . 1995 ) . most of these studies suffer from a failure to treat all of the bulk property variables in an unbiased and quantitative way . astronomers typically examine univariate or bivariate distributions , sometimes constructing composite variables ( such as hardness ratios ) with pre - determined relationships to include one or two additional variables . but it is quite possible that the complex astrophysics producing grbs will not manifest themselves in simple bivariate plots , just as the division between short - hard and long - soft bursts is not evident in spectral variables alone ( pendleton et al . grb catalogs , like most multiwavelength astronomical catalogs , are multivariate databases and should be treated with multivariate statistical methods that can objectively and effectively uncover structure involving many variables ( feigelson & babu 1997 ) . two previous studies take a fully multivariate approach to understanding grb bulk properties . baumgart ( 1994 ) constructs a neural network taxonomy of 99 grbs from the pvo satellite using 26 variables representing both bulk burst properties and detailed temporal characteristics ( e.g. number of peaks , fractal dimension , wavelet transform crossings ) and finds two or three distinct grb classes . bagoly et al . ( 1997 ) perform principal components and factor analyses of nine bulk property variables using 625 grbs from the batse 3b catalog . they find that the relationships in the database are determined principally by only three variables : an appropriately weighted fluence , a weighted burst duration , and ( to a lesser extent ) flux in the highest energy bin . we note , however , that it can be dangerous to look for correlations prior to classifying ( or establishing the homogeneity of ) the population . while the anticorrelation between hardness ratio and burst duration seen in full samples ( k93 ) may be the manifestation of a single astrophysical process , it may alternatively reflect differences between distinct processes . the latter possibility is suggested by a reported hardness - duration positive correlation within the long - soft class of bursts ( dezalay et al . 1996 ; horack & hakkila 1997 ) . most multivariate analyses thus begin with a study of homogeneity and classification , and then investigate the variance - covariance structure ( i.e. correlations ) within each class . this paper describes a multivariate analysis of grbs from the third batse catalog ( meegan et al . 1996 ) . after defining the sample ( 2 ) , we start with a simple statistical description of the variables and their bivariate relationships for the entire dataset ( 3 ) . we then seek distinct types of clusters in two ways . first , a standard nonparametric agglomerative hierarchical clustering analysis is performed ( 4 ) which reveals three distinct classes . the statistical significance of the third cluster is validated , under gaussian assumptions , with manova tests . second , a parametric maximum likelihood model - based clustering procedure is adopted which reveals the same three groups and indicates strong evidence for the presence for three rather than two groups ( 5 ) . the variance - covariance structure of each group is then examined ( 6 ) . results are synthesized in the discussion ( 7 ) . throughout the paper , we discuss our mathematical techniques to help the reader understand the complexities of multivariate analysis . from the vast literature in this subject , we recommend the following monographs for interested readers : johnson & wichern ( 1992 ) and jobson ( 1992 ) for overviews of applied multivariate analysis ; hartigan ( 1975 ) , jain & dubes ( 1988 ) and kaufman & rousseeuw ( 1990 ) for multivariate clustering algorithms ; murtagh & heck ( 1987 ) and , more briefly , babu & feigelson ( 1996 ) and feigelson & babu ( 1997 ) , for multivariate methodology in astronomy . our sample is drawn from the third catalog of the _ burst and transient source experiment _ ( batse ) on board the compton gamma ray observatory . this 3b catalog has 1122 grbs detected by batse between 1991 april 19 and 1994 september 19 . the catalog is presented and fully described by meegan et al . our database was extracted from the on - line database www.batse.msfc.nasa.gov/data/grb/catalog in may 1996 , which provides many properties of each burst . there are roughly eleven variables of potential astrophysical interest : two measures of location in galactic coordinates , @xmath5 and @xmath6 ; two measures of burst durations , the times within which 50% ( @xmath7 ) and 90% ( @xmath1 ) of the flux arrives ; three peak fluxes @xmath8 , @xmath9 and @xmath10 measured in 64 ms , 256 ms and 1024 ms bins respectively ; and four time - integrated fluences @xmath11-@xmath12 in the 20 - 50 kev , 50 - 100 kev , 100 - 300 kev and @xmath13 kev spectral channels respectively . researchers commonly consider three composite variables : the total fluence , @xmath14 , and two measures of spectral hardness derived from the ratios of channel fluences , @xmath15 and @xmath16 . due to the limitations of available multivariate statistical techniques , we ignore other variables of potential relevance including the heteroscedastic measurement errors of each quantity ( i.e. , errors that differ from point to point ) and truncation values associated with batse triggering operations . of the 1122 listed bursts , 807 have data on all the variables described above . ten bursts listed with zero fluences were eliminated . our sample thus has 797 batse grbs . for some analyses , we also used a subset of 644 bursts with ` debiased ' durations , @xmath17 . here the durations are modified to account for the effect that brighter bursts will have signal above the noise for longer periods than fainter bursts with the same time profiles ( j. norris , private communication ) . statistical analyses in 3 , 4 and 6 were conducted within the statistical analysis system _ sas / stat _ , a very large and widely distributed commercial statistical software package ( sas institute inc . _ sas / stat _ procedures _ cluster _ , _ glm _ , _ princomp _ and _ varclus _ were used . the analysis in 5 was performed with the _ mclust _ software ( banfield & raftery 1993 ; fraley 1998 ) , which is interfaced to the splus statistical package ( splus version 3.4 ; mathsoft , inc . 1996 ) and its extensions . further information is provided at http://stat.washington.edu/fraley/software.html and http://lib.stat.cmu.edu/general/mclust . for multivariate data visualization , we used the xgobi ( swayne , cook & buja 1991 ) program , available from the _ statlib _ software archive at http://lib.stat.cmu.edu/general/xgobi . hypertext links to a variety of public domain software for multivariate analysis , classification and visualization are available at the penn state _ statcodes _ web site , http://www.astro.psu.edu/statcodes . we are faced with a multivariate database of 797 objects and 15 variables ( 11 variables from the catalog , 3 composite variables , and @xmath18 ) . two initial problems are frequently faced in analyses of multivariate databases . first , variables with incompatible units and ranges must be compared . units can be removed by normalization ( e.g. replacing @xmath11 by @xmath19 ) , by standardization ( e.g. replacing @xmath11 by @xmath20 where @xmath21 is the sample standard deviation ) , or by taking logarithms . second , the dimensionality of the problem should be reduced , as many of the variables are closely interrelated either by construction or by astronomical circumstance . although there are no mathematical rules regulating reduction of dimensionality , it can usefully be guided by a correlation matrix showing bivariate relationships and a principal components analysis showing multivariate relationships that are mainly responsible for structured variance in the data . scientific reasoning can also be used to eliminate consideration of variables . we conducted a preliminary examination of data representations , correlation matrices and bivariate plots , and principal components analyses to facilitate choice of variables . when no mathematical preference arose , we selected variables most commonly used by previous researchers to facilitate comparison of results . our choices were as follows . we use log variables , rather than normalized or standardized variables . we kept information on burst fluence and spectra through @xmath22 and hardness ratios rather than through the original fluences @xmath11-@xmath12 . we initially eliminated @xmath8 and @xmath10 from consideration , and later eliminated @xmath9 when we found that its main contribution to the clustering process was noise . we chose to remove the location variables @xmath23 , already established by other researchers to be random for the entire sample , but use them later to test for isotropy of subsamples . the debiased @xmath24 is used only in special tests . our analysis was thus performed in six or fewer dimensions using log @xmath7 , log @xmath1 , log @xmath22 , log @xmath9 , log @xmath25 and log @xmath26 . tables 1 and 2 give basic statistics for these six variables : means , standard deviations , and bivariate values of pearson s linear correlation coefficient @xmath27 . for @xmath28 and assuming bivariate normal populations , any @xmath29 implies that a correlation between the two variables exists at a two - tailed significance level @xmath30 ( beyer 1968 , pp . 389 and 283 ) . but from an astrophysical perspective , we might consider any relationship with @xmath31 to be of little interest . figure 1 shows the bivariate scatter plots . the correlation structure of the entire sample ( table 2 ) shows that the two measures of duration and the two measures of spectral hardness have correlation near unity , indicating that they are nearly redundant . @xmath22 and @xmath9 are quite dissimilar : @xmath22 shows a strong correlation with burst duration ( e.g. lee & petrosian 1997 ) and no relation to hardness , while @xmath9 shows no relation to duration but is mildly correlated with hardness ( mallozzi et al . burst duration is anticorrelated with hardness ( k93 ; fenimore et al . the cosmological anticorrelation between duration and peak flux reported by norris et al . ( 1995 ) is statistically significant , but accounts for only a few percent of the variance between these variables . the correlation matrix based on the debiased @xmath18 values yields very similar results . however , the scatter plots ( figure 1 ) show a more complex story . first , many plots show inhomogeneous distributions inconsistent with the unimodal multinormal ( i.e. multivariate gaussian ) population assumed by pearson s @xmath27 . the distributions often seem bimodal with asymmetrical non - gaussian shapes . one outlier burst is also seen in several projections . we therefore consider the hypothesis that the sample consists of two or more distinct classes , and proceed to find the ` clusters ' using well - established methods . agglomerative hierarchical clustering is a procedure based on the successive merging of proximate pairs of clusters of objects . it produces a clustering tree or dendrogram starting with n clusters of 1 member ( or a coarse partition based on prior knowledge ) and ending with one cluster of n members . unfortunately , there are many possible ways to proceed ; mathematics provides little guidance among the choices and no probabilistic evaluation of the results without the imposition of additional assumptions . the scientist must make four decisions to fully define the clustering procedure : 1 . creating _ unit - free variables _ is essential for meaningful treatment of objects in multivariate space ( 3 ) . a favorite choice by statisticians is standardization , where each variable is normalized by the standard deviation of the sample . astronomers more commonly make logarithmic transformations or construct ratios of variables sharing the same units . we follow the tradition of grb researchers by measuring spectral hardness with ratios of fluences having the same units , and making logarithmic transformations of all variables . the _ metric _ defines the meaning of proximity between two objects or clusters . common choices are the simple euclidean distance between unit - free variables and the squares of euclidean distances . we chose the former option for most of the analyses in this section . 3 . several _ merging procedures _ can be used . one might begin by merging the clusters with the nearest neighbors . this is called single linkage clustering and is most familiar in astronomy where it is frequently called the friends - of - friends algorithm . it tends to produce long stringy clusters , and is equivalent to a well - known divisive clustering procedure known as pruning the minimal spanning tree . complete linkage proceeds by maximizing the distance between clusters , and leads to evenly bifurcating dendrograms . for most of our analysis , we choose average linkage where the distance between two clusters is the average of the distances between pairs of observations , where each member of the pair comes from a different cluster . this is a compromise between single and complete linkage and tends to give compact clusters . specifically , the distance between clusters k and l is given by ( e.g. sas institute inc . 1989 , pp . 529ff ; johnson & wichern 1992 , pp . @xmath32 where the bar indicates an unweighted mean , @xmath33-th cluster . another popular choice is ward s minimum variance criterion where the distance between the two clusters is the anova ( analysis of variance ) sum of squares between two clusters added up over all variables ( ward 1963 ) , @xmath34 if the sample is generated by a mixture of multinormal ( i.e. multidimensional gaussian ) distributions where each distribution has covariance matrix of the form @xmath35 , this method joins clusters to maximize the likelihood at each level of the hierarchy , and so is a special case of the model - based clustering methodology to be discussed in 5 . 4 . as the procedure gives a hierarchy from @xmath36 clusters with 1 object down to one cluster with @xmath36 objects , the user must choose _ how many clusters to report _ as scientifically important clusters . this choice can be assisted by examination of two statistics . the squared correlation coefficient , @xmath37 , states the fraction of the total variance accounted for by a partition into @xmath38 clusters , @xmath39 the squared semi - partial correlation coefficient , @xmath40 measures the difference in the variance between the resulting cluster and the immediate parent clusters normalized by the total sample variance , @xmath41 @xmath37 thus tells how much of the scatter is explained by a given level of clustering , and @xmath40 tells how much improvement is achieved between levels . we emphasize again that there is no mathematically ` best ' choice , although extensive experience with problems in many fields has led to a preference for certain combinations ( e.g. standardized variables and ward s minimum variance criterion ) . we conducted extensive experiments with different choices . the last several levels of the clustering tree for the 797 grbs using the six unit - free variables shown in table 1 , average linkage and a euclidean metric are shown in figure 2 ( left panel ) with details in table 3a . the action taken at each level is indicated in column 2 of table 3 , which may refer to a level higher in the tree which ( for brevity ) is not shown here . two types of mergers are seen : the incorporation of ` twigs ' of one or a few grbs into a large preexisting ` trunk ' ( levels 1 , 3 , 4 , 5 and 7 ) ; and the union of two substantial branches into a single larger trunk ( levels 2 , 6 and 8) . the first type has little effect on the variance of the sample with @xmath42% . the single grb brought into the main trunk at level 1 is the distant outlier seen in several panels of figure 1 . the level 2 merger of clusters with 190 and 606 members is clearly the most important structure , accounting for roughly 53% of the variance of the entire sample . this is the bifurcation of the sample into two classes easily seen in figure 1 and noted by k93 and others . the principal finding that is not immediately obvious from figure 1 is the structure indicated at level 6 . the main trunk of 599 bursts ( plus a few twigs to be merged later ) is divided into groups of 93 and 506 bursts . this division accounts for 10% of the total variance of the sample , indicated in both the @xmath37 and @xmath40 values . we found that the twigs in the tree structure disappear if the peak flux @xmath9 variable is omitted and the analysis is made in 5-dimensional space ( figure 2 , right panel , and table 3b ) . here the largest cluster of 593 members is formed by the union of clusters with 107 and 486 bursts , again accounting for 10% of the sample variance . it is possible that @xmath9 is a nuisance variable irrelevant to the basic astrophysics of grbs , producing noisy ` twigs ' seen in table 3a and figure 2 ( left panel ) . we tested many variants of hierarchical clustering . we replaced average linkage hypothesis with complete linkage , single linkage and ward s minimum variance criterion . the ward s criterion computation , for example , gave three clusters with 468 , 145 and 184 bursts . we clustered using nonparametric density estimation based on the 100 nearest neighbors , and clustered using the principal components rather than the observed variables . various methods were tried with both the observed @xmath1 values and debiased @xmath18 values , with little effect on the results . all methods showed two strong clusters and the outlier but , in some cases , the third cluster appeared only weakly . to proceed further , we choose a single clustering structure for detailed study : the 5-dimensional average linkage analysis ( table 3b ) with three clusters class i with 486 bursts , class ii with 203 bursts , class iii with 107 bursts . class iv consisting of the single outlier is ignored because of independent evidence that its properties are due to data of poor quality ( 7 ) . the membership of these clusters is given in table 4 , and four projections of the clusters onto two - dimensional scatter plots are shown in figure 3 . these are frames from the ` grand tour ' movie of the 5-dimensional dataset provided by the xgobi software where each cluster is ` brushed ' with a different symbol . note that , in general , there is no reason why classification structure should be most evident in projections parallel to the variable axes shown in figure 1 . it is more important that the clusters show cohesion in many projections of the dataset . the grand tour of the 797 grbs shows that classes i , ii and the outlier are very distinct in most projections . class iii often lies between classes i and ii ( e.g. top panels of figure 3 ) , but in other projections is offset from the line between classes i and ii ( e.g. bottom panels of figure 3 ) . it also appears elongated along some projections , while the larger classes i and ii appear roughly hyperspherical . this analyses described here provide considerable evidence for three major clusters and an outlier . but as some nonparametric clustering procedures did not find a strong third cluster , there is some some worry that class iii is simply a group of bursts with properties intermediate between the classes i and ii . while nonparametric hierarchical clustering methods can not address this question , it can be investigated with parametric methods . mathematically well - founded methods for evaluating the statistical significance of a proposed multivariate classification scheme are available under the assumption that the population is a multinormal mixture ; that is , the objects of each class are drawn from multivariate gaussians . all relationships between the variables must thus be linear ( as in table 2 ) , although the relationships may differ between clusters . there is no requirement of sphericity , so that clusters may have shapes akin to pancakes or cigars with arbitrary orientations in multidimensional space . the separate existence of each of the postulated subpopulations can be tested using multivariate analysis of variance ( manova ) . the results of our manova calculations are summarized in table 5 . the columns give the values of the three manova statistics , followed by details for the wilks @xmath0 : the corresponding value of the @xmath58 statistic , the numerator and denominator degrees of freedom for that @xmath58 value ; and the resulting @xmath60 value . details for pillai s and hotelling - lawley s traces are omitted , but give similar results in all cases . the first row tests the null hypothesis that the classes i , ii and iii have the same mean , the second row tests the equality of classes i and ii , and so forth . the @xmath58 values are very high in all cases , indicating that the clusters are different with extremely high statistical significance ( @xmath61 ) . as the tails of the distribution are poorly determined unless the sample size is extremely large . ] this is a clear _ quantitative _ demonstration that at least two clusters exist among grbs ( for a univariate test , see ashman , bird & zepf 1994 ) , which was _ qualitatively _ reported by delazay et al . ( 1992 ) and k93 . the other rows in table 5 test the hypotheses that each proposed class has the same mean as each other class . one problem with these manova tests is that they are conditional on the classification that has been found using the clustering algorithm . because the clustering algorithm is constructed to find groups that are different from one another , tests such as these tend to be biased towards finding structure , perhaps where none exists . although the manova results seem to indicate very strong evidence of structure , they can not be taken as definitive for this reason . tests arising from model - based clustering can overcome this problem , as discussed in the following section . in the previous section , we conducted a hierarchical clustering analysis without making assumptions regarding the shapes of the clusters , but needed the parametric assumption of normality to estimate the statistical significance of the resulting classification . it is reasonable to conduct the entire analysis , both clustering and validation , within a model - based framework . we report here an analysis of this type again assuming that the grb population consists of a mixture of multivariate gaussian classes . early development of this model for clustering is discussed in mclachlan & basford ( 1988 ) ; we use more recent developments here . first , an initial classification for each possible number of clusters is found via agglomerative hierarchical clustering ( murtagh & raftery 1984 , banfield & raftery 1993 , fraley 1998 ) . next , the em ( expectation - maximization ) algorithm is used to refine partitions obtained from hierarchical clustering ( celeux & govaert 1995 , dasgupta & raftery 1998 ) . finally , the bayesian information criterion is used to select the ` best ' partition among those associated with different numbers of clusters ( dasgupta & raftery 1998 ) . in the model considered here , the @xmath43-dimensional observations @xmath62 are drawn from @xmath38 multinormal groups , each of which is characterized by a vector of parameters @xmath63 for @xmath64 . our goals are : to determine the number of grb types , @xmath38 ; to determine the cluster assignment of each burst ; and to estimate the mean @xmath65 and covariance matrix @xmath66 for each cluster . following fraley ( 1998 ) , the density of an observation @xmath67 from the @xmath68th subpopulation is expressed as follows : @xmath69 where @xmath70 means multivariate normal . we estimate the parameters using the principle of maximum likelihood . in the hierarchical clustering phase , we use the classification likelihood @xmath71 where @xmath72 represents the observations and @xmath73 is the cluster assignment : @xmath74 when @xmath75 comes from the @xmath68-th group . equivalently , @xmath76 where @xmath77 is the set of indices corresponding to observations belonging to the @xmath68-th group . the method used here for maximizing the likelihood ( fraley 1998 ) and implemented in the _ mclust _ code involves parameterization of the @xmath78 matrices in terms of their eigenvectors and eigenvalues ( analogous to a principal components analysis ) , and iterative relocation of the clusters using the em algorithm . the em ( estimation - maximization ) algorithm ( dempster , laird & rubin 1977 ) , one of the most successful methods in modern statistics , is a procedure for iteratively maximizing likelihoods in a wide variety of circumstances . for example , the lucy - richardson algorithm in astronomical image restoration is the em algorithm . in the present application , we apply em to the mixture likelihood @xmath79 where @xmath80 are mixing probabilities associated with each group . for a given number @xmath38 of components in the mixture , we use em to estimate the conditional probability that observation @xmath62 belongs to the @xmath68-th group for each @xmath45 and selected @xmath68 via maximum likelihood . although the computational procedure has some limitations ( e.g. convergence of the em iterations is not guaranteed ; clusters can not be extremely small ) , it is generally efficient and effective for gaussian clustering problems when started from reasonable partitions such as those produced by hierarchical agglomeration . we use the bayes factor to assess the evidence for a given number of clusters against a different number of clusters . the bayes factor , defined in the context of bayesian statistics , is the posterior odds for one model against the other when the prior odds are equal to one ( i.e. , when one does not favor one model over the other _ a priori _ ) . kass & raftery ( 1995 ) review the use of bayes factors in adjudicating between competing scientific hypotheses on the basis of data . the bayes factor for a model @xmath81 against a competing model @xmath82 ( say , for three @xmath83 two classes of grbs ) is defined as @xmath84 where @xmath85 ) for @xmath86 is obtained by integrating the likelihood times the prior density over the parameters of the model . it can be viewed as a likelihood ratio , but it differs from the usual frequentist ratio that underlies the likelihood ratio test in that the latter is obtained by maximizing ( rather than integrating ) the likelihood over the model parameters . twice the logarithm of the bayes factor can be approximated by the bayesian information criterion or bic ( schwarz 1978 ) , @xmath87 where @xmath88 is the likelihood and @xmath89 is the number of parameters for one mixture model , and similarly for @xmath90 and @xmath91 . the bic measures the balance between the improvement in the likelihood and the number of model parameters needed to achieve that likelihood . while the absolute value of the bic is not informative , differences between the bic values for two competing models provide estimates of the evidence in the data for one model against another . conventionally , bic differences @xmath92 represent weak evidence , differences between 2 and 6 represent positive evidence , @xmath93 strong evidence , and @xmath94 very strong evidence ( jeffreys 1961 , appendix b ; kass & raftery 1995 ) . the use of the bic in choosing clusters in a mixture or clustering model is discussed by roeder & wasserman ( 1997 ) and dasgupta & raftery ( 1998 ) . bayes factors and bic have the advantage that they can be used to assess the evidence _ for _ a null hypothesis , unlike standard significance tests which can only reject a null hypothesis . they can also easily be used to compare non - nested models , again unlike standard significance tests which require competing models to be nested . to reduce the dimensionality of the problem and the complexity of the calculation , we eliminated the highly redundant @xmath7 and @xmath26 variables ( see figure 1 ) and considered only the 3 variables @xmath1 , @xmath22 and @xmath25 for the sample of 797 batse grbs . the _ mclust _ model - based clustering procedure described above was run for trials of @xmath95 groups . the resulting values of bic(@xmath38 ) are plotted in figure 4 . the maximum bic is achieved for three classes . most importantly , the bic value for @xmath96 is @xmath97 units above that for @xmath98 . this corresponds to strong evidence indeed for the presence of three groups rather than two . this result strongly confirms the analysis in 4 indicating the existence of three clusters , and this time is free of the problem that the manova tests are conditional on the estimated partition . the result here takes account of the fact that the partition is not known in advance . we have also calculated the bic for @xmath99 with various constraints on the covariance matrix * @xmath100 * such as hypersphericity and uniformly shaped ellipsoids . spherical clusters give poor fits . uniform ellipsoids give good fits with 4 and 8 clusters . but in all cases , the maximum likelihood assuming 2 clusters is much lower than the likelihood of @xmath101 clusters . the cluster assignment vector @xmath102 for the @xmath96 model with unconstrained * @xmath100 * is given in table 4 . over 85% of the assignments are the same as those obtained from the nonparametric hierarchical clustering procedure in 4 , so that we note only differences between the two clustering results with @xmath103 and @xmath104 markings . all but one of the 96 assignment differences move bursts from classes i and ii into class iii . the close agreement between the cluster assignments in the two methods reinforces confidences in the conclusions from both of them . we can now examine the properties of grbs within each cluster with reasonable confidence that the populations are distinct from each other but internally homogeneous . these properties become inputs to astrophysical theories seeking to explain grb bulk properties . table 6a lists the means and standard deviations of the principal variables for each cluster based on both the nonparametric and model - based clustering procedures . the two methods give very similar results . the three types are well - separated in the burst duration variables : cluster i bursts have the longest durations around @xmath105-@xmath106 seconds , cluster ii bursts have the shortest durations below 1 second , and cluster iii bursts have intermediate durations around 2 seconds . this is shown clearly in figure 5 , which projects each class onto the univariate @xmath1 axis . cluster iii bursts are also intermediate in their fluences , although their fluence distribution overlaps that of the fainter class ii bursts . the hardness ratios of all three clusters overlap considerably , but class iii bursts have the softest spectra and class i bursts have intermediate spectra . we can thus classify the types in the three principal dimensions duration / fluence / spectrum ( table 6c ) : class i is long / bright / intermediate , class ii is short / faint / hard , and class iii is intermediate / intermediate / soft . a major constraint for the astrophysical interpretation of grbs has been the remarkable isotropy of their spatial distribution in the celestial sphere . it is possible that , while the bulk of grbs are isotropic and have an inferred extragalactic origin , some class of grbs have significant anisotropy which would reflect a galactic origin ( see lamb 1995 ) . we apply four statistical tests for isotropy discussed by briggs ( 1993 ) and applied by briggs et al . ( 1996 ) to various subsamples of the 3rd batse catalog of grbs . the statistics are : @xmath107cos@xmath108 , where @xmath109 is the angle between a burst and the galactic center ; @xmath107sin@xmath110 , where @xmath6 is the galactic latitude ; rayleigh - watson @xmath111 ; and bingham @xmath112 . @xmath107cos@xmath108 tests the dipole moment around the galactic center , @xmath107sin@xmath110 tests the quadrapole moment with respect to the galactic plane , @xmath111 tests the dipole moment around any point in the celestial sphere , and @xmath112 tests the quadrapole moment around any plane or two poles . the expected values for the four statistics assuming random isotropic distribution on the sphere are 0 , 0 , 3 and 5 respectively . the asymptotic distributions of these statistics are known . table 6b shows the results of this analysis for clusters i - iii , kindly calculated for us by michael briggs . no deviations from isotropy are found . the @xmath107cos@xmath108 and @xmath107sin@xmath113 values lie within one standard deviation of the expected value for a random distribution . the @xmath111 and @xmath112 values must be larger than the expected value to indicate anisotropy . the only such case , class ii with @xmath1147.32 , has a deviation with very low significance ( @xmath115 ) . we thus do not confirm belli s ( 1997 ) report of significant differences in spatial distributions of burst classes i and iii , although we did not specifically test the galactic latitude distribution . in principle , the relative populations of the three classes may be an important constraint on astrophysical theory . we find that class i contains more than half of the bursts with the remainder divided between class ii and class iii ( table 6c ) . but we do not believe our analysis gives a precise census for two reasons . first , the exact assignments of individual bursts to clusters depend on the detailed assumptions of the clustering algorithms . for example , class ii is larger than class iii in the 5-dimensional nonparametric procedure but is smaller in the 3-dimensional model - based procedure . second , the numbers of weaker bursts in classes ii and iii are strongly dependent on the details of the batse instrument s burst triggering process which produces a complicated truncation bias for fainter bursts . we look for structure within each of the clusters by computing correlation coefficients similar to those in 3 for the entire sample . results are given in table 7 . here we see a systematic difference between the two clustering methodologies : nonparametric average linkage clustering tends to give stronger correlations between the variables than the model - based clustering . for example , in the nonparametric analysis we find significant positive correlations between total fluence and hardness in classes i and ii , and a correlation between duration and fluence in class ii . however , we attach more credence to the model - based results for this purpose than the average linkage results because the former method is specifically designed to provide optimal estimates of the within - group covariances given the clustering model . the model - based results do not give strong evidence for any non - zero correlations between variables , suggesting that the partition into three clusters explains all of the correlation between variables in the full dataset . we thus find , using clustering and validation methods with different mathematical underpinnings , that three classes of grbs are present in our large subset of the third batse catalog . most of the structure can be found using three fundamental burst properties , duration / fluence / spectrum . the class properties and relation to previous research can be briefly summarized as follows : class i : : these long / bright / intermediate bursts correspond to the well - known populous long - soft class of k93 and others . within this group , we do not confirm a hardness - duration correlation reported by dezalay et al . ( 1996 ) and horack & hakkila ( 1997 ) . class ii : : this short / faint / hard group corresponds to the short - hard burst type of k93 and others . fluence - duration and fluence - hardness correlations may tentatively be present within the class . note that while the mean location of this type is consistent in the two clustering schemes , its size and population ( e.g. 1/2 or 1/4 that of class i ) differs between clustering algorithms . class iii : : the discovery of this group with intermediate / intermediate / soft properties is the principal result of this study . the group is easily distinguished in the projections of figure 3 , but can also be discerned in some panels of figure 1 . for example , it lies between classes i and ii in the @xmath116 , @xmath117 and @xmath118 scatter plots . in the univariate @xmath1 distribution shown in figure 5 , class iii accounts for most , but not all , of the bursts in the small peak around @xmath119 sec between the major short and long duration peaks . it is possible that our class iii is related to the class of no - high - energy ( nhe ) burst and peaks discussed by pendleton et al . these bursts have unusually weak @xmath12 emission , soft @xmath120 kev spectra , and low @xmath22 . however , the nhe class does not appear to exhibit a clear duration segregation from other bursts as we find for class iii . class iii does not appear to be the third cluster found by baumgart ( 1994 , see his table 3 ) , but the high dimensionality of his analysis prevents a simple comparison with our low dimensionality study . outlier : : batse trigger event 2757 , burst 3b 940114 , is the outlier in the nonparametric analysis of 4 and is clearly visible in many projections in figures 1 and 3 . it has an exceedingly soft hardness ratio and short burst duration . but examination of the original batse database shows that the @xmath121 fluxes are very weak with large measurement uncertainties . the published 3b catalog gives only an upper limit to its total fluence and no estimate of its hardness ratio ( meegan et al . the unusual properties of this burst are thus illusory and are due to its very weak fluence . the multivariate analysis described here is not comprehensive and may not have uncovered all of the structure in the third batse catalog of bulk grb properties . our reduction of dimensionality may have been too severe , omitting , for example , the potentially important @xmath12 as a distinct variable ( pendleton et al . 1997 ; bagoly et al . 1997 ) . many methodological options were not exercised . for example , it would be valuable to repeatedly apply the @xmath68-means partitioning algorithm to the database under the assumption that three clusters are present ( see murtagh 1992 for an astronomical application of this method ) , check for skewness or kurtosis in the clusters , and undertake an oblique decision tree analysis to give analytical formulation to hyperplanes separating the clusters ( see white 1997 ) . codes for these and many other multivariate techniques are publicly available through the web metasite _ statcodes _ at www.astro.psu.edu/statcodes . however , the efforts described here are far more capable of finding and quantifying clustering in the database than most previous analyses ( 1 ) . previous studies have been based on qualitative rather than quantitative procedures for identifying structures , and provide no statistical validation of their claims . it is thus not surprising that we uncovered structure missed by previous researchers . in particular , our confidence in the presence of a third cluster , class iii , is strong . two completely independent mathematical procedures ( 4 and 5 ) found very similar structure , each validated with high statistical confidence . it is possible that the clustering reported here is indeed present in the database , but does not have an astrophysical origin . the complex triggering mechanism of the batse instrument mechanism and biases in bulk property values at low signal - to - noise ratio are two problems that probably affect the multivariate structure . we have investigated one manifestation of the latter effect using @xmath18 and found no effect on our results . instrumental biases generally affect the number of bursts found in some regions of the multivariate hyperspace ( thereby biasing log @xmath36 - log @xmath122 distributions ) and may alter the location of clusters , but are unlikely to cause the appearance of clustering that is not present in the underlying population . nonetheless , since the batse instrument identifies bursts on three separate timescales , it is possible that the third cluster here is related to a selection effect associated with the batse triggering mechanisms . we conclude that the third batse catalog shows three statistically significant types of bursts ( duration / fluence / spectrum ) : class i grbs are long / bright / intermediate , class ii grbs are short / faint / hard , and class iii grbs are intermediate / intermediate / soft . unless the separation of class iii from the other types is due to some subtle batse instrumental effect , these types are likely to be real and their existence should be considered an important input into astrophysical theories for grbs . for example , the three types may reflect different types of external environments and internal shocks in relativistic fireball models ( mszros & rees 1993 ; panaitescu & mszros 1998 ) . note that statistical anlaysis is unable to determine whether burst types represent fundamentally different astrophysical processes or distinct conditions within a single astrophysical model . our results can be confirmed and extended in two fashions . first , the analysis described here can be validated with several hundred more bursts collected by batse since the september 1994 cutoff in the database used here . second , following baumgart ( 1994 ) , the dimensionality of the problem can be enlarged to include detailed characteristics of the burst temporal behaviors . burst smoothness @xmath83 peakiness , characteristic wavelet scales , spectral evolution , and other parameters can be included . with this enlarged database , one can perform both an unsupervised exploratory cluster analysis similar to that described here , and manova - type analyses that assume the existence of the three groups to determine whether the clusters have distinctive temporal properties . this work was supported by nasa grant nagw-2120 and nsf grant dms-9626189 funding astrostatistical research at penn state , and onr grants n-00014 - 96 - 1 - 0192 and n-00014 - 96 - 1 - 0330 funding statistical research at washington . sm would like to thank the departments of astronomy & astrophysics and statistics at penn state for hospitality while this work was underway . we are very grateful to jay norris ( nasa - gsfc ) for providing debiased duration values and michael briggs ( nasa - msfc ) for performing isotropy calculations . we also thank drs . briggs , norris and peter mszros ( penn state ) for sharing many valuable insights into grb astronomy . pendleton , g. n. , paciesas , w. s. , briggs , m. s. , preece , r. d. , mallozzi , r. s. , meegan , c. a. , horack , j. m. , fishman , g. j. , band , d. l. , matteson , j. l. , skelton , r. t. , hakkila , j. , ford , l. a. , kouveliotou , c. & koshut , t. m. 1997 , apj , 489 , 175 lrc & & + & mean & s.d@xmath123 + log @xmath7 ( sec ) & 0.55 & 0.92 + log @xmath1 ( sec ) & 0.96 & 0.92 + log @xmath22 ( erg @xmath124 ) & -5.61 & 0.76 + log @xmath9 ( photons s@xmath125 @xmath126 & 0.16 & 0.45 + log @xmath25 & 0.25 & 0.33 + log @xmath26 & 0.48 & 0.30 + lrrrrrr & & & & & & + & log @xmath7 & log @xmath1 & log @xmath22 & log @xmath9 & log @xmath25 & log @xmath26 + log @xmath7 & 1.00 & & & & & + log @xmath1 & 0.97 & 1.00 & & & & + log @xmath22 & 0.63 & 0.66 & 1.00 & & & + log @xmath9 & -0.01 & 0.04 & 0.59 & 1.00 & & + log @xmath25 & -0.36 & -0.36 & 0.02 & 0.24 & 1.00 & + log @xmath26 & -0.35 & -0.35 & -0.00 & 0.19 & 0.96 & 1.00 + rcrrr & & & & + level & merger & members & @xmath40 & @xmath37 + + 8 & 10 + 15 & 506 & 0.08 & 0.65 + 7 & 14 + 137 & 93 & 0.00 & 0.65 + 6 & 8 + 7 & 599 & 0.10 & 0.55 + 5 & 9 + 266 & 188 & 0.00 & 0.55 + 4 & 5 + 26 & 190 & 0.00 & 0.55 + 3 & 6 + 12 & 606 & 0.01 & 0.54 + 2 & 3 + 4 & 796 & 0.53 & 0.01 + 1 & 2 + 616 & 797 & 0.00 & 0.00 + & & & & + + 6 & 15 + 21 & 107 & 0.01 & 0.70 + 5 & 10 + 8 & 486 & 0.01 & 0.69 + 4 & 7 + 20 & 203 & 0.01 & 0.68 + 3 & 6 + 5 & 593 & 0.10 & 0.58 + 2 & 3 + 4 & 796 & 0.58 & 0.00 + 1 & 2 + 616 & 797 & 0.01 & 0.00 + lcrrr & & & & + variable & method@xmath127 & + & & & & + + log @xmath7 & np & 1.13 @xmath128 0.44 & -0.80 @xmath128 0.41 & 0.33 @xmath128 0.27 + log @xmath1 & np & 1.55 @xmath128 0.40 & -0.42 @xmath128 0.44 & 0.71 @xmath128 0.32 + & mb & 1.22 @xmath128 0.39 & -0.91 @xmath128 0.35 & 0.29 @xmath128 0.41 + log @xmath22 & np & -5.21 @xmath128 0.59 & -6.37 @xmath128 0.57 & -6.11 @xmath128 0.37 + & mb & -5.13 @xmath128 0.58 & -6.46 @xmath128 0.54 & -5.93 @xmath128 0.47 + log @xmath25 & np & 0.19 @xmath128 0.27 & 0.51 @xmath128 0.27 & 0.08 @xmath128 0.40 + & mb & 0.21 @xmath128 0.26 & 0.52 @xmath128 0.28 & 0.16 @xmath128 0.40 + log @xmath26 & np & 0.43 @xmath128 0.23 & 0.70 @xmath128 0.26 & 0.35 @xmath128 0.39 + & & & & + + @xmath107cos @xmath109@xmath4 & np & 0.015 & -0.041 & 0.010 + @xmath107sin@xmath129 & np & -0.012 & -0.025 & 0.028 + rayleigh - watson & np & 0.39 & 1.28 & 1.80 + bingham & np & 2.02 & 7.32 & 1.95 + & & & & + + number & np & 486 & 203 & 107 + & mb & 426 & 170 & 201 + duration & & long & short & intermediate + fluence & & bright & faint & intermediate + spectrum & & intermediate & hard & soft + + lrrrrr & & & & & + & log @xmath7 & log @xmath1 & log @xmath22 & log @xmath25 & log @xmath26 + + + log @xmath7 & 1.00 & & & & + log @xmath1 & 0.88 & 1.00 & & & + log @xmath22 & 0.10 & 0.22 & 1.00 & & + log @xmath25 & -0.11 & -0.08 & 0.39 & 1.00 & + log @xmath26 & -0.11 & -0.08 & 0.38 & 0.97 & 1.00 + & & & & & + + + log @xmath7 & 1.00 & & & & + log @xmath1 & n / a & n / a & & & + log @xmath22 & n / a & 0.01 & 1.00 & & + log @xmath25 & n / a & -0.01 & 0.06 & 1.00 & + log @xmath26 & n / a & n / a & n / a & n / a & n / a + & & & & & + + + log @xmath7 & 1.00 & & & & + log @xmath1 & n / a & n / a & & & + log @xmath22 & n / a & -0.03 & 1.00 & & + log @xmath25 & n / a & -0.05 & 0.02 & 1.00 & + log @xmath26 & n / a & n / a & n / a & n / a & n / a + & & & & & + + + log @xmath7 & 1.00 & & & & + log @xmath1 & 0.86 & 1.00 & & & + log @xmath22 & 0.02 & 0.06 & 1.00 & & + log @xmath25 & -0.24 & -0.34 & -0.16 & 1.00 & + log @xmath26 & -0.22 & -0.32 & -0.22 & 0.95 & 1.00 + & & & & & + + + log @xmath7 & 1.00 & & & & + log @xmath1 & n / a & n / a & & & + log @xmath22 & n / a & 0.03 & 1.00 & & + log @xmath25 & n / a & -0.01 & 0.07 & 1.00 & + log @xmath26 & n / a & n / a & n / a & n / a & n / a +
in qed , the landau - pomeranchuk - migdal ( lpm ) effect interpolates between the bethe - heitler and factorization limit for the radiation spectrum of a charged particle undergoing multiple , @xmath4 say , scatterings in a medium . if the separation between scattering centers is very large , then the radiation off these centers reduces to a sum of radiation spectra for @xmath4 small - angle scatterings - this is the bethe - heitler limit . in the opposite limit , when the scattering centers sit too close together to be resolved by the emitted photon , the observed radiation factorizes into a product of a single scattering bethe - heitler spectrum for momentum transfer @xmath5 and the elastic cross section for the momentum transfer @xmath6 accumulated over @xmath4 small - angle scatterings . as first noted by landau and pomeranchuk @xcite , the relevant length scale for the interpolation between bethe - heitler and factorization limit is the coherence length ( formation length ) @xmath7 , determined by the longitudinal momentum transfer @xmath8 , @xmath9 this characterizes the longitudinal scale on which the radiated particle becomes distinguishable from its radiating parent . scattering amplitudes for the radiation off different scattering centers interfer destructively if their separation is less than @xmath7 : the _ coherent _ factorization limit is suppressed in the ultrarelativistic limit with respect to the _ incoherent _ bethe - heitler limit . for a quantitative description of the lpm interference effect , the relative phases of the different contributions to the @xmath4-fold scattering amplitude matter . these depend on the transverse energies and thus require knowledge about the transverse motion of the radiating particle in the medium . migdal @xcite was the first to develop a dynamical description to this aim , employing a two - dimensional fokker - planck transport equation @xcite for the hard parton . in the limit of an infinite medium , his well - known result shows a characteristic @xmath10 low frequency suppression of @xmath11 compared to the constant dependence in the bethe - heitler limit , i.e. , the coherent factorization limit for @xmath12 vanishes . this is , however , only a very special feature of migdal s limiting case , where the formation length goes to infinity for @xmath12 , but never exceeds the ( infinite ) extension @xmath13 of the medium . for a medium of finite size , the formation length does exceed the system size below some critical frequency @xmath14 , and the bethe - heitler limit is finite . migdals @xmath15-dependence is hence not valid for @xmath16 . moreover , additional effects become important for the radiation spectrum at lower frequencies @xcite . most notably , this is the transition radiation and the ter - mikaelian effect @xcite of dielectric suppression . the renewed interest in the lpm - effect has at least two reasons : on the one hand , fourty years after discovering the theoretical principles @xcite , the first precision measurements of the lpm - effect @xcite ( and the ter - mikaelian effect @xcite ) were made recently by the slac-146 collaboration . on the other hand , with the advent of a new generation of relativistic heavy ion colliders at rhic and lhc , the understanding of the non - abelian analogue becomes important . in the qed case , the experiment explored relatively thin targets with @xmath17 on the order 0.1 to 10 , in which the transition between bethe - heitler and factorization limit occurs . for a quantitative understanding , a realistic theory has to account for the finite extension of the target , the multiple elastic scatterings in the target , multiple photon emission , and possibly additional complications like radiation off structured targets . there are at least three modern approaches , which can account in principle for the lpm - effect in realistic targets . they implement the eikonal approximation for the radiating hard particle in different ways : i ) blankenbecler and drell @xcite ( see also baier and katkov @xcite ) started from the solution of the klein - gordon equation for the charged particle in the presence of an external field . the solution was approximated to order @xmath18 in a high energy expansion in which the eikonal path of the radiating particle is recovered . ii ) zakharov @xcite proposed a light - cone path integral formulation which describes in a coordinate - space representation the transverse momentum kicks on the eikonal path , mimicking the elastic scatterings by an effective dipole cross section . zakharov s work @xcite provides the most accurate description of the measured data so far . as we shall see in section [ sec2c ] , his starting point is closely related to a high energy expansion of the dirac equation in the presence of an external potential . iii ) a third approach is due to r. baier et al . ( bdmps ) @xcite , who started from the radiation amplitudes for @xmath4-fold scattering in the eikonal approximation . for qed , the consistency of their approach and the work of blankenbecler and drell can be checked diagrammatically @xcite . a discussion of the lpm effect also exists for non - equilibrium conditions @xcite . in the qcd case , high @xmath19 gev jets will be one of the new probes of the dense matter produced in relativistic heavy ion collisions . deviations from factorized perturbative qcd due to final state rescattering are sensitive to the density and coupling in the plasma and will affect key observables like the high momentum tails of single particle spectra . the corresponding study of qcd radiative energy loss due to final state rescattering was initiated by gyulassy and wang @xcite . recently , it has been extended most notably in a series of papers by bdmps @xcite using equal time perturbation theory . also , zakharov has pointed out that his formalism can be adopted to qcd bremsstrahlung with a dipole prescription @xcite , and the equivalence of zakharov s formalism with the work of bdmps was sketched @xcite . all these calculations of the lpm - effect in qcd , however , are ( 1 ) limited to the cases of infinitely many or very few ( @xmath20 ) rescatterings of the parton , and ( 2 ) they mainly focus on the angular integrated energy loss @xmath21 in which the transverse momentum dependence of the radiation pattern is averaged out . this energy loss , however , is not a good observable for qcd , because the parton shower is not directly observable due to hadronization . in addition , the qcd bremsstrahlung of hard jets must compete with the hard radiation associated with the jet production . to detect modifications of this hard vacuum bremsstrahlung spectrum due to final state rescattering requires knowledge about the angular distribution of the spectrum . in qcd , the transverse momentum dependence of the lpm - effect is thus indispensable for a quantitative understanding of radiative energy loss . recently , kopeliovich , schfer and tarasov ( kst ) @xcite have used the furry approximation of the dirac equation in order to account for the transverse momentum dependence of the radiation spectrum emitted in a multiple scattering process . for the case of qcd bremsstrahlung radiation , they translate their qed results into qcd via an _ a posteriori _ dipole prescription ( [ 5.1 ] ) . this strategy is frequently used not only for the calculation of the radiative energy loss of high @xmath22 partons @xcite , but also for the description of nuclear shadowing in the target rest frame @xcite and for related problems of diffractive dissociation of virtual photons @xcite . as we show in what follows , an extension of the kst - formalism has great potential for the calculation of radiative energy loss in realistic scenarios since ( 1 ) it provides a smooth interpolation between the cases of infinitely many and very few rescatterings and ( 2 ) it allows to compare qed - inspired calculations of qcd radiative energy loss with perturbative qcd results . the present work focuses on the general formalism , its region of validity , and a qualitative discussion of its generic features . it gives expressions which allow for the numerical calculation of the radiation spectrum as a function of the medium density and extension , but it leaves phenomenological applications to further publications . our work is organized as follows . in section [ sec2 ] , we derive the starting point of our discussion , the radiation spectrum ( [ 2.22 ] ) . we discuss how a technical complication , the regularization of this spectrum , can be dealt with analytically , and we calculate the corresponding integrated energy loss . section [ sec3 ] focuses on limiting cases of the general radiation formula ( [ 2.22 ] ) . we derive the radiation spectra for @xmath23 , @xmath1 and @xmath2 scatterings and we show that these reproduce the bethe - heitler and the factorization limit . in section [ sec4 ] , the dipole approximation of the radiation spectrum ( [ 2.22 ] ) is discussed . this gives access to true in medium properties of the radiation spectrum which can not be obtained from an expansion to finite order in the coupling constant . finally , we discuss in section [ sec5 ] how this method can be extended to qcd and how it compares to results from perturbative qcd . our main results are summarized in the conclusion . an expansion of the lpm - radiation cross section in orders of the coupling constant is essentially an expansion in the number of elastic scatterings , since each elastic scattering mott cross section is proportional to @xmath24 . in contrast , a high energy expansion of the solution of the dirac equation in the presence of an external field takes to leading order in @xmath18 an arbitrary number of elastic scatterings into account . in this section , we derive the corresponding high energy limit of the qed radiation cross section for a hard electron , traversing a medium of longitudinal density @xmath25 . the physics contained in our main result ( [ 2.22 ] ) will be discussed in the following sections [ sec3 ] - [ sec5 ] . our approach in [ sec2a ] parallels to a large extent that of kopeliovich , schfer and tarasov ( kst ) in ref . we present the derivation in full detail to introduce our notation and to discuss all the approximations involved in the calculation . in contrast to ref . @xcite , our radiation cross section ( [ 2.22 ] ) contains an @xmath26-regularization whose treatment is discussed in subsection [ sec2b ] . then we show that the kst - formalism results in an angular integrated energy loss formula from which modifications to zakharov s formalism @xcite can be obtained . we consider a relativistic electron undergoing multiple small - angle scattering in a spatially extended medium , described by an external potential @xmath27 . the angular dependence of the radiation spectrum for emitted photons carrying away a fraction @xmath28 of the incident electron energy @xmath29 is given by the differential cross section @xmath30 where @xmath31 and @xmath32 denote the transverse momenta of the photon and outgoing electron , respectively . the radiation amplitude for transversely polarized photons is given in terms of the ingoing and outgoing electron wavefunctions @xmath33 and @xmath34 , @xmath35 here , @xmath36 , and @xmath37 is the adiabatic switching off of the interaction term at large distances . this term plays an important role in what follows since the @xmath38 limit does not commute with the longitudinal @xmath39-integration . the wavefunctions @xmath40 solve the dirac equation in the external potential @xmath27 : @xmath41\ , \psi(x , p_{1,2 } ) = 0\ , . \label{2.3}\ ] ] following kopeliovich et al . @xcite , we rewrite the dirac equation @xmath42 \psi(x , p_{1,2 } ) \nonumber \\ & & = \left [ -i\ , \bbox{\alpha}\cdot\left(\bbox{\nabla}u({\bf x})\right ) - u^2({\bf x})\right ] \psi(x , p_{1,2})\ , . \label{2.4}\end{aligned}\ ] ] this allows for an expansion to order @xmath18 , treating the right hand side as a small perturbation , solving the corresponding homogeneous problem and including in a first iteration the derivative term @xmath43 . the result is the furry approximation @xcite : @xmath44 which is exact to order @xmath45 and @xmath46 . the building blocks of this solution are the differential operators @xmath47 and the functions @xmath48 whose properties we explain now : to determine the functions @xmath48 , we neglect the second derivative @xmath49 in ( [ 2.4 ] ) . this is allowed since the longitudinal distances in a multiple scattering problem are much larger than the transverse ones . the left hand side of ( [ 2.4 ] ) reduces then to a two - dimensional schrdinger equation , with corresponding retarded green s function @xmath50\ , g(z_2,{\bf r}_2;z_1,{\bf r}_1|{\bf p } ) \nonumber \\ & & \qquad \qquad = i\,\delta(z_2-z_1)\,\delta({\bf r}_2-{\bf r}_1 ) \label{2.7}\ , , \end{aligned}\ ] ] satisfying @xmath51 and vanishing for @xmath52 . the functions @xmath48 in ( [ 2.5 ] ) , ( [ 2.6 ] ) are solutions of this two - dimensional schrdinger equation , @xmath53 for very early and very late times , i.e. for far forward and far backward longitudinal distances @xmath54 and @xmath55 respectively , they satisfy the boundary conditions @xmath56 this ensures that the wavefunctions @xmath57 approximate plane waves at asymptotic distances . the differential operators @xmath58 are obtained by including the derivative term @xmath59 of ( [ 2.4 ] ) in a first iteration of this solution , @xmath60 we now explain how to obtain from the furry approximation an explicit expression of the differential radiation cross section ( [ 2.1 ] ) . we work in the ultrarelativistic limit , @xmath61 , @xmath62 , @xmath63 . the longitudinal axis is redefined to be parallel to the emitted photon @xmath64 , see figure [ fig4 ] . the transverse momenta of the ingoing ( @xmath65 ) and outgoing ( @xmath66 ) electron for small @xmath28 are therefore @xmath67 = 7.5 cm for the calculation of @xmath68 , we use the following four steps : + 1 ) we do the time - integral in ( [ 2.2 ] ) which ensures energy conservation , @xmath69 . + 2 ) we determine the @xmath39-dependent phase factor @xmath70 for ( [ 2.2 ] ) by observing that the @xmath39-dependent phases in ( [ 2.5 ] ) , ( [ 2.6 ] ) come with the modulus of the spatial momentum @xmath71 . using energy - momentum conservation , one finds to leading order @xmath18 @xmath72 the inverse momentum transfer @xmath73 is often refered to as photon formation length @xmath7 . however , contributions from transverse energies @xmath74 dominate the @xmath39-dependent phase factor of @xmath75 , as will be seen below . hence , @xmath7 gives only a rough upper estimate for the longitudinal size over which interference effects suppress photon emission . + 3 ) we absorb in ( [ 2.2 ] ) the spinor structure of the wavefunctions @xmath76 , @xmath77 in the interaction vertex @xmath78 depending on the electron spin of the ingoing ( @xmath79 ) and outgoing ( @xmath80 ) electron and the photon helicity ( @xmath81 ) , this vertex takes the form @xmath82 \nonumber \\ & & \times \left ( { \partial\over \partial r_x } -i\ , \lambda_\gamma\ , { \partial\over \partial r_y } \right)\ , , \label{2.15 } \\ \widehat\gamma_r\bigl(\lambda_e=-\lambda_{e ' } , \lambda_\gamma\bigr ) & = & 2\ , m_e\ , x\ , \lambda_\gamma\ , \delta_{\lambda_\gamma\ , , 2\lambda_e } \ , , \label{2.16}\end{aligned}\ ] ] where the spin - flip contribution ( [ 2.16 ] ) is only non - vanishing for @xmath83 . the differential operators in ( [ 2.14 ] ) act to the right on the two different transverse components of @xmath84 . in what follows , we are only interested in the spin- and helicity - averaged combination @xmath85 which takes the simple form @xmath86 \frac{\partial}{\partial{\bf r } } \cdot \frac{\partial}{\partial{\bf r'}}\ , + 2m_e^2x^2\ , . \label{2.17}\ ] ] 4 ) we split up the green s functions @xmath87 at longitudinal distances @xmath88 in such a way that contributions from the amplitude ( [ 2.2 ] ) and its complex conjugate part can be paired , see eg . ( [ 2.18 ] ) and fig . [ fig1 ] . after these four steps , the radiation probability can be expressed as a fourier transform over pairs of green s functions on whose transverse coordinates the interaction vertex ( [ 2.17 ] ) is acting : @xmath89 to turn this into an explicit expression , it is necessary to i ) calculate the averages @xmath90 over the scattering centers in the medium and ii ) evaluate the green s functions @xmath91 which are given by path integrals @xmath92 \right\ } \label{2.19}\end{aligned}\ ] ] with boundary conditions @xmath93 , @xmath94 . we show in appendix [ appa ] that the in medium averages involved in ( [ 2.18 ] ) result in a factor which depends on the cross section @xmath95 times the density @xmath25 of scattering centers as follows @xmath96 \right\}\bigg\rangle \nonumber \\ & & \qquad = \exp\left\ { - \int\limits_{z}^{z'}{\it d}\xi\ , \sigma\bigr(\xi,\left[\bbox{r}(\xi)-\bbox{r}'(\xi)\right]\bigl ) \right\}\ , , \label{2.20 } \\ & & \sigma\bigl(\xi,\bbox{\rho}(\xi)\bigr ) = \frac{1}{2}\ , n(\xi)\ , % \theta_a\ , \sigma\bigl(\bbox{\rho}(\xi)\bigr)\ , . \label{2.21 } \end{aligned}\ ] ] an explicit expression for @xmath97 , supporting its interpretation as a mott cross section , is also derived in appendix [ appa ] . most importantly , the resulting expression @xmath98 depends only on the relative distance @xmath99 between the paths @xmath100 and @xmath101 . as we explain in appendix [ appb ] , this makes it possible to carry out a large number of the integrals and path - integrals in ( [ 2.18 ] ) analytically . one obtains ( see appendix [ appb ] for further details ) @xmath102 { \cal k}\bigl(z',\bbox{\rho}_2;z,\bbox{\rho}_1|\mu\bigr ) \ , . \label{2.22 } \end{aligned}\ ] ] this differs from the result in @xcite by including the regularization @xmath103 . the physical radiation spectrum is obtained by removing @xmath38 _ after _ doing the @xmath39 and @xmath88-integrations and taking @xmath104 and @xmath105 . to simplify the notation , we define @xmath106 which plays the role of mass , and the factor @xmath107 which accounts for a combination of recurring kinematical prefactors . the coefficients of the spin - flip ( @xmath108 ) and non - flip ( @xmath109 ) contributions of the interaction vertex ( [ 2.17 ] ) read @xmath110 the remaining path integral @xmath111 in ( [ 2.22 ] ) is given by @xmath112 \right\}\ , . \label{2.24}\end{aligned}\ ] ] in this path integral , @xmath113 plays the role of an imaginary potential , while its physical interpretation is up to a minus sign that of an elastic mott cross section times the density of scattering centers , cf . equation ( [ a.12 ] ) . depending on the context , we shall hence refer to @xmath114 as a potential or a cross section . the following sections are devoted to a study of the differential cross section ( [ 2.22 ] ) . we focus on analytically accessible limiting cases which illustrate the physics contained in ( [ 2.22 ] ) and we discuss approximations which allow for its numerical analysis . the regularization prescription in ( [ 2.22 ] ) can not be neglected since the @xmath38 limit does not commute with the @xmath39- and @xmath88-integrations . this complicates practical applications of ( [ 2.22 ] ) : even if the green s function @xmath111 is known explicitly in some approximation ( see sections [ sec3 ] and [ sec4 ] below ) , the expression ( [ 2.22 ] ) is not suited for numerical calculations since one can not control numerically the @xmath38-limit _ after _ carrying out the @xmath39-integrations . to solve this problem , we determine the @xmath38 limit of the radiation spectrum ( [ 2.22 ] ) analytically . we consider a medium of arbitrary but finite longitudinal extension which is positioned along the longitudinal axis between 0 and the finite distance @xmath13 . the @xmath39- and @xmath88-integrations of ( [ 2.22 ] ) can be split into six parts , @xmath115 to shorten the following calculations , we omit the spin - flip contribution @xmath108 to the radiation spectrum . this contribution is negligible in the relativistic limit @xmath116 on which we focus in the following . [ if needed , the @xmath108-contributions to all our results can be recovered without additional technical difficulties . ] the differential cross section receives six contributions which we label according to ( [ 2.25 ] ) in an obvious way : @xmath117 since the medium does not extend outside the interval @xmath118 $ ] , we can replace the in - medium propagator @xmath119 outside @xmath118 $ ] by the free propagator @xmath120 , given in ( [ 3.3 ] ) below . e.g. , for @xmath121 , we can write @xmath122 this allows to do all `` infinite '' integrals @xmath123 and @xmath124 in ( [ 2.25 ] ) analytically . one is left with the `` finite '' integrals @xmath125 for which the @xmath38 limit commutes with the @xmath39-integration . doing the @xmath39-integrations first , then taking @xmath126 and @xmath127 respectively , and finally adiabatically switching off the regularization , @xmath128 , we find @xmath129 here , the factors @xmath130 contain the explicit solutions of those @xmath39-integrals which extend to @xmath54 or @xmath55 respectively . using the notational shorthands @xmath131 we obtain @xmath132 and analogously , @xmath133 the differential radiation spectrum ( [ 2.26 ] ) given by the equations ( [ 2.28 ] ) - ( [ 2.38 ] ) is suitable for a numerical evaluation since the @xmath38-limit is taken . previous numerical investigations @xcite of ( [ 2.22 ] ) were based on a non - regularized expression which did not include the term @xmath103 in the integrand . the difference to the above results is found by determining the above @xmath39-integrals without regularization : @xmath134 the differential cross - section ( [ 2.22 ] ) can be related to the unregularized one @xmath135 : @xmath136\ , + \ , \theta_{\rm corr}^{\rm osc}[z_-,z_+]\ , . \label{2.44}\end{aligned}\ ] ] here , the correction term @xmath137 $ ] summarizes those oscillating contributions which vanish if one averages suitably in a numerical calculation over the boundaries @xmath55 and @xmath54 of the longitudinal integration . in addition , there are two terms which do not depend on the boundaries @xmath55 , @xmath54 . they stem from the terms @xmath138 and @xmath139 which differ by an overall factor 2 from the regularized expressions @xmath140 , @xmath141 . these terms provide the proper subtraction of the weizscker - williams fields of the ingoing and outgoing electron . as will become clear below , they are non - negligible , and ensure e.g. the consistency of ( [ 2.22 ] ) with the bethe - heitler spectrum in the appropriate limiting cases . the total energy loss cross section is calculated by integrating ( [ 2.22 ] ) over transverse momenta . to this aim , we rewrite the differential cross section as a function of the relative momentum transfer @xmath142 and the transverse momentum @xmath31 of the emitted photon . since there is no kinematical boundary on @xmath143 , the integration over the transverse momentum of the electron is simple : @xmath144 { \cal k}\bigl(z',\bbox{\rho}_2=0;z,\bbox{\rho}_1|\mu\bigr ) \ , . \label{2.45 } \end{aligned}\ ] ] the integration over the photon transverse momentum is more complicated due to the finite kinematical boundaries for the photon , @xmath145 : @xmath146 \nonumber \\ & & \qquad \times { \cal k}\bigl(z',\bbox{\rho}_2=0;z,\bbox{\rho}_1|\mu\bigr ) \ , . \label{2.46}\end{aligned}\ ] ] it is interesting to compare this energy loss formula to that derived by zakharov @xcite : @xmath147 \nonumber \\ & & \times \left [ { \cal k}\bigl(z',0;z,0|\mu\bigr ) - { \cal k}_0\bigl(z',0;z,0|\mu\bigr ) \right]\ , . \label{2.47 } \end{aligned}\ ] ] here , @xmath120 is the free , non - interacting path integral , given explicitly in ( [ 3.3 ] ) below . if we had ignored the kinematical boundary , extending the @xmath148-integration in ( [ 2.46 ] ) up to infinity , we would have regained the expression of zakharov except for the term proportional to @xmath120 . we note that zakharov s arguments leading to ( [ 2.47 ] ) are very different from our derivation of ( [ 2.46 ] ) . the transverse momentum dependence of the radiation enters in no intermediate step of his calculation @xcite . also , the @xmath120-term in ( [ 2.47 ] ) does not result from a derivation : it is rather subtracted _ a posteriori _ as a ` renormalization prescription ' in order to cancel a singularity in the integrand of ( [ 2.47 ] ) for small @xmath149 . in this sense , our derivation differs from zakharov s result ( [ 2.47 ] ) by : i ) taking the finite phase space of the emitted photon properly into account , ii ) containing a proper @xmath26-regularization of the radiation amplitude and iii ) not employing a subtraction of a singular contribution a posteriori . information about the target medium enters the radiation cross section ( [ 2.22 ] ) via the product @xmath150 , which measures the elastic cross section times the density of scattering centers in the medium . since @xmath151 @xmath25 , an expansion of ( [ 2.22 ] ) in powers of @xmath24 is an expansion in powers of the opacity @xmath152 , where @xmath153 is the effective elastic single scattering cross section and @xmath154 @xmath152 measures the average number of scatterings for an electron traversing a medium of length @xmath13 . since the elastic cross section in ( [ 2.22 ] ) is not a geometrical quantity but depends on the integration variables , our expansion of ( [ 2.22 ] ) will be formally in powers of @xmath155 , and we shall refer by a slight abuse of language to @xmath156 as opacity . of course , after all integrals are done , each power of @xmath155 will be accompanied by a power of the elastic cross section . the @xmath4-th term in this expansion of ( [ 2.22 ] ) is of order @xmath157 and corresponds to the multiple scattering off exactly @xmath4 external potentials . an expansion in the opacity @xmath156 is thus an expansion in the number of rescatterings . below , we study this low opacity expansion of the radiation spectrum up to third order . this leads for equation ( [ 2.22 ] ) to a number of consistency checks which any radiation spectrum including in medium effects should satisfy . moreover , this provides the basis for a discussion of the corresponding qcd radiation spectrum in section [ sec5 ] . the derivation of the radiation spectrum ( [ 2.22 ] ) relies on the approximation that the distribution of scattering centers in the medium can be described by the average ( [ 2.20 ] ) . this uses explicitly that the size of the medium is much larger than the debye radius of a single scattering potential . a priori , it may hence seem unclear to what extent one can still recover from ( [ 2.22 ] ) the correct radiation spectrum for a single scattering process where the extension of the potential is the extension of the target . however , in the low opacity limit , when the distance between scattering centers is much larger than the photon formation length , multiple scattering should not affect the radiation pattern . the radiation spectrum should converge to the single scattering bethe - heitler cross section times the opacity factor @xmath156 . deriving this is an important consistency check for our formalism . our expansion of the integrand of the cross section ( [ 2.22 ] ) in powers of @xmath151 @xmath25 , uses the corresponding expansion of the path integral @xmath158 where from now on , we suppress the explicit @xmath159-dependence in @xmath111 and in the corresponding free green s function @xmath120 , @xmath160 from the two potentials in the third line of ( [ 2.22 ] ) and from the propagator @xmath111 , the radiation spectrum ( [ 2.22 ] ) receives three contributions to order @xmath161 : @xmath162 here , @xmath163 and @xmath164 denote the derivatives of the zeroth and first order contributions to the propagator ( [ 3.2 ] ) , @xmath165 we consider a medium of finite longitudinal extension @xmath13 and homogeneous density @xmath166 . then , all integrals in ( [ 3.5 ] ) - ( [ 3.7 ] ) can be done analytically . taking the limit @xmath38 after the @xmath39-integrations , we arrive at @xmath167 the result is particularly transparent in the relativistic limit , when all mass dependencies can be neglected and @xmath168 . for a medium characterized by a set of yukawa potentials with electric charge @xmath169 and debye screening mass @xmath170 ( see appendix [ appa ] for a derivation ) , @xmath171 denotes the elastic scattering cross section with @xmath143 given by ( [ 2.12 ] ) . the three terms ( [ 3.10])-([3.12 ] ) then combine to @xmath172 this is the high - energy limit of the bethe - heitler radiation cross section for scattering off a yukawa potential times the opacity factor @xmath173 . the exact result [ see e.g. equation ( 5.150 ) of ref . @xcite in the coordinate system fig . [ fig4 ] used here . ] differs from the above approximation only by terms of order @xmath174 and @xmath175 . the main properties of the qed radiation spectrum are seen clearly from ( [ 3.14 ] ) . the spectrum vanishes for vanishing momentum transfer @xmath176 , it shows the @xmath177-dependence characteristic for the qed radiation spectrum peaking at forward rapidity , and it shows the correct @xmath178 @xmath179 dependence . we mention as an aside that the evaluation of the cross section ( [ 3.4 ] ) without regularization prescription leads to terms @xmath180 and @xmath181 which are a factor 2 larger while the interference term remains unchanged . the @xmath26-regularization of ( [ 2.22 ] ) is thus necessary to regain the bethe - heitler spectrum . in complete analogy to the derivation of the bethe - heitler limit , one can expand the cross section ( [ 2.22 ] ) to second order . all contributions to the radiation cross section depend now on the product of two elastic mott cross sections @xmath182 , whose combined momentum transfers sum up to @xmath143 . for this we introduce the shorthand @xmath183 also , we introduce shorthands for the transverse momenta and the corresponding transverse energies , @xmath184 the second order contribution of the radiation spectrum ( [ 2.22 ] ) consists of six terms . after integrating out all transverse coordinates , it takes the form @xmath185\ , n^2_0\ , . \label{3.16}\end{aligned}\ ] ] the first three terms stem from the expansion of the exponential term @xmath186 in ( [ 2.22 ] ) to second order in @xmath114 , the fourth term is from the second order expansion of the green s function @xmath111 , and the remaining two terms are contributions from the first order in @xmath111 times the first order of the exponential term . the variables @xmath187 stand for the remaining longitudinal integrals over phase factors . in general , they are involved , e.g. @xmath188 for a medium of homogeneous density @xmath189 and finite length @xmath13 , they reduce to very simple forms : @xmath190 from this , we read off simple limiting cases : in the limit of a very thin target of fixed opacity , we can move the two scattering centers so close together that the photon can not resolve them . the spectrum is then indistinguishable from a single scattering process @xmath191 this is the coherent factorization limit that corresponds to one single effective bethe - heitler radiation spectrum which depends only on the initial and final momenta . the characterisitc momentum dependence of the bethe - heitler radiation spectrum ( [ 3.14 ] ) is now combined with the convolution of two mott cross sections ( [ 3.15 ] ) . the first @xmath13-dependent correction to this factorization limit is proportional to @xmath192 and takes the form @xmath193\ , , \label{3.25}\end{aligned}\ ] ] which is mainly of formal interest since the @xmath143-integral diverges logarithmically . in the opposite limit , @xmath194 , we can study for fixed opacity @xmath195 the case of two well - separated scattering centers : @xmath196\ , . \label{3.26}\end{aligned}\ ] ] this is the incoherent bethe - heitler limit in which the radiation spectrum is the sum of two independent bethe - heitler contributions associated with the scattering off the first and the second external potential with momentum transfers @xmath197 and @xmath198 , respectively . it provides the starting point for a comparison of the corresponding qcd radiation spectrum with a full @xmath3 perturbative calculation in section [ sec5 ] . the pattern emerging in this expansion to second order in the opacity is fully confirmed in the case of @xmath199 scattering centers . in appendix [ appc ] , we derive the corresponding radiation cross section and check the bethe - heitler and factorization limit . to calculate the radiation spectrum ( [ 2.22 ] ) to all orders in opacity , an approximation scheme for the path integral @xmath111 is needed . the low opacity expansion , studied in section [ sec3 ] , becomes rapidly more complicated for increasing number of scattering centers . it is only useful for the description of ultrathin media where very few elastic scatterings have to be taken into account . for the generic case of a hard projectile particle undergoing many small angle scatterings , the dipole approximation of @xmath111 is standard @xcite . this dipole approximation is based on the observation that for small transverse distances @xmath200 , the cross section @xmath201 of the yukawa potential ( [ a.11 ] ) has a leading quadratic dependence : @xcite @xmath202\ , , \label{4.2}\end{aligned}\ ] ] where @xmath203 denotes eulers constant . the main contribution to the radiation spectrum ( [ 2.22 ] ) comes from small values of @xmath204 , where @xmath205 shows only a slow logarithmic dependence on @xmath204 and can be approximated by a constant , @xmath206 . for a quadratic dependence @xmath207 , where this logarithmic dependence is weak enough to be neglected , the path integral @xmath111 in ( [ 2.24 ] ) is that of a harmonic oscillator @xcite @xmath208 \over 2\ , \sin(\omega\delta z ) } \right\ } \label{4.3}\end{aligned}\ ] ] with the oscillator frequency @xmath209 in the following , we exploit the consequences of this approximation . in section [ sec2b ] , we have determined the six contributions ( [ 2.28 ] ) - ( [ 2.33 ] ) to the radiation spectrum ( [ 2.22 ] ) . in the dipole approximation , the transverse integrations in ( [ 2.28 ] ) - ( [ 2.33 ] ) reduce to gaussian integrals and can be done analytically . here we discuss the resulting expression @xmath210 obtained from ( [ 2.26 ] ) . for three of the six terms @xmath211 , all integrals can be done analytically : @xmath212 these terms show a characteristic dependence on the coupling constant : in the dipole approximation , @xmath213 is constant and measures the size of the elastic single scattering mott cross section , see eq . ( [ 4.2 ] ) . thus , @xmath213 and @xmath214 are proportional to @xmath24 , and the three terms @xmath180 , @xmath215 and @xmath216 are non - analytic in the coupling constant . for small @xmath24 , their leading dependence is @xmath217 @xmath218 . we show in the next subsection that these contributions combine to a radiation cross section with a leading dependence @xmath219 @xmath218 . the factor @xmath220 responsible for the remaining non - analyticity of the radiation cross section has an intuitive physical interpretation as molere @xmath143-broadening . this broadening is proportional to the density of the medium @xmath189 , the path length @xmath13 inside the medium and the probability @xmath213 that a scattering center in the medium interacts with the hard electron . as expected from a particle undergoing random motion in the @xmath143-plane , the accumulated average @xmath221 grows proportional to @xmath13 . the corresponding contribution ( [ 4.9 ] ) is seen explicitly in @xmath180 and @xmath215 , but it can also be recovered from @xmath216 in an expansion to lowest order in @xmath24 , when @xmath222 we now turn to the remaining three contributions of the radiation cross section ( [ 4.5 ] ) , which are given in terms of integrals over the longitudinal extension of the medium : @xmath223\ , \times \nonumber \\ & & \quad \exp\left\{{-x^2\over 4\ , c_2 } \left [ b_2^{(1)}\ , { \bf p}_{1\perp}^2 -2\,{\bf p}_{1\perp}\cdot { \bf p}_{2\perp } + b_2^{(2)}\ , { \bf p}_{2\perp}^2 \right ] \right\}\ , , \label{4.11 } \\ & & i_5^{({\rm osc } ) } = { { \rm re}\over q_2 } \int\limits_0^l { \it d}z\ , e^{i\,\bar{q}\,(z - l)}\ , { \pi\ , x^2\over c_5 ^ 2}\ , \nonumber \\ & & \quad\ , \left[a_5^{(2)}\ , { \bf p}_{2\perp}^2 + a_5^{(12)}\ , { \bf p}_{1\perp}\cdot{\bf p}_{2\perp } \right]\ , \times \nonumber \\ & & \quad \exp\left\{{-x^2\over 4\ , c_5 } \left [ b_5^{(1)}\ , { \bf p}_{1\perp}^2 -2\,{\bf p}_{1\perp}\cdot { \bf p}_{2\perp } + b_5^{(2)}\ , { \bf p}_{2\perp}^2 \right ] \right\}\ , , \label{4.12}\\ & & i_4^{({\rm osc } ) } = -\ , { \rm re } \int\limits_0^l { \it d}z\ , \int\limits_z^l { \it d}z'\ , e^{-i\,\bar{q}\,(z'-z ) } { \pi\ , x^2\over c_4 ^ 3}\ , \times \nonumber \\ & & \quad \left [ a_4^{(0 ) } + a_4^{(1)}\ , { \bf p}_{1\perp}^2 + a_4^{(12)}\ , { \bf p}_{1\perp}\cdot{\bf p}_{2\perp } + a_4^{(2)}\ , { \bf p}_{2\perp}^2\right]\ , \times \nonumber \\ & & \quad \exp\left\{{-x^2\over 4\ , c_4 } \left[b_4^{(1)}\ , { \bf p}_{1\perp}^2 - 2\ , { \bf p}_{1\perp}\cdot { \bf p}_{2\perp } + b_4^{(2)}\ , { \bf p}_{2\perp}^2\right ] \right\}\ , . \label{4.13}\end{aligned}\ ] ] to make the structure of these terms transparent , we have introduced several shorthands @xmath224 , @xmath225 and @xmath226 . the factors @xmath225 and @xmath226 show simple limiting behaviour : @xmath227 this allows us to recover the molire @xmath143-broadening term ( [ 4.9 ] ) to leading order in the coupling constant in all six contributions @xmath211 of the radiation spectrum ( [ 4.5 ] ) . the explicit form of the functions @xmath226 are : @xmath228 + \sin[\omega\ , z ] \right\}\ , , \label{4.16 } \\ c_5 & = & { i\ , \mu\over 2}\ , \omega\ , \left\ { \omega\ , z\ , \cos[\omega(l - z ) ] + \sin[\omega(l - z ) ] \right\}\ , , \label{4.17 } \\ c_4 & = & { i\ , \mu\over 2}\,\omega\ , \left\ { -\omega^2\,z\,(l - z')\ , \sin[\omega(z'-z ) ] + \sin[\omega(z'-z ) ] \right . \nonumber \\ & & \left . + \omega\,(l - z'+z)\ , \cos[\omega(z'-z ) ] \right\}\ , , \label{4.18}\end{aligned}\ ] ] from which the limit ( [ 4.14 ] ) is recovered easily . the functions @xmath225 are given by @xmath229 - \omega\ , ( l - z)\ , \sin[\omega\ , z]\ , , \label{4.19 } \\ b_2^{(2 ) } & = & \cos[\omega\ , z]\ , , \label{4.20 } \\ b_5^{(1 ) } & = & \cos[\omega\ , ( l - z)]\ , , \label{4.21 } \\ b_5^{(2 ) } & = & \cos[\omega\ , ( l - z ) ] - \omega\ , z\ , \sin[\omega\ , ( l - z)]\ , , \label{4.22 } \\ b_4^{(1 ) } & = & \cos[\omega\ , ( z'-z ) ] - \omega\ , ( l - z')\ , \sin[\omega\ , ( z'-z ) ] \ , , \label{4.23}\\ b_4^{(2 ) } & = & \cos[\omega\ , ( z'-z ) ] - \omega\ , z\ , \sin[\omega\ , ( z'-z ) ] \ , . \label{4.24}\end{aligned}\ ] ] from which the limiting form ( [ 4.15 ] ) is also clear . finally , the prefactors of the integrands in ( [ 4.11])-([4.13 ] ) are : @xmath230\ , , \label{4.25 } \\ a_5^{(2 ) } & = & { \mu\over 2}\ , \omega^2 z\ , , \qquad a_5^{(12 ) } = { \mu\over 2}\ , \omega\ , \sin[\omega\ , ( l - z)]\ , , \label{4.26 } \\ a_4^{(0 ) } & = & \mu^2\ , \omega^4\ , c_4\ , { 1\over x^2}\ , z(l - z')\ , , \label{4.27 } \\ a_4^{(1 ) } & = & { \mu^2\ , \omega^4\over 4}\ , \left ( ( l - z')^2\ , c + { { l - z'}\over \omega}\ , s\right)\ , , \label{4.28 } \\ a_4^{(2 ) } & = & { \mu^2\ , \omega^4\over 4}\ , \left ( z^2\ , c + { z\over \omega}\ , s\right)\ , , \label{4.29 } \\ a_4^{(12 ) } & = & { \mu^2\ , \omega^4\over 4}\ , \left\ { z(l - z ' ) + \left(z\ , c + { s\over \omega}\right ) \right . \nonumber \\ & & \qquad \qquad \qquad \times \left . \left((l - z')\ , c + { s\over \omega}\right ) \right\}\ , , \label{4.30}\end{aligned}\ ] ] where we have used @xmath231 $ ] and @xmath232 $ ] . the expressions given here allow for a numerical calculation of the radiation spectrum , once the extension @xmath13 of the medium , its density @xmath189 , and the measure @xmath213 of the elastic cross section of a single scatterer are given . based on these expressions , we discuss in the next subsections i ) an analytically accessible limiting case of the spectrum ( [ 4.5 ] ) and ii ) a bethe - heitler limit which allows to determine the constant @xmath213 phenomenologically . we now turn to a limiting case of the radiation spectrum ( [ 2.22 ] ) in which a characteristic in - medium effect can be isolated in a simple analytical expression . this limit focusses on the kinematical region @xmath233 the first condition ( [ 4.31 ] ) is satisfied in the high energy limit when @xmath28 is small and the photon is radiated under a small angle with respect to the beam . it is characteristic for relativistic kinematics that the transverse momentum freed by multiple interactions can be significantly larger than the total transverse momentum @xmath234 transfered by the medium , i.e. , @xmath235 and @xmath236 . according to ( [ 4.32 ] ) and ( [ 4.33 ] ) , the length of the medium has to exceed one formation length significantly , while the last condition @xmath237 can be realized e.g. by choosing a target of sufficiently low density . we note that ( [ 4.33 ] ) is by far the most stringent condition . as a consequences of ( [ 4.32 ] ) and ( [ 4.33 ] ) , the terms @xmath238 , @xmath239 are rapidly oscillating as function of @xmath13 . if we assume a target with varying extension @xmath13 , e.g. due to an unpolished surface , then contributions to the radiation spectrum ( [ 4.5 ] ) proportional to these oscillating terms will be averaged out by the experiment . an expansion of the terms in ( [ 4.5 ] ) in powers of the coupling constant leads then to @xmath240 e^ { - { { \bf q}_{\perp}^2\over 2\ , n\ , c\ , l}}\ , , \label{4.36 } \\ i_3^{({\rm osc } ) } & = & 0 \label{4.37 } \\ i_4^{({\rm osc } ) } & = & \left [ { 2\,\pi\over n\ , c\ , l}\ , { { \bf p}_{*\perp}^2\over q^2 } + { 2\pi\over q^2 } + { { 4\ , \pi}\over q^2}{1\over l^2\ , q^2 } \right ] e^ { - { { \bf q}_{\perp}^2\over 2\ , n\ , c\ , l}}\ , , \label{4.38 } \\ i_5^{({\rm osc } ) } & = & \left [ { -\ , 2\,\pi\over n\ , c\ , l}{{\bf p}_{*\perp}^2\over q^2 } - { 4\pi\over q^2}{1\over l^2\ , q^2 } \right ] e^ { - { { \bf q}_{\perp}^2\over 2\ , n\ , c\ , l}}\ , , \label{4.39 } \\ i_6^{({\rm osc } ) } & = & { 1\over 2}\ , { { \bf p}_{*\perp}^2\over q^2}\ , { 2\,\pi\over n\ , c\ , l}\ , \exp\left\ { - { { \bf q}_{\perp}^2\over 2\ , n\ , c\ , l}\right\}\ , . \label{4.40 } \end{aligned}\ ] ] here , we have used ( [ 4.32 ] ) to neglect the @xmath143-dependent terms except for the leading @xmath241 molire factor ( [ 4.9 ] ) , and we have approximated @xmath242 . corrections to these terms are of order @xmath243 and @xmath244 . neglecting @xmath245-contributions , the radiation spectrum ( [ 4.5 ] ) takes the simple form @xmath246 this molire limit is the main result of the present subsection . in accordance with the bethe - heitler single scattering cross section ( [ 3.14 ] ) , we still find the characteristic @xmath177 rapidity dependence , as well as the @xmath247 fall off of the spectrum in the kinematical regime @xmath248 . in contrast to the single scattering bethe - heitler cross section , however , the spectrum now peaks at vanishing total momentum transfer @xmath249 rather than to vanish for @xmath176 . the reason is that the radiation from many small angle scatterings adds up to a finite contribution , while the sum @xmath250 of many small random momentum transfers @xmath251 peaks at zero . also the dependence of ( [ 4.41 ] ) on the coupling constant is easy to understand . the cross section is proportional to @xmath219 , since one photon is radiated off . the broadening of the @xmath143- distribution is determined by the probability that the hard electron undergoes an interaction . this probability is given by the elastic single scattering mott cross section , and hence the exponent shows a characteristic @xmath217-dependence . the radiation cross section ( [ 4.5 ] ) obtained in the dipole approximation , is clearly more complex than the limiting case ( [ 4.41 ] ) . the latter , however , illustrates most clearly that generic in medium effects are retained in the dipole approximation which can not be obtained from a calculation to fixed order in the coupling constant . in the dipole approximation , the expansion of the radiation cross section ( [ 2.22 ] ) in powers of the coupling constant does not converge . the spectra for @xmath252 scatterings derived in section [ sec3 ] , are proportional to the @xmath4-th power of the elastic scattering cross section @xmath253 which is essentially the fourier transform of @xmath254 . while the fourier transform of the analytic expression ( [ a.11 ] ) for @xmath254 is well - defined , the fourier transform of its dipole approximation @xmath255 diverges . this failure of the dipole approximation to allow for an expansion of ( [ 2.22 ] ) in powers of the opacity @xmath156 does not affect its validity for the calculation of medium effects as given e.g. in the last subsections . this can be seen e.g. from the factor @xmath256 in ( [ 2.22 ] ) . its fourier transform is well - defined and it is well approximated in the dipole approximation @xmath255 by a gaussian of appropriate width . it is only the expansion of this factor in @xmath219 whose fourier transform does not receive the main contributions from small values around @xmath257 . this is the reason why the dipole approximation of @xmath254 for small values of @xmath258 can not be combined without difficulties with an expansion in the coupling constant . for the @xmath143-integrated spectrum ( [ 2.45 ] ) , this technical difficulty does not exist . the reason is that in the integrand of ( [ 2.45 ] ) , the first order term @xmath259 comes multiplied by @xmath260 which is to lowest order in the coupling constant a gaussian in @xmath261 . this ensures that the main contribution to the integral ( [ 2.45 ] ) comes from small values of @xmath258 where the dipole approximation is valid . the corresponding bethe - heitler spectrum in which the dipole approximation is employed _ after _ integrating out the @xmath143-dependence , reads : @xmath262 in the calculation of ( [ 4.42 ] ) we have not neglected the @xmath263-dependence of the radiation spectrum ( [ 2.22 ] ) . the consequence is the apprearance of the very small term @xmath264 which regulates the @xmath265 singularity . this allows to calculate the @xmath31-integrated bethe - heitler energy loss formula in the dipole approximation @xmath266 we note that except for the small correction from the kinematical boundary @xmath267 of the @xmath31-integral , expression ( [ 4.43 ] ) is consistent with the bethe - heitler term derived by zakharov in the dipole approximation @xcite . equations ( [ 4.42 ] ) or ( [ 4.43 ] ) allow to determine the only free parameter @xmath213 from a comparison with well - tabulated bethe - heitler scattering cross section @xcite . once , @xmath213 is fixed , the radiation spectrum ( [ 4.5 ] ) in the dipole approximation provides thus a parameter free prediction of the measured in - medium energy loss . zakharov has pursued this strategy to fix @xmath213 in his @xmath143- and @xmath31- integrated spectrum ( [ 2.47 ] ) , and this has lead to a very successful description of the slac-146 data on radiative energy loss @xcite . to calculate the integrated qcd radiative energy loss of a hard coloured parton , zakharov @xcite has used a very simple substitution in his energy loss formula ( [ 2.47 ] ) . this qcd dipole prescription consists in replacing the dipole cross section @xmath268 in the calculation of the qed radiation spectrum by a combination of three dipole cross sections @xcite : @xmath269 here , we have chosen @xmath270 proportional to the elastic @xmath6-@xmath6 mott cross section , cf . subsection [ sec5a ] . the heuristic argument for the prescription ( [ 5.1 ] ) starts from the representation of the projectile quark in the light cone frame as a superposition of the bare quark and higher fock states , @xmath271 if all fock components interact with the external potentials with the same amplitude , then the coherence between these amplitudes is not disturbed , and no bremsstrahlung is generated . the radiation amplitude depends hence on the difference between the elastic scattering amplitudes of different fluctuations . the dipole cross section @xmath272 contains information about this difference since it arises from averaging in ( [ 2.20 ] ) the ( part of the ) amplitude @xmath68 for the `` ingoing '' @xmath273 with the ( part of the ) complex conjugated amplitude @xmath274 for the `` outgoing '' higher fock state @xmath275 . in the transverse plane , the separation of the @xmath275 fluctuation from the ingoing @xmath6 can be estimated via the uncertainty relation : @xmath276 and @xmath277 . the transverse distance @xmath258 is linked to the transverse momentum in ( [ 2.22 ] ) by a fourier transform and we may think of @xmath258 as the transverse size of the @xmath263-@xmath278 fluctuation . with the transverse center of mass of this fluctuation at the position of the ingoing quark @xmath273 , see figure [ fig2 ] , the transverse distance between the charged components @xmath6 and @xmath263 is then @xmath279 , and it is the dipole of this size which determines the radiation spectrum . for related arguments , see also ref . @xcite . = 7.5 cm in qcd , the emitted gluon is charged too , and aside from the @xmath6-@xmath263 dipole of size @xmath279 , there is a @xmath280-@xmath263 dipole of size @xmath258 and a @xmath280-@xmath6 dipole of size @xmath281 . the prefactors introduced for these dipoles in the prescription ( [ 5.1 ] ) stem from interpolating between the limits @xmath282 in the limit @xmath283 , the @xmath284 pair is indistinguishable from a pointlike color - octet charge and the ratio @xmath285 of the octet to triplet couplings arises . in the opposite limit @xmath286 , the gluon - quark pair at vanishing separation is indistinguishable from a pointlike quark . the question to what extent this intuitive physical picture leads to the correct qcd radiation cross section was addressed recently in ref . there it is argued that zakharov s integrated energy loss formula ( [ 2.47 ] ) coincides for qcd with the result of their calculation based on time - ordered perturbation theory . for a discussion of the terms neglected in this approach , see ref . @xcite . here , we contribute to this discussion by comparing the transverse momentum dependence resulting from the qcd dipole prescription ( [ 5.1 ] ) to the transverse momentum dependent radiation cross sections for @xmath0 and @xmath3 scattering centers , calculated @xcite in time - ordered perturbation theory . in the present subsection , we test the qcd dipole prescription ( [ 5.1 ] ) for the case of @xmath0 scattering . we start from the @xmath0 qed bethe - heitler spectrum ( [ 3.14 ] ) in the form @xmath287 to specify the absolute size of @xmath270 in ( [ 5.1 ] ) , we relate the elastic qed @xmath288-@xmath288 cross section to the corresponding qcd one by @xmath289 where @xmath226 denotes the colour factor , @xmath290 @xmath291 , @xmath292 , @xmath293 , for @xmath6-@xmath6 , @xmath6-@xmath280 , @xmath280-@xmath280 . also , to change to the qcd - case , we replace the coupling to the emitted photon in ( [ 5.5 ] ) by @xmath294 , where the casimir @xmath295 of the adjoint representation accounts for the emitted gluon . with this input , we follow the dipole prescription and substitute ( [ 5.1 ] ) in the bethe - heitler cross section ( [ 5.5 ] ) . after rescaling @xmath143 @xmath296 on both sides of the equation , we find ( note that @xmath297 ) @xmath298 this is the bertsch - gunion radiation spectrum @xcite times the elastic mott cross section for a debye - screened coloured potential . for @xmath0 , the qcd dipole prescription turns to leading order @xmath299 the bethe - heitler qed radiation spectrum _ exactly _ into the bertsch - gunion qcd radiation spectrum . this success does not depend on the particular combination of colour dipoles introduced in ( [ 5.1 ] ) . it is rather due to the scaling property of the dipole size in ( [ 5.5 ] ) : replacing in the argument of @xmath300 the transverse extension @xmath301 modifies both : ( i ) the @xmath177 rapidity distribution of the qed spectrum into the flat rapidity distribution of the qcd spectrum and ( ii ) the characteristic @xmath302-dependence of the qed spectrum into the characterisitc @xmath143-dependence of the qcd radiation spectrum . the interference terms of qcd multiple scattering bremsstrahlung radiation are more complicated than their qed analogue . especially , since the gluon is charged , it can rescatter on additional potentials , and these cascading contributions do not factorize from the gluon production amplitudes . this effect does not exist for the qed bremsstrahlung : there , the only in medium modification of the produced photons is the dielectric suppression which is clearly a final state effect which factorizes from the production amplitude . this qualitatively different structure of qcd rescattering processes makes it interesting to study the accuracy of the qcd dipole prescription for @xmath3 scatterings . to this aim , we compare the @xmath3 radiation spectrum ( [ 3.15 ] ) to the result of the full @xmath3 qcd radiation spectrum calculated in time ordered perturbation theory in ref . @xcite : @xmath303 where @xmath304 , @xmath305 , @xmath306 . equation ( [ 5.8 ] ) is obtained by squaring the amplitude in eq . ( 2.20 ) of ref . @xcite which corresponds to the seven diagrams in fig . we have used the following shorthands : @xmath307 the @xmath3 radiation spectrum ( [ 5.8 ] ) is obtained for the case of two scattering centers placed at _ fixed _ longitudinal positions @xmath308 , @xmath309 with @xmath310 . in contrast , the @xmath3 calculation presented in section [ sec3b ] considers two scattering centers at _ arbitrary _ longitudinal positions distributed according to a homogeneous density @xmath189 within a longitudinal extension @xmath13 . in fact , the interference terms in ( [ 5.8 ] ) depend in a different way on the kinematical variables than the interference terms in ( [ 3.16 ] ) . moreover , we can not compare the @xmath311 limit of both expressions , since ( [ 5.8 ] ) is obtained in time ordered perturbation theory where @xmath312 is used to neglect diagrammatic contributions . this leaves us for a direct comparison of both calculations with only the limiting case @xmath313 which corresponds to two scattering centers at arbitrary large relative distance . one checks that the qcd dipole prescription leads for the @xmath3 rescattering result ( [ 3.26 ] ) in the @xmath314 limit @xmath315 due to the symmetry of the @xmath197- and @xmath198-integrations , one can exchange @xmath316 in ( [ 5.13 ] ) . also , for a direct comparison with the @xmath3 case ( [ 5.12 ] ) , the opacity factor @xmath317 that corresponds to the probability of two scatterings , should be dropped . to discuss the differences between the result ( [ 5.13 ] ) of the qcd dipole prescription and the @xmath3 pqcd result ( [ 5.12 ] ) , we present in figure [ fig3 ] the seven feynman amplitudes contributing to the radiation spectrum ( [ 5.8 ] ) . for each of the diagrams , we denote by @xmath318 the entire contribution , by @xmath319 the part of the contribution whose phase factor depends only on @xmath320 , by @xmath321 the part whose phase factor depends on @xmath320 and @xmath322 , etc . based on calculations reported in ref . @xcite , one finds @xmath323 as can be seen from figure [ fig3 ] , the terms @xmath324 and @xmath325 are the bertsch - gunion radiation contributions off the first and second scattering center respectively . the term @xmath326 corresponds to gluon emission around @xmath320 and rescattering of the gluon at @xmath322 . since it is the gluon and not the quark which scatters off the second coloured potential , the term @xmath326 is enhanced by a factor @xmath327 in ( [ 5.12 ] ) . these three sets of diagrams leading to @xmath324 , @xmath325 and @xmath326 can also be seen to form the building blocks of an effective current in the bdmps - approach , see e.g. fig . 4 of ref . @xcite . = 7.5 cm one might have expected naively that the qcd dipole prescription maps the two bethe - heitler terms of ( [ 3.26 ] ) onto the two bertsch - gunion terms @xmath324 and @xmath325 . this is only the case for the @xmath31-integrated cross section ( i.e. , the energy loss ) if the kinematical boundary @xmath328 is ignored . then , the @xmath31-integration allows to shift @xmath329 which changes @xmath330 in ( [ 5.13 ] ) . the @xmath3 low opacity expansion allows a quantitative check of qed - inspired qcd calculations . the deviation of the limiting case ( [ 5.13 ] ) from the perturbative qcd result is clearly non - negligible : for the @xmath31-integrated case , where the @xmath331 and @xmath332-contributions have the same size , one can read off a correction factor @xmath333 with which the @xmath3 dipole cross section ( [ 5.13 ] ) has to be multiplied to regain the pqcd radiation spectrum . [ in fact , for qcd , the @xmath31-integrals are divergent in the @xmath334 limit due to the collinear singularity that must be regulated by an in - medium screening scale . ] for the @xmath31- differential spectrum or for finite @xmath13 , corrections are clearly more complicated , but the important message is that at least for simple few - rescattering cases , the resulting expressions can be compared to pqcd results and the deviations can be quantified . the fact that the qcd dipole prescription tends to underestimate the full pqcd radiation cross section by a significant amount ( at least for @xmath3 ) may indicate that it is of use as a lower bound for the possible radiative energy loss . in the present work , we have extended the kst - formalism @xcite for quantum electrodynamics in several ways : ( i ) we have supplemented it with a regularization prescription which correctly removes the weizscker - williams fields of the asymptotic states from the radiation spectrum and thus warrants the correct limiting behaviour of the formalism for a set of few - scattering scenarios . ( ii ) we have calculated the corresponding angular integrated energy loss and shown that it implies corrections to zakharov s energy loss formula ( [ 2.47 ] ) . ( iii ) we have derived a properly regularized radiation spectrum in the dipole approximation which allows for a straightforward numerical calculation , as well as for further analytical studies in limiting cases . most importantly , the kst - formalism describes the angular dependence of the radiative energy loss which was shown in section [ sec4 ] to have a very rich non - analytical structure , incorporating intuitive properties of in - medium propagation as e.g. the molire @xmath143-broadening . in the qcd case , the situation is much more complex due to the colour structure and the rescattering of the gluon . we have tested the qcd dipole prescription , used in various previous applications @xcite , against few - scattering pqcd results @xcite . the good news is that this prescription recovers the @xmath0 qcd bertsch - gunion gluon radiation spectrum . for multiple collisions , however , even the angular integrated distributions differ from the exact result by a sizeable factor which approaches 17/8 at large separation . this discrepancy is due to the increase in the effective elastic cross section because both the jet and the gluon undergo multiple scattering . clearly , the qcd problem still needs further work and the present formalism needs to be extended to properly take into account non - abelian features . the introduction of the low opacity expansion in this paper was shown to be a powerful test of any proposed extension of qed calculations to qcd . we thank boris kopeliovich and andreas schfer for several discussions , especially about the problem of regaining the bethe - heitler limit from the radiation spectrum of ref . @xcite . after informing them about the importance of the regularization prescription discussed in section [ sec2b ] , boris kopeliovich noted that alexander tarasov had come to the same conclusion on the basis of earlier calculations of one of us . we also thank peter levai and ivan vitev for many discussions about radiative energy loss calculations in the bdmps - formalism and for sharing with us the unpublished result ( [ 5.8 ] ) whose derivation will appear in @xcite . finally , we thank u. heinz for helpful comments regarding the manuscript . this work was supported by the director , office of energy research , division of nuclear physics of the office of high energy and nuclear physics of the u.s . department of energy under contract no . de - fg-02 - 92er-40764 . in this appendix , we give details of the calculation of in medium averages , used in the derivation of the radiation probability ( [ 2.22 ] ) . we start from a set of @xmath4 static external potentials @xmath335 at positions @xmath336 , @xmath337 the in medium average @xmath338 averages over the elementary scattering centers in a volume of longitudinal extension @xmath339 and transverse radius @xmath340 according to @xmath341 we are interested in the combinations @xmath342 of external potentials appearing in the squared radiation amplitude ( [ 2.18 ] ) , @xmath343 = \bigg\langle \exp\left\ { \sum_{i=1}^n\ , w_i\bigl ( { \bf r},{\bf r}';z',z\bigr ) \right\}\bigg\rangle\ , , \label{a.3 } \\ & & w_i\bigl ( { \bf r},{\bf r}';z',z\bigr ) = \nonumber \\ & & \quad i\int\limits_{z}^{z'}{\it d}\xi\ , \left [ u\bigl({\bf r}-{\bf r}_i,\xi - z_i\bigr ) - u\bigl({\bf r}'-{\bf r}_i,\xi - z_i\bigr ) \right]\ , . \label{a.4 } \end{aligned}\ ] ] with a density @xmath344 of scattering centers ( which is @xmath345 for the homogeneous case ) , the average ( [ a.3 ] ) can be rewritten in the form @xmath346 the notational shorthand @xmath347 introduced here does _ not _ grow linearly with the number of scattering centers @xmath4 . this is crucial for the @xmath348 limit in ( [ a.5 ] ) to work . so far , we have not specified the remaining functional dependence of @xmath347 in ( [ a.6 ] ) . for homogeneous targets of sufficiently large transverse size , it is clear that @xmath347 can only depend on the relative path difference @xmath349 . we now demonstrate for a specific model that @xmath347 takes the form @xmath350 = - \int\limits_{z}^{z ' } { \it d}\xi\ , \sigma\bigl(\xi,\bbox{\rho}(\xi)\bigr)\ , , \label{a.7}\ ] ] where @xmath98 @xmath351 @xmath97 is defined in terms of the so - called dipole cross section @xmath254 and the longitudinal density @xmath25 of scattering centers , see ( [ 2.21 ] ) . the model under consideration is the abelian version of the gyulassy - wang model which takes for the elementary scattering centers in ( [ a.1 ] ) yukawa potentials with debye mass @xmath170 , @xmath352 we have included in the definition of these potentials an extra power in the coupling constant to take the coupling of the potential to the passing electron into account . in the following calculation , we assume that the average range @xmath353 of the potentials @xmath335 is much smaller than the size of the medium , @xmath354 . also , we use that density fluctuations are negligible on the scale @xmath353 . expanding @xmath347 to leading non - vanishing order in the coupling constant @xmath355 , we find @xmath356 \ , , \label{a.9}\end{aligned}\ ] ] where @xmath357 the integrals in ( [ a.9 ] ) can be done analytically , leading to zakharov s result @xcite @xmath358 in our study of the @xmath31-dependence of the radiation probability , we shall encounter the fourier transform of @xmath359 $ ] with respect to the transverse momentum transfer @xmath143 from the medium to the electron , @xmath360 for non - vanishing momentum transfer , the first term of ( [ a.12 ] ) is the mott cross - section for electron scattering off the single external potential ( [ a.1 ] ) in the relativistic limit . the size of the prefactor of the @xmath361 contribution is such that the @xmath143-integral over ( [ a.12 ] ) vanishes . this warrants that the dipole cross section @xmath254 vanishes at vanishing dipole size @xmath362 . one can trace back the origin of this second term : in the expansion of ( [ a.4 ] ) to second order in the coupling constant , it arises from contributions which are second order in @xmath363 but zeroth order in @xmath364 , or vice versa . this amounts to multiplying the second order term of @xmath365 with the zeroth order term @xmath366 of the complex conjugated amplitude . the contribution @xmath365 @xmath367 to the radiation cross section vanishes however due to energy momentum conservation . calculating such a contribution from the furry approximation ( [ 2.5 ] ) , ( [ 2.6 ] ) , this energy momentum conservation appears in the form of a @xmath368-function @xmath361 , resulting in an additional contribution to the mott cross section . for a comparison with experimental data , one has to assume a transverse momentum transfer @xmath369 which is always finite but can become arbitrarily small . then the second term in ( [ a.12 ] ) drops out in all physical quantities . in this appendix , we give details of the solutions of the path integrals , used for the derivation of the radiation probability @xmath370 in ( [ 2.22 ] ) . we start from the observation that with the help of the averaging ( [ 2.20 ] ) @xmath371\ , , \label{b.1 } \\ & & \bbox{\hat\rho}(z ) = \bbox{r}(z ) - \bbox{r}'(z)\ , , \qquad \bbox{\bar\rho}(z ) = \bbox{r}(z ) + \bbox{r}'(z)\ , . \label{b.2}\end{aligned}\ ] ] here , the dot denotes a derivative with respect to the longitudinal direction and the integral @xmath372 is trivial . it results in a @xmath368-function which constrains all possible paths @xmath373 by the mid - point rule to a straight line @xmath374 the final result is @xmath375 \over { 2\ , ( z'-z ) } } \right\}\ , , \label{b.4 } \end{aligned}\ ] ] we note that this ` eikonal path ' @xmath376 does not result from an eikonal approximation . it is rather obtained from a calculation on the cross section level , once the ingoing and outgoing states @xmath57 of ( [ 2.5 ] ) , ( [ 2.6 ] ) are adopted as starting point : the furry approximation is a systematic expansion in @xmath18 and thus closely related to the eikonal approximation . = 8.5 cm the explicit expression ( [ b.4 ] ) allows to simplify the radiation probability ( [ 2.18 ] ) by using @xmath377 % \nonumber \\ % & & \qquad \times & & = - \exp\left\{-i\bbox{p}_{2\perp}\cdot \bbox{\rho}_2 - \int_{z'}^{z_+ } \sigma(\xi,\bbox{\rho}_2)\ , { \it d}\xi\right\}\ , , \label{b.5 } \\ & & \int d{\bf r}_1\ , d{\bf r}'_1\ , e^{i{\bf p}_{1\perp}\cdot ( { \bf r}_1-{\bf r}'_1 ) } s(z,\bbox{r},\bbox{\rho}';z_-,\bbox{r}_1,\bbox{r}'_1|p_1 ) \nonumber \\ % & & = -\int { \it d}\bbox{\rho}_1\ , \ , \ , % \delta^{(2)}\bigl [ \bbox{\rho}_1 - ( \bbox{r}-\bbox{\rho}')\bigr ] % \nonumber \\ % & & \qquad \times & & = - \exp\left\{i\bbox{p}_{1\perp}\cdot \bbox{\rho}_1 - \int_{z_-}^{z } \sigma(\xi,\bbox{\rho}_1)\ , { \it d}\xi\right\}\ , , \label{b.6 } \end{aligned}\ ] ] where we have introduced the coordinates @xmath378 the integrals ( [ b.5 ] ) , ( [ b.6 ] ) depend only on the relative distances @xmath261 , @xmath379 . also , due to the boundary conditions , the path @xmath380 turns out to be @xmath381-independent in ( [ b.5 ] ) and ( [ b.6 ] ) . equation ( [ b.4 ] ) provides an explicit expression for the in - medium average of pairs of green s functions as long as both green s functions are for particles with the same momentum @xmath382 . to calculate the radiation probability ( [ 2.17 ] ) , we have to extend this calculation to the case of two different momenta . to this end , we use the identity @xmath383 where the norm @xmath384 is given in terms of the boundary conditions for the path integral over @xmath385 , @xmath386 this allows us to write @xmath387 \right\}\ , , \label{b.11 } \end{aligned}\ ] ] where @xmath388 and @xmath389 . introducing the new coordinates @xmath390 the exponent of the path integral ( [ b.10 ] ) reads @xmath391 and @xmath342 appears with the argument @xmath392 . in complete analogy to the calculation of ( [ b.4 ] ) , this renders the path integral over @xmath393 trivial and reduces the support of @xmath394 to a straight line . with the choice of boundary conditions ( [ b.9 ] ) , we find @xmath395 \right\ } \nonumber \\ & & \times \int { \cal d}\bbox{r}_b \exp\left\ { \frac{i\mu}{2}\int_z^{z'}\bbox{\dot r}_b^2 - \int_z^{z ' } \sigma\bigl(\xi,\bbox{\hat \rho}_f + x\bbox{r}_b\bigr ) \right\}\ , , % { \cal f}(\bbox{\hat \rho}_f + x\ , \bbox{r}_b)\ , , \label{b.15}\end{aligned}\ ] ] where @xmath396 and @xmath397 we can now simplify the radiation probability @xmath370 of ( [ 2.18 ] ) with the help of ( [ b.5 ] ) , ( [ b.6 ] ) and ( [ b.15 ] ) . to this aim , we observe first that with a partial integration , the derivative @xmath398 @xmath399 of the interaction vertex ( [ 2.17 ] ) changes into @xmath400 @xmath401 . then we change to the integration variables @xmath402 the only @xmath403-dependence of the integrand of @xmath370 is in the exponential of the second line of ( [ b.15 ] ) . rewriting this in the new integration variables ( [ b.7 ] ) and ( [ b.17 ] ) , we find @xmath404 \over { 2(z'-z ) } } \right\ } \nonumber\\ & & = \frac{- 8\pi i\ , ( z'-z)}{p\ , x } \exp\left\ { \frac{ip\ , ( 1-x)}{2(z'-z)x } \left(\bbox{\rho}_1-\bbox{\rho}_2\right)^2 \right\}\ , . \label{b.18}\end{aligned}\ ] ] the radiation probability reads now @xmath405 \frac{\partial}{\partial\bbox{\rho}_1 } \cdot % \frac{\partial}{\partial\bbox{\rho}_2}\ , % + 2m_e^2x^2 \right\ } % \nonumber\\ & & \quad \times \widehat{\gamma}_{-{\rho}_1}\ , \widehat{\gamma}_{-{\rho}_2}^*\ , \exp\left\ { \frac{ip\ , ( 1-x)}{2(z'-z)x } \left(\bbox{\rho}_1-\bbox{\rho}_2\right)^2 \right\ } \nonumber \\ & & \quad \times \int { \cal d}\bbox{r}_b\ , \exp\left\ { \frac{i\ , p\ , ( 1-x)\ , x}{2 } \int_z^{z'}\bbox{\dot r}_b^2(\xi)\ , { \it d}\xi \right\ } \nonumber \\ & & \qquad \quad \times \exp\left\ { - \int_z^{z ' } \sigma\bigl(\xi,\bbox{\hat \rho}_f + x\bbox{r}_b\bigr)\ , { \it d}\xi \right\}\ , . \label{b.19}\end{aligned}\ ] ] this is proportional to a transverse area @xmath406 which we devide out in what follows : the corresponding radiation spectrum is given per unit transverse area . the above expression can be simplified further if one observes that @xmath407 after a transformation @xmath408 , and @xmath409 , this allows to combine in ( [ b.19 ] ) the gaussian exponent and the path integral into a new path integral over @xmath410 with boundary conditions @xmath411 the resulting expression is given in ( [ 2.22 ] ) and differs from the result of ref . @xcite by the regularization @xmath412 of the integrand only . the extension of the formalism to spin- and helicity dependent quantities is obtained , by inserting for @xmath413 and @xmath414 in ( [ b.19 ] ) the spin- and helicity dependent vertex functions ( [ 2.15 ] ) and ( [ 2.16 ] ) , rather than the average ( [ 2.17 ] ) . for the case of @xmath415 scatterings , the combined momentum transfer @xmath143 from the medium is desribed by the folding of three elastic mott cross sections @xmath182 , @xmath416 with the shorthands @xmath417 the radiation spectrum ( [ 2.22 ] ) expanded to third order reads @xmath418\ , n^3_0\ , . \label{3.29}\end{aligned}\ ] ] the variables @xmath419 stand again for the longitudinal integrals over phase factors . for a homogeneous medium of density @xmath189 and length @xmath13 they read @xmath420 from this one finds again the simple limiting cases . we confirm the factorization limit @xmath421 and the bethe - heitler limit @xmath422\ , . \label{3.41}\end{aligned}\ ] ] also , we have checked in complete analogy to the case of @xmath3 scatterings that the lowest order correction to the factorization limit ( [ 3.40 ] ) is proportional to @xmath192 times @xmath423 . the @xmath143-integral of this correction diverges logarithmically . this further corroborates the conclusions drawn in subsection [ sec3b ] . landau and i.ya . pomeranchuk , dokl . sssr * 92 * ( 1953 ) 535 , 735 . berestetski , e.m . lifshits and l.p . pitaevski , _ landau course of theoretical physics vol . iv _ , oxford , pergamon press , 1979 . a.b . migdal , phys . rev . * 103 * ( 1956 ) 1811 . bell , nucl . * 8 * ( 1958 ) 613 . s. klein , hep - ph/9802442 , submitted to _ review of modern physics_. m.l . ter - mikaelian , jetp * 25 * ( 1953 ) 289 . slac-146 coll . , p. anthony et al . , phys . * 75 * ( 1995 ) 1949 . slac-146 coll . , p. anthony et al . , phys . d 56 * ( 1997 ) 1373 . slac-146 coll . , p. anthony et al . , phys . * 76 * ( 1996 ) 3550 . r. blankenbecler and s.d . drell , phys . d 53 * ( 1996 ) 6265 . r. blankenbecler , phys . rev . * d55 * 190 . baier and v.m . katkov , phys . rev . * d57 * ( 1998 ) 3146 ; v.n . baier and v.m . katkov , phys . rev . * d59 * ( 1999 ) 056003 . baier and v.m . katkov , hep - ph/9902436 . zakharov , jetp letters * 63 * ( 1996 ) 952 , * 65 * ( 1997 ) 615 . zakharov , hep - ph/9805271 . b.g . zakharov , phys . * 61 * ( 1998 ) 838 [ yad . fiz . * 61 * ( 1998 ) 924 ] , hep - ph/9807540 . r. baier , y.l . dokshitzer , a.h . mueller , s. peign and d. schiff , nucl . phys . * b478 * ( 1996 ) 577 . i.m . dremin and c.s . lam , mod . a13 * ( 1998 ) 2789 . j. knoll and d.n . voskresensky , annals phys . * 249 * ( 1996 ) 532 ; phys . * b351 * ( 1995 ) 43 . m. gyulassy and x .- wang , nucl . b420 * ( 1994 ) 583 . wang , m. gyulassy and m. plmer , phys . rev . * d51 * ( 1995 ) 3436 . r. baier , y.l . dokshitzer , a.h . mueller , s. peign and d. schiff , nucl . phys . * b483 * ( 1997 ) 291 . r. baier , y.l . dokshitzer , a.h . mueller , s. peign and d. schiff , nucl . phys . * b484 * ( 1997 ) 265 . r. baier , y.l . dokshitzer , a.h . mueller and d. schiff , phys . rev . * c58 * ( 1998 ) 1706 . r. baier , y.l . dokshitzer , a.h . mueller and d. schiff , nucl . phys . * b531 * ( 1998 ) 403 . kopeliovich , a. schfer and a.v . tarasov , phys . * c59 * ( 1999 ) 1609 . kopeliovich , j. raufeisen and a.v . tarasov , phys . * b440 * ( 1998 ) 151 . j. raufeisen , a.v . tarasov and o.o . voskrenskaya , hep - ph/9812398 , eur . j. c in press . nikolaev and b.g . zakharov , jetp * 78 * ( 1994 ) 598 . c. itzykson and j .- b . zuber , _ quantum field theory _ , new york , mcgraw - hill , 1985 . zakharov , sov . * 46 * ( 1987 ) 92 . brodsky , a. hebecker , and e. quack , phys . * d55 * ( 1997 ) 2584 . m. gyulassy , p. levai and i. vitev , hep - ph/9907461 . y .- s . tsai , rev . * 46 * ( 1974 ) 815 . gunion and g. bertsch , phys . * d25 * ( 1982 ) 746 .
while most observed protoplanetary discs display accretion rates of order @xmath6 m@xmath7 yr@xmath8 @xcite , some objects , in particular fu orionis systems ( e.g. * ? ? ? * ) , have exhibited episodes of enhanced accretion up to @xmath9 m@xmath7 yr@xmath8 @xcite over observed time - scales of 10 - 100 years . it has been hypothesised that _ all _ protostellar systems actually undergo such episodic accretion , equating to a small fraction ( the `` duty cycle '' ) of their total lifetime , possibly concentrated in their early stages @xcite . indeed this would alleviate the apparent discrepancy between the low observed accretion rates and that required to achieve the final mass of the protostars @xcite . although the mechanism underlying fu orionis outbursts has not been established to date @xcite , a presently popular class of model uses as its only ingredients the two currently favoured drivers of transport in protoplanetary disks , namely the magnetorotational instability ( mri ; @xcite ) and the gravitational instability @xcite . the `` gravo - magnetic cycle '' , as it has come to be called @xcite , relies on the fact that a disc with a `` normal '' density is typically too cold ( @xmath10 k ) and too dense for its mid - plane to be sufficiently ionised to trigger the mri , giving rise to a _ dead zone_. as the disc is gravitationally stable at such densities , accretion will only be possible in the surface layers where x - ray / uv ( or cosmic ray ) ionisation keeps the mri alive @xcite . however , a steady - state situation can not be achieved and mass accumulates in the dead zone , until gravitational instability sets in , raising the disc s mid - plane temperature ( @xmath11 ) . near @xmath12 k , thermal ionisation triggers the mri throughout the entire thickness of the disc , resulting in an accretion burst which lasts @xmath13 years ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? once the previously accumulated mass is dumped on the proto - star , the disc returns to the initial `` low state '' ( or quiescent state ) . the cycle may then resume , with a duty cycle @xmath14 % at low infall rates ( @xmath15 m@xmath7 yr@xmath8 ) up to @xmath16 at high infall rates ( @xmath17 m@xmath7 yr@xmath8 ) . the significant changes in the disc s temperature and luminosity during such an accretion burst can chemically process a large fraction of the disc material . several previous studies have shown the impact of accretion bursts on the chemical evolution of the envelope and disc . @xcite studied the impact of accretion busts on the co chemistry in the envelopes , in particular the large increase in luminosity , and found that repeated bursts could lead to repeated freeze - out cycles . these evolutionary co cycles lead to possible observational evidence of accretion outbursts in young star forming cores @xcite . additionally , @xcite used ` single - point ' models to study the impact of accretion bursts on the local evolution of various chemical species in the envelope , indicating that the outcome depends delicately on the balance between burst and chemical time - scales . however , what has not been considered in great detail is that this processed material can then be transported to smaller radii in the disc by advection and to larger ones by turbulent diffusion . these effects , which may persist for significantly longer than the duration of the bursts , may lend themselves to observable diagnostics ; whether for extrasolar systems , e.g. with the increased resolution and sensitivity offered by _ alma _ and _ herschel _ , or our own solar system , taking advantage of the rich body of cosmochemical evidence . to this end , in this series we set out to investigate the chemical consequences of accretion bursts . an astrochemically abundant species of choice in this regard is water @xcite , and in particular its d / h ratio @xcite . observations of the interstellar medium and pre - stellar cores suggest that water starts deuterium - rich ( d / h@xmath18 ; @xcite ) . yet bulk chondrites and comets ( with a further _ in situ _ measurement to come for comet churyumov - gerasimenko @xcite ) in our solar system display lower ( d / h)@xmath0 ratios ( around @xmath19 ) , although some micron - scale d - rich hotspots in clays of the semarkona chondrite do reach ( d / h)@xmath0 @xmath20 @xcite . this in turn is much higher than the protosolar value ( essentially that of molecular hydrogen ) of @xmath3 @xcite , and reaction kinetics prohibit the conversion of low ( d / h)@xmath0 ratios to higher values in the protoplanetary disc @xcite . the current interpretation is that the solar system inherited its water from the parent molecular cloud @xcite , and that this water _ partially _ exchanged its deuterium with the nebular gas @xcite . coupling this deuterium exchange with mixing throughout the disc produced the range of intermediate ( d / h)@xmath0 values we now observe @xcite . the bottom line is that the ( d / h ) ratio of water is a sensitive tracer of infall , temperature and turbulent mixing conditions . all previous studies of the ( d / h)@xmath0 ratio assumed a constant turbulence level ( often parametrised in terms of the shakura & sunyaev @xmath21 , @xcite ) , and thus could not reproduce the gravo - magnetic cycle , and had no accretion outbursts . additionally , only @xcite considered the effects of infall , during which most of the outbursts are expected to take place @xcite , along with the subsequent disc evolution . @xcite presented steady - state analytical calculations , indicating that in such types of model , a high efficiency of turbulent diffusion - with a schmidt number @xmath22 of 0.2 or smaller - until the snow line and beyond would be required to account for the smallness of the ( d / h)@xmath0 ratio of carbonaceous chondrites and the earth relative to oort cloud comets ; but as yet the realism of such a small schmidt number is unclear @xcite . however , as isotopic exchange is most efficient for temperatures above @xmath23 k , clearly , outbursts , which may reach temperatures in excess of 1000 k , could significantly impact the ( d / h)@xmath0 profile of protoplanetary disc and possibly account for the low ( d / h)@xmath0 ratios of carbonaceous chondrites . in this work , we examine the effect of accretion outbursts arising from the gravo - magnetic limit cycle on the evolution of the ( d / h ) ratio of water using numerical models presented in section [ sec:1dmodel ] . since this process has not been studied before , to gain insight we perform numerical experiments where the disc experiences a steady infall rate for the entire simulation ( section [ sec : steady ] ) . secondly , we perform more realistic evolutionary calculations which include a disc building phase for a few hundred - thousand years followed by a stage of isolated evolution for several million years , that are meant to represent the evolution of an actual protoplanetary disc , e.g. the solar nebula ( section [ sec : building ] ) . the implications are then discussed in section [ sec : discussion ] and the conclusions summarized in section [ sec : conclusion ] . in order to investigate the effect of accretion bursts on the chemical evolution of the ( d / h)@xmath0 ratio we adopt a simple one - layer approach @xcite . our model is a modified version of that originally presented by @xcite to study the impact of dead - zones in protoplanetary discs and associated accretion bursts , where we additionally include radial transport of heat by radiation and turbulence ( see * ? ? ? * ) . here we extend the model to include the transport of water due to advection and turbulent diffusion as well as the chemical evolution of the ( d / h)@xmath0 ratio . the evolution of the surface density @xmath24 for a disc around a forming star , with a dead - zone is given by ( e.g. * ? ? ? * ) : @xmath25\right\}+\dot{\sigma}^g_{\rm in}\label{eqn : gas}\ ] ] where @xmath26 is the kinematic viscosity due to the mri or self - gravity which we detail in section [ sec : viscosity ] and @xmath27 is the rate of material falling onto the disc . @xmath28 is the surface density that is ` mri active ' and is given by : @xmath29 for all our models where we include a dead - zone ( and hence accretion bursts ) we choose @xmath30k and @xmath31 g @xmath32 . in simulations where we do not include a dead - zone ( and thus no accretion bursts ) we set @xmath33 . the evolution of the temperature at the mid - plane of the disc is given by @xcite : @xmath34\nonumber\\&&-\left(\frac{2h}{\sigma^g rc_p}\right)\frac{\partial}{\partial r}\left(rf_r\right)\label{eqn : temp}\end{aligned}\ ] ] where @xmath35 is the heat capacity at constant pressure , @xmath36 is the ratio of specific heats and @xmath37 is the scale height of the disc , with @xmath38 the keplerian angular velocity and @xmath39 the isothermal sound speed . @xmath40 is the ` potential temperature ' of the disc s mid - plane and following @xcite we define it as : @xmath41 noting that equation [ eqn : temp ] is independent of the choice of scaling for the potential temperature.the net heating rate ( @xmath42 ) is given by : @xmath43 where @xmath44 is the stefan - boltzmann constant and @xmath45 is the vertical optical depth , with @xmath46 the opacity . the two fluxes @xmath47 & @xmath48 are the radial heat fluxes due to turbulence and radiation respectively ; which are given by : @xmath49 where @xmath50 is the prandtl number ( in most calculations we take @xmath51 , the preferred value suggested by @xcite ) , and @xmath52 with @xmath53 the radiation constant , @xmath54 the speed of light and @xmath55 the mid - plane density . for our model we adopt the @xcite opacities and equation [ eqn : temp ] is integrated as described in @xcite . our heating term ( equation [ eqn : gamma ] ) does not account for stellar irradiation , which is known to dominate the heating in the outer disc ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? therefore , we do not allow the disc s temperature to drop below what would be obtained for a passively heated disc given by : @xmath56 we set @xmath57 to : @xmath58^{1/4}\ ] ] where the mass scaling is identical to that adopted by @xcite , based on a fit to a t tauri star s birth - line from @xcite . finally , the temperature is also restricted to be no less than 10 k at very large radius , to account for heating from the surrounding star - forming region . in all calculations we assume an ideal equation of state with @xmath59 and a mean molecular weight ( @xmath60 ) of 2.35 amu as the majority of the disc material we are interested in remains molecular and neutral . we treat the viscosity using an ` alpha ' model such that : @xmath61 for mri turbulence we set @xmath62 and for angular momentum transport due to self - gravity we set @xmath63 to @xmath64 where @xmath65 is the toomre parameter @xcite given by : @xmath66 the form of @xmath67 is not important and is merely designed to keep @xmath68 ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) as suggested by simulations of self - gravitating discs ( e.g. * ? ? ? * ; * ? ? ? we treat the water component of the disc as a passive scalar species , such that it is advected along with the gas and can diffuse with respect to the gas due to turbulence . thus , the evolution of the water component is given by ( e.g. * ? ? ? * ; * ? ? ? * ) : @xmath69+\dot{\sigma_i}\label{eqn : s_water}\end{aligned}\ ] ] where we evolve equation [ eqn : s_water ] for the two water species h@xmath70o and hdo . @xmath71 represents both infall onto the disc and the chemical evolution of the two species . since the purpose of this work is to understand whether accretion bursts affect the evolution of the ( d / h)@xmath0 ratio rather than make detailed predictions , we do not attempt to model a full chemical network but adopt the simple model proposed by @xcite based on @xcite , where the ( d / h)@xmath0 ratio in water evolves as : @xmath72\label{eqn : water_chem}\ ] ] where @xmath73 is the gas number density at the disc s mid - plane , @xmath74 is the rate constant for the deuteration of water given by @xcite @xmath75 \ ] ] and ( d / h)@xmath76 is the equilibrium d / h ratio at a given temperature , which we take from @xcite assuming the d / h ratio in the background molecular hydrogen dominated disc is @xmath77 . the latter values were found by @xcite to agree reasonably well with the values obtained from their more detailed chemical network . in all calculations we assume , like @xcite , that the ( d / h)@xmath0 ratio of the in - falling primordial water is @xmath78 . the infall profile adopted depends on the evolutionary model ( steady infall vs. disc building calculations ) . for the steady infall problem we use a gaussian infall profile which peaks at a radius @xmath79 and has standard deviation @xmath80 ( e.g. * ? ? ? for the full evolutionary calculations , where we account for disc building from a uniformly rotating collapsing core , we follow @xcite and adopt the @xcite infall profile : @xmath81,\ ] ] where the instantaneous centrifugal radius ( @xmath82 ) of the infalling material is given by angular momentum balance : @xmath83 where @xmath84 is the core s rotation rate ( considered to be constant in space and time ) , @xmath85 is the temperature of the core and @xmath86 is the mass of the star & disc . finally , we assume the core follows an isothermal inside - out collapse ( e.g. * ? ? ? * ) and use a constant infall rate of @xmath87 until the core mass is fully depleted , where @xmath88 is the sound speed for gas at a temperature of @xmath89 in all calculations ( as in * ? ? ? our numerical code is based on the method detailed in @xcite where advective fluxes are reconstructed using a second order method and a van - leer limiter . equations [ eqn : temp ] & [ eqn : gas ] are integrated explicitly using a scheme that is second order in space and first order in time using a staggered mesh and finite volume operators . in order to evolve the transport and chemistry of the water content we make use of operator splitting : first , equation [ eqn : s_water ] without the chemical evolution terms is integrated explicitly using a scheme that is second order in space and first order in time . secondly the chemical evolution is evolved by integrating equation [ eqn : water_chem ] using an implicit update , such that : @xmath90\end{aligned}\ ] ] such a scheme is stable for equation that are both stiff and non - stiff over a simulation domain and thus appropriate for our problem . the simulations are initialised with the gas surface density set to the numerical floor of @xmath91 g @xmath32 and the temperature set to 10 k. we use a non - uniform radial grid , using power - law spacing @xcite . at the inner boundary we use a zero - torque boundary condition for the gas surface density and a free outflow boundary for the water species . for our steady infall calculations we use 200 cells spaced between 0.1 and 66.8 au and at the outer boundary we use a zero - flux boundary condition for both the gas and water species . whereas for the disc building calculations we use 400 cells spaced between 0.047 au and 6685 au , at the outer boundary we use a zero shear boundary condition for the gas and a free - outflow boundary condition for the water . in appendix a , we benchmark our simplified scheme against the results of the more complicated chemical network used by @xcite and find very good agreement . therefore , our simple scheme is appropriate for analysing the impact of accretion bursts on the d / h ratio in water . in order to better understand the effects of accretion bursts we consider a setup with steady infall ; where the star s mass does not grow with time and is fixed to @xmath92 m@xmath7 . such calculations have been previously used to study the properties of the accretion bursts themselves @xcite as they reach a steadily repeating limit cycle . initially , only gas falls onto the disc and it reaches a steady - limit cycle after approximately 0.3 - 0.6 myr . after 1 myr we then allow both water and gas to fall onto the disc and evolve the calculation for a further 3 myr . we perform large sets of calculations , where in all sets we vary the infall rate from @xmath93 to @xmath94 m@xmath7 yr@xmath8 and the infall radius from 0.32 to 32 au . we run one set of calculations with no dead - zone and thus no accretion bursts . additionally , we run five further sets which include accretion bursts : three have a prandtl number of 10 ( the theoretically preferred value of @xcite ) and vary the schmidt number taking @xmath95 , 1.0 & 3.0 ; finally we run two sets with a schmidt number of 1.0 but prandtl numbers of 1 & 100 . in figure [ fig : dh_profile1 ] , we show the radial distribution of the d / h ratio in water 0.5 myr after the water has been introduced into the disc ( we plot the profile during quiescence as the d / h ratio shows a transient during the burst ) for a simulation with @xmath96 m@xmath7 yr@xmath8 and @xmath97 au in the six sets of calculations . profile after 0.5 myr of evolution in a disc with a steady infall rate of @xmath96 m@xmath7 yr@xmath8 occurring at @xmath97 au . the thin solid line shows the result with no dead zone ( and therefore no accretion bursts ) . the thick lines show simulations with accretion bursts with schmidt numbers of ( 0.3 - dashed , 1.0 - solid & 3.0 - dot - dashed ) . the thin dotted lines represent calculations with accretion bursts and a schmidt number of 1.0 but prandtl numbers of 1.0 ( black ) & 100.0 ( grey ) . ] these profiles clearly show that accretion bursts have a strong impact on the d / h ratio distribution . with accretion bursts the radius at which the d / h ratio reaches the fully equilibrated value of @xmath98 ( the `` equilibration radius '' ; where the equilibration time - scale is comparable to the transport time - scale c.f . @xcite ) increases from roughly 1 au to 3 au . additionally , it decreases the d / h ratio at large radius in the disc from the unprocessed value ( @xmath78 ) with no accretion bursts to @xmath99 with accretion bursts and a schmidt number of 0.3 . finally , we note that the value of the prandtl number makes little difference . this is likely due to the fact that while the prandtl number does affect the individual properties of an accretion burst ( e.g. duration ) it does not effect the duty - cycle of the burst significantly @xcite and hence the fraction of disc material processed by the accretion burst when averaged over a time - scale longer than the burst time - scale . in figures [ fig : space_time_nb ] & [ fig : space_time_b ] we show the space - time evolution of the temperature ( right - panel ) and ( d / h)@xmath0 ( middle - panel ) , along with the accretion rate evolution ( left - panel ) . figure [ fig : space_time_nb ] shows that the ( d / h)@xmath0 ratio quickly reaches a steady state distribution where the equilibration radius occurs at 1 au ( the temperature here is essentially due to viscous dissipation ) , in agreement with the analytic models of @xcite . however , figure [ fig : space_time_b ] shows that , in the presence of accretion bursts , the ( d / h)@xmath0 ratio never reaches a steady state ; but like the accretion rate reaches a steadily repeating pattern , where incoming material is accreted into the dead zone region with a ( d / h)@xmath0 @xmath100 , which is then equilibrated in the dead zone region when the disc goes through an accretion burst , which raises the temperature well above the equilibration temperature to @xmath101 k. thus , the dead zone is able to thermally process a larger fraction of the disc material , such that the equilibration radius is considerably further out than predicted by models that do not include accretion bursts . furthermore , as well as being carried by advection along with the gas , this newly equilibrated water in the dead zone region can diffuse out to large radius ( @xmath102 au ) in the disc lowering the ( time - independent ) ( d / h)@xmath0 ratio . in the case with accretion bursts , this is more efficient than without because equilibration occurs at a larger radius . in figures [ fig : panel_nb ] and [ fig : panel_sc1 ] we show the ( d / h)@xmath0 ratio measured at different locations in the disc after 0.5 myrs of evolution for the whole range of @xmath103 and @xmath79 examined . figure [ fig : panel_nb ] shows the ( d / h)@xmath0 ratio at 1.0 , 3.0 , 10 & 30 au for the simulation set with a schmidt number of 1.0 and no accretion bursts . this shows that at high accretion rates and small infall radii , most of the in - falling water is equilibrated before it can be transported to large radii resulting in a fully equilibrated disc . at low accretion rates @xmath104 m@xmath7 yr@xmath8 the disc is too cold to equilibrate the water except at very small radii ( @xmath14 au ) and at intermediate values one gets a profile similar to that shown in figure [ fig : dh_profile1 ] where the water is equilibrated inside some radius , the equilibration radius , which is an increasing function of mass accretion rate . essentially , the warmer the disc region onto which the infalling matter arrives , the more equilibrated the water of the disc as a whole will be . the simulation set with accretion bursts and schmidt number of 1.0 is shown in figure [ fig : panel_sc1 ] with the ( d / h)@xmath0 ratio measured at 1.0 , 3.0 , 10.0 & 30.0 au after 0.5 myr of evolution . this shows a similar trend with infall rate and infall radius as previously ; however , with accretion bursts it is much more pronounced . similarly , the general trend of larger equilibration radius and lower ( d / h)@xmath0 ratio at large radius in the specific case discussed above holds for the majority of the parameter space . comparing these results with those found in figure [ fig : panel_nb ] indicates that for the majority of the parameter space of interest the terrestrial region will have equilibrated water . however , in order for the ( d / h)@xmath0 ratio to be reduced from primordial values at large radius requires infall rates @xmath105 m@xmath7 yr@xmath8 , which is expected to occur at early times ( e.g. * ? ? ? however , the reduced ( d / h)@xmath0 ratio at large radius does appear to sensitively depend on the infall radius , with @xmath106 au resulting in fully equilibrated water at large radius : essentially all the incoming water is thermally processed in the dead - zone before it is transported to large radius . whereas with @xmath107 au the ( d / h)@xmath0 ratio is only reduced from primordial values by a factor of @xmath108 , where the equilibrated water at large radius comes from the turbulent diffusion of material processed in the dead - zone that has then be transported to larger radius . s@xmath8 and a schmidt number of unity . ] we now consider calculations where we build the disc from a uniformly rotating core . initially the star has a mass of 0.2 m@xmath7 and the cloud core contains 0.8 m@xmath7 . the cloud core then falls onto the disc as described in section [ sec : infall ] . any material that falls inside the inner boundary is automatically added to the star , whose mass is subject to an explicit update . since we have seen above that the prandtl number makes little difference to the results , we only perform calculations with @xmath51 . we then perform a series of simulations where we vary the cloud s rotation rate @xmath84 between @xmath109 & @xmath110 s@xmath8 , the schmidt number between 0.1 & 10 and whether or not the disc experiences accretion bursts . in figure [ fig : mass_evolve ] , we show how the individual components ( star , cloud core & disc ) in our disc building calculations evolve during the simulation for a model with @xmath111 s@xmath8 . the top panel shows a model with no accretion burst and the bottom panel shows a simulation with accretion bursts . in this case infall takes approximately 0.2 myr . in the case with no accretion bursts the star rapidly approaches 1 m@xmath7 after infall ceases and the disc remains below 0.1 m@xmath7 for the majority of the evolution and steadily declines in a power - law fashion . however , in the case with accretion bursts , the dead zone traps material in the disc . this results in a more massive disc around 0.1 - 0.2 m@xmath7 for the majority of the evolution , which accretes material onto the star in rapid bursts . this evolution is similar to that found previously ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) where discs that contain dead zones are generally more massive ( and thus hotter ) than discs that do not . in figures [ fig : db_st_nb ] & [ fig : db_st_b ] we show the evolution of the accretion rate ( left panel ) , centrifugal radius and snow - line radius taken at @xmath112 k ( thick dashed line and thick solid line respectively in middle and right panels ) , ( d / h)@xmath0 ratio ( middle panel ) and disc temperature ( right panel ) as a function of time of the calculations shown in figure [ fig : mass_evolve ] . figure [ fig : db_st_nb ] shows the space - time evolution for a calculation with @xmath111 s@xmath8 and no accretion burst . the accretion rate evolution shows that once the infall onto the disc ceases the accretion rate drops dramatically ; this coincides with a large decrease in disc temperature , which then slowly declines with time . the middle panel shows that the ( d / h)@xmath0 ratio has a plateau around 10 au and declines at both large and small radius . this evolutionary track was seen by @xcite and is also shown in our benchmark calculation discussed in the appendix . at large radius this disc is primarily composed of material that fell at small radius into the hot disc and then was transported to large radius by decretion ( e.g. * ? ? ? * ; * ? ? ? the water that fell at early times into the hot inner disc was equilibrated and then transported to large radius resulting in a ( d / h)@xmath0 ratio close to the equilibrated value at large radius ( although this radius increases with time ) . at small radius the disc is hot enough to rapidly equilibrate the water , although the equilibration radius moves inwards with time as the disc temperature falls as the disc s mass decreases . figure [ fig : db_st_b ] shows the same calculation , but this time with accretion bursts . in this case , once infall ceases the accretion rate reaches @xmath15 m@xmath7 yr@xmath8 , which then rapidly burst to @xmath9 m@xmath7 yr@xmath8 occasionally . while the evolution of the temperature and ( d / h)@xmath0 ratio is similar at very large radius to the calculation without accretion bursts with a region of equilibrated water that is transported to large radius by decretion . however , we see that the presence of a dead - zone and accretion bursts dramatically changes the ( d / h)@xmath0 ratio profile in the inner regions of the disc . we also see an individual accretion burst has a transient effect that lasts @xmath1130.5 myr . each accretion burst processes a large fraction of water in the disc , resulting in a lower peak ( d / h)@xmath0 ratio ( as also seen in the steady infall calculations ) and results in a equilibration radius that on average remains constant with time ( at @xmath114 au ) and does not move in with time like in the case without accretion bursts . the evolution of the ( d / h)@xmath0 profile during an accretion burst is shown in figure [ fig : pre_post ] . profile just before ( 0.795 myr , solid ) and after ( 0.819 myr , dashed ) an accretion burst . shown for the simulation with @xmath115 and @xmath111 s@xmath8 where the accretion burst occurs at @xmath1130.8 myr ( see figure [ fig : db_st_b ] ) . ] where the solid line shows the ( d / h)@xmath0 profile just prior to an accretion burst and the dashed line shows the ( d / h)@xmath0 profile just after an accretion outburst . we see that the outburst equilibrates water in the dead zone ( @xmath116 au ) , decreasing the overall ( d / h)@xmath0 ratio of the entire disc . this newly equilibrated disc material is then transported by advection to smaller radius and to large radius by turbulent diffusion . this diffusion of newly equilibrated water to large radius results in a lower ( d / h)@xmath0 ratio at large radii in the disc , explaining lower peak ( d / h)@xmath0 ratios compared to burst - free models . in figure [ fig : dh_omega ] , we show how the core s rotation rate effects the ( d / h)@xmath0 profile . we plot the ( d / h)@xmath0 ratio for calculations with a schmidt number of 1 , both with ( dashed ) and without ( solid ) accretion bursts at 1 , 3 , 10 & 30 au after 1 myr of evolution . these show that there is a critical rotation rate below which the entire disc is equilibrated of @xmath117 s@xmath8 for the calculations without accretion bursts , and @xmath118 s@xmath8 for the calculations with accretion bursts , corresponding to maximum centrifugal radii of @xmath5 au and @xmath119 au , respectively ( see equation ( [ rc ] ) ) . below these critical rotation rates all infalling material lands in regions of the disc where the temperature is above the equilibration temperature . in the cases with accretion bursts at all rotation rates we find that at radii of @xmath5 au the water is equilibrated , unlike the case without accretion burst where above the critical core rotation rate it returns to near primordial values . at large radii the trend seen from the steady infall calculations is also repeated here , where accretion bursts results in a lower ( d / h)@xmath0 ratio ; however , the effect is less pronounced . we note the variability seen in the 1.0 & 3.0 au is indicative of the variability in the ( d / h)@xmath0 ratio with time , in the case with accretion bursts . ratio as a function of radius for calculations with @xmath120 s@xmath8 for schmidt numbers of 0.1 ( dashed ) , 1.0 ( solid ) & 10.0 ( dot - dashed ) . the top panel shows calculations without accretion bursts the bottom panel shows calculations with accretion bursts . ] finally , in figure [ fig : profile_sc ] we show the impact of the schmidt number on the ( d / h)@xmath0 profile , where we plot our the ( d / h)@xmath0 ratio as a function of radius in our calculations with @xmath121 s@xmath8 after 1 myr of evolution with schmidt numbers of 0.1 ( dashed ) , 1.0 ( solid ) & 10.0 ( dot - dashed ) . the top panel shows the calculation without accretion bursts and the bottom panel shows the calculations with accretion bursts . at small schmidt numbers there is enhanced transport of equilibrated water to large radius resulting in a lower ( d / h)@xmath0 ratio at large radius ( as seen by @xcite ) . conversely , larger schmidt numbers result in sharper transitions from the equilibrated water to the primordial value outward as expected from the reduced efficiency of turbulent diffusion ( e.g. * ? ? ? * ; * ? ? ? our calculations have shown that the presence of dead zones in protoplanetary discs , which can give rise to accretion bursts have a significant impact on the evolution of the ( d / h)@xmath0 ratio in water . this is not surprising since the ( d / h)@xmath0 ratio equilibrates at temperatures @xmath122 k @xcite which lies in - between the quiescent temperature of the dead - zone of @xmath123 k and the dead - zone temperature during an accretion burst of @xmath101 k. hence , un - equilibrated material that resides in the dead zone prior to an accretion burst is brought closer to equilibration by the higher disc temperatures . our calculations have shown that the cloud core s rotation rate plays a strong role in setting the ( d / h)@xmath0 ratio at large radius in the disc . thus , given the several order of magnitude spread in observed core rotation rates , with typical values @xmath124 s@xmath8 @xcite then one would expect to observe a large spread in the ( d / h)@xmath0 values measured at large radii in protoplanetary discs . therefore , herschel ( e.g. * ? ? ? * ) , alma or tmt measurements of the ( d / h)@xmath0 ratio would allow inferences about the initial angular momentum of the discs . additionally , if the ( d / h)@xmath0 ratio could be spatially resolved then the properties of accretion bursts could be probed or spatial variations in the ( d / h)@xmath0 ratio then the probe could be used to place constraints on the schmidt number , similar to the method detailed by @xcite for spatial variations at various molecular snow lines . furthermore , combined observations of the crystalline silicate fraction ( e.g. * ? ? ? * ) and ( d / h)@xmath0 ratio would allow the outward transport of crystalline materials by decretion during disc building hypothesis ( e.g. * ? ? ? * ) to be tested . indeed discs with high crystallinity fractions at large radius should also possess lower ( d / h)@xmath0 ratios . thus , future observations of the ( d / h)@xmath0 ratio should be able to probe the angular momentum distribution of the disc during the its building and formation . additionally , spatially resolved observations of the ( d / h)@xmath0 ratio should be able to determine whether the majority of protoplanetary discs experienced accretion bursts during their early evolution . to first order , the d / h systematics in the solar system may be subsumed in two salient features @xcite : * most oort cloud comets show ( d / h)@xmath0 values of @xmath125 ( with some comets showing lower values ) , lower than expected in molecular clouds . * carbonaceous chondrites display d / h ratios significantly lower than cometary ( around the terrestrial value @xmath126 ) . both domains of variation are significantly higher than the protosolar value ( @xmath3 ) and significantly lower than the value observed in molecular clouds ( @xmath18 ) . in our framework , fact ( i ) indicates that infall has occurred on relatively small radii to allow significant equilibration before outward transport , which , as mentioned in the previous subsection , would also account for the presence in comets of crystalline silicates , also deemed to have originated in the inner solar system @xcite . at the same time , the relatively large d / h ratio of cometary water compared to the protosolar value suggests that equilibration has not been complete . judging from figure [ fig : dh_omega ] , that would at face value constrain the rotation rate to be within a factor of several of @xmath127 s@xmath8 ( with a factor of @xmath108 systematic difference between the outburst - free and outburst - laden cases ) . of course , the infall model is very idealized , so the situation may not be as fine - tuned as it may seem ( whether with or without outbursts , incidentally ) . it is also possible that comets , whose ( d / h)@xmath0 is somewhat variable , do not mark the _ peak _ ( d / h)@xmath0 value of the disc , and whatever that may be , this latter value may still be consistent with the lower end of those determined for prestellar cores ( e.g. * ? ? ? the importance of fact ( ii ) may be grasped from examination of figure [ fig : dh_temp ] which plots ( d / h)@xmath0 against the disc temperature for various disc models , along with the aforementioned solar system data . the dashed line shows the curve for a simulation without outbursts @xmath128 s@xmath8 and a schmidt number of 1 , for which the maximum d / h ratio roughly matches that displayed by comets at large radii ( @xmath102 au ) . it is seen that at the snow line ( taken here to correspond to a temperature of 170 k ) , ( d / h)@xmath0 is not very different from this maximum value , that is , in this case diffusion of equilibrated water from the inner solar system is too inefficient to induce a substantial deviation . however , carbonaceous chondrites , which generally show ample evidence of aqueous alteration and therefore initial presence of ice , should have formed beyond the snow line , and yet their d / h ratio is much lower than this . this @xcite pointed out as a general problem of standard disc models with @xmath129 . however , @xcite noted that assuming a schmidt number of about 0.2 ( more efficient turbulent diffusion ) , would alleviate this discrepancy , as illustrated by their analytic profile ( in a quasi - static approximation ) plotted as a red dot - dashed line in the figure ( at higher temperatures ) there . ] . while non - mri and non - ideal turbulence may produce sc values @xmath14 ( e.g. * ? ? ? * ; * ? ? ? * ) , it is unclear how realistic such low schmidt numbers are , and the required sc values could be yet lower if the d / h values displayed by carbonaceous chondrite clays were compromised by isotopic exchange with organic matter as suggested by @xcite . if we now allow for bursts , the situation changes : the solid blue line in figure [ fig : dh_temp ] stands for a simulation with bursts ( @xmath130 s@xmath8 ) also matching the cometary ( d / h)@xmath0 value at large radii , but this time matching the chondritic value at the snow line , despite a schmidt number of 1 . this is because accretion bursts have been able to process water in the inner disc , affecting the snow line , but the ( d / h)@xmath0 value at large radii has not evolved in pace with it ( see also figure [ fig : pre_post ] ) . so this may be an alternative way to account for the low d / h ratios of carbonaceous chondrites . to summarize , in order to explain the distribution of ( d / h)@xmath0 ratios measured in carbonaceous chondrites and comets we need that either : ( i ) the turbulent transport occurs with a rather low schmidt number @xmath131 , and the disc did not experience accretion bursts @xcite ; or ( ii ) turbulent transport occurs with a nominal schmidt number of @xmath5 and the disc experiences several accretion bursts during the initial stages of its evolution . ratio shown against temperature after 1 myr of evolution compared to several solar - system constrains : protosolar ( e.g. * ? ? ? * ) , carbonaceous chondrites ( e.g. * ? ? ? * ) and the majority of oort cloud comets ( e.g * ? ? ? * ; * ? ? ? the snow - line at a temperature of @xmath132 k is marked with the thin vertical dotted line . the thick solid line shows a simulation with @xmath115 , accretion bursts and @xmath111 s@xmath8 . the thick dashed line shows a simulation with @xmath115 , no accretion bursts and @xmath128 s@xmath8 . the thick dot - dashed line corresponds to the best fit solution of @xcite with @xmath133 . ] while we have shown that accretion bursts can have a significant impact on the evolution of the ( d / h)@xmath0 ratio , our results can not be considered precise predictions . rather , our simple models have allowed us to explore a large parameter space . thus , we must discuss the impact our simplifications may have had on the evolution . in particular , we adopted a rather simple model for the chemical evolution of the water and integrated one rate equation instead of a large chemical network ; however , we have shown our method agrees well with the more detailed calculations of @xcite , although reactions not considered by the latter ( e.g. ion - neutral ) could be important @xcite . perhaps the biggest simplification is the use of a one - dimensional , one - layer model , where the evolution is controlled by the properties at the mid - plane of the disc . @xcite performed a series of simulations where they considered the impact of the vertical temperature structure on the evolution of the ( d / h)@xmath0 ratio and found it to be negligible . however , this may not be true for the layered structure in the dead - zone where the active surface layers that contain little mass , have a temperature different from the more massive dead zone @xcite and there may be turbulent mixing between the active and dead layers , with the possibility that the dead zone itself contains residual levels of turbulence ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? such a 2d structure would be worth taking into account . we have also assumed throughout the paper that water was tightly coupled to the gas . however , this may not be true at large distances ( where water is solid ) and late stages of the disc s evolution , especially if water becomes incorporated in growing aggregates of mm to m size ( e.g. * ? ? ? . it might be _ a priori _ expected that the effect will be to make outward transport of equilibrated water more difficult @xcite but simulations including gas - solid drift , accretion and fragmentation would be worthwhile to address the effect on the ( d / h)@xmath0 ratio . finally , since we have shown ( in agreement with @xcite ) that the radius and time at which infalling material reaches the disc plays an important role in the late time ( d / h)@xmath0 profile , then more realistic disc formation models should be considered that account for differential rotation and turbulence in the cloud core , as well as the two- ( and possibly three- ) dimensional nature of the infall onto the disc @xcite , along with the associated angular momentum mis - match that can occur between the keplerian disc rotation and the rotation rate of the infalling material . we have used a 1d model of a layered protoplanetary disc that undergoes accretion bursts due to the gravo - magnetic limit cycle ( a commonly invoked mechanism to explain fu orionis outbursts ) to study their impact on the ( d / h)@xmath0 ratio . we have performed two types of calculations ; firstly , a set of calculations with a steady infall rate onto the disc where we vary the infall rate , infall radius , schmidt number , prandtl number and whether or not the disc contains a dead zone . secondly , we perform a series of calculation where we build the disc and star from a uniformly rotating isothermal cloud core , where we vary the rotation rate , schmidt number and whether the disc contains a dead zone . in all these calculations we find that accretion burst play a major role in equilibrating water with the surrounding hydrogen gas . thus , the ( d / h)@xmath0 ratio carries an imprint of whether the disc experienced accretion bursts or not . our main findings are as follows : 1 . during an accretion burst , water in the dead zone region is equilibrated , lowering its d / h ratio to @xmath98 , as the dead zone temperature reaches @xmath101 k during the accretion burst , higher than the equilibration temperature . discs that undergo accretion bursts have , in general , lower ( d / h)@xmath0 ratios as the burst thermally process a larger fraction of the disc material . equilibrated water from the accretion burst can be carried to large radius by diffusion resulting in lower d / h ratios at 10 - 100s of au . discs that undergo accretion bursts have larger equilibration radii ( 1 - 5 au ) compared to discs that do not experience accretion bursts . 4 . the rotation rate of the parent cloud core plays an important role in setting the large scale ( d / h)@xmath0 profile as suggested by @xcite , indicating there should be a significant variation in the d / h ratio in protoplanetary discs . agreement with the solar - system constraints requires , either : ( 1 ) the solar nebula experienced no accretion bursts and had a low schmidt number @xmath4 ; or ( 2 ) the solar nebula experienced accretion bursts and had a schmidt number close to `` nominal '' ( i.e. unity ) . given the important role of accretion bursts in the evolution of ( d / h)@xmath0 , we suggest more detailed modelling of their impact on the chemical and physical evolution of protoplanetary discs be performed . finally , coupling these models with future observations , the ( d / h)@xmath0 ratio and profile could be used as probe for the prevalence of accretion bursts in the earliest stages of protoplanetary disc evolution . we thank the referee for comments which improved the paper . we are grateful to phil armitage for useful discussions . the numerical calculations were performed on the sunnyvale cluster at cita , which is funded by the canada foundation for innovation . since we have adopted a simplified approach to the chemical evolution in view of the numerous chemical reactions that may govern evolution of the deuterium fraction in water , we need to test whether our method is appropriate for studying the effect of accretion bursts . therefore , we repeat the ` standard model ' of disc evolution calculation of @xcite who include a much more complete chemical network with 14 species and 48 reactions . the model parameters are described in table 3 of @xcite ( i.e. @xmath134 s@xmath8 , @xmath135 k & @xmath136 ) . however , we do not simplify our temperature solver and opacity structure to that used by @xcite , but rather use the more detailed method described in section [ sec:1dmodel ] . the results of our calculation are shown every 0.05myr in the left hand panel of figure [ fig : yang_bench ] and the results from the yang et al.s original calculation are shown in the right - hand panel at 0.1 , 0.2 , 0.3 , 0.5 & 1.0 myr . we see that the two evolutionary paths agree very well , particularly at late times ( which we are primarily interested in from the point of view of observables ) . this indicates that our simple method for calculating the evolution ( d / h)@xmath0 ratio produces results comparable with the more detailed chemical method used by @xcite . this is not surprising since @xcite performed a detailed exploration of the different approaches for the chemical evolution and found they gave similar results . therefore , we are confident that our method is suitable for exploring the effect of accretion bursts on the d / h ratio in water . ceccarelli c. , caselli p. , bockelee - morvan d. , mousis o. , pizzarello s. , robert f. , semenov d. , 2014 , to appear in protostars and planets vi , eds beuther , r. klessen , c. dullemond , th . henning , university of arizona press ( arxiv:1403.7143 )
unstable particles are a perfect testing ground for theories that try to unify the principle of relativity with quantum mechanics for ( at least ) two reasons . first , an unstable particle provides a very simple example of a non - trivial interacting quantum system . due to the specific initial condition characteristic to this problem ( there is just one particle in the initial state ) , a rigorous description of the decay is possible in a small hilbert space that contains only states of the particle and its decay products , so the solution can be obtained in a closed form . second , unstable particles played an important role in confirming predictions of einstein s special relativity . it was demonstrated experimentally that the decay of moving particles slows down in a good agreement with einstein s time dilation formula @xcite . for a long time it was believed that relativistic quantum mechanics and quantum field theory must reproduce exactly the einstein s time dilation formula for unstable particles @xcite . however , a detailed quantum relativistic calculation in @xcite has shown that there are small corrections to this formula in the case of particles with definite momentum . this result was later confirmed in other studies @xcite . although surprising , these findings did not challenge directly the applicability of special relativity to unstable systems . this is because relativistic transformations connect results of measurements of two inertial observers moving with constant velocities with respect to each other . then a fair comparison with the time dilation formula requires consideration of unstable states having definite values of velocity for both observers . such states do not have a definite momentum for , at least , one observer . their decay laws were considered in recent article @xcite which alleges that the decay accelerates when particle s velocity increases , instead of being slowed down as experiment shows . in the present article we resolve this controversy by performing a detailed calculation of the decay laws of particles with narrow distributions of velocities observed from moving reference frames . we found that the result of @xcite is based on an incorrect identification of the subspace of states of the unstable particle in the full hilbert space of the system . in sections [ ss : general ] - [ ss : instant ] we present a rigorous quantum relativistic framework required for the description of decays . the exact formula for the time dependence of the non - decay probability in a moving inertial frame of reference is derived in section [ sc : moving - frame ] . particular cases of this formula relevant to unstable particles with sharply defined momenta or velocities are considered in sections [ sc : mov ] and [ sc : general ] , respectively . section [ sc : numerical ] is devoted to numerical calculations of the differences between the accurate quantum mechanical result and the standard time dilation formula ( [ eq:1b ] ) . although these differences are much smaller than the resolution of modern experiments , their presence is in sharp contradiction with special relativity which does not tolerate even small deviations from the einstein s time dilation . the decay of unstable particles is described mathematically by the _ non - decay probability _ which has the following definition . suppose that we have a piece of radioactive material with n unstable nuclei prepared simultaneously at time @xmath0 and denote @xmath1 the number of nuclei that remain undecayed at time @xmath2 . then the non - decay probability indicates the rapidity of the observer that measures the non - decay probability , as described below . ] @xmath3 ( also called the _ non - decay law _ in this paper ) is defined as the fraction of nuclei that survived the decay in the limit of large @xmath4 @xmath5 so , at each time point the piece of radioactive material can be characterized by its _ composition _ @xmath6 , where @xmath7 is the share of undecayed nuclei in the sample and @xmath8 is the share of nuclei that were transformed to the _ decay products_. in this paper , in the spirit of quantum mechanics , we will treat @xmath4 unstable particles as an ensemble of identically prepared systems and consider @xmath3 as a property of a single particle ( nucleus ) , i.e. , the probability of finding this particle in the undecayed state . in this paper we are concerned with comparison of decay observations made by two observers @xmath9 and @xmath10 that move with respect to each other . without loss of generality we will assume that observer @xmath10 moves with respect to @xmath9 with velocity @xmath11 along the @xmath12-axis and that at time @xmath0 measured by both observer s clocks the origins of their coordinate systems coincide and all three pairs of coordinate axes are parallel ( @xmath13 , etc . ) . for discussion purposes we will say that @xmath9 is at rest while @xmath10 is moving . to simplify formulas we will use the _ rapidity _ parameter @xmath14 instead of velocity @xmath15 . for example , in this notation the famous relativistic factor takes the form @xmath16 . if from the point of view of @xmath9 the decay is described by the function @xmath3 , then the non - decay law of the same particle seen by the moving observer @xmath10 will be denoted by @xmath17 . denotes time measured by the clock belonging to the observer @xmath10 . ] calculation of this function and , in particular , the relationship between @xmath17 and @xmath3 is the major goal of this work . the purpose of any relativistic theory is to describe the relationships between measurements made by different _ inertial observers _ or from different _ inertial reference frames _ , i.e. , reference frames moving in space with constant velocities along straight lines and without rotation . the special _ principle of relativity _ tells us that these reference frames are exactly equivalent * * postulate i. * experiments identically arranged and performed in two different inertial reference frames @xmath9 and @xmath18 always yield the same results . for any two observers @xmath9 and @xmath18 there is an _ inertial transformation _ that connects @xmath18 to @xmath9 , i.e. , the set of rules that allows one to change from the reference frame @xmath9 to the reference frame @xmath18 . each inertial transformation is a combination of space and time translations , rotations and boosts . a composition of two inertial transformations is again a valid inertial transformation . this composition obeys the associativity law . the inverse of an inertial transformation is also an inertial transformation . therefore , they form a 10-parameter lie group . the structure of this group is specified in the second postulate * * postulate ii . * inertial transformations ( space and time translations , rotations , and boosts ) form the _ poincar group_. obviously , two different observers obtain different results by measuring observables of the same physical system . most problems in physics can be understood as translations of descriptions of the physical system between different reference frames . for example , if we have a full description of the system from the point of view of observer @xmath9 , then the time evolution is obtained by answering the question `` what is the description of the same system from the point of view of observer @xmath19 that is shifted in time with respect to @xmath9 ? '' it is important to realize that inertial transformations between observers can be divided into two groups : _ kinematical _ and _ dynamical_. kinematical transformations are those whose action on observables does not depend on the interaction . for example , if two stationary observers look at the unstable particle from different points in space , they would assign the same non - decay probability to the particle . the same is true for stationary observers having different orientations in space . from this follows , these forms are not appropriate for the description of non - decay laws of unstable particles . ] * * postulate iii . * space translations and rotations are kinematical . on the other hand , time translations produce non - trivial changes in the system . the composition of the unstable system looks different for observers @xmath9 and @xmath19 shifted in time with respect to each other . the exact action of time translations on observables of the physical system should be obtained as a result of solution of dynamical equations which depend on the interaction acting in the system . hence the following postulate is true for all types of isolated interacting systems . * * postulate iv . * time translations are dynamical , i.e. , interaction - dependent . the above postulates i - iv have overwhelming experimental support . they are assumed to be valid throughout this paper . however they are not sufficient for a full description of the unstable particle in a moving reference frame . such a description requires also knowledge of the nature of boosts . here we have a choice between two paths forward . one path is to postulate certain properties of boosts . this path was taken by einstein . it leads to special relativity and to the time dilation formula ( [ eq:1b ] ) as explained in the next section . in this paper we will argue in favor of choosing another path : keep postulates i - iv , add to them well - established postulates of quantum mechanics , and see what are the implications for the transformations of observables ( in particular , the non - decay probability ) with respect to boosts . this approach is employed starting from section [ ss : general ] throughout the paper . in addition to postulates i - iv , einstein s special relativity makes two more assumptions regarding the nature of boosts * * assumption v. * boosts are kinematical . by postulating the kinematical character of boosts , special relativity insists that the internal composition of a compound system does not depend on the velocity of the observer . for example , if @xmath9 and @xmath10 are two observers moving with respect to each other , then , according to special relativity , both observer will measure the same composition of the unstable system at time @xmath0 . these statements are often considered as self - evident in discussions of special relativity . for example , r. polishchuk writes in @xcite _ `` any event that is `` seen '' in one inertial system is `` seen '' in all others . for example if observer in one system `` sees '' an explosion on a rocket then so do all other observers . '' _ in addition to the universality and interaction - independence of boost transformations , special relativity also postulates the exact transformation laws of physical observables with respect to boosts . they are referred to as _ lorentz transformations_. the most fundamental are lorentz transformations for space - time coordinates of events . loses its localization from the point of view of the moving observer @xmath10 @xcite . so , assumption vi can be applied only in the classical limit . ] * * assumption vi . * if from the point of view of observer @xmath9 an event is localized in a space point @xmath20 at time @xmath21 , then from the point of view of observer @xmath10 the same event has space - time coordinates @xmath22 given by lorentz formulas + @xmath23 let us now demonstrate that in special relativity the postulates i - iv and assumptions v - vi are sufficient to unambiguously describe particle decay in different reference frames without invoking any information about the interaction governing the decay . suppose that from the point of observer @xmath9 the unstable system is prepared in the state with composition @xmath24 at rest in the origin @xmath25 at time @xmath0 . then observer @xmath9 may associate the space - time point @xmath26 with the event of preparation . in accordance with the dynamical character of time translations , the non - decay probability @xmath3 decreases with time . from experiment and quantum mechanical calculations ( see , e.g. , section [ sc : numerical ] ) it is known that the non - decay law has an ( almost ) exponential shape ) here . ] @xmath27 where @xmath28 is the _ lifetime _ of the unstable particle . at time @xmath29 the non - decay probability is exactly @xmath30 , so that the composition is @xmath31 . this `` one lifetime '' event has coordinates @xmath32 according to the observer @xmath9 . let us now take the point of view of the moving observer @xmath10 . due to the assumption v , both @xmath9 and @xmath10 agree that at the `` preparation '' and `` lifetime '' events the non - decay probabilities are 1 and @xmath30 , respectively . however , observer @xmath10 may not agree with @xmath9 about the space - time coordinates of these events . substituting ( [ eq : event1 ] ) and ( [ eq : event2 ] ) in ( [ eq : lorentz - time ] ) - ( [ eq : lorentz - x ] ) we see that from the point of view of @xmath10 , the `` preparation '' event has coordinates @xmath33 , and the `` lifetime '' event has coordinates @xmath34 . therefore , the time elapsed between these two events is @xmath35 times longer than in the reference frame @xmath9 . this also means that the decay law of the particle is exactly @xmath35 slower from the point of view of the moving observer @xmath10 . this is reflected in the famous `` time dilation '' formula @xmath36 this formula was confirmed in numerous experiments @xcite , most accurately for muons accelerated to relativistic speeds in a cyclotron @xcite . these experimental findings were certainly a triumph of einstein s theory . however , as we see from the above discussion , in special relativity eq . ( [ eq:1b ] ) can be derived only under two assumptions v and vi , which lack proper justification . therefore , a question remains whether eq . ( [ eq:1b ] ) is a fundamental exact result or simply an approximation that can be disproved by more accurate measurements ? our goal in the rest of this paper is to demonstrate that this classical result of special relativity is not exact . in section [ sc : numerical ] we will calculate corrections to the formula ( [ eq:1b ] ) . let us now turn to the description of an isolated unstable system from the point of view of relativistic quantum theory . in our approach we will keep postulates i - iv . however , we are not going to use assumptions v and vi . we will consider a model theory with particles @xmath37 , @xmath38 , and @xmath39 , so that particle @xmath37 is massive spinless and unstable , while its decay products @xmath38 and @xmath39 are stable and their masses satisfy the inequality @xmath40 which makes the decay @xmath41 energetically possible . in order to simplify calculations and avoid being distracted by issues that are not relevant to the problem at hand we neglect the spin of the particle @xmath37 and assume that there is only one decay mode of this particle , i.e. , into two decay products @xmath38 and @xmath39 . for our discussion , the nature of the particle @xmath37 is not that important . for example , this could be a muon or a radioactive nucleus or an atom in an excited state . when measuring the non - decay law , experimentalists simply count the number of particles and determine their types . for example , observations of a muon may result in only two outcomes . one can find either a non - decayed muon or its decay products ( an electron , a neutrino , and an antineutrino ) . thus the number of particles is a legitimate observable , and for our model system @xmath42 we can introduce hermitian operators for the number of particles @xmath43 , @xmath44 , and @xmath45 . since these observables can be measured simultaneously , we can assume that these three operators commute with each other . two combinations of their common eigenvalues are allowed in our system : @xmath46 hence the hilbert space of the unstable system should be represented as a direct sum of two orthogonal subspaces , and @xmath47 are allowed to take any values from zero to infinity . however , for most unstable particles the interaction between the decay products in the final state can be ignored , and considering the subspace ( [ eq : tensor - product ] ) of the full fock space is a reasonable approximation . ] @xmath48 where @xmath49 is the subspace of states of the unstable particle @xmath37 which corresponds to the set of eigenvalues ( [ eq : eigenval1 ] ) , and @xmath50 is the orthogonal subspace of the decay products which corresponds to the set of eigenvalues ( [ eq : eigenval2 ] ) . we can now introduce a hermitian operator @xmath51 ( also known as `` yes - no experiment '' ) that corresponds to the observable `` particle @xmath37 exists '' . the operator @xmath51 can be fully defined by its eigensubspaces and eigenvalues . when a measurement performed on the unstable system finds it in a state corresponding to the particle @xmath37 ( i.e. , the state vector is within @xmath52 ) , the value of @xmath51 is 1 . when the decay products @xmath53 are observed ( the state vector lies in @xmath54 ) , the value of @xmath51 is 0 . apparently , @xmath51 is a projection operator on the subspace @xmath52 . for each normalized state vector @xmath55 the probability of finding the unstable particle @xmath37 is given by the expectation value of the observable @xmath51 @xcite @xmath56 in relativistic quantum mechanics , the dynamics of the system is described by a unitary representation @xmath57 of the poincar group in the hilbert space @xmath58 @xcite . it is convenient to express representatives @xmath57 as exponential functions of _ generators _ @xmath59 , and @xmath60 . then a general unitary operator from the set @xmath61 can be always written as a product @xmath62 where @xmath63 is the rotation vector , @xmath64 is the velocity of the boost , @xmath65 is the vector of space translation , and @xmath21 is the amount of time translation . generators are hermitian operators and correspond to certain _ total _ observables of the system . space translations are generated by the vector of the total linear momentum @xmath66 . the generator of rotations @xmath67 is the operator of the total angular momentum . the generator of time translations @xmath68 is the hamiltonian ( total energy ) . the generator of boosts @xmath69 is called the _ boost operator_. these generators must satisfy the commutation relations of the poincar group lie algebra . in this paper we will need , in particular , the following commutators @xmath70 = -i\frac{\hbar}{c^2 } h \delta_{ij } \label{eq:5.55}\ ] ] @xmath71 = -i\hbar p_i \nonumber\ ] ] where @xmath72 . they imply the following useful relationships @xmath73 \label{eq:6.2 } \\ e^ { \frac{i}{\hbar } \mathbf{k } c \vec{\theta } } h e^ { -\frac{i}{\hbar } \mathbf{k } c \vec{\theta } } & = & h \cosh \theta + c ( \mathbf{p } \cdot \frac{\vec{\theta}}{\theta } ) \sinh \theta \label{eq:6.3}\end{aligned}\ ] ] the representation ( [ gen - elem ] ) allows us to relate results of measurements in different reference frames . let us first take the point of view of the observer @xmath9 and consider a vector @xmath74 the first argument is the rapidity parameter of the reference frame from which the state @xmath75 is observed and the second argument is the time of observation . this is consistent with the convention adopted for @xmath17 in section [ intro ] . ] that describes a state in which the unstable particle @xmath37 is found with 100% certainty . @xmath76 then the time evolution of the state vector @xmath77 in the reference frame @xmath9 is given by @xmath78 and the non - decay law is given by @xmath79 from this equation it is clear that the hamiltonian @xmath68 describing the unstable system should not commute with the projection @xmath51 @xmath80 \neq 0 \label{eq:44b}\end{aligned}\ ] ] otherwise , the subspace @xmath52 of states of the particle @xmath37 would be invariant with respect to time translations and the particle @xmath37 would be stable ( @xmath81 for all @xmath21 ) . the moving observer @xmath10 describes the initial state ( at @xmath0 ) by the vector @xmath82 the time dependence of this state is @xmath83 then the non - decay law from the point of view of @xmath10 is @xmath84 where the last equation follows from the property @xmath85 of the operator @xmath51 . before calculating the non - decay law in the moving reference frame ( [ eq:3b ] ) , let us first consider a simpler case when the interaction responsible for the decay is `` turned off '' . then the representation of the poincar group @xmath86 acting in @xmath58 is _ non - interacting_. this representation is constructed in accordance with the structure of the hilbert space ( [ eq : tensor - product ] ) as @xmath87 where @xmath88 are unitary irreducible representations of the poincar group corresponding to particles @xmath37 , @xmath38 , and @xmath39 , respectively . the generators of the representation ( [ eq : non - int ] ) are denoted by @xmath89 , @xmath90 , @xmath91 , and @xmath92 . the operator of non - interacting _ mass _ @xmath93 commutes with @xmath89 , @xmath90 , @xmath91 , and @xmath92 . according to ( [ eq : masses ] ) , the operator @xmath94 has a continuous spectrum in the interval @xmath95 and a discrete point @xmath96 embedded in this interval . from definition ( [ eq : non - int ] ) it is clear that the subspaces @xmath49 and @xmath97 are invariant with respect to @xmath86 . moreover , the particle number operator @xmath43 and the projection operator @xmath51 commute with the non - interacting generators @xmath98 = [ t,\mathbf{j}_0 ] = [ t,\mathbf{k}_0 ] = [ t , h_0]=0 \label{eq : t - comm}\end{aligned}\ ] ] this implies that the particle @xmath37 is stable with respect to time translations and boosts , as expected @xmath99 the primary reason for considering the representation @xmath86 is that it allows us to build a convenient basis in the subspace @xmath52 . let us denote @xmath100 a vector in @xmath52 that corresponds to the particle @xmath37 with zero momentum . @xmath101 then we find that the vector @xmath102 describes the particle @xmath37 with definite momentum @xmath103 . indeed , using ( [ eq:6.2 ] ) we obtain @xmath104 ) | \mathbf{0 } \rangle \\ & = & \frac{\vec{\theta}}{\theta } m_ac \sinh \theta e^{-\frac{ic}{\hbar}\mathbf{k}_0 \vec{\theta } } | \mathbf{0 } \rangle \\ & = & \frac{\vec{\theta}}{\theta } m_ac \sinh \theta | \mathbf{p } \rangle \end{aligned}\ ] ] since particle @xmath37 is spinless by our assumption , the eigenvectors @xmath105 of the total momentum operator @xmath89 form a full basis in the subspace @xmath52 , so that @xmath106 then any state @xmath107 of the particle @xmath37 can be represented by a linear combination of these basis vectors @xmath108 where @xmath109 is the wave function in the momentum representation . in order to ensure the normalization @xmath110 , the function @xmath111 must satisfy @xmath112 apparently , vectors @xmath113 themselves are not normalized . if we want to study the non - decay law of a state with definite momentum @xmath114 , we should use a state vector ( which we denote by @xmath115 to distinguish it from @xmath116 ) that has a normalized momentum - space wave function sharply localized near @xmath114 . in order to satisfy eq . ( [ eq : normalization ] ) such a wave function may be formally represented as a square root of the dirac s delta function @xmath117 the action of boosts on the basis vectors @xmath113 and @xmath118 is known from wigner s theory of irreducible unitary representations of the poincar group ( see , e.g. , @xcite ) @xmath119 where @xmath120 using eq . ( [ eq:129a ] ) and the property @xmath121 we can find boost transformations for an arbitrary state vector of the form ( [ eq : expansion ] ) @xmath122 these transformations can be viewed as transformations of the corresponding momentum - space wave function , e.g. , @xmath123 so that the boost operator can be represented as a differential operator in the momentum space @xmath124 for further calculations we will need to define the _ newton - wigner position _ operator @xcite in @xmath58 @xmath125 which has the property @xmath126 & = & i\hbar \delta_{ij } \label{eq : heis}\end{aligned}\ ] ] that can be verified by direct substitution . according to eq . ( [ eq:7.21 ] ) @xmath127 therefore operator @xmath128 acts as a translation operator in the momentum space . in particular , we can write @xmath129 where @xmath130 . on the other hand , applying the boost transformation ( [ eq : boost - tr ] ) to the momentum eigenfunction we obtain is the jacobian of the transformation @xmath131 . ] @xmath132 this suggests that momentum eigenvector ( [ eq : mom - eigen2 ] ) has another useful representation @xmath133 let us now `` turn on '' the interaction responsible for the decay and discuss the interacting representation @xmath57 of the poincar group in @xmath58 with generators @xmath66 , @xmath67 , @xmath69 , and @xmath68 . according to our postulates iii and iv , the generators of space translations and rotations are interaction - free , @xmath134 while the generator of time translations ( the hamiltonian @xmath68 ) contains an interaction - dependent term @xmath135 . @xmath136 from the commutator ( [ eq:5.55 ] ) it then follows @xcite that the generators of boosts must be interaction - dependent as well @xmath137 where @xmath138 . this means that we are working in the so - called dirac s _ instant form of dynamics_. we will further assume that the interacting representation @xmath57 belongs to the bakamjian - thomas form of dynamics @xcite , which is characterized by the property that the interacting operator of mass @xmath139 commutes with the newton - wigner position operator . ] @xmath140 = 0 \label{eq : nw - comm}\end{aligned}\ ] ] our next goal is to define the basis of common eigenvectors of commuting operators @xmath89 and @xmath141 in @xmath58 . , there are other independent operators in the mutually commuting set containing @xmath89 and @xmath141 . these are , for example , the operators of the square of the total angular momentum @xmath142 and the projection of the total angular momentum on the @xmath12-axis @xmath143 . therefore a unique characterization of any basis vector requires specification of all corresponding quantum numbers as @xmath144 . however these quantum numbers are not relevant for our discussion and we omit them . ] these eigenvectors must satisfy conditions @xmath145 they are also eigenvectors of the interacting hamiltonian @xmath146 @xmath147 where @xmath148 . that depends on the eigenvalue @xmath149 of the interacting mass operator and @xmath150 in eq . ( [ eq : omega ] ) that depends on the fixed value of mass @xmath96 of the particle @xmath37 . ] in the zero - momentum eigensubspace of the momentum operator @xmath89 we can introduce a basis @xmath151 of eigenvectors of the interacting mass @xmath141 @xmath152 then the basis @xmath153 in the entire hilbert space @xmath58 can be built by formula ( cf . ( [ eq : mom - eigen2 ] ) ) @xmath154 . these eigenvectors are normalized to delta functions @xmath155 the actions of inertial transformations on these states are well - known @xcite . in particular , for boosts along the @xmath12-axis ( cf . ( [ eq:129a ] ) ) and time translations we obtain @xmath156 where @xmath157 next we notice that due to eqs . ( [ eq : heis ] ) and ( [ eq : nw - comm ] ) vectors @xmath158 also satisfy eigenvector equations ( [ eq : mom - eigen ] ) - ( [ eq : mass - eigen ] ) , so they must be proportional to the basis vectors @xmath159 @xmath160 where @xmath161 is an unimodular factor . unlike in ( [ eq : alpha ] ) , we can not conclude that @xmath162 , because , generally , the action of @xmath163 on eigenvectors @xmath164 involves multiplication by a unimodular scalar in addition to the shift of momentum . however , if the interaction is not pathological we can assume that the factor @xmath161 is smooth , i.e. , without rapid oscillations . this property will be used in derivation of eq . ( [ eq : appr_dec ] ) . obviously , vector @xmath165 can be expressed as a linear combination of zero - momentum basis vectors @xmath166 , so we can write and @xmath39 , i.e. , discrete eigenvalues of @xmath141 below @xmath167 . then the spectrum of @xmath141 ( similar to the spectrum of @xmath94 ) is continuous in the interval @xmath95 , and integration in ( [ eq : integral ] ) should be performed from @xmath168 to infinity . ] @xmath169 where function @xmath170 describes the _ mass distribution _ of the unstable particle . for most realistic unstable systems the mass distribution @xmath171 has the breit - wigner form ( see fig . 1 ) ) could be different from the mass @xmath96 that particle @xmath37 has in the absence of decay interaction . however , we will disregard this possibility here . ] @xmath172 where parameter @xmath173 is a normalization factor required to ensure that the mass distribution ( [ eq:63 ] ) - ( [ eq:64 ] ) is normalized to unity . we now use eqs . ( [ eq : alpha ] ) and ( [ eq : integral ] ) to expand the vector @xmath113 in the basis @xmath174 @xmath175 then , from ( [ eq:28a ] ) we obtain a useful formula @xmath176 suppose that vector @xmath75 in ( [ eq : expansion ] ) describes a state of the unstable particle @xmath37 from the point of view of the observer @xmath9 . the time dependence of this state seen from the moving reference frame @xmath10 is obtained by applying eqs ( [ eq:44a ] ) , ( [ eq:29a ] ) , ( [ eq:29b ] ) , and ( [ eq : p - expansion ] ) @xmath177 the inner product of this vector with @xmath178 is found by using ( [ eq:35a ] ) @xmath179 introducing new integration variables @xmath180 and taking into account ( [ eq : invar ] ) , this equation can be rewritten as @xmath181 the non - decay probability in the reference frame @xmath10 is then found by substituting ( [ eq : projector ] ) in eq . ( [ eq:3b ] ) @xmath182 which is an exact formula valid for all values of @xmath183 and @xmath21 . in the reference frame at rest ( @xmath184 ) , formula ( [ eq:50 ] ) simplifies @xmath185 let us consider a particular case of this expression when the unstable particle has a well - defined momentum , i.e. , described by the state vector @xmath186 . if the particle @xmath37 is in the state @xmath187 with zero momentum then @xmath188 and from the above equation we obtain @xmath189 which is the standard formula for the non - decay law of a particle at rest . , this formula can be compared , for example , with eq . ( 3.8 ) in @xcite . ] in particular , if we substitute ( [ eq:63 ] ) - ( [ eq:64 ] ) for the mass distribution @xmath171 we obtain approximate exponential decay ( [ eq : expo ] ) where the lifetime is given by @xmath190 if the particle @xmath37 has a definite non - zero momentum @xmath191 , then @xmath192 and @xmath193 in a number of works @xcite it was noticed that if one interprets the state @xmath118 ( where @xmath194 ) as a state of unstable particle moving with definite speed @xmath195 , then the decay law of the moving particle ( [ eq:35 ] ) can not be connected with the non - decay law of the particle at rest ( [ eq:34 ] ) by einstein s formula ( [ eq:1b ] ) , i.e. , @xmath196 this observation prompted authors of @xcite to question the applicability of einstein s special relativity to particle decays . however , at a closer inspection it appears that this result does not contradict einstein s time dilation formula ( [ eq:1b ] ) directly . formula ( [ eq:1b ] ) refers to observations made on the same particle from two frames of reference moving with respect to each other . if from the point of view of observer @xmath9 the particle is described by the state vector @xmath197 which has zero momentum and zero velocity , then from the point of view of @xmath10 this particle is described by the state @xmath198 which is an eigenstate of the velocity operator[multiblock footnote omitted ] @xmath199 but is not an eigenstate of the momentum operator @xmath89 . therefore , strictly speaking , its non - decay law is not described by @xmath200 . in order to compare with einstein s formula ( [ eq:1b ] ) , we need to calculate the decay law in the moving frame of reference @xmath201 . this is done in section [ sc : general ] . let us now discuss the non - decay probability @xmath202 at @xmath0 in the moving frame of reference . instead of eq . ( [ eq:50 ] ) , it is more convenient to use the general definition ( [ eq:3d ] ) which expresses @xmath203 as the square of the norm of the projection of @xmath204 on the subspace @xmath52 . we are going to prove that for a normalized @xmath77 this probability can not be equal to 1 for all non - zero @xmath183 . suppose that this statement is wrong , so that ( in agreement with assumption v ) for any @xmath205 and any @xmath206 , the vector @xmath207 belongs to @xmath52 . then the subspace @xmath52 is invariant under action of boosts @xmath208 and operator @xmath209 commutes with the projection @xmath51 . then from the poincar commutator ( [ eq:5.55 ] ) and @xmath210= 0 $ ] it follows by jacobi identity that @xmath211 & = & \frac{ic^2}{\hbar}[t,[k_z,(p_0)_z ] ] \\ & = & \frac{ic^2}{\hbar } [ k_z,[t , ( p_0)_z ] ] - \frac{ic^2}{\hbar}[(p_0)_z,[t , k_z ] ] \\ & = & 0\end{aligned}\ ] ] which contradicts eq . ( [ eq:44b ] ) . this contradiction implies that the state @xmath207 does not correspond to the particle @xmath37 with 100% probability ; this state must contain contributions from the decay products even at @xmath0 @xmath212 this is the `` decay caused by boost '' . thus we found that assumption v from section [ ss : relativity ] is not correct . sometimes this assumption is formulated as @xcite _ `` flavor is the quantum number that distinguishes the different types of quarks and leptons . it is a lorentz invariant quantity . for example , an electron is seen as an electron by any observer , never as a muon . '' _ although the above statement about the electron is correct ( because the electron is a stable particle ) , this is not true about the muon which , according to ( [ eq : not - in ] ) can be seen as a single particle by the observer at rest and as a group of three decay products ( an electron , a neutrino , and an antineutrino ) by the moving observer . unfortunately exact evaluation of eq . ( [ eq:50 ] ) for @xmath213 is not possible , so we need to make approximations . let us discuss properties of the initial state @xmath77 in more detail . first , in all realistic cases this state is not an eigenstate of the total momentum operator , so the wave function is not localized at one point in the momentum space ( as assumed in ( [ eq : sqrt ] ) ) but has a spread ( or uncertainty ) of momentum @xmath214 and , correspondingly an uncertainty of position @xmath215 . on the other hand , the state @xmath216 is not an eigenstate of the mass operator @xmath141 , and @xmath77 is characterized by the uncertainty of mass @xmath217 ( see fig . 1 ) that is related to the lifetime of the particle @xmath218 by formula ( [ eq : lifetime ] ) . it is important to note that in all cases of practical interest the mentioned quantities are related by inequalities @xmath219 in particular , the latter inequality means that the uncertainty of position is mush less than the distance passed by light during the lifetime of the particle . for example , in the case of muon @xmath220s and , according to ( [ eq : gg2 ] ) , the spread of the wave function in the position space must be much less than 600 m , which is a reasonable assumption . therefore , we can safely assume that the factor @xmath171 in ( [ eq:50 ] ) has a sharp peak near the value @xmath221 . then we can move the value of the smooth function @xmath222 at @xmath223 outside the integral by @xmath149 . ) that the spread of the momentum - space wave function is much less than @xmath224 we could have moved the value of the smooth function @xmath225 outside the integral by @xmath149 . this would mean that the non - decay law in the moving frame of reference is controlled by the spread of the particle wavefunction in the momentum space rather than by its mass uncertainty which disagrees with experimental observations . ] @xmath226 now we are going to use formula ( [ eq : appr_dec ] ) to calculate the decay law for an approximate eigenstate of momentum in @xmath52 whose wave function @xmath227 is localized near zero momentum @xmath228 , though inequality ( [ eq : gg ] ) still remains valid . we denote this state vector by symbol @xmath229 $ ] . the state @xmath230 $ ] is an approximate eigenstate of the velocity operator ( [ eq : velocity ] ) for all values of @xmath183 . therefore @xmath231}(\theta , t)$ ] in eq . ( [ eq : omega-0-t ] ) can be understood as the non - decay law of a particle with definite speed @xmath232 . note that the second factor in the integrand in ( [ eq : appr_dec ] ) is a slowly varying function of @xmath191 . therefore , we can set @xmath233 in eq . ( [ eq : appr_dec ] ) and obtain @xmath234}(\theta , t ) & \approx & \vert \int \limits_{m_b + m_c}^{\infty } dm \vert \mu(m)\vert^2 e^{-\frac{it}{\hbar}\sqrt{m^2 c^4 + m_a^2 c^4\sinh^2 \theta } } \vert^2 \label{eq : omega-0-t}\end{aligned}\ ] ] if we approximately ) . ] identify @xmath235 with the momentum @xmath236 of the particle @xmath37 from the point of view of the moving observer @xmath10 then @xmath234}(\theta , t ) & \approx & e^{-\frac{i}{\hbar}\omega_{\mathbf{p } } t}|^2 \label{eq:62}\end{aligned}\ ] ] so , in this approximation the non - decay law ( [ eq:62 ] ) in the frame of reference @xmath10 moving with the speed @xmath237 takes the same form as the non - decay law ( [ eq:35 ] ) of a particle moving with momentum @xmath238 with respect to the stationary observer @xmath9 . in this section we are going to perform numerical calculations of the differences between the classical einstein s formula ( [ eq:1b ] ) @xmath234}^{class}(\theta , t ) = \omega_{|\mathbf{0}]}(0 , \frac{t}{\cosh \theta } ) \label{eq : time - dilation}\end{aligned}\ ] ] and the actual non - decay law ( [ eq : omega-0-t ] ) of a moving particle having either definite momentum or sharply defined speed . in these calculations we assumed that the mass distribution @xmath171 of the unstable system ( see eqs . ( [ eq:63 ] ) - ( [ eq:64 ] ) and fig . 1 ) is centered at the value of mass @xmath239 mev/@xmath240 , the total mass of the decay products was @xmath241 mev/@xmath240 , and the width of the mass distribution was @xmath217= 20 mev/@xmath240 . these values do not correspond to any real particle , but they are typical for strongly decaying baryons . it is convenient to measure time in units of the classical lifetime @xmath242 . denoting @xmath243 , we find that einsteinian non - decay laws ( [ eq : time - dilation ] ) for any value of @xmath183 are given by the same universal function @xmath244 . this function was calculated for values of @xmath245 in the interval from 0 to 6 with the step of 0.1 the calculations were performed by direct numerical integration of eq . ( [ eq:34 ] ) using the _ mathematica _ program shown below .... gamma = 20 mass = 1000 theta = 0.0 do[print[(1/0.9375349 ) abs[nintegrate [ gamma/(2 pi ) / ( gamma^2/4 + ( x - mass)^2 ) exp [ i t sqrt [ x^2 + mass^2 ( sinh [ theta])^2 ] cosh [ theta ] / gamma ] , { x , 900 , 1010 , 1100 , 300000 } , minrecursion - > 3 , maxrecursion - > 16 , precisiongoal - > 8 , workingprecision - > 18]]^2 ] , { t , 0.0 , 6.0 , 0.1 } ] .... as expected , function @xmath244 is very close to the exponent @xmath246 . for small values of the argument @xmath245 are too large to be shown on the scale of fig . next we used eq . ( [ eq : omega-0-t ] ) and the above _ mathematica _ program to calculate the non - decay laws @xmath247}(\theta , \chi ) $ ] in moving reference frames for three values of the parameter @xmath183 ( = ` theta ` ) , namely 0.2 , 1.4 , and 10.0 , that correspond to velocities of 0.197c , 0.885c , and 0.999999995c , respectively . these calculations qualitatively confirmed the time dilation formula ( [ eq : time - dilation ] ) to the accuracy of better than 0.3% . however , they also revealed important differences @xmath248}(\theta , \chi ) - \omega ^{class}(\chi ) $ ] which are plotted as thin lines in fig . 2 . the lifetime of the particle @xmath37 considered in our example ( @xmath249 s ) is too short to be observed experimentally . so , calculated corrections to the einstein s time dilation law have only illustrative value . however , from these data we can estimate the magnitude of corrections for particles whose time - dependent non - decay laws can be measured in a laboratory , e.g. , for muons ( @xmath250s , @xmath251 , @xmath252 ) . taking into account that the magnitude of corrections is roughly proportional to the ratio @xmath253 @xcite , we can expect that for muons the maximum correction should be about @xmath254 which is much smaller than the accuracy of modern experiments ( about @xmath255 @xcite ) . so , experimental observation of the predicted corrections requires significant improvements of existing experimental techniques . note that formulas ( [ eq : omega-0-t ] ) and ( [ eq:62 ] ) for the particle seen from the moving frame are approximate in the sense that they ignore the `` decay caused by boost '' expressed by eq . ( [ eq : not - in2 ] ) . nevertheless , our major approximation ( [ eq : gg ] ) is well justified and it can not explain the discrepancy of our results ( einsteinian time dilation of the decay plus small interaction - dependent corrections ) with the conclusion of ref . @xcite about the acceleration of the decay of moving particles . in our view this conclusion does not refer to the experimentally measured non - decay law that is defined as the probability of finding one unstable particle . instead of eq . ( [ eq:3b ] ) , the non - decay probability was defined in @xcite by formula @xmath256 whose physical meaning remains unclear . in this article we assumed that the interaction responsible for the particle decay has the bakamjian - thomas instant form of dynamics . this assumption was used to simplify calculations , but there is no good reason to believe that real interactions have this form . then , naturally , one may ask a question : `` is there another form of dynamics in which einstein s time dilation formula ( [ eq:1b ] ) is exactly true ? '' the answer is _ no_. in any instant form of dynamics ( including non - bakamjian - thomas instant forms of dynamics ) the boost operators contain interaction terms , so the results ( [ eq : not - in ] ) - ( [ eq : not - in2 ] ) are still valid , and eq . ( [ eq:1b ] ) holds only approximately . in the point form dynamics and rotations @xmath90 are kinematical , while generators of space translations @xmath66 and time translations @xmath68 contain interaction terms . ] the subspace @xmath52 of the unstable particle is invariant with respect to boosts , so there can be no boost - induced decays ( [ eq : not - in ] ) . however , due to the interaction - dependence of the total momentum operator @xmath66 , one should expect decays induced by space translations @xmath257 this prediction is not confirmed by experiments which show that the composition of an unstable particle is not affected by space translations of the observer . times instead of experimentally observed slowdown @xcite . ] therefore , the point form of dynamics is not acceptable for the description of decays . similarly , never observed translation- and/or rotation - induced decays are characteristic for all non - instant forms of dynamics , e.g. , the front form . therefore only the instant form of dynamics is appropriate for the description of particle decays . note , however , that the decay slowdown and the absence of translation - induced decays are firmly established only for unstable systems whose lifetimes are long enough to be measured experimentally , i.e. , those governed by electromagnetic and weak interactions . for strongly interacting resonances these properties are beyond experimental resolution , and the non - instant forms of dynamics can not be ruled out . in this paper we analyzed the relationships between the non - decay laws in the moving reference frame and in the reference frame at rest . we used a rigorous quantum relativistic approach that is applicable to any unstable system independent on the nature of interaction governing the decay . a complete description of dynamics in different reference frames was obtained by using relativistic postulates i - vi and rules of quantum mechanics only . we found that assumptions v and vi of special relativity are not needed . moreover , their consequences ( the universal and exact time dilation of the decay of moving particles ) are in contradiction with rigorous calculations . although the time dilation ( [ eq:1b ] ) of special relativity was qualitatively confirmed by our results , we also found small corrections to this formula that depend on the strength of interaction . in a broader sense our results indicate that clocks viewed from the moving reference frame do not go exactly @xmath258 slower . the exact amount of time dilation depends on the physical makeup of the clock and on interactions responsible for the operation of the clock . b. rossi and d. b. hall , _ variation of the rate of decay of mesotron with momentum _ , phys . rev . * 59 * ( 1941 ) , 223 ; d. s. ayres , a. m. cormack , a. j. greenberg , r. w. kenney , d. o. cladwell , v. b. elings , w. p. hesse , and r. j. morrison , _ measurements of the lifetime of positive and negative pions _ , d. * 3 * ( 1971 ) , 1051 ; c. e. roos , j. marraffino , s. reucroft , j. waters , m. s. webster , e. g. h. williams , a. manz , r. settles , and g. wolf , @xmath259 _ lifetimes and longitudinal acceleration _ , nature * 286 * , ( 1980 ) , 244 . j. bailey , k. borer , f. combley , h. drumm , f. kreinen , f. lange , e. picasso , w. von ruden , f. j. m. farley , j. h. field , w. flegel , and p. m. hattersley , _ measurements of relativistic time dilatation for positive and negative muons in a circular orbit _ , nature * 268 * ( 1977 ) , 301 ; f. j. m. farley , _ the cern ( g-2 ) measurements _ , z. phys . c * 56 * ( 1992 ) , s88 . b. bakamjian and l. h. thomas , _ relativistic particle dynamics . ii _ , phys . rev . * 92 * ( 1953 ) , 1300 ; f. coester and w. n. polyzou , _ relativistic quantum mechanics of particles with direct interactions _ , phys . rev . d * 26 * ( 1982 ) , 1348 .
homo and heteroclinic bifurcations constitute the core of our understanding of complicated recurrent behaviour in dynamical systems . the history goes back to poincar in the late @xmath0 century , with major subsequent contributions by the schools of andronov , shilnikov , smale and palis . these schools have been founded on a combination of analytical and geometrical tools which developed a quite good understanding of the qualitative behaviour of those dynamics . differential equations modelling physical experiments frequently have parameters which appear in the differential equations , and it is known that qualitative changes may occur in the solution structure of these systems as the parameters vary . the dynamical behaviour of systems can be strongly influenced by special geometric or analytic invariant structures appearing in the equations ( _ e.g. _ divergence - free , preserving or reversing symmetries ) . one may ask which dynamical behaviour we would expect to see in the presence of a given invariant structure . generically , this is a hard question to answer , but some questions could be partially answered , by considering local and global bifurcations of low codimension . a bykov cycle on a three - dimensional manifold is a heteroclinic cycle between two hyperbolic saddle - foci of different morse index , where one of the connections is transverse and the other is structurally unstable . there are two different possibilities for the geometry of the flow around the one - dimensional connection depending on the direction solutions turn around it . all literature about the michelson system @xcite considered that in the neighbourhoods of the two saddle - foci , trajectories wind in the same direction . an immediate question arises : * what happens if trajectories wind with opposite directions near each node ? the first author and duarte proved in @xcite that the set of @xmath1-divergence - free vector fields defined in a compact three - dimensional riemannian manifold without boundary , has a @xmath1-residual set such that any vector field inside it is anosov or else , the flow associated to it has dense elliptic solutions in the phase space . furthermore , in @xcite , also in this context , the authors proved that if the vector field is not anosov , then it can be @xmath1-approximated by another divergence - free vector field exhibiting homoclinic tangencies . however , a conservative vector field whose flow has a persistent bykov cycle may lie outside these residual / dense subsets and the developed theory can not be applied for this degenerated class of systems . since bykov cycles can not be anosov ( due to the presence of equilibria ) , the dichotomies in @xcite suggest the following problem : * could we perform a @xmath1-perturbation within the set of vector fields whose flow has a bykov cycle in such a way that the elliptic periodic solutions are dense and/or tangencies occur ? in this paper , we partially answer the questions * ( q1)**(q2)*. we present a mechanism for the coexistence of hyperbolic and non - hyperbolic dynamics arising in a neighbourhood of a conservative bykov cycle where trajectories turn in opposite directions near the two saddle - foci - see figure [ conservativo2 ] . we show that in a @xmath1-open class of divergence - free vector fields , tangencies of the invariant manifolds of two hyperbolic saddle - foci densely occur . we prove that the global dynamics is persistently dominated by heteroclinic tangencies and by the existence of infinitely many elliptic points coexisting with hyperbolic dynamics arising from @xmath1transversality . tangencies are strongly connected with the cocoon bifurcations . we describe an extended version of these bifurcations which can be detected numerically with the _ time delay function _ introduced by lau @xcite . see also @xcite . we also find the relations between these four dynamical phenomena : chirality , tangencies , generalized cocoon bifurcations and elliptic solutions . in the conservative setting , cocoon bifurcations are connected with unfoldings of a hopf - zero singularity , which has been shown to occur in climatological models with seasonal forcing . represents the set of @xmath1-divergence free vector fields on @xmath2 . on the dashed line , we find the vector fields whose flow has a given bykov cycle and its continuation . the set @xmath3 is the open set ( with the induced topology ) in which theorem [ main1 ] is valid . , height=113 ] we recall some important results about heteroclinic tangencies for general and conservative systems . the systematic study of bifurcations of tangencies was started by gavrilov and shilnikov @xcite in the seventies , for the case of two - dimensional dissipative diffeomorphisms . the authors established that diffeomorphisms with homoclinic tangencies might separate systems with regular and chaotic dynamics , _ i.e. _ they belong to the boundary of morse - smale systems and the transition through this boundary corresponds to an @xmath4explosion . the theorem on the cascade of periodic sinks / sources near a tangency plays an important role see newhouse @xcite : in the dissipative case , for any one parameter family that unfolds generically a quadratic homoclinic tangency , there exists a sequence of intervals of values of the parameter such that the corresponding diffeomorphism has sinks / sources . one of the fundamental results in homoclinic bifurcations was established in newhouse @xcite : the existence of regions of the space of two - dimensional diffeomorphisms where tangencies are dense . these regions are called _ newhouse regions _ ; the result has been generalized for the multidimensional case by gonchenko @xcite . dynamics of systems within these regions is complex and rich and , as claimed in @xcite , it is impossible to give the complete description of bifurcations of such systems . the majority of these results was obtained for the broader case of general systems without any restriction on the preservation of some invariant structure . they can not be directly applied to volume - preserving and/or reversible systems because they require special considerations . nevertheless , the main geometric and analytical arguments can be also used for systems with additional structures , having special care . in this direction , important results on the birth of elliptic periodic points in area - preserving maps under bifurcations of tangencies were obtained in @xcite . see also the paper by lamb and stenkin @xcite about _ mixed dynamics . _ bifurcations of single - round periodic solutions were studied in @xcite for two - dimensional symplectic maps close to a map having a quadratic homoclinic tangency . the corresponding first return maps have been derived , bifurcations of their fixed points have been studied and bifurcation diagrams for one - parameter general unfoldings have been constructed . using renormalization results , the existence of one - parameter cascades of elliptic single - round periodic solutions has been proved . let @xmath2 be a three - dimensional closed and connected @xmath5 riemaniann manifold without boundary , endowed with a volume - form and let @xmath6 denote the lebesgue measure associated to it . let @xmath7 be a @xmath8 vector field , with @xmath9 . by picard s theorem on the existence and uniqueness of solutions of differential equations , @xmath10 integrates into a complete @xmath8 flow @xmath11 in a sense that @xmath12 we say that @xmath13 is _ divergence - free _ if its divergence is equal to zero _ i.e. _ if @xmath14 computed in local coordinates @xmath15 , see @xcite . equivalently , @xmath10 is divergence - free if , accordingly to the liouville formula , the volume - measure @xmath6 is invariant for the associated flow . in this case we say that the flow is _ conservative _ , _ volume - preserving _ or _ incompressible _ and is such that @xmath16 , for all @xmath17 and for all @xmath18 . we denote by @xmath19 , @xmath9 , the space of @xmath8 divergence - free vector fields on @xmath2 and we endow this set with the usual @xmath8 whitney topology . let also denote by @xmath20 the space of @xmath8 general vector fields on @xmath2 without no divergence - free constraint . we say that an equilibrium @xmath21 of @xmath10 ( _ i.e. _ @xmath22 ) , is _ hyperbolic _ if @xmath23 has neither pure complex nor zero as an eigenvalue . a hyperbolic saddle - equilibrium with a pair of conjugated non - real eigenvalues is called a _ saddle - focus_. for regular points , ( _ i.e. _ non - equilibrium points ) we define hyperbolicity with respect to the poincar map in a standard way see e.g. @xcite . given a hyperbolic equilibrium point @xmath24 with respect to a vector field @xmath10 , we denote by @xmath25 its unstable / stable global manifold . local unstable / stable manifolds are denoted by @xmath26 . when there is no ambiguity we omit the letter @xmath10 in these notations . open class of divergence - free vector fields defined on a three - dimensional compact manifold , tangencies of the invariant manifolds of two hyperbolic saddle - foci occur densely.,height=264 ] in the present paper , all equilibria are hyperbolic saddle - foci . the dimension of the unstable manifold of a saddle - focus will be called the _ morse index _ of the saddle . given two equilibria @xmath27 and @xmath28 , a _ heteroclinic trajectory _ from @xmath27 to @xmath28 , denoted @xmath29 $ ] , is a solution of @xmath30 contained in @xmath31 . there may be more than one solution from @xmath27 to @xmath28 . a heteroclinic cycle is a finite collection of equilibria together with a set of heteroclinic trajectories connecting the equilibria in a cyclic way . more details in field @xcite . in a three - dimensional smooth manifold , a _ bykov cycle _ is a heteroclinic cycle associated to two hyperbolic saddle - foci with different morse indices , in which the one - dimensional manifolds coincide and the two - dimensional invariant manifolds have a transverse intersection @xcite . the terminology _ bykov cycle _ is because it was bykov @xcite who studied dynamical properties from the existence of similar cycles in a more general context than conservative ones . this type of cycle , or rather its bifurcation point , is also called by _ t - point _ by glendinning and sparrow @xcite . in this section , we state the main theorems and we relate them with other results in the literature . our object of study is the dynamics around an autonomous divergence - free smooth vector field whose flow has a bykov cycle , for which we give an accurate description here . specifically , we study a @xmath1vector field:@xmath32 whose flow has the following properties ( see figure [ conservativo2 ] ) : 1 . [ h1 ] there are two hyperbolic saddle - foci equilibria @xmath27 and @xmath28 . assuming that , for @xmath33 , the morse index of @xmath34 is @xmath35 , the eigenvalues of @xmath23 are : 1 . @xmath36 and @xmath37 , where @xmath38 and @xmath39 , for @xmath40 ; 2 . @xmath41 and @xmath42 , where @xmath43 and @xmath44 , for @xmath45 . [ h2 ] there is a heteroclinic cycle @xmath46 consisting of @xmath27 , @xmath28 and two one - dimensional heteroclinic connections : @xmath47 $ ] and @xmath48 $ ] . the solution @xmath47 $ ] is called the _ fragile connection_. 3 . [ h3 ] at the heteroclinic connection @xmath48 $ ] , the two - dimensional manifolds @xmath49 and @xmath50 meet transversely . property * ( h3 ) * is generic . indeed , the transverse intersection of @xmath49 and @xmath50 of * ( h3 ) * persists under @xmath1perturbations , whereas the fragile connection @xmath47 $ ] does not . although bykov cycles appear naturally in systems with some symmetry as in knobloch _ et al _ @xcite , michelson @xcite , rodrigues and labouriau @xcite , unless it is explicitly said , in this paper we state general results without any kind of restriction about the symmetry . there are two possibilities for the geometry of the solutions around the cycle , depending on the direction they turn around the one - dimensional heteroclinic connection @xmath29 $ ] , in the neighbourhoods of the saddle - foci . under the context of hypotheses * ( h1)**(h3 ) * , we introduce the following definition adapted from @xcite . let @xmath51 and @xmath52 be small disjoint neighbourhoods of two saddle - foci @xmath27 and @xmath28 with disjoint boundaries @xmath53 and @xmath54 , respectively . typical trajectories starting at @xmath55 near @xmath50 go into the interior of @xmath51 in positive time , then follow the connection @xmath56 $ ] , go inside @xmath52 , and then come out at @xmath54 . let @xmath57 be a piece of the described trajectory from @xmath53 to @xmath54 . now , join its starting point to its end point by a line segment , forming a closed curve . this curve is called the _ loop of _ @xmath58 . [ chi ] the two saddle - foci @xmath27 and @xmath28 in @xmath46 have the _ same chirality _ if the loop of every trajectory in @xmath53 is linked to @xmath46 in the sense that , for every @xmath58 close to @xmath59 , the loop of @xmath58 and @xmath46 can not be disconnected by an isotopy . otherwise , we say that @xmath27 and @xmath28 have _ different chirality_. in contrast with the findings of @xcite , if the two nodes have different chirality , then the rotations may cancel out . this is a key idea for the proof of the main results , which will be formalized via hypothesis * ( h4)*. 1 . [ h4 ] the saddle - foci @xmath27 and @xmath28 have different chirality . a good overview of the previous hypothesis has been considered in @xcite . it corresponds to the _ non - concatenation property _ stated in @xcite . property * ( h4 ) * means that there are two open neighbourhoods of the equilibria such that for any trajectory going from the first to the second , the direction of its turning around the heteroclinic connection @xmath47 $ ] is different . a direct corollary is the following : the condition * ( h4 ) * is persistent under _ isotopies _ : if it holds for @xmath10 , then it holds for continuous one - parameter families containing it , as long as the fragile connection remains . in order to prove our main results , we need some additional assumptions on the set of parameters @xmath60 that determine the linear part of the vector field @xmath10 at the equilibria . for any @xmath61 , let @xmath62 be the subset of parameters given by : @xmath63 the constant @xmath64 is related with the construction of the global map around the cycle and will be clarified later . observe that condition ( [ final_formulae ] ) in the definition of @xmath62 is satisfied by an open set of parameters @xmath65 , @xmath66 , @xmath67 , @xmath68 and does not involve the quantities @xmath37 and @xmath69 . the set @xmath70 has full lebesgue measure within the set of parameters @xmath62 . hereafter , let @xmath71 denote the set of vector fields on @xmath2 satisfying * * ( h1)***(h5 ) * , which is a baire space @xcite . as depicted in figure [ scheme1 ] , the set @xmath72 may be seen as an open set in the space @xmath73 , endowed with the induced topology . let @xmath74 . the vector field @xmath75 in dumortier _ et al _ @xcite associated to the michelson system satisfies * ( h1)**(h3 ) * but condition * ( h4 ) * does not hold . @xmath76-reversibility where @xmath77 is a natural obstacle to * ( h4)*. in the next section we show a way to construct theoretically a divergence - free vector field displaying a bykov cycle and satisfying * * ( h1)****(h5)**. this is what we call a _ conservative bykov cycle_. [ bykovexists ] any closed , connected , riemannian three - dimensional manifold @xmath2 supports a bykov heteroclinic cycle satisfying properties * ( h1)**(h5 ) * associated to some divergence - free vector field @xmath78 , @xmath9 . let us consider a riemannian closed , connected and three - dimensional manifold @xmath2 . we let @xmath79 be two hyperbolic saddle - foci associated to linear divergence - free vector fields @xmath80 and @xmath81 and complying the hypotheses * * ( h1 ) * * and * ( h5 ) * , and moreover in such a way that * ( h4 ) * also holds . now , we would like to plunge in @xmath2 the open sets containing @xmath27 and @xmath28 . by the poincar - hopf theorem @xcite , we know that the sum of the indices of the equilibria of a vector field equals the euler characteristic of @xmath2 . so , we carefully choose a finite number @xmath82 of traceless equilibria , _ i.e. _ equilibria associated to divergence - free linear vector fields @xmath83 and fulfilling the poincar - hopf theorem . using moser s charts @xcite we send ( locally ) the vector fields @xmath84 and then , using the pasting lemma @xcite we extend to a divergence - free vector field in the whole manifold . let @xmath85 be the obtained vector field and we still denote by @xmath86 be two hyperbolic saddle - foci with the stable manifold of @xmath34 having dimension @xmath35 , @xmath87 . using @xcite we consider @xmath88 @xmath1-close to @xmath10 with a dense trajectory with initial condition @xmath89 , @xmath90 . since @xmath90 is dense , it passes arbitrarily close to @xmath91 and to @xmath92 . by the @xmath1-connecting lemma @xcite we consider @xmath93 @xmath1-close to @xmath94 such that @xmath95 intersects @xmath96 . since these manifolds are two - dimensional and @xmath2 is three - dimensional we can assume that the intersection is transversal . using again @xcite , we consider @xmath97 @xmath1-close to @xmath98 with some dense trajectory for the flow @xmath99 and with @xmath100 still intersecting @xmath101 transversely . since the solution is dense , it passes arbitrarily close to @xmath102 and to @xmath103 . again , by the @xmath1-connecting lemma , we consider @xmath104 @xmath1-close to @xmath105 such that @xmath106 intersects @xmath107 , non - transversely of course , and with @xmath108 still intersecting @xmath109 transversely . this satisfies the hypotheses * ( h2 ) * and * ( h3 ) * and therefore we obtain a bykov heteroclinic cycle associated to the divergence - free vector field @xmath110 . by technical reasons , in section [ sectionlocaldynamics ] , our main results ask for at least @xmath111-regularity of the vector field @xcite . nevertheless , if we start with an element in @xmath112 exhibiting a bykov cycle and since the hyperbolic equilibria are @xmath1-stable , similar arguments to those used in proposition [ bykovexists ] can ensure that @xmath1-arbitrarily close to the initial vector field , there exists an element in @xmath113 satisfying * ( h1)**(h5)*. this is due to the fact that @xmath114 is @xmath1-dense in @xmath112 by ( * ? ? ? * theorem 2.2 ) . we address the reader to the statements of ( * ? ? ? * theorem 3.1 and 3.2 ) and the toolboxes used in the @xmath1-closing and connecting lemmas which do not affect the regularity of the initial vector field . by @xcite , the existence of hyperbolic equilibria implies that the flow is not @xmath1near an anosov one and thus , @xmath1close to the flow , tangencies are expected . furthermore , by @xcite , arbitrarily @xmath1close to a conservative flow displaying a bykov cycle , we may find an incompressible flow with a dense set of elliptic periodic solutions . nonetheless , these approximated flows might not have a bykov cycle anymore , due to the fragile connection @xmath115 $ ] . we start with some definitions that make our results more precise . let @xmath116 , @xmath117 be a small tubular neighbourhood of the heteroclinic cycle @xmath46 and @xmath118 be a cross section to @xmath48 $ ] . we say that the two - dimensional invariant manifolds @xmath49 and @xmath50 have a tangency of order @xmath119 inside @xmath117 if the following conditions hold : * the manifolds @xmath49 and @xmath50 meet tangencially ; * there is a solution contained in @xmath120 which lies entirely in @xmath117 and has exactly @xmath121 intersection points with @xmath118 . a tangency of order 1 inside @xmath117 is also called by _ primary tangency_. our first result is the following : [ main1 ] there exists a @xmath1dense subset @xmath122 of the @xmath123open set @xmath3 such that for any @xmath124 and any tubular neighbourhood @xmath117 of the bykov cycle , the flow of ( [ general ] ) displays a primary tangency inside @xmath117 between the manifolds @xmath49 and @xmath50 . the perturbations of theorem [ main1 ] are performed within a degenerate subclass of incompressible flows exhibiting bykov cycles . the density of flows exhibiting tangencies is achieved without breaking the bykov cycle . the proof of this result will be addressed in section [ sectiont - pointhetero ] , where we concentrate our attention on the geometric intersection between the invariant manifolds . now , we consider a generalized version of the cocoon bifurcations described in @xcite , which can be seen as a series of global and local bifurcations that create an infinite number of heteroclinic connections . the authors of @xcite called cocoon to these series of bifurcations because it forms a cocoon like structure in a section transverse to @xmath46 . [ definition_cocoon ] let @xmath125 . we say that @xmath10 exhibits a _ generalized cocooning cascade of heteroclinic tangencies _ if , for every @xmath126 large , there exists a closed two - dimensional torus @xmath127 such that : 1 . for @xmath128 , @xmath129 the vector field @xmath10 can be @xmath1-approximated by @xmath130 whose flow displays a tangency between @xmath49 and @xmath50 , intersecting twice @xmath127 and having length greater than @xmath131 within @xmath127 . definition [ definition_cocoon ] is different to the general case given in @xcite in which the authors considered generic unfoldings of @xmath10 . in @xcite , when following a given sequence of vector fields converging to @xmath10 , a saddle - node bifurcation takes place , creating a pair of elliptical and hyperbolic periodic solutions , followed by an infinite cascade of period doubling bifurcations . since there are infinitely many heteroclinic tangencies occurring near the primary tangency , in section [ cocoon_section ] we prove that : [ main2 ] let @xmath3 be the @xmath1-open set of theorem [ main1 ] . for any @xmath132 , and any tubular neighbourhood @xmath117 of the bykov cycle , the vector field @xmath10 exhibits a generalized cocooning cascade of heteroclinic tangencies . the proof is based on @xcite and on the @xmath1-connecting lemma for volume - preserving flows @xcite . these heteroclinic bifurcations are responsible for the folding and for the fractal structure of the two - dimensional invariant manifolds in the cross section @xmath118 previously described by lay @xcite . the cocoon bifurcations might be observed numerically with the _ time delay function_. a discussion of the mechanism explaining the scape of points is given in ( * ? ? ? * section 4 ) . , emphasizing the region in which we apply the @xmath1-connecting lemma . the effect of moving the invariant two - dimensional manifolds corresponds to the motion suggested by the arrow.,height=226 ] in the context of dissipative diffeomorphisms containing homoclinic points , newhouse @xcite introduced the term _ wild attractor _ for hyperbolic sets whose invariant manifolds have a tangency . newhouse described what happens in a one - parameter unfolding , when a tangency splits , and discovered nontrivial , transitive and hyperbolic sets whose stable and unstable invariant manifolds have persistent nondegenerate tangencies . a given tangency may be removed by a small perturbation , however one can not avoid the appearance of new tangencies . this theory can be extended to heteroclinic tangencies see ( * ? ? ? * section 3.2 ) . [ elliptic ] let @xmath76 be a diffeomorphism defined on a compact section @xmath2 transverse to @xmath46 . we say that @xmath24 is an _ elliptic periodic orbit _ of period @xmath133 for the diffeomorphism @xmath76 if the following conditions hold : * @xmath134 and @xmath135 , for all @xmath136 and * the eigenvalues of the map @xmath137 are non - real and have modulus equal to 1 . from now on , the map @xmath76 of definition [ elliptic ] can be thought as the first return map to @xmath118 ( induced by the flow ) . by the generalised stokes theorem @xcite , the return map of a volume - preserving vector field to any transverse section @xmath118 preserves an ` area - measure ' , and so it may be seen as an area - preserving diffeomorphism see lemma [ area - preserving ] . newhouse s results near tangencies have been extended to conservative maps in @xcite : any area - preserving map with a heteroclinic tangency can be @xmath138approximated by an open domain in the space of area - preserving maps exhibiting persistent tangencies . generic diffeomorphisms on such a domain exhibit infinitely many elliptic islands , which take the role of attractors in the dissipative context . let @xmath76 be the first return map to a section transverse to the cycle @xmath46 . a 1-periodic solution of the flow of ( [ general ] ) is a periodic solution associated to a fixed point of @xmath76 ( also called by _ single round periodic solution _ ) . we obtain the following result : [ corollaryperioddoubling ] let @xmath132 . the map @xmath76 defines a sequence of conservative horseshoes accumulating on @xmath46 . there are persistent heteroclinic tangencies of the invariant manifolds associated to periodic solutions and infinitely many 1-periodic elliptic solutions nearby . the realization of @xmath76 as return map associated to a divergence - free vector field @xmath1-close to the original follows the same lines as ( * ? ? ? * section 3 ) . * the method . * our analysis follows a classic procedure . we first construct a model of the dynamics near a bykov cycle in terms of poincar maps between neighbourhoods of the saddle - foci , and then we analyze the algebraic bifurcation equations . near each equilibrium we derive local maps by assuming that the flow is governed by appropriate linearized vector fields in these regions . the flow near the connections is approximated by linearizing about each one , which enables us to derive global maps . appropriate composition of the local and global maps yields the desired return map . our aim is to provide a direct and intuitive geometric picture and to uncover the scaling laws of the codimension - one bifurcations that occur near the cycle . in this section , we establish local coordinates near the equilibria @xmath27 and @xmath28 and define some notation that will be used in sequel . the starting point is an application of a quite useful conservative version of sternberg s theorem for @xmath1linearization proved by banyaga , de la llave and wayne @xcite to @xmath1linearize the vector field around the saddle - foci and to introduce cylindrical coordinates around them . for each saddle , we obtain the expression of the local map that sends points in the border where the flow goes in , into points in the border where the flows goes out in positive time _ i.e. _ by forward iteration . we also establish a convention for the transition maps from one neighbourhood to the other see also ( * ? ? ? * lemma 3.4 ) where it is proved the conservative flowbox theorem . since the open and dense non - ressonance conditions of ( * ? ? ? * theorem 1.2 ) are satisfied , then around the saddle - foci the vector field @xmath10 is @xmath1conjugate to its linear part . in cylindrical coordinates @xmath139 the linearizations at @xmath140 and @xmath141 take the form , respectively : @xmath142 and @xmath143 rescaling coordinates , we may consider cylindrical neighbourhoods of @xmath140 and @xmath141 in @xmath144 of radius @xmath145 and height @xmath146 that we denote by @xmath51 and @xmath52 , respectively , as illustrated in figure [ conservativo3 ] . their borders have three components : * the _ wall _ parametrized by @xmath147 and @xmath148 with the usual cover : @xmath149 here @xmath58 represents the height of the cylinder and @xmath150 is the angular coordinate , measured from the point @xmath151 in the heteroclinic connection @xmath152 $ ] . * two disks , the top and the bottom of the cylinder . we assume the fragile heteroclinic connection @xmath153 $ ] goes from the top of one cylinder to the top of the other , and we take a polar covering of the top disk : @xmath154 where @xmath155 and @xmath156 . on these cross sections , we will define the poincar maps to study the dynamics near the cycle . consider the cylinder wall of @xmath51 , that meets the stable manifold of @xmath27 , denoted by @xmath157 , on the circle parametrized by @xmath158 . the top part @xmath159 of the wall near @xmath27 will be denoted by @xmath160 . by construction , trajectories starting at interior points of @xmath161 go into the cylinder in positive time and come out at the cylinder top , denoted @xmath162 . solutions starting at interior points of @xmath162 go outside the cylindrical neighbourhood in negative time . in these coordinates , the invariant manifold @xmath163 is the @xmath164axis , intersecting @xmath162 at the origin of the polar coordinates . reversing the time , we get dual statements for @xmath28 . after linearization , the one - dimensional invariant manifold of @xmath28 , @xmath165 , is the @xmath164axis , intersecting the top of the cylinder , @xmath166 , at the origin of its coordinates . trajectories starting at interior points of @xmath166 go into @xmath52 . trajectories starting at interior points of the cylinder wall @xmath167 go into @xmath52 in negative time . the set @xmath168 is parametrized by @xmath158 . trajectories that start at @xmath169 leave the cylinder @xmath52 at @xmath167 . and @xmath51 near @xmath28 and @xmath27 , respectively . the relative configuration of the saddle - foci is the same as in figure [ conservativo2].,height=309 ] let @xmath89 be a regular point , @xmath170 and @xmath171 be the poincar map , where @xmath172 , @xmath173 , is the surface contained in @xmath2 whose tangent space at @xmath174 is generated by @xmath175 , @xmath173 , and @xmath176 is a small neighborhood in @xmath177 of @xmath150 . we can guarantee the existence of a continuous time-@xmath178 arrival function @xmath179 from @xmath176 into @xmath180 . if @xmath181 is sufficiently small then : @xmath182 is a thin flowbox with base @xmath183 and top @xmath184 . due to the conservative flow - box theorem ( see ( * ? ? ? * lemma 3.4 ) ) , with a volume - preserving change of coordinates , we can trivialize @xmath185 into a cylinder and thus we may obtain an expression for the global map @xmath186 . next lemma is the conservative version of a result stated in bykov @xcite and its proof directly follows by applying @xcite and @xcite : [ perturb ] there are a volume - preserving change of coordinates in @xmath187 such that the linear part of the global map @xmath188 has the form : @xmath189 for some @xmath61 . consider the transition maps : @xmath190 by lemma [ perturb ] , the first order ( linear ) approximation of @xmath191 , up to a change of coordinates , it is equal to : @xmath192 the transition @xmath193 can be seen as a rotation by an angle @xmath194 . as in @xcite , we assume that @xmath195 , simplifying our computations . the flow is transverse to the previous cross sections and the boundaries of @xmath51 and of @xmath52 may be written as the closures of the disjoint unions @xmath196 and @xmath197 , respectively . the trajectory of the point @xmath198 in @xmath199 leaves @xmath51 at @xmath162 at : @xmath200 where @xmath201 . in a dual way , points @xmath202 in @xmath203 leave @xmath52 at @xmath167 at : @xmath204 where @xmath205 . the minus sign in the equation @xmath206 of suggests the condition stated on hypothesis * ( h4 ) * about different chirality . now we describe the geometry associated to the local dynamics near each equilibrium . we start with some useful definitions adapted from @xcite . a continuous real valued map defined on an interval @xmath207 $ ] is _ quasi - monotonic _ if it is monotonic for a subinterval @xmath208\subseteq [ 0,1]$ ] where @xmath209 . we shall define three objects to be used in the sequel : segments , spirals and helices : 1 . [ def_segmento ] a _ segment _ @xmath210 in @xmath160 or @xmath167 is a smooth regular parametrized curve of the type @xmath211\rightarrow in(\sigma_1 ) \qquad \text{or } \qquad \beta : ( 0,1]\rightarrow out(\sigma_2)\ ] ] that meets @xmath212 or @xmath213 transversely at the point @xmath214 only and such that , writing @xmath215 both @xmath150 and @xmath58 are quasi - monotonic and bounded functions of @xmath216 and @xmath217 is bounded . [ def_spiral ] a _ spiral _ in @xmath162 ( resp . : @xmath166 ) around a point @xmath218(resp . : @xmath219 is a curve @xmath220\rightarrow out(\sigma_1 ) \qquad \text{or } \qquad \alpha : ( 0,1]\rightarrow in(\sigma_2)\ ] ] satisfying @xmath221 and such that , if @xmath222 are its expressions in polar coordinates @xmath223 around @xmath224 , then @xmath225 and @xmath226 are quasi - monotonic , with @xmath227 , i.e. , winds infinitely many times around the point @xmath224 . 3 . consider a cylinder @xmath228 parametrized by a covering @xmath229 $ ] , with @xmath230 where @xmath231 is periodic . helix _ in the cylinder @xmath228 _ accumulating on the circle _ @xmath232 is a curve @xmath233\rightarrow c$ ] such that its coordinates @xmath234 satisfy @xmath235 and the map @xmath236 is quasi - monotonic . the next result summarizes some basic ideas on the spiralling geometry of the flow near the saddle - foci and has been proved in ( * ? ? ? * section 6 ) . since in its original form the authors assumed implicitly that the property * ( h4 ) * does not hold , we make the necessary reformulations . [ structures ] a segment @xmath210 : 1 . in @xmath160 is mapped by @xmath237 into a spiral in @xmath162 around @xmath238 ; 2 . in @xmath167 is mapped by @xmath239 into a spiral in @xmath166 around @xmath240 ; 3 . [ item3 ] in @xmath160 is mapped by @xmath237 into a spiral in @xmath162 around @xmath238 , that is mapped by the conservative transition @xmath241 into another spiral , around @xmath242 . this new spiral suffers the effect of the hyperbolic saddle distortion given by lemma [ perturb ] . 4 . if * ( h4 ) * does not hold , the spiral defined on ( [ item3 ] ) is mapped by @xmath243 into a helix in @xmath167 accumulating uniformly on the circle @xmath244 . the complete study where hypothesis * ( h4 ) * does not hold has been done in @xcite for dissipative systems . we assume that @xmath245 and @xmath246 define vertical lines across the cylinder walls of @xmath51 and @xmath52 , @xmath160 and @xmath167 , respectively . let @xmath247 , @xmath248 $ ] , be a parametrization of @xmath245 , where @xmath249 are the coordinates of the point @xmath48 \cap in(\sigma_1)$ ] . let @xmath250 be the coordinates of @xmath251 . the set @xmath252 ( @xmath253 ) defines an oriented curve in @xmath167 . under hypothesis * ( h4 ) * , it changes the direction of its turning around the cylinder @xmath167 at infinitely many points where it has a vertical tangent . in what follows , we obtain the expression for @xmath254 , @xmath253 . the proof runs along the same lines as in @xcite . [ x , y ] let @xmath255 , @xmath248 $ ] , be a segment in @xmath160 parametrized by @xmath216 . then , for @xmath253 : @xmath256)+ \phi(\varphi)\\ \\ y_2(s ) = s [ c(\varphi)]\\ \end{array } \right.\ ] ] where @xmath257 and @xmath258 with the argument @xmath259 taken in the interval @xmath260 $ ] , @xmath261 , that contains @xmath57 . from now on , denote by @xmath262 the following map @xmath263 ( height ) and ( b ) ; the coordinate @xmath264 ( angular coordinate ) , with @xmath265 , using maple . parameters : @xmath266 , @xmath267 and @xmath268.,height=264 ] if @xmath269 , then @xmath270 and @xmath271 . moreover , @xmath272 if @xmath273 as in the reversible case reported in @xcite , then @xmath274 is constant and the segment @xmath275 is mapped under @xmath262 into a vertical segment . as suggested in figure [ maple1](a ) , if @xmath265 , then the coordinate @xmath276 is not a monotonic function of @xmath216 but @xmath277 . moreover , for any @xmath278 , we have : @xmath279 and @xmath280 under hypotheses * * ( h1)***(h5 ) * , we show that the coordinate map @xmath274 is not a monotonic function of @xmath216 , since the curve @xmath281 reverses the direction of its turning around @xmath167 infinitely many times see figure [ maple1](b ) . this is the notion suggested by the following definition . we say that the vector field @xmath10 has the _ dense reversals property _ if for the vertical segment @xmath282,\ ] ] the projection into @xmath283 of the points where @xmath284 has a vertical tangent is dense in @xmath285 . the dense reversals property is the key step in the proof of theorem [ main1 ] . in order to prove it , we use the assumption * ( h5 ) * on the parameters @xmath286 that determine the linear part of the vector field @xmath10 at the equilibria . , we observe the line @xmath49 reversing the orientation of its angular coordinate . the set @xmath285 is diffeomorphic to a circle.,height=188 ] the proof will be divided into five parts , * ( i ) * * ( v)*. the first part of the proof runs along the same lines to that of @xcite , which gives an optimal condition for the existence of tangencies . let @xmath132 . * ( i ) * we need to compute the coordinate @xmath264 at the points where @xmath287 has a vertical tangent . differentiating the expression of lemma [ x , y ] with respect to @xmath216 , we get : @xmath288 and then @xmath289.\ ] ] therefore @xmath290 has solutions if and only if @xmath291 define now the @xmath292periodic map : @xmath293 * ( ii ) * now we are going to prove that the left part of ( [ eqdxw ] ) has a zero if and only if @xmath294 indeed , the expression @xmath291 is equivalent to @xmath295 using the definition of @xmath296 and @xmath297 , the last equality yields the expression : @xmath298 which is the same as @xmath299 therefore the differential equation @xmath290 has solutions if and only if @xmath300 corresponding to the condition that defines @xmath70 . if @xmath301 $ ] is a solution of ( [ eqdxw ] ) , then there are infinitely many given by @xmath302 where @xmath303 . since @xmath304 , then @xmath290 has solutions @xmath305 where @xmath306 . see figure [ reversions ] . * ( iii ) * for any @xmath307 and @xmath308 , we have @xmath309 , we observe the @xmath49 reversing the orientation of its angular coordinate . two consecutive reversion points lie in region of the graph with different concavities , diving rise to different types of tangencies ( and therefore to different types of dynamics ) . , height=151 ] using the expressions we get that @xmath310 also , if @xmath311 $ ] , then @xmath312 lies in the same interval and @xmath313 the result follows , using the expression for @xmath264 in lemma [ x , y ] . in part * ( iii ) * , we have shown that @xmath314 hence , if the genericity condition @xmath315 in holds , then the points @xmath316 are dense in the circle defined by @xmath317 and thus @xmath10 has the dense reversals property . with respect to the coordinate @xmath274 , in the wall of the cylinder @xmath52 , two consecutive reversion points lie in region of the graph with different concavities , as illustrated on figure [ wall ] . this phenomenon implies the existence of chaotic dynamics near the tangency as we proceed to explain . * ( iv ) * if @xmath318 is the projection of a point where @xmath284 has a vertical tangent , then @xmath50 is tangent to @xmath49 at the corresponding point . otherwise an arbitrarily small modification of the vector field @xmath10 in a neighbourhood of @xmath319 will move @xmath318 , creating the tangency see figure [ cocoon ] . suppose @xmath320 is a curve close to a vertical segment and that it is parametrized by @xmath321with @xmath322 then , changing @xmath76 near @xmath319 if necessary , we may assume that @xmath323 meets @xmath281 at a point where the last curve has a vertical tangent . the perturbation is achieved in part * ( v)*. * ( v ) * in @xcite is it used an unfolding argument to obtain tangencies from the dense reversals property . since we have no guarantee that a generic unfolding process holds in the volume - preserving setting , we should tackle the problem using another approach . the key step is the volume - preserving hayashi s connecting lemma ( * ? ? ? * and theorem e ) , whose formulation may be adapted for flows as follows : ( * ? ? ? * adapted ) [ wex ] let @xmath324 be a non - periodic solution of @xmath325 . for any @xmath1-neighbourhood @xmath326 of @xmath10 , there exists @xmath327 and @xmath328 such that for all @xmath329 $ ] and any two points @xmath330 outside the tube @xmath331}\varphi(t,(b(z,\delta))$ ] , if the forward trajectory of @xmath224 intersects @xmath332 and the backward trajectory of @xmath333 intersects @xmath332 , then there exists @xmath334 such that @xmath335 outside @xmath331}\varphi(t,(b(z,\delta))$ ] and there exists a one - dimensional connection @xmath336 $ ] . the previous result says that if two distinct initial conditions @xmath224 and @xmath333 have solutions that visit a given neighbourhood of a point @xmath337 and the initial conditions @xmath224 and @xmath333 are sufficiently far away from a piece of the backward solution of @xmath164 , then we can find a divergence free vector field for which @xmath224 and @xmath333 are heteroclinically related . the support of the perturbation lies on @xmath338}\varphi(t,(b(z,\delta)).\ ] ] by * ( i)(iii ) * , we clearly have two points @xmath339 and @xmath340 in the conditions of theorem [ wex ] . it is obvious that the solution associated of @xmath164 can not intersect the heteroclinic connection @xmath29 $ ] . therefore , we can always choose @xmath341 sufficiently small in order that the sets @xmath342}\varphi(t,(b(z,\delta))$ ] and @xmath29 $ ] are disjoint . then , we can perform the volume - preserving perturbation given by theorem [ wex ] to obtain the tangencial intersection between @xmath49 and @xmath50 by making a connection near @xmath164 . if the intersection is tangential the proof is completed . otherwise , we consider a smooth path @xmath343}$ ] of elements on @xmath3 such that @xmath344 and @xmath345 . since @xmath346 displays no intersection between @xmath347 and @xmath348 , @xmath349 displays a transversal intersection between @xmath350 and @xmath351 , and the accumulation is of a quadratic type , then there must be @xmath352 such that @xmath353 displays a tangencial intersection between @xmath354 and @xmath355 . the classic cocoon bifurcation introduced in @xcite ` begins ' when the two - dimensional invariant manifolds of @xmath27 and @xmath28 have a tangency . after the first tangency , when moving slightly a given one - dimensional parameter , we obtain two structurally stable heteroclinic connections . it is the beginning of a sequence of heteroclinic bifurcations . saddle - node bifurcations and elliptic solutions have been detected numerically in @xcite . we would like to prove that , for any @xmath356 , and any tubular neighbourhood @xmath117 of the bykov cycle @xmath46 , for every @xmath126 , there exists a closed two - dimensional torus @xmath127 such that @xmath357 and @xmath10 can be @xmath1-approximated by @xmath130 whose flow displays a tangency between @xmath49 and @xmath50 , intersecting twice @xmath127 and having length greater than @xmath131 within @xmath127 . theorem [ main1 ] shows that the vector field of ( [ general ] ) can be @xmath1-approximated by another vector field such that the associated flow displays a primary tangency inside @xmath117 between the manifolds @xmath49 and @xmath50 . beyond the tangency , there are infinitely many pieces of @xmath49 accumulating on @xmath358 . therefore the set @xmath359 accumulates on @xmath360 . repeating recursively the argument of the proof of theorem [ main1 ] in the appropriate direction , we get curves in @xmath167 with the dense reversals property . using again the @xmath1-connecting lemma as in * ( v ) * , we may construct vector field @xmath130 such that its flow exhibits tangencies of order @xmath361 , with @xmath362 large , and then we may define a torus @xmath127 , not containing the saddle - foci , inside which heteroclinic tangencies of order @xmath361 spend most of the time ( say larger than @xmath126 ) . an illustration of this phenomenon has been given in figure [ cocoon ] . the main tool for detecting the manifolds @xmath49 and @xmath50 is the time delay function , defined by : @xmath363 where @xmath364 corresponds to integrate along the trajectory @xmath365 the time delay function measures how long it takes for a given solution to escape from the sphere of radius @xmath366 . trajectories starting at the regions between two consecutive lines of figure [ cocoon ] ( corresponding to heteroclinic tangencies ) will turn one more time around the equilibria @xmath28 - more details in ( * ? ? ? * section 7 ) . this is why the map @xmath364 acquires a castle shape with a higher level between pairs of equilibria of the time delay map . in the present paper , we can not use the arguments of @xcite in which lebesgue almost all solutions leave @xmath117 and hence will escape from a given sphere centered on the origin , with radius @xmath367 . the later paper uses explicitly that hypothesis * ( h4 ) * does not hold . by detecting logarithmic equilibria of this map , one locates the invariant manifolds . a typical time delay function has several singularities as the one depicted in ( * ? ? ? 11 ) , corresponding to the heteroclinic tangencies . the time delay function acts as a numerical mechanism , searching along lines in a cross section , for pieces of the invariant manifolds of the two nodes . heteroclinic connections form a skeleton for the structure of periodic and bounded aperiodic orbits . according to @xcite , when elliptic periodic solutions are detected , the time delay function has flat tops . numerical results have been presented in ( * ? ? ? * section 4 ) . the analytical existence of these solutions is the core of next section . the organizing centre for the dynamics may be characterized by the presence of a hyperbolic set which is flow - invariant , indecomposable and topologically conjugate to a bernoulli shift space with finitely many symbols . the union of all of these sets is not uniformly hyperbolic , and countably many of these sets are destroyed under small generic perturbations . homoclinic cycles of shilnikov type and subsidiary cycles are expected to occur @xcite . in this section we study the existence of elliptic fixed points for the map @xmath76 . we will make use of the references @xcite . the next result will be useful to make the bridge between conservative vector fields and conservative first return maps . we thanks santiago ibez for the hints on the next proof . [ return_map ] [ area - preserving ] the return map of a volume - preserving vector field to any transverse section @xmath118 preserves the area @xmath368 ( for some density @xmath369 ) , and so is an area - preserving diffeomorphism . let @xmath10 be a volume preserving vector field , _ i.e. _ such that @xmath370 . take a transverse section @xmath118 with a well defined poincar map @xmath371 , where @xmath372 . now , take a simply connected subset @xmath373 ( assume it is a disk to make it simple ) . consider the solid @xmath2 obtained by saturation of the flow from @xmath374 to @xmath375 . the boundary of @xmath2 , @xmath376 , can be written as @xmath377 , where @xmath378 is the saturation by the flow of the one - dimensional boundary @xmath379 of @xmath373 until it reaches @xmath380 . by the ostragadsky - stokes theorem @xcite , the triple integral of @xmath381 on @xmath2 equals the surface integral of @xmath382 on @xmath376 . more formally : @xmath383 since @xmath384 , the left part of the previous equality is zero . moreover , @xmath385 because @xmath10 is tangent to @xmath378 at each point . hence @xmath386 in the above expression , the normal vector on @xmath374 is the opposite to the normal vector on @xmath375 because the normal vector must have a continuous variation on @xmath376 . by taking the normal vector pointing in the same direction the conclusion is that @xmath387 meaning that there exists ` an adapted ' measure that is preserved ( by @xmath388 ) . we will borrow the arguments of @xcite to show the existence of suspended horseshoes near the heteroclinic tangency . the latter authors compute the first return map @xmath76 and apply the rescaling lemma for studying bifurcations of fixed points . the rescaling lemma says that we may reduce the study of bifurcations of the first return maps to the classic analysis of bifurcations of the two - dimensional conservative hnon - like maps . by @xcite , we may find @xmath389 and a union of strips @xmath390 , @xmath391 , accumulating on the segment @xmath392 as @xmath393 such that @xmath394 has a conservative horseshoe near the cycle , for @xmath395 . for small perturbations of @xmath76 , infinitely many bifurcations of horseshoes creation or destruction occur , including the birth and disappearance of elliptic periodic points . these horseshoe bifurcations must have different scenarios depending on a type of the initial tangency . the character of reciprocal position of the strips @xmath396 and their images @xmath397 are essentially defined by the signs of two parameters that governs the linear part of the global map ; the authors of @xcite selected six different cases of maps with quadratic homoclinic tangencies . for these parameters and for @xmath119 large , the position of all the strips @xmath396 and @xmath397 have regular intersections . in particular , if @xmath398 , then the first return map associated to any vector field @xmath1-arbitrarily close to @xmath10 has a conservative horseshoe accumulating on @xmath399 . moreover , there are persistent heteroclinic tangencies of the invariant manifolds of periodic solutions . in a small section transverse to the cycle ( not containing it ) , since the number of symbols of the bernoulli shift approaches @xmath400 as the strip approaches the cycles , a cascade of bifurcations occurs , associated to the creation and annihilation of horseshoes @xcite . there are no guarantees that such perturbations still have the bykov cycle in its flow . if @xmath401 is a linear map such that the @xmath402 , then @xmath403 if and only if its eigenvalues are complex ( non - real ) and conjugated . let @xmath356 . any neighbourhood of the points where we observe dense reversals property contains nontrivial hyperbolic subsets including infinitely many horseshoes , accumulating on the tangency . in particular , there are infinitely many 1-periodic solutions accumulating on the tangency . let @xmath404 and @xmath405 be the first return map of a vector field on @xmath2 satisfying ( * h1*)(*h5 * ) . the points @xmath406 for which we observe reversals satisfy @xmath407 observing that @xmath408 and @xmath409 is bounded , it follows that : @xmath410 near the initial conditions in @xmath160 where we observe the dense reversals property , condition ( [ tang_final ] ) holds and thus , we conclude that @xmath411 as illustrated in figure [ trace ] , for @xmath412 , if @xmath413 is a neighbourhood in @xmath167 of a point for which we observe a reversal , then there is a line where the trace is zero and that splits @xmath413 into two connected components where the trace is positive or negative . in particular , we may define a strip @xmath414 in which the trace belongs to the interval @xmath415 , where all 1-periodic solutions should be elliptic . intersecting the set of 1-periodic solutions that accumulate on the tangencies with the strip @xmath118 , yields infinitely many elliptic 1-periodic solutions accumulating on the tangency . yields infinitely many elliptic 1-periodic solutions accumulating on the tangency . , height=113 ] dynamics near an elliptic point is reminiscent of the dynamics near a center type equilibrium : under regularity assumptions on the vector field , a generic elliptic point is surrounded by a family of closed kam - curves which occupy a set of almost full lebesgue measure in a small neighbourhood of the fixed point . on each kam - curve the map @xmath76 is conjugate to an irrational rotation . as the distance to the fixed point increases , the rotation number changes in a monotonic way . the kam - curves do not fill the neighbourhood of the elliptic point . rational rotation numbers correspond to resonant birkhoff zones between the closed curves and the dynamics in these regions is typically chaotic . this gives rises to chaotic dynamics between tori regions coexisting with hyperbolic horseshoes and the bykov cycle . adding the reversibility to the problem , symmetric elliptic trajectory may be seen as the limit of other elliptic periodic points see gonchenko _ et al _ @xcite . when the fragile connection is broken , near the bykov cycle there are a wide range of curious homoclinic bifurcations as reported in @xcite . some of them do not depend on the chirality and are generated by the swirling pattern due to the existence of saddle - foci . they can be viewed as spiralling structures in a parameter diagram . conservative structures might be comprised of elliptic islands separated by saddles . there is another self - similar organization of embedded saddles and centers on smaller scales . the closer one approaches a center of the rings , the greater the number of twists the outgoing heteroclinic connection makes before returning to the saddle see @xcite . in this paper , trying to answer the questions * ( q1)**(q2 ) * , we have proved that the extended version of the cocoon bifurcations for conservative systems , the heteroclinic tangencies and the elliptic periodic solutions are strongly related . more precisely , in the present paper we have considered divergence - free vector fields whose flows have bykov cycles , we have found heteroclinic tangencies leading to the description of elliptic periodic solutions for the first return map to a cross section defined near the tangency . within a codimension 1 set in @xmath73 , tangencies between the invariant manifolds of the two saddle - foci densely occur . the proof of the existence of elliptic solutions near the horseshoes does not run along the same lines of gonchenko and shilnikov @xcite , in which the authors used renormalisation to the hnon map . along the proof of our results , we use the topological notion of chirality , making our findings different from those of @xcite . we asked for the @xmath1 volume - preserving connecting lemma @xcite to perform local perturbations . similar heteroclinic bifurcations have been considered in @xcite , where generic flat perturbations do not preserve the cycle . the non - wandering dynamics near the bykov cycle is dominated by hyperbolic horseshoes , conjugate to a full shift over a finite alphabet , that accumulate on the cycle . hyperbolic behaviour can not be separated from the ellitic behaviour by isotopies . our underlying hypothesis on the maintenance of the bykov cycle is inspired in the @xmath76-reversible case , with @xmath416 , where the mantainance holds _ per se_. preserving the bykov cycle is the most important consequence from the anti - symmetry ; this is why we decided to reach general results without the reversibility hypothesis . the divergence - free character is crucial because it guarantees the stability of elliptic periodic solutions in our three - dimensional context . in higher dimensional case , the divergence freeness is no longer sufficient and we should consider hamiltonian vector fields in order to get the stability of elliptic periodic solutions . in the present paper , we are able to prove the existence of heteroclinic tangencies between @xmath49 and @xmath50 . our findings are richer that those of @xcite in the sense that we can explicitly give the manifolds for which we observe tangency . note that the works @xcite proved the existence of a @xmath1-dense / residual set inside which tangencies and elliptic solutions occur via a mechanism developed by mora and romero @xcite to create open sets having a dense set of maps with tangencies nearby . different chirality of the nodes is a new topological concept and has never been studied in the conservative setting . this topological property has profound effects on the dynamics near the bykov cycle . of course , in agreement with kaloshin s theorem on the prevalence of kupka - smale systems , these tangencies can not occur for a full lebesgue measure on @xmath2 . the density of tangencies near the cycle is markedly different from the results about the michelson system proved in @xcite where the tangencies ( near the bykov cycle ) are rare . if @xmath77 , the @xmath76-reversibility prevents different chirality of the nodes . the context of the proof of theorem [ main1 ] is different from that described in knobloch _ et al _ @xcite using spirals , a work that has been motivated by the michelson system . instead of restricting the flow to the wall of the cylinder ( as we do in theorem [ main1 ] ) , their conclusions , like bykov s @xcite , are supported by the study of spirals intersections . in ( * lemmas 6.6 and 6.7 ) it is shown that two logarithmic spirals with a common centre can have at most two tangencies , assuming an irrational ratio of the real part of the complex eigenvalues see figure [ spirals](a ) . although these spirals are only the leading order terms of the bifurcation equations , the density of tangencies can not occur near the cycle , as shown in @xcite . without breaking the fragile connection , assuming the same chirality of the nodes , tangencies may occur but not near the cycle . bykov implicitly assumed that the chiralities of two nodes are different in formulas ( 3.4),(3.5 ) of @xcite . indeed , observe the same sign of the exponent of @xmath417 in the formulas that represent the evolution of the angular component of the spirals : @xmath418 where @xmath419 represents the angular coordinates in @xmath162 and @xmath166 and the nonreal eignevalues are given by @xmath420 , @xmath421 . the same happens in formula ( 3.1 ) of bykov @xcite . after taking logarithms of both sides , from equality ( 3.3 ) on , bykov neglects the assumption about the chiralities of the nodes , assuming that they are different . therefore , in the cross sections , the spirals corresponding to the two - dimensional invariant manifolds of the saddle - foci are oriented in the same way . this explains the orientation of the spirals of figure 2 of @xcite in contrast to those depicted in figure 11 of @xcite . these spirals are revived in figure [ spirals ] . bykov never comments on the chiralities of the nodes , assuming implicitly that they are different . and @xmath422 have : case ( a ) : an infinite number of transversal intersections and at most two tangential intersections in total @xcite . case ( b ) : an infinite number of transversal intersections and , for small perturbations which do not break the fragile connection , an infinite tangential intersections.,height=170 ] the cocoon bifurcation is a set of rich bifurcation phenomena numerically observed by lau @xcite in the michelson system . the authors in kokubu _ et al _ @xcite proved the existence of the cocoon bifurcations for the michelson system , which arise as the travelling wave equation of the kuramoto - sivashinky equation and also as a part of the limit family of the unfolding of the three - dimensional nilpotent singularity of codimension 3 see @xcite . numerical results show that , when a given parameter varies , the system exhibits an infinite sequence of heteroclinic bifurcations due to the tangency of the two - dimensional invariant manifolds of saddle - foci . each bifurcation creates new transverse heteroclinic connections . in @xcite , these bifurcations have been studied from a general point of view ; they have been explained as a consequence of an organizing center called the _ cusp transverse heteroclinic chain_. fixing the fragile connection , the cocoon bifurcations play the same role as the newhouse bifurcations ( generic unfoldings for the homoclinic tangency ) in the dissipative category . thick cantor sets appear near the tangency , which are @xmath8-persistent , @xmath423 . in the present paper , we prove the existence of cocoon bifurcations within a degenerate set of divergence - free vector fields characterized by different chirality of the nodes . in contrast to the findings of @xcite , in which the authors could break the cycle , our perturbations have been performed without breaking the fragile connection . furthermore , in @xcite , the reversibility is used to get saddle - nodes with both stable and unstable sets of dimension 1 and the volume - preserving character plays no role . we may use our results to illustrate the expected behaviour of the simple bifurcations of symmetric periodic solutions , namely saddle - node and period doubling via the creation and annihilation of horseshoes in a similar phenomenon to that described in @xcite . although we deal with a degenerate class of vector fields , our results hold for any generic unfolding of the cycle . the greatest difficulty arises when we want to preserve the bykov cycle , a property that emerge in @xmath76-reversible systems with @xmath416 . in a broader way , our results hold for any kind of cycles having a one - dimensional solution connecting two saddle - foci . in fact , following the approach in @xcite , it should be possible to show that vector fields containing bykov cycles are always @xmath1-close to another with a tangency ( and even a cocconing cascade of tangencies ) , without making any assumption on the chirality . nevertheless , there is no guarantee that the cycle is preserved . although a degenerate class of systems satisfy * ( h1)(h5 ) * , the present work may be seen as the description of a simple and general mechanism to create non - hyperbolic behaviour in a robust fashion near a singular cycle . since the existence of cycles involving equilibria is not a generic phenomenon , they may exist outside the residual sets given by the classical dichotomies for divergence - free vector fields @xcite . several questions remain to be solved for this kind of cycles , in particular about the existence and sign of the lyapunov exponents . the lebesgue measure of the points that remain for all time in the neighbourhood of the cycle is an interesting point to be considered . in the spirit of @xcite , conservativeness - breaking is a another direction that should be investigated . a first attempt to understand this phenomenon may be found in dumortier _ et al _ ( * ? ? ? * section 4 ) . the present article should be seen as a starting point for further related studies ; the ergodic characterization of the dynamics near these cycles ( even in the dissipative case ) is still far for being completely understood . gonchenko , s. v. , shilnikov , l. p. & turaev , d. m. , homoclinic tangencies of an arbitrary order in newhouse domains , _ itogi nauki tekh . , sovrem . mat . _ , * 67 * , ( 1999 ) 69128 [ english translation in j. math . , 105:1738 - 1778 , 2001 ] . labouriau , i . s. ; & rodrigues , a. a. p. , `` partial symmetry breaking and heteroclinic tangencies '' , in : s. ibez , j.s . prez del ro , a. pumario , j.a . rodrguez ( eds . ) , progress and challenges in dynamical systems , in : springer proc . math . stat . , springer - verlag , ( 2013 ) , 281299 .
in this paper we continue the exploration of the stellar content of obscured galactic giant h ii regions begun by ( * ? ? ? * hereafter paper i ) . @xmath0 , @xmath1 , and @xmath2 images are used to make a broad assessment of the stellar content of obscured star forming regions in the milky way . infrared spectroscopy follows , providing details of the brightest cluster members which can be used to make distance , mass , and luminosity estimates . the spectra are placed in proper context by comparison to new infrared spectral classification systems for massive stars @xcite . the known hot star content of the galaxy is rapidly expanding on the strength of the sophisticated infrared detector gains of the last 10 years . @xcite have recently published a very detailed account of the most massive stars in the ( relatively ) nearby giant h ii region m17 . the present project ( paper i , this work , and future work ) seeks to provide a large sample of massive star clusters with which to study the young and massive stellar content in the galaxy . this sample builds on the detailed visual studies of the galactic ob associations @xcite and provides a bridge to studies of young stellar objects in star forming regions . finally , the investigation of a large sample of clusters in galactic giant h ii regions will be important in understanding the massive star clusters in the galactic center @xcite which may have formed under different conditions than are typical in the disk of our galaxy @xcite . w42 is located in the fourth galactic quadrant at _ @xmath4 25.4 , 0.2 . @xcite determined w42 to be at the `` near '' kinematic distance ( 3.7 kpc for @xmath5 8 kpc ) and hence somewhat less luminous than earlier estimates @xcite . in this series of papers we shall follow a suggestion of dr . robert kennicutt ( private communication ) that `` giant '' means that more than @xmath6 lyman continuum ( @xmath4 lyc ) photons are inferred to be emitted per second from the h ii region . this is about ten times the luminosity of the orion nebula and roughly the number emitted from the hottest _ single _ o3-type star . as these stars are not found in isolation , there is an implication that a `` giant '' h ii region contains some minimum of _ multiple _ o - type stars . in light of our new distance estimate ( see 4 ) , w42 probably falls below this limit being perhaps a few times more luminous than the trapezium in orion . our target list was originally based on eleven of the most luminous giant h ii regions from the study of @xcite who tabulated their lyc output derived from radio continuum measurements and kinematic distance estimates . w42 is not seen in visual images , and we estimate a foreground extinction of @xmath7 @xmath4 10 mag ( see below ) . @xmath0 ( @xmath8 , @xmath9 ) , @xmath1 ( @xmath10 , @xmath9 ) , and @xmath2 ( @xmath11 , @xmath12 ) images of w42 were obtained on the nights of 29 august 1998 and 01 , 02 may 1999 with the f/14 tip tilt system on the cerro tololo interamerican observatory ( ctio ) 4 m blanco telescope using the two facility imagers cirim ( 1998 data ) and osiris , the ohio state infrared imager / spectrometer ( 1999 data ) . spectroscopic data were obtained on the night of 02 may 1999 using the f/14 tip tilt system at the blanco telescope with osiris . cirim and osiris are described in the instrument manuals found on the ctio web pages ( www.ctio.noao.edu ) . for osiris , see also @xcite . the tip - tilt system is described by @xcite . the tip tilt system uses three piezo electric actuators to move the secondary mirror at high frequency in a computer controlled feed back loop which corrects the natural image centroid motion . osiris employs 0.16@xmath13 pixels and cirim 0.21@xmath13 pixels . all basic data reduction was accomplished using iraf . each image / spectrum was flat fielded using dome flats and then sky subtracted using a median combined image of five to six frames . for w42 itself , independent sky frames were obtained five to ten arcminutes south of the cluster . standard stars used the median combination of the data for sky . the osiris 1999 may images were obtained under photometric conditions and in @xmath14 0.5@xmath13 to 0.6@xmath13 fwhm seeing ( with the tip tilt correction ) . total exposure times were 270 s , 135 s , and 135 s at @xmath0 , @xmath1 , and @xmath2 , respectively . the individual @xmath0 , @xmath1 , and @xmath2 frames were shifted and combined ( figure [ w42i ] ) , and these combined frames have point sources with fwhm of @xmath14 0.6@xmath13 , 0.7@xmath13 , and 0.6@xmath13 at @xmath0 , @xmath1 , and @xmath2 , respectively . dophot @xcite photometry was performed on the combined images . the flux calibration was accomplished using standards 9170 and 9172 from @xcite which are on the las campanas observatory standard system ( lco ) . the lco standards are essentially on the cit / ctio system @xcite , though color transformations exist between the two systems for redder stars . the standards were taken just before the w42 data and within 0.16 airmass of the airmass for w42 ; no corrections were applied for these small differences in airmass . aperture corrections using 11 pixel radius apertures were used to put the instrumental magnitudes on a flux scale . stars brighter than about 10th magnitude are expected to be a few percent non - linear on the osiris images . we have included the 1998 cirim data for such stars since the count levels for the cirim images were in the fully linear regime ( no linearity correction needed ) . the zero point to the cirim photometry was determined by comparing the instrumental magnitudes of stars in common to the osiris and cirim images . the cirim @xmath15band photometry includes a color correction term in order to make a transformation on to the cit / ctio system ( paper i ) . the osiris data was placed on the cit / ctio system by making small corrections ( @xmath16 10 @xmath17 ) to the @xmath0 , @xmath1 and @xmath2 magnitudes based on linear fits to the magnitude differences as a function of color for stars in common to the cirim and osiris images . uncertainties for the final @xmath18 magnitudes include the formal dophot error added in quadrature to the error in the mean of the photometric standards ( including the transformation to osiris magnitudes for the cirim data ) , and the error in aperture corrections used in transforming from the dophot photometry to osiris magnitudes . the latter errors dominate and were derived from the scatter in the measurements of four to seven relatively uncrowded stars on the mosaic frames . the sum ( in quadrature ) of the aperture correction and standard star uncertainties is @xmath19 0.018 , @xmath19 0.023 , @xmath19 0.019 mag in @xmath0 , @xmath1 , and @xmath2 , respectively . the dophot errors ranged from approximately @xmath19 0.01 mag to an arbitrary cut off of 0.2 mag ( stars with larger errors were excluded from further analysis ) . the flat field illumination was not uniform . a smooth gradient with full range of about 10@xmath17 was present . corrections for this gradient were made based on observations of a standard star taken over a 49 position grid covering the array . the spectra of three of the brightest four stars in the center of w42 were obtained with a 0.48@xmath13 wide slit ( oriented ew ) in 0.7@xmath13 fwhm seeing and divided by the spectrum of hr 6813 ( a1v ) to remove telluric absorption features . br@xmath20 absorption in hr 6813 was removed by eye by drawing a line across it between two continuum points . one dimensional spectra were obtained by extracting and summing the flux in a @xmath19 2 pixel aperture ( 0.64@xmath13 wide ) . the extractions include background subtraction from apertures centered 1.0@xmath13 on either side of the object . the wavelength calibration was accomplished by measuring the positions of bright oh@xmath21 lines from the @xmath3band sky spectrum @xcite . lines are identified by their relative differences between one and another . the measured dispersion is 0.0003683 pix@xmath22 . the spectral resolution at 2.2 is @xmath23 3000 . a spectacular stellar cluster is revealed at the heart of w42 in our near infrared images ( figure [ w42i ] and figure [ w42c ] ) . apparently , the cluster has just emerged from the edge of the molecular cloud from which it formed . this is confirmed below through @xmath3band spectroscopy which shows that the central massive star has largely cleared away its birth cocoon , but two of the next brightest stars have not . @xcite observed w42 in the mid and far infrared . their analysis revealed two distinct sites of star formation toward w42 . one , g25.4se is located at 3.7 kpc from the sun ( accounting for a sun to galactic center distance of 8 kpc ) , while the other , g25.4nw is probably located at about 9.6 kpc . lester et al . found a 10 unresolved source located at ra ( 2000 ) @xmath24 , dec ( 2000 ) @xmath25@xmath26 which is coincident with g25.4se . this mid ir source is associated with the stellar cluster we have observed . comparison of the position of the bright foreground star in the se corner of our @xmath3band image to the same star on the image from the digitized sky survey ( dss ) results in a position for the ( @xmath3band ) bright central star in the cluster ( w42 # 1 ; see 3.2 ) of ra ( 2000 ) @xmath27 , dec ( 2000 ) @xmath25@xmath28 . the @xmath29 color@xmath30magnitude diagram ( cmd ) for the region toward w42 is shown in figure [ cmd ] . a cluster sequence is evident at 0.6 @xmath29 1.5 along with stars with much redder colors and a probable foreground sequence . the @xmath31 vs. @xmath29 color color diagram is presented in figure [ cc ] . as expected from the morphology and range of colors in figure [ w42c ] , the strong effects of differential reddening can be seen in the color color diagram . typical reddening lines are shown for m giants @xcite and early o stars @xcite . the latter color was transformed as described in 3.1.3 . the cluster stars defined below in 3.1.1 and shown as _ open circles _ in figure [ cc ] may have a slight unexplained systematic offset relative to the reddening line for normal o stars . this possible offset will not affect the conclusions in this paper regarding the cluster stars . the relationship for the intrinsic colors of classical t tauri stars ( pre main sequence stars ) @xcite is also shown for reference . the adopted reddening law is from @xcite . by separating the stars in figure [ cmd ] based on radial position , we can better define the central cluster relative to the surrounding field . in figure [ surf ] , we plot the radial surface density of stars centered on the position of the bright central star , w42 # 1 ( 3.2 ) . we also plot the radial surface density for stars with @xmath2 @xmath32 14 mag , for which the number counts are more nearly complete ( see below ) . the surface density becomes approximately uniform at a radius of @xmath33 , continuing up to 50@xmath13 . it then begins to fall rapidly at the edge of the array ( _ dashed vertical line _ ) as expected due to the rectangular shape of the field . taking these radii as representative , we divide the cmd into regions with 0@xmath34 30@xmath13 , 30@xmath34 50@xmath13 , and @xmath35 50@xmath13 , as shown in figure [ cmdrad ] . these represent the central cluster , a background annulus , and the region of the array where edge effects are important to the radial number counts . the range of color and brightness in figure [ cmdrad ] overlaps for stars in the cluster and background regions . a distinct blue sequence can be seen in the 0@xmath34 30@xmath13 cluster cmd which appears to merge smoothly with other stars in the 30@xmath34 50@xmath13 cmd . clearly the cluster can not be completely extracted from the surrounding field based on radial position alone . in order to further enhance the cluster sequence , we defined a background cmd using the 30@xmath34 50@xmath13 region and accounting for the area difference between this annulus and the central region . the background cmd was binned in 0.5 mag color magnitude bins . we then randomly selected and subtracted stars from the cluster in equal numbers from bins matching the background cmd . the resulting cmd is shown in figure [ cmdsub ] . the corresponding colors for these stars are plotted as _ open circles _ in figure [ cc ] . most of the brightest stars in the central cluster fall along an essentially vertical track as expected for hot stars . the average color of the brightest seven of these stars is @xmath29 @xmath4 0.637 which corresponds to an extinction at 2.2 of @xmath36 1.07 mag ( @xmath7 @xmath37 10 mag ) using the interstellar reddening curve of @xcite and an intrinsic @xmath29 @xmath4 @xmath38 @xcite . many other stars appear more reddened , typically up to @xmath29 @xmath14 2 with some as high as 3.8 . a star with @xmath29 @xmath4 2 and intrinsic @xmath29 @xmath4 0 would have @xmath36 3.2 mag ( @xmath7 @xmath37 32 mag ) . it will be useful in discussing the w42 cluster below to have an estimate for the zero age main sequence ( zams ) transformed into the @xmath2 vs. @xmath29 plane . we have constructed such an estimate using the model results of @xcite . the models give bolometric luminosities ( ) and effective temperatures ( ) for stars of a given mass as they begin their evolution on the zams . using relationships for spectral type vs. @xcite , visual bolometric correction ( bc@xmath39 ) vs. effective temperature @xcite , @xmath40 vs. spectral type @xcite , and @xmath29 vs. spectral type @xcite , we transform the and to the @xmath2 vs. @xmath29 cmd . a small correction has been made in an attempt to place the @xmath29 colors from @xcite onto the cit / ctio system . the @xcite colors are basically referred to the system of @xcite with newly defined colors involving @xmath1 ( johnson used no @xmath1 filter ) . in addition , @xcite found that interpolating an @xmath1 magnitude from the observed @xmath0 and @xmath2 values gives a very good estimate of the observed @xmath1 magnitude . @xcite found no discernible color correction between the saao system and that of @xcite . we have therefore used the transformation of @xcite between the saao system and cit / ctio to transform the @xmath29 colors given by @xcite . these corrections are at most one percent and hence negligible compared to the measurement uncertainties and scatter due to differential reddening . the zams cmd is shown in figure [ cmdsub ] for a particular distance ( 2.2 kpc ) which is discussed in 4 . the values plotted in figure [ cmdsub ] are listed in table [ tabzams ] . these values may be compared to those given by @xcite . there is generally good agreement as a function of . this is expected since the same models and colors were used in both cases . there are systematic differences at the @xmath14 @xmath19 0.3 mag level for @xmath41 as a function of spectral type since the spectral type vs. relation adopted here @xcite is different than that used by @xcite who used the relation given by @xcite . these differences do not affect the conclusions of 4.3 since the resulting @xmath41 for the spectral type in question ( o5o6 ) is within 0.1 mag in both cases . the @xmath3band luminosity function ( klf ) is shown in figure [ klf ] . neglecting the last five bins where the counts appear to be incomplete and the first 3 bins where the counts may be better described as uniform , the @xmath3band counts are well fit by a power law with index 0.4 ( log@xmath42n @xmath4 0.4 @xmath43 @xmath2 + constant ) . we estimated the completeness of the klf by performing artificial star experiments . initially , we attempted to add stars to the original @xmath3band image , but even adding 10@xmath17 of the observed klf resulted in recovered luminosity functions much less complete than the observed one . this is due to the high spatial density of the cluster . to avoid this problem , we constructed complete artificial frames and analyzed them in the same way as the original frame . anticipating the result , we used the spatial density distribution of the stars on the @xmath3band image with @xmath44 14 mag ( figure [ surf ] ) to generate the positions of stars on the artificial frames . the input luminosity function was constructed from two components , a uniform distribution from 8.5 @xmath45 10.5 , and a power law for 10.5 @xmath46 19.5 . the total number of stars was set by the flux on the @xmath3band image , less a uniform background component determined from the sky frame . this should be conservative regarding the number of stars since the strong nebular flux is included . this latter aspect may be balanced somewhat by the fact that the artificial frames did not include variable extinction and nebular emission . the total @xmath3band flux in figure [ w42i]@xmath47 is 6.0 mag , not including the bright saturated star in the se corner of the frame . the total flux of the average actual input luminosity function ( figure [ fake ] ) was 5.9 mag which can be compared to the total flux in observed stars ( figure [ klf ] ) , @xmath48 6.18 mag . we constructed 10 artificial frames by randomly sampling the spatial distribution and luminosity functions . dophot was run on each frame and the recovered stars matched to the input lists . the average recovered luminosity function , input luminosity function , and completeness fraction are shown in figure [ fake ] . the general shape and distribution of the recovered stars in figure [ fake ] suggest the experiments are a fair test of completeness of the original frame . from the same input and recovered star lists , we also show the luminosity functions and completeness fraction for the stars located at 0@xmath3430@xmath13 . see figure [ fake30 ] . tests were also made to check whether somewhat steeper power laws were also consistent with the observed klf . this might be the case if the crowding were sufficient to `` hide '' many fainter stars . artificial star experiments analogous to those above ( including the same uniform component ) but with a power law component of 0.5 are not consistent with the data . a single power law component of this steepness produces more star light than is necessary to account for the observed total flux and produces too many stars in the recovered luminosity function at fainter magnitudes if it is required to produce approximately the correct numbers at brighter magnitudes . the spectra of the targets w42 # 1 , # 2 , and # 3 are shown in figure [ spec ] . the final signal to noise in the three spectra is typically 80105 , 7595 , and 5578 for w42 # 1 , w42 # 2 , and w42 # 3 , respectively and is higher on the red end than the blue end for all three . the brightest star in the central cluster , w42 # 1 ( @xmath49 8.8 mag ) , shows characteristic o star features @xcite . these include ( 2.069 and 2.078 ) emission , ( 2.1155 ) emission , br@xmath20 ( 2.1661 ) absorption , and ( 2.1891 ) absorption . comparison to the standards presented by @xcite results in @xmath3band spectral type of ko5-o6 . these stars typically have mk spectral types of o5 to o6 . the present classification system laid out by @xcite does not have strong luminosity class indicators . still , the ( 2.06 ) and br@xmath20 features can be used to approximately distinguish between dwarf+giants on the one hand , and supergiants on the other : the supergiants tend to have emission or weak absorption in these lines . the spectrum of w42 # 1 shown in figure [ spec ] has been background subtracted with nearby apertures to account for the nebular emission seen in projection toward the star . the apparent absorption feature at the position of br@xmath20 and absence of a feature at ( 2.06 ) ( which might indicate a poor subtraction of the nebular contribution ) , suggests that w42 # 1 is a dwarf or giant star . in contrast to w42 # 1 , the spectra of w42 # 2 and # 3 show only emission features at ( 2.06 ) and br@xmath20 . these spectra have also been background subtracted with nearby ( 1.0@xmath13 ) background apertures . we believe the strong emission remaining after this subtraction is related to the local environment in these stars ( see 4.1 ) . the dense stellar cluster evident in figure [ w42i ] surrounded by intense nebulosity leaves no doubt that this is a young object still emerging from its birth environment at the edge of its parent molecular cloud . the cluster appears to have emerged by clearing the foreground material to the west ; darker regions with fewer stars remain toward the east . there is a suggestion of photoevaporated regions on the edge of the cloud to the east similar to those seen in m16 @xcite and ngc 3603 @xcite . this is particularly evident in figure [ w42i]@xmath50 . this picture is consistent with the co line maps of @xcite , who found the peak brightness temperature of the associated molecular cloud to be offset to the east of the w42 ( g25.4se ) h ii region ( see their figure 8) . higher spatial resolution images and in nebular lines are in order to further study the interaction and impact of the ionizing cluster on the molecular cloud interface . the process of clearing the local environment is still on going , clumps of dark material are seen in projection against the photoionized h ii region ( e.g. sw of the cluster center ) . in 3.1.2 , the extinction to the brightest stars on the cluster sequence was found to be @xmath51 @xmath4 1.07 mag . this value can be taken as representative of the foreground extinction to the cluster , and indicates a @xmath52 magnitude of approximately 18 for w42 # 1 . this is consistent with the non detection of the cluster on the dss plates . a few stars remain after subtracting the `` background '' component from the cmd ( 3.1.1 ) , but these are more likely in the foreground since the extinction toward them is as low as @xmath51 @xmath14 0.3 mag . there are a host of very red objects indicated in figures [ w42i ] , [ cmdrad ] , and [ cmdsub ] . such colors are suggestive of young stellar objects ( yso ) in the context of an embedded stellar cluster emerging from its parent cloud . the observed colors of known populations of pre main sequence stars ( pms ) and more heavily embedded protostars appear to lie on / occupy rather well defined sequences / regions in observational color space ; and their colors , which are redder than usual stellar photospheres have been successfully modeled as arising from excess emission produced by circumstellar disks @xcite and envelopes @xcite . like normal stars , pms objects may be seen through foreground extinction , further reddening their colors . deeply embedded protostars may have colors with or without excess emission , but typically suffer large extinction due to the dense envelope surrounding them @xcite . obviously , protostars too may suffer additional foreground extinction . the @xmath31 vs. @xmath29 color color diagram is most useful in assessing whether any particular star may have excess emission . the diagram distinguishes between normal stellar colors which are seen through a column of dust , hence making them redder , and a contribution which is due to emission ( reprocessed stellar light from circumstellar envelope and/or accretion luminosity from a circumstellar disk ) . @xcite and @xcite have shown that disks can provide a source of excess emission to the normal stellar colors . in figure [ cc ] , we have plotted the classical t tauri sequence ( ctts ) from @xcite along with reddening lines for the interstellar reddening law of @xcite . the former locus may be understood as arising from disk luminosity and projection effects ( e.g , lada & adams 1992 and references therein ) . the near infrared colors of weak lined t tauri stars ( wtts ) are similar to normal stars @xcite . these stars exhibit spectra which suggest they have disks , but the disk contribution to the near infrared colors is negligible . the herbig aebe stars which are higher mass analogs to the ctts show similar colors to the ctts but extend to generally larger @xmath29 excess and with a larger contribution to the colors from circumstellar extinction @xcite . the colors of these objects have been fit by circumstellar envelope models in which the excess emission arises from reprocessed stellar light via dust heating and re - radiation @xcite , or alternatively , from circumstellar disks with central holes @xcite . we also show the reddening lines for a main sequence o star and m giant ( see s3.1 ) which can be taken as approximate guides for the expected colors for normal stars along sight lines to the inner galaxy . the _ black squares _ in figure [ cc ] represent all the stars in figure [ w42i ] for which @xmath0 , @xmath1 , and @xmath2 colors were measured . the _ open circles _ represent the stars in the `` background subtracted '' central cluster ( 3.1.1 ) . the w42 cluster stars ( including w42 # 1 ) occupy a rather tight sequence at modest reddening in figure [ cc ] as expected for young massive stars seen through different columns of obscuring material . at larger reddening ( @xmath29 , @xmath31 @xmath4 1,2 ) , there is a larger dispersion of colors . there are also a number of stars which lie significantly beyond the reddening band for normal stars in the region of pms stars and protostars as discussed above . two of these objects are w42 # 2 and # 3 . spectra for these objects are shown in figure [ spec ] . neither spectrum shows stellar absorption features , though both exhibit emission at br@xmath20 or ( 2.06 ) after background subtraction ( 2 ) . accounting for the foreground extinction to w42 # 1 , both w42 # 2 and # 3 lie in a region of the color color plot occupied by luminous protostars and herbig aebe stars @xcite . @xcite suggest that herbig aebe stars might be recently `` uncovered '' protostars given the large overlap in the color color plot for the two objects . we believe that the position of w42 # 2 and # 3 in the color color plot combined with their emission line spectra , strongly suggests they are also luminous ysos . the absence of absorption features is consistent with veiling due to the excess emission seen in figure [ cc ] . in the case of w42 # 2 , the residual br@xmath20 emission may be due to a circumstellar disk ; spherical or envelope distributions to the circumstellar material are also possible . higher spectral resolution data on the br@xmath20 line in this object will be required to rule out models of one nature or the other . in w42 # 3 , the presence of emission but no br@xmath20 could be due to imperfect subtraction . w42 # 3 has closer neighbors making the subtraction of the background more difficult . if the emission seen in projection toward w42 # 3 is due to a compact h ii region , then the emission may be coming from a region closer to the star than the br@xmath20 , the latter naturally subtracting off better on these angular scales . we plan to obtain @xmath53band spectra of w42 # 2 and # 3 where the excess emission should be less allowing for an improved picture of the nature of the embedded objects which give rise to the emission spectra seen at @xmath2 . there are several other stars with similar or slightly lower brightness indicated in figure [ cmdsub ] , but with apparently normal colors for hot stars . this suggests a mixture of massive objects with and without circumstellar ( possibly disk ) signatures as seen in m17 @xcite . like m17 , the ysos indicate a very young age for the cluster . the situation in w42 may be compared to other stellar clusters in giant h ii regions . for m17 , @xcite report that the vast majority of cluster stars they studied have infrared excesses . this result should be confirmed with higher spatial resolution images ; the photometry presented by @xcite was performed by summing the flux in 4.8@xmath13 diameter apertures on their 0.8@xmath13 images . @xcite note that their data set was analyzed with profile fitting by @xcite who obtain similar results , and @xcite clearly show that at least some of the massive stars in m17 have disk like spectroscopic features . @xcite find far fewer stars with excess emission in the orion trapezium cluster and attribute this to a mixture of older and younger stars . however , @xcite reach a different conclusion regarding the ages of the orion cluster stars , and we will discuss this further in the next section . in w43 ( paper i ) , only a handful of objects appear in the excess emission region of the color color plot and none of the three spectroscopically classified hot stars do . the brightest object in w43 exhibits wolf rayet features in its @xmath3band spectrum suggesting an older age relative to w42 . this is similar to ngc 3603 , r136 , and the arches ( see the discussion in paper i and references therein ) , and the implication is that while such stars may still be core hydrogen burning , they are not on the zams . these comparisons need to account for the fact that the extinction is generally greater in w42 and ( much so ) in w43 , so that many objects are not detected at @xmath0 or @xmath1 and hence do not appear in the color color plot . clearly , high spatial resolution , homogeneous data sets , each analyzed in detail for for completeness at @xmath0 , @xmath1 , and @xmath2 would be useful in assessing the intrinsic fractions of stars with excess emission in embedded clusters in giant h ii regions . without spectra , it is not possible to identify the nature of the remaining objects in the color excess region of figure [ cc ] , though in the context of this young cluster , it is likely that some are pms stars or protostars . it is also clear from figure [ cc ] that some ysos can occupy the same region of color space as normal stars . some of the cluster stars in figure [ cc ] near ( @xmath29 , @xmath31 @xmath4 1,2 ) where the dispersion in color is larger might be a combination of wtts with essentially normal stellar colors and ctts / herbig aebe stars with smaller excess emission , the actual nature depending also on the luminosity of any particular object . alternately , these could be normal main sequence stars . finally , some could be normal k or m giants in the background , though it is unlikely that many are , given the large number of stars in close association with the central region of the cluster and the fact that we have subtracted a background component from the cmd . inspection of figures [ cmd ] and [ cc ] indicates that w42 # 3 has varied in brightness and color between the 1998 and 1999 observations presented here , becoming fainter at @xmath2 and redder . color and brightness variations in ysos can be explained by a combination of changes in the accretion luminosity produced by the circumstellar disk and changes in the extinction toward the source . @xcite monitored a sample of ysos for color changes , finding the slopes of the colors ( measured over day to year timescales ) in the @xmath31 vs. @xmath29 color color diagram were intermediate between the ctts locus ( excess color arising from disk luminosity ) and pure variations along the interstellar reddening lines such as might occur when clumps of obscuring material pass in front of the line of sight to the yso . @xcite found some yso variations were consistent with one or the other effect dominating . the color variation for w42 # 3 appears most consistent with a change in the local extinction . the klf has been used as the basis for mass function determinations in young embedded or obscured massive star clusters . @xcite found that the m17 cluster klf had a slope consistent with the @xcite initial mass function ( imf ) value if the m17 stars were normal main sequence type , but noting that this was difficult to reconcile with their finding that essentially all the m17 cluster stars had infrared excesses . only in the case where such excesses arise in passive disks could their mass function determination still follow from the klf . @xcite present a klf for the trapezium cluster in orion ( m42 ) . they found the slope of the klf to be inconsistent with that for the @xcite initial mass function ( imf ) , hypothesizing that there are substantial numbers of older stars in m42 as well as very young ones . however , @xcite found that the klf in m42 was consistent with a population of only very young stars if pms evolutionary tracks were used instead of assuming the cluster stars were on the main sequence . @xcite show that deuterium burning on the pms can cause peaks in the klf which are a function of age thus producing luminosity functions different from expected if main sequence mass luminosity relations are assumed . more recently , the detailed optical imaging / spectroscopic investigation of @xcite exquisitely details the young main sequence and pms stellar population in the orion trapezium cluster , clearly demonstrating the overall youth of the cluster . @xcite presented a mass function determination for the arches cluster located near the galactic center ( gc ) . they find that the arches cluster may have upwards of 100 o stars and an age of @xmath14 2 myr . using their observed klf and a mean value for @xmath51 , these authors determined the mass function by relating the @xmath2 magnitudes to stellar mass using the @xcite stellar evolutionary models . @xcite find a mass function in the arches which is significantly flatter than @xcite . they attribute this result to the different pre conditions for star formation in the central few hundred parsecs of the galaxy which they argue should favor higher mass stars . this is in contrast to the situation elsewhere in the galaxy and in the large magellanic cloud ( lmc ) @xcite . @xcite find from a combination of spectroscopy and imaging that the imf is similar for ob associations and dense massive star clusters ( most notably r136 in the lmc ) and in agreement with a @xcite imf . the preceding discussion indicates that care must be taken in transforming the klf into a mass function , including the effects of pms evolution . in this sense , the determination of a mass function for w42 is premature given the implied young age and lack of transformations between pms models and near infrared colors ( which should include the effects of associated disk emission ) . however , a comparison to the different klfs is warranted . in 3.1 we derived the klf for the inner 30@xmath13 central cluster in w42 and showed that it should be complete to 80 @xmath17 for @xmath54 15 mag . the measured slope of the cumulative counts in the central 30@xmath13 klf ( @xmath54 15 , correcting for incompleteness ) is 0.38 @xmath19 0.016 . if we consider only stars in the background subtracted cmd ( figure [ cmdsub ] , the measured slope ( @xmath54 15 ) is 0.36 @xmath19 0.011 . the slope in our background annulus ( 30@xmath13 to 50@xmath13 , @xmath54 15 , correcting for incompleteness ) is 0.51 @xmath19 0.017 . these values may be compared to the results of @xcite , @xcite , and @xcite . for convenience , we compare the power law slope to the _ cumulative _ klf which @xcite also calculated for their data . in m17 , @xcite report a slope of 0.26 , which they claim is consistent with the @xcite imf . we have calculated approximate slopes for @xcite and @xcite from figures of the published luminosity functions ; these are 0.39 @xmath19 0.016 and 0.28 @xmath19 0.013 , respectively . @xcite conducted completeness detailed tests and corrected the published counts . @xcite conducted more rudimentary checks on completeness and report the klf for the magnitude range they believe to be complete . @xcite and @xcite conducted no completeness tests using artificial stars , to the best of our knowledge , though @xcite claim the klf is complete . all the klfs discussed here are defined for the central clusters in the respective star forming regions , and none have been corrected for reddening . thus our comparison is only for the slopes of the klfs and does not account for possible differences resulting from differential reddening . the klf in w42 appears to be more similar to that in orion ( i.e. relatively steep ) than in m17 or the arches . given the likely presence of ysos as discussed above , we may be seeing the effects of pms evolution on the number counts as is the case for the trapezium . both the arches and m17 clusters clearly have more massive stars than are present in the central cluster of w42 or the trapezium @xcite . the apparent klfs in these latter two clusters appear at least superficially similar , though @xcite and @xcite reach rather different conclusions when converting to mass functions . @xcite derive an imf which is much flatter ( @xmath14 factor of 2 ) than @xcite , while @xcite find a mass function which is possibly consistent with @xcite . in 3.2 we classified the spectrum of the bright central star , w42 # 1 , as an early to mid o type ( @xmath37 o5o6 ) . several lines of evidence presented above suggest that the w42 cluster is quite young . if we take w42 # 1 to be on the zams then its apparent brightness would give a distance to the cluster of 2.2 kpc , considerably closer than the radio distance ( 3.7 kpc , see 3 ) . in 3.2 , we argued that the spectrum of w42 # 1 was most similar to those for the dwarf or giant luminosity classes . using the average @xmath55 from @xcite for dwarf stars and colors from 3 , gives 2600@xmath56 kpc . if the giant star @xmath55 from @xcite is used , then the distance estimate becomes 3400@xmath57 pc . the uncertainty in the distance estimate is completely dominated by the luminosity class assumed and the scatter in the intrinsic @xmath55 of o stars ( @xmath19 0.67 mag ) . the uncertainties in the reddening and apparent magnitude are negligible in comparison : a few percent for @xmath58 and 10 @xmath17 for @xmath51 , where the largest part of the uncertainty is in the choice of reddening law ( see mathis 1990 ) . these distance estimates are in agreement with @xcite who argued that w42 ( g25.4se ) was at the near distance given by the radio recombination line velocity and galactic rotation model . depending on the true luminosity of w42 # 1 , the cluster may even be somewhat closer . @xcite estimated the lyc luminosity of w42 to be 8.2@xmath59 assuming the far kinematic distance ( 13.4 kpc ) . adopting the zams distance ( 2.2 kpc ) as indicated by the young nature of the cluster ( stars w42 # 2 and # 3 ) considerably reduces the expected ionizing flux to 2@xmath60 . in figure [ cmdsub ] , we have plotted a zero age main sequence ( zams ) derived from the models of @xcite ( z @xmath4 0.02 ) ; see 3.1.3 . we have placed the zams in figure [ cmdsub ] at 2200 pc , assuming that the w42 # 1 is on the zams . in this case , the ( incomplete ) cluster sequence reaches early k type stars . for the trapezium @xcite claim that the klf shows an intrinsic peak well above the sensitivity of their @xmath3band images . this peak is due to the pms evolution of the lower mass stars and is sensitive to age on the pms . for the close distance to w42 implied if w42 # 1 is on the zams , the similarity found above for the klf of w42 compared to the trapezium then suggests a similar effect on the klf may be at work . i.e. , the fainter magnitudes may correspond to lower mass pms stars . in the present case , we are not claiming to see a real turn over in the klf ( as zinnecker et al . do in orion ) since this occurs where the @xmath3band counts are demonstrably incomplete due to crowding . for a distance of 2.2 kpc , the cluster has a radius of 0.32 pc at 30@xmath13 . the surface density in the inner 10@xmath13 ( figure [ surf ] ) is then within a factor of two for that in the trapezium ( * ? ? ? * @xmath14 6500 pc@xmath61 ) not accounting for the substaintial incompleteness in the present case . w42 is thus quite dense , and this is another indication of its young age @xcite . the zams clearly demonstrates the lack of any sensitivity to temperature in the hot stars at near infrared wavelengths . this means that traditional methods of cluster fitting ( in the observational color magnitude plane ) for age or distance can not be used and only through suitable numbers of infrared spectra ( resulting in and ) will the cluster properties truly emerge . the lack of temperature sensitivity in the near infrared colors does , however , lead to accurate estimates of the foreground extinction . we have presented high spatial resolution @xmath0 , @xmath1 , and @xmath2 images of the massive star cluster at the heart of the giant h ii region w42 . our @xmath3band spectra of three of the brightest four stars in the central 30@xmath13 of the cluster indicate a very young population . the brightest star w42 # 1 is classified as ko5o6 based on the system of @xcite . such stars are typically associated with mk o5o6.5 dwarfs . # 2 and w42 # 3 show no stellar absorption features . this fact , combined with their position in the excess emission band of the @xmath31 vs. @xmath29 color color plot , leads us to classify them as ysos . the klf was computed and compared to other massive star clusters . the klf in w42 appears more similar to that of the trapezium than to the more massive clusters in m17 and the arches . the steepness of the klf in the trapezium has been attributed to the pms stars present there , and it is possible that a similar effect is present in w42 given our spectroscopic and imaging evidence for ysos . our spectrum of w42 # 1 confirms the results of @xcite that w42 ( g25.4se ) can not be at the far kinematic distance . earlier estimates of the lyc output from this giant h ii region must therefore be revised downward ( by @xmath14 an order of magnitude ) . if w42 # 1 is on the zams , as we argue based on the presence of young stellar objects , then the cluster is even closer than the near kinematic distance ( 3.7 kpc ) . we estimate 2.2 kpc . in this case , w42 should not be considered a giant h ii region as defined in 1 . our spectrum of w42 # 1 is not sensitive to luminosity class , though it suggests a dwarf or giant classification . psc appreciates continuing support from the national science foundation . we wish to acknowledge the continuing excellent support received from the ctio mountain staff , particularly from night assistants hernn tirado , patricio ugarte , and alberto zuiga . lrrrrrrr o3 & 51230 & 89.7 & @xmath3010.28 & @xmath305.73 & @xmath304.80 & @xmath300.93 & @xmath300.05 + o4 & 48670 & 65.1 & @xmath309.74 & @xmath305.34 & @xmath304.41 & @xmath300.93 & @xmath300.05 + o4.5 & 47400 & 56.4 & @xmath309.50 & @xmath305.17 & @xmath304.24 & @xmath300.93 & @xmath300.05 + o5 & 46120 & 49.3 & @xmath309.24 & @xmath305.00 & @xmath304.07 & @xmath300.93 & @xmath300.05 + o5.5 & 44840 & 43.6 & @xmath308.97 & @xmath304.82 & @xmath303.89 & @xmath300.93 & @xmath300.05 + o6 & 43560 & 38.9 & @xmath308.70 & @xmath304.62 & @xmath303.69 & @xmath300.93 & @xmath300.05 + o6.5 & 42280 & 34.9 & @xmath308.41 & @xmath304.43 & @xmath303.50 & @xmath300.93 & @xmath300.05 + o7 & 41010 & 31.5 & @xmath308.12 & @xmath304.23 & @xmath303.30 & @xmath300.93 & @xmath300.05 + o7.5 & 39730 & 28.6 & @xmath307.84 & @xmath304.04 & @xmath303.11 & @xmath300.93 & @xmath300.05 + o8 & 38450 & 26.0 & @xmath307.55 & @xmath303.85 & @xmath302.92 & @xmath300.93 & @xmath300.05 + o8.5 & 37170 & 23.7 & @xmath307.28 & @xmath303.68 & @xmath302.75 & @xmath300.93 & @xmath300.05 + o9 & 35900 & 21.6 & @xmath307.01 & @xmath303.51 & @xmath302.62 & @xmath300.89 & @xmath300.05 + o9.5 & 34620 & 19.7 & @xmath306.75 & @xmath303.36 & @xmath302.49 & @xmath300.87 & @xmath300.05 + b0 & 33340 & 17.9 & @xmath306.49 & @xmath303.21 & @xmath302.36 & @xmath300.85 & @xmath300.05 + b0.5 & 32060 & 16.3 & @xmath306.23 & @xmath303.07 & @xmath302.28 & @xmath300.79 & @xmath300.04 + b1 & 21500 & 7.2 & @xmath303.52 & @xmath301.70 & @xmath300.94 & @xmath300.76 & @xmath300.04 + b2 & 18000 & 5.4 & @xmath302.31 & @xmath300.92 & @xmath300.25 & @xmath300.67 & @xmath300.04 + b3 & 15500 & 4.3 & @xmath301.39 & @xmath300.36 & 0.21 & @xmath300.57 & @xmath300.03 + b5 & 13800 & 3.6 & @xmath300.74 & 0.09 & 0.52 & @xmath300.43 & @xmath300.02 + b7 & 12200 & 3.0 & @xmath300.07 & 0.59 & 0.89 & @xmath300.30 & @xmath300.02 + b9 & 10600 & 2.5 & 0.70 & 1.15 & 1.29 & @xmath300.14 & @xmath300.01 + a0 & 9850 & 2.2 & 1.13 & 1.44 & 1.44 & 0.00 & 0.00 + a2 & 9120 & 2.0 & 1.59 & 1.76 & 1.63 & 0.13 & 0.01 + a5 & 8260 & 1.8 & 2.23 & 2.15 & 1.80 & 0.35 & 0.02 + a7 & 7880 & 1.7 & 2.55 & 2.42 & 1.97 & 0.45 & 0.02 + f0 & 7030 & 1.4 & 3.36 & 3.23 & 2.44 & 0.79 & 0.04 + f2 & 6700 & 1.3 & 3.72 & 3.61 & 2.68 & 0.93 & 0.05 + f5 & 6400 & 1.2 & 4.07 & 3.98 & 2.97 & 1.01 & 0.06 + f8 & 6000 & 1.1 & 4.58 & 4.52 & 3.40 & 1.12 & 0.06 + g0 & 5900 & 1.1 & 4.71 & 4.67 & 3.45 & 1.22 & 0.07 + g2 & 5770 & 1.1 & 4.89 & 4.86 & 3.46 & 1.40 & 0.08 + g5 & 5660 & 1.0 & 5.05 & 5.04 & 3.49 & 1.55 & 0.08 + g8 & 5440 & 1.0 & 5.37 & 5.41 & 3.81 & 1.60 & 0.09 + k0 & 5240 & 0.9 & 5.69 & 5.76 & 4.01 & 1.75 & 0.10 + k2 & 4960 & 0.8 & 6.15 & 6.30 & 4.05 & 2.25 & 0.13 +
the neutron lifetime @xmath1 is an important parameter in tests of the standard model of particle physics . it also affects the rate of helium production in the early universe and the energy production in the sun . the current particle data group ( pdg ) average is @xmath1 = 880.1@xmath2 1.1 s @xcite . however , the value of one experiment @xcite , which reported the lowest measurement uncertainty of @xmath3 0.8 s , is @xmath3 3.5 s lower than the bulk of other data in the pdg collection @xcite , that are grouped consistently around 882.0 s ( @xmath2 1.0 s ) @xcite . therefore the actual uncertainty of @xmath1 to be used in cosmological calculations may be of the order of 2 seconds or more . as a possible way of advancing this field , storage of polarized ultracold neutrons ( ucns ) in a magnetic trap has been pioneered by paul _ _ @xcite and is currently being pursued vigorously by several groups worldwide @xcite . one advantage of magnetic ucn storage versus storage in material bottles , the method used in a number of previous neutron lifetime measurements @xcite , is the potential absence of losses due to effects other than @xmath4-decay . there are no wall losses , the slow loss due to quasi - stable orbits is serious but believed to be manageable by avoiding regular orbits @xcite , and the potential loss due to depolarization , defined as spin flip relative to the local field direction , is commonly assumed to be negligible . for systems using permanent magnets the question of gradual demagnetization over time appears to have found little attention so far . until recently ucn depolarization estimates @xcite were based on majorana s quasi - classical result of 1932 @xcite for a free polarized particle with magnetic moment moving with constant velocity vector through a non - uniform static magnetic field of specific form . only its spin state was assumed to be affected by the magnetic field . this model predicted a depolarization probability @xmath5 for one passage through the field . this value decreases exponentially with the adiabaticity parameter @xmath6 , where @xmath7 is the frequency of rotation of the field as seen from the reference frame of the moving particle , in the critical region where the field rotates fastest while the magnitude @xmath8 of the magnetic field may be small . @xmath9 is the larmor frequency . for magnetic field parameters as currently used or proposed for ucn storage , @xmath10 would be of order @xmath11 , thus immeasurably small . recently , walstrom _ et al . _ @xcite pointed out that the values of @xmath10 for confined , rather than freely moving , neutrons are much larger . for a ucn moving along a vertical path in the storage system proposed by them , @xmath10 was estimated to be in the range @xmath12 to @xmath13 . this is much larger than the majorana value but still negligible in any actual or projected neutron lifetime experiment . using a simplified model of magnetic field distribution we extend the theory of @xcite to include arbitrary ucn motion with both vertical and horizontal velocity components , confined to the vertical space between upper and lower turning points that depend only on the ucn energy for vertical motion . in our model ( introduced in sec . [ sec : ii ] ) the magnetic field magnitude @xmath8 is uniform within any horizontal plane , so there is no horizontal component of magnetic force . therefore the neutron moves with constant velocity in the horizontal @xmath14- and @xmath15- directions . we show that @xmath10 could reach a level approaching the tolerance limit for a high precision neutron lifetime measurement unless precautions are taken . as is well known the most critical issue is the choice of a stabilization field perpendicular to the magnetic mirror field , of sufficient strength so that the depolarization rate will be negligible in a neutron lifetime experiment . our model field is close to the `` bathtub configuration '' of ref . @xcite but the lateral confinement of ucns , achieved there by double curvature of the magnetic mirror , is simulated differently . the magnetic mirror is horizontal and extends to infinity in both lateral dimensions . however , one could imagine the presence of ideal vertical mirrors reflecting the ucns back and forth in the horizontal directions without any change in the analysis . more specifically , we use an infinite ideal planar halbach array @xcite , which is free of the field ripples present in actual realizations @xcite . in the design of ref . @xcite , the ripples are important only within about 1 mm of the surface of the magnets . this region is not reached by the ucns whose maximum energy ( for vertical motion ) is @xmath345 nev ( for the parameters in @xcite ) , since they reverse their flight direction before entering this zone . using this model of field distribution we have also studied the problem of depolarization of ucns in reflection from a non - magnetic mirror immersed in a non - uniform magnetic field . this question is important as a mechanism that may limit the efficiency of ucn polarizers based on transmission through a magnetic field . for a sufficiently strong field , neutrons in only one spin state can pass the field to proceed to the experiment . otherwise they are reflected . however , following the polarizer the ucns are usually reflected on trap or guide walls exposed to the stray field of the polarizer and thus may lose their 100% polarization if the reflection process involves depolarization . moreover , a possible depolarization on non - magnetic trap walls in a magnetic field is highly relevant in measurements of the neutron decay asymmetry parameter @xmath0 using ultracold neutrons @xcite . this problem has first been investigated by pokotilovski @xcite who used an adaptation of the majorana model to the reflection geometry . in the present work we study certain aspects of this problem by imagining a horizontal lossless non - magnetic ucn mirror inserted at a variable height into the magneto - gravitational storage space . the net depolarization per bounce on this mirror will be compared to the depolarization for one bounce in the magnetic field in the absence of the mirror to obtain an estimate for the depolarization effect of the mirror . the topic of ucn depolarization in magnetic storage or in mirror reflection in a magnetic field raises interesting questions of quantum interpretation . fig . 3 of ref . @xcite and our fig . [ fig : two ] ( to be discussed in sec . [ sec : iii.b.2 ] ) show the probability for the neutron to be in the spin - flipped state ( relative to the field direction ) as a function of position of the neutron as it moves through the magnetic storage space . the curve is strongly peaked at the critical level where the field rotates fastest in the reference frame of the moving neutron . this behavior is the same as displayed by the majorana result @xcite ( where it is more difficult to deduce since the author used a quantization axis fixed in space rather than rotating together with the field ) . in a semi - classical interpretation , as the ucn starts moving from one turning point , say the upper one , down toward the lower one , the spin vector rotates away from the quantization axis ( which was chosen parallel to the local magnetic field vector in ref . it reaches a certain maximum angle around the critical zone ; then this rotation is reversed and ends at a much smaller value at the next turning point for ucn motion . this indicates that an analyzer of neutron polarization placed at different heights would show a variation of depolarization by many orders of magnitude ( @xmath16 8 decades for the example shown in fig . 3 of ref . @xcite ) over the vertical range of the storage space . the depolarization rate expected for an actual ucn magnetic storage experiment , without any polarization analyzer intersecting the beam , is determined by the current of ucns in the `` wrong '' spin state , i.e. of high - field seekers leaving the system at the lower and upper turning points while the `` correct '' ( high - field repelled ) state is reflected and returns to the storage space . this association of net depolarization with loss currents is consistent with the following interpretation : at the turning points a measurement is performed ( in the sense of quantum mechanics ) , conceptually by neutron detectors placed just below the bottom and just above the top of the storage region for a given ucn energy for vertical motion . these detectors would intersect the ucns in the `` wrong '' spin state as they exit the storage system . in the copenhagen interpretation , such a measurement ( actual or hypothetical ) resets the ucn wave function to a pure state of high - field repelled neutrons . the spin state then evolves as described by the spin - dependent schrdinger equation ( or its semi - classical analog ) until the next `` measurement '' takes place at the following turning point and the process of wave collapse and wave evolution is repeated . alternative interpretations are conceivable but we will use the picture outlined above . following ref . @xcite we use the wentzel - kramers - brillouin ( wkb ) approximation to solve the spin - dependent schrdinger equation . this appears justified since the spatial variation of field variables ( gravitational potential and magnetic field * b * ) is much slower than the variation of ucn wave function . the scales are of order cm for gravity and * b * , and of order @xmath17 m or less for the neutron wavelength . we are aware of the fact that an exact treatment of ucn depolarization in magnetic storage may involve quantum electrodynamics since the moving neutron , in its reference frame , is affected by a time - dependent electromagnetic field , i.e. by low - energy photons . we will also neglect temporal fluctuations of the field due to mechanical vibrations or , if electromagnets are used for field generation , ac components of the current supply . we are not aware of any work on time - dependent effects of this kind in magnetic ucn storage . as it rotates in the ( @xmath18)-plane . its magnitude @xmath19 decreases exponentially with height @xmath20 and is represented by the arrow length using a log scale . the angle @xmath21 of the halbach field is also shown . the superimposed stabilization field @xmath22 in the @xmath14-direction increases slowly with @xmath20 as in ref . @xcite and is symbolized by the crosses of variable size.,width=291 ] we consider a halbach array @xcite of permanent magnets of thickness @xmath23 covering the infinite @xmath24-plane , where the @xmath15-axis points to the right , the @xmath20-axis points up ( against gravity ) and the @xmath14-axis toward the front ( fig . [ fig : one ] ) . we choose @xmath25 at the upper magnet surface and will closely follow the description in ref . @xcite , apart from this choice of system of coordinates . this choice will allow us to use the pauli matrices in their standard form . in the limit of infinitely fine division of magnet blocks in the @xmath15-direction , let the magnetization vector have a constant magnitude @xmath26 but , viewed along the positive @xmath14 direction , rotate clockwise in the @xmath27-plane with periodicity @xmath28 in @xmath15-direction : @xmath29 using a complex quantity @xmath30 , eq . ( [ ( 1 ) ] ) can also be written @xmath31 . @xmath32 and @xmath33 are unit vectors . we choose the same parameters as for the design in ref . @xcite , thus @xmath34 = 5.2 cm and @xmath35 @xmath36 . the special feature of the halbach system is that it generates a magnetic field only on one side of the magnets , namely in the half space @xmath37 for clockwise sense of rotation . this halbach field is @xmath38 or , in complex notation , @xmath39 . @xmath40 is determined by the remanent field @xmath41 and the block depth @xmath23 = 2.54 cm . the magnitude of @xmath42 , @xmath43 , only depends on the vertical coordinate @xmath20 . the field distribution is shown schematically in fig . [ fig : one ] . in the actual scheme @xcite , the uniform rotation is replaced by dividing the rotation period @xmath34 into four blocks , each of length @xmath44 and with the same magnetization @xmath45 , but with an angle of 90@xmath46 between the directions of @xmath47 in adjacent blocks ( schematically represented as @xmath48 . alternative designs are in the form of vertical or horizontal cylinders where the magnets are assembled along the cylindrical surface : in a dipolar way in @xcite , with adjacent blocks magnetized in the peripheral direction with equal magnetic poles facing each other ( schematically : @xmath49 ) . in project @xcite a cylindrical octupole halbach magnet is used where 32 blocks are distributed uniformly over the perimeter and the direction of magnetization advances by 56.25@xmath50 from one block to the next . in these systems the magnitude @xmath8 of magnetic field increases sharply near the magnetic wall . a superconducting quadrupole system of magnetic ucn storage for a neutron lifetime experiment is used in ref . @xcite while in ref . @xcite the ucn are stored in superfluid helium using a ioffe system with horizontal electromagnetic quadrupole . for a quadrupole the field magnitude increases linearly with radial distance from the axis . at least one magnetic end cap is required for all cylindrical systems ; on the upper side of vertical systems gravitational confinement can be used . the field distribution for systems using permanent magnets with magnetization direction advancing in steps from block to block may be expressed as a fourier series , as in eq . ( 7 ) of ref . @xcite . the first term of the expansion is dominant and has the form ( [ ( 2 ) ] ) with constant @xmath51 somewhat smaller than @xmath52 . for the planar quadrupole halbach system the reduction factor is @xmath53 @xcite and the field in the lower half space @xmath54 no longer vanishes . the higher fourier components generate a ripple field in the @xmath27-plane , which is significant within @xmath55 mm from the magnet surface ( and even induces a logarithmic divergence in the field gradient within @xmath56 m ) . but this space is not accessible to the ucns if we choose a spectrum soft enough to ensure that all neutrons approaching the magnet from above are reflected back up before reaching the ripple zone . we use the value of 0.64 t for the magnetic field at a safe distance 2 mm , which corresponds to @xmath57 t at the surface . thus , for the halbach array generated field we assume the form ( [ ( 2 ) ] ) with @xmath57 t and strict confinement of the vector @xmath42 in the @xmath27-plane . we ignore the small field ripple in the @xmath14-direction considered in @xcite since it also decays strongly with distance from the magnet surface . these small perturbations are not expected to affect the depolarization results obtained below in a significant way . our value of @xmath58 is @xmath59% lower than the design value of @xcite to take into account partial demagnetization , over time , of the ndfeb permanent magnets exposed to large fields . a common feature of the various magnetic ucn storage schemes is the requirement of a bias field @xmath60 perpendicular to the main field . it ensures that the field magnitude @xmath61 exceeds a certain minimum value everywhere in the storage volume , especially at critical positions where the field rotates fast in the neutron s moving reference frame . the main purpose of the present work is to provide an estimate of this minimum field for typical field parameters , as those in ref . @xcite where the field @xmath60 also serves the purpose of guiding the decay electrons out of the storage space to a detector as a way to monitor the neutron decay rate in real time . @xmath60 is generated by a toroidal electromagnet , and it is oriented along the longitudinal direction of the `` bathtub surface '' which corresponds to the @xmath14-direction in our model with a planar , rather than curved halbach magnet . we use the same @xmath20-dependence as in ref . @xcite , @xmath62 with @xmath63 m. the magnitude of @xmath64 is uniform on a horizontal plane , and since the halbach field magnitude @xmath19 is uniform for given height @xmath20 no horizontal force acts on the stored ucn . in ref . @xcite a value of 0.05 to 0.1 t was proposed for @xmath65 . we will consider field strengths down to the mt range since this range seems to be closer to the bias field used in ref . @xcite . in this latter work the value used was not given but it was estimated , on the basis of the majorana formula , that a minimum field @xmath66 of @xmath30.001 t was required for their neutron lifetime measurement . our analysis of the evolution of spin - flip probability for ucns moving in our model magnetic field distribution differs from that of ref . @xcite as follows . it is not restricted to purely vertical motion but assumes that the ucns can also have arbitrary horizontal velocity components @xmath67 and @xmath68 . @xmath67 and @xmath68 are constant since neither gravity nor the net magnetic field @xmath69 of our model exert a horizontal force on the neutron . as noted earlier , both @xmath19 and @xmath66 are uniform at given height @xmath20 , and @xmath42 is perpendicular to @xmath60 , thus @xmath70 depends on @xmath20 only . as a result , the magnetic force , which is determined by the gradient of @xmath71 , has no horizontal components and the equation of motion is separable in three dimensions . we use three different methods of analysis , both for purely magnetic confinement of ucns with arbitrary 3d velocity components , and for a system involving a non - magnetic mirror placed into our model magnetic field distribution : ( a ) in sec . [ sec : iii ] a quantum approach using the wkb approximation , ( b ) in sec . [ sec : iv ] a quasi - classical approximation and ( c ) in sec . [ sec : v ] direct numerical integration of the equations of motion . in the quantum approach the stationary spin - dependent schrdinger equation is solved using the wkb approximation , extending the method of ref . @xcite to 3d motion . the semi - classical majorana method @xcite which was adapted to magnetically confined ucns in ref . @xcite will also be extended to 3d . using both methods we will also analyze ucn reflection on a non - magnetic mirror in a magnetic field . in sec . [ sec : v ] we will show that some analytic results can be obtained as well by direct numerical integration of the wave equation for the spin - flipped component of the wave function . this is only feasible because the magnetically trapped ucns have relatively long wavelengths , in the @xmath17 m range , so the number of wave oscillations for the entire integration path is not too large . the wave function for a ucn moving in the magneto - gravitational field of the trap is a linear superposition of the two eigenstates of the magnetic moment interaction hamiltonian @xmath72 where @xmath73 is the neutron magnetic moment in terms of the nuclear magneton @xmath74 j / t , @xmath75 is the pauli spin operator , and @xmath76 is the local magnetic field . the two eigenstates @xmath77 and @xmath78 of @xmath79 satisfy the eigenvalue equations @xmath80 and correspond , respectively , to neutron spin parallel to @xmath81 with spin energy @xmath82 , and to antiparallel spin with energy @xmath83 . these spin eigenfunctions are obtained by spin rotation from the @xmath14-axis to the direction of @xmath76 through angles @xmath84 and @xmath85 . the polar field angle is @xmath86 , where @xmath87 is due to the bias field @xmath88 and @xmath89 is the magnitude of the halbach field @xmath42 . the azimuthal angle in the @xmath27-plane is @xmath90 . ( [ ( 2 ) ] ) shows that for the halbach field configuration @xmath91 thus @xmath85 depends only on @xmath15 ( not on @xmath20 or @xmath14 ) , while @xmath84 depends only on @xmath20 . these properties will simplify the analysis considerably . exact correspondence between the system of coordinates @xmath92 used here and the system @xmath93 used in ref . @xcite ( with @xmath94 pointing up ) is established if we add the constant @xmath95 to the right - hand side of eq . ( [ ( 5 ) ] ) . performing the spin rotation through angles @xmath84 and @xmath85 we obtain for the spin basis vectors @xcite with quantization axis along @xmath81 @xmath96 where @xmath97 , @xmath98 and @xmath99 . we write the dependence of the wave function on position and spin in the form @xmath100 where we have used the superscript ( 3 ) to indicate that @xmath101 are functions of the three space coordinates while the corresponding functions @xmath102 and @xmath103 , introduced below , depend on @xmath20 only . @xmath104 satisfies the eigenvalue equation @xmath105 \chi\ ] ] for a neutron of mass @xmath106 with constant total energy @xmath107 moving in a uniform gravitational field of magnitude @xmath108 and a non - uniform magnetic field @xmath109 . using subscripts to denote partial differentiation , the laplace operator acting on the wave function gives @xmath110 where for the second and third term the indicated permutations are performed . the basis vectors for quantization along the fixed @xmath14-axis can be expressed in terms of the basis vectors @xmath77 and @xmath78 : @xmath111 using eq . ( [ ( 10 ) ] ) and noting that , from ( [ ( 5 ) ] ) , @xmath112 , @xmath113 , @xmath114 and also @xmath115 , we obtain @xmath116 as in ref . @xcite we will use the wkb approximation @xcite and keep only those terms in eq . ( [ ( 9 ) ] ) that contain the derivatives of the field variables ( @xmath84 and @xmath85 ) in lowest order since those change on the scale of centimeters while the waves in real space , @xmath117 and @xmath118 , vary on the micrometer scale , i.e. @xmath119 times faster . this implies that all second derivatives of @xmath77 and @xmath78 are dropped , along with other small terms . ( using the numerical integration described in sec . [ sec : v ] we have performed test runs where the small terms were retained . the results were the same within the precision of numerical integration . ) assuming that the ucn started out from a pure ( + ) spin state and keeping only the dominant terms in eq . ( [ ( 9 ) ] ) we obtain @xmath120 \chi^- , \end{aligned}\ ] ] where we have used @xmath121 , @xmath122 , and the fact that in practice @xmath123 . the functions in real space multiplying @xmath77 and @xmath78 can be simplified by noting that the @xmath15 and @xmath14 dependence of @xmath117 has the plane wave form @xmath124 and @xmath118 is proportional to @xmath125 . the wave numbers @xmath126 and @xmath127 are constant and @xmath128 represents a bloch - wave modulation due to the periodicity of the halbach field . in practice , @xmath126 and @xmath127 are of order @xmath17m@xmath129 , thus much larger than @xmath130 and @xmath131 , both of which are of order @xmath36 . thus we can factor eq . ( [ ( 12 ) ] ) in the form @xmath132\chi^+ \\ & + \mathrm{e}^{-ikx}[\beta '' - ( k_x^2+k_z^2)\beta - ( \theta'\alpha'+kk_x\alpha\sin\theta)]\chi^-\},\nonumber\end{aligned}\ ] ] simplifying the notation . in eq . ( [ ( 13 ) ] ) and henceforth , @xmath102 and @xmath103 stand for the @xmath20-dependent parts of the wave function only , and differentiation with respect to @xmath20 is denoted by primes . we also drop the subscript @xmath20 from the @xmath20-component of the wave vector . thus , @xmath133 and @xmath134 . the terminology @xmath135 conforms to that used in ref . @xcite where motion in horizontal directions was not taken into account in the depolarization calculations . inserting eq . ( [ ( 13 ) ] ) into the eigenvalue equation ( [ ( 8) ] ) gives @xcite two coupled equations , one for spinor @xmath77 ( i.e. , for low - field seeking ucns with spin parallel to @xmath136 , which can be stored ) and the other for @xmath78 ( i.e. , for the fraction of ucns whose spin has flipped relative to @xmath136 and which therefore can escape from the trap ; the probability of flipping twice is negligible ) : @xmath137+mgy\alpha+|\mu_{\textrm{n}}|b\alpha\ ] ] and @xmath138 \nonumber \\ & + mgy\beta-|\mu_n|b\beta.\end{aligned}\ ] ] in the framework of the wkb approximation , the solution of ( [ ( 14 ) ] ) is @xcite @xmath139 where @xmath140 @xmath141 is the magnitude of the @xmath20-component of local wave vector for the storable ( + ) spin state ( parallel to @xmath81 ) and @xmath142 is that for the ( @xmath143 ) spin state . in eq . ( [ ( 17 ) ] ) , @xmath144 is the greatest height a neutron of energy @xmath107 and given @xmath126 and @xmath127 would achieve in the gravitational field if the magnetic field were switched off . in eq . ( [ ( 16 ) ] ) , @xmath145 is the phase angle , for the + spin state , accumulated between the start of vertical motion and the position @xmath20 . the initial height @xmath146 for motion upward is assumed to be that of the lower turning point , thus @xmath147 , and for motion downward the initial level is taken at the upper turning point , @xmath148 . the additional + or @xmath143 sign in the argument of the exponential function in ( [ ( 16 ) ] ) , in front of @xmath149 , refers to this direction of the motion ; plus for upward and minus for downward , as in @xcite . the wkb wave function ( [ ( 16 ) ] ) is normalized to a constant particle flux @xmath150 in the @xmath20-direction . for the spin - flipped ucns , the flux in the @xmath20-direction is the measure of the probability of depolarization , as shown below . at the classical turning points , where @xmath151 , the wkb form ( [ ( 16 ) ] ) diverges and has to be replaced by the airy function , as shown in @xcite , but the wkb form is still valid almost all the way to the turning point , except for the last @xmath17 m or so , since it correctly represents the asymptotic behavior of the airy function in this region this is an important feature of the approximation used in ref . @xcite and in the present work . it is made more explicit as follows : the asymptotic form of the airy function in the region of real waves ( rather than the exponentially decaying wave on the other side ) is @xcite @xmath152 . \nonumber\end{aligned}\ ] ] the wave number @xmath153 is determined by the local acceleration @xmath154 at the turning point @xmath146 , and @xmath155 . for upward motion the wkb wave approximation ( [ ( 16 ) ] ) for @xmath102 , which is valid between the turning points , is matched to the first term inside the braces of ( [ ( 19 ) ] ) , and for downward motion to the second term by adjusting the constant multiplying the airy function . it follows from eq . ( [ ( 15 ) ] ) that the wave function @xmath156 for the spin flipped component is determined by the inhomogeneous second - order differential equation @xmath157 having separated off the @xmath15 and @xmath14 dependence allows us to choose the same wkb form for @xmath103 as in ref . @xcite : @xmath158 where the function @xmath159 modulating the wkb wave represents the amplitude of spin flip . apart from the modulation @xmath159 , @xmath103 is constructed in the same way as @xmath102 . the phase accumulated since the start at a turning point , @xmath160 always has a larger magnitude than the phase @xmath161 for @xmath102 since @xmath162 is greater than @xmath163 ( except in zero magnetic field ) . summarizing , the governing equation for @xmath103 is the second - order differential equation @xmath164 \alpha(y),\end{aligned}\ ] ] where the right - hand side represents an inhomogeneous term and @xmath102 is given by eq . ( [ ( 16 ) ] ) while @xmath103 has the form @xmath165 in the second expression on the right - hand side of ( [ ( 23 ) ] ) we have carried out the differentiation of @xmath102 , using the wkb rule of considering the slow - varying terms as constant , with the result @xmath166 where the + sign applies to upward motion and the @xmath143 sign to downward motion . this replacement is valid except within a few @xmath17 m of the turning points . ( [ ( 23 ) ] ) is the same as eq . ( 28 ) of ref . @xcite except for the additional , @xmath126 dependent term on the right - hand side . it is present because we include motion with finite lateral momentum @xmath167 , whereas the analysis in ref . @xcite was restricted to the special case @xmath168 . we will show that this new term makes the major contribution to ucn depolarization in magnetic storage . we also note that eq . ( [ ( 23 ) ] ) does not depend on @xmath127 , thus motion exactly along the @xmath14-direction does not induce depolarization . this is understandable since neutrons moving along the @xmath14-axis during the short periods of horizontal motion at the turning points move in a uniform @xmath81-field . on the other hand , in motion along the @xmath15-direction they are exposed to the strong field ripple due to the rotating halbach field . these features are expected to hold , on a qualitative basis , also for the `` bathtub system '' of ref . @xcite , where the direction perpendicular ( parallel ) to the curved halbach array corresponds to our @xmath14-axis ( @xmath15-axis ) . next we will solve eq . ( [ ( 23 ) ] ) to obtain the depolarization rate for ucn storage in our model field distribution . we will first consider a neutron of given energy and fixed values of @xmath126 and @xmath127 , moving downward from the upper turning point @xmath169 . using the wkb rule , the second derivative @xmath170 is obtained from ( [ ( 24 ) ] ) as @xmath171 inserting @xmath172 and @xmath102 from ( [ ( 16 ) ] ) into eq . ( [ ( 23 ) ] ) gives @xmath173 \exp\big(-i\phi(y)\big),\ ] ] where @xmath174 and @xmath175 . defining a function @xmath176 through @xmath177 we obtain from ( [ ( 26 ) ] ) @xmath178 \exp\big(-i\psi(y)\big),\ ] ] where @xmath179 . the phases @xmath180 and @xmath181 are fast - varying quantities , while the field variables @xmath182 and @xmath183 vary slowly with @xmath20 and will therefore be considered constant in all differentiations . in integrations , as needed to obtain @xmath176 from eq . ( [ ( 27 ) ] ) , this `` wkb rule '' directly corresponds to performing the integral of products of slow- and fast varying terms by parts and neglecting the second term which contains the derivative of the slow - varying factor . it has been shown numerically in ref . @xcite that for the parameters of magnetic ucn storage at hand this procedure gives approximations with precision in the range @xmath184 . thus we obtain from ( [ ( 27 ) ] ) @xmath185\exp(-i\psi')dy ' \nonumber \\ & = \int_{\psi_s}^\psi \frac{iu(y')+v(y')}{k_-+ k_+ } \exp(-i\psi ' ) d\psi ' \\ & = \frac{i[iu(y ) + v(y)]}{k_-(y ) + k_+(y ) } \exp\big(-i\psi(y)\big ) . \nonumber\end{aligned}\ ] ] the third step in ( [ ( 28 ) ] ) is an integration by parts , where only the leading term is kept . the lower limit of the @xmath20-integration in ( [ ( 28 ) ] ) is the upper turning point . carrying out the integration in the last step of eq . ( [ ( 28 ) ] ) we should expect a contribution from this lower limit of integration ( @xmath186 or @xmath187 ) . however , such a term does not appear in ( [ ( 28 ) ] ) for the following reasons : first we note that in the quantum treatment involving the airy function the turning `` point '' is blurred within a range of order @xmath17 m . second , as mentioned in sec . [ sec : i ] , we assume , as the authors of ref . @xcite did , that at a turning `` point '' ( here the region around @xmath169 ) the neutron starts out in a pure low - field seeking spin state ( + ) , i.e. from @xmath188 , @xmath189 . this implies that @xmath102 and the functions @xmath190 and @xmath176 , derived from @xmath191 and @xmath192 through eq . ( [ ( 23 ) ] ) , tend to 0 as @xmath193 ( in practice , for @xmath20 just a few @xmath17 m above the classical turning point ) . it was also mentioned earlier , that the wkb function used here for @xmath102 is just the asymptotic representation of the airy function @xmath194 which does satisfy the initial condition without any constant added since @xmath194 and @xmath195 vanish for @xmath193 . as a result , there is no lower - limit contribution to @xmath196 in ( [ ( 28 ) ] ) , and the same is true also for the functions @xmath159 and @xmath103 derived below by further integration ( in equations ( [ ( 29 ) ] ) and ( [ ( 32 ) ] ) ) . we can also argue that , due to the factor @xmath197 in @xmath198 ( defined following ( [ ( 26 ) ] ) ) , a wave containing a term derived from a finite integration constant in ( [ ( 28 ) ] ) would diverge at the endpoint @xmath199 of integration , where @xmath200 , and therefore must be zero , in the same way as in total reflection a wave increasing exponentially inside the medium must have amplitude zero . the singularity is avoided only by setting the integration constant in ( [ ( 28 ) ] ) equal to zero . remembering the definition @xmath201 , we integrate eq . ( [ ( 28 ) ] ) once more to obtain @xmath202 in ( [ ( 29 ) ] ) , we have defined @xmath203 it follows from the definition of @xmath163 and @xmath162 in eq . ( [ ( 17 ) ] ) that @xmath204 depends only on the magnitude @xmath205 of the local magnetic field . from the symmetry of the problem it follows that motion in the opposite direction , from the lower turning point at @xmath199 upward to @xmath169 gives the same function @xmath159 as in the last form of eq . ( [ ( 29 ) ] ) , except that the phase term @xmath206 is replaced by @xmath207 and @xmath208 is replaced by its complex conjugate @xmath209 . @xmath210 is a phase factor of unit amplitude , which arises due to the shift of reference point for @xmath211 from @xmath169 to @xmath199 when we change from downward to upward motion . none of these differences affect the squared magnitude @xmath212 which measures the probability to find the neutron in the spin - flipped state at height @xmath20 . an explicit expression for @xmath210 will be given following eq . ( [ ( 44 ) ] ) . we can compare our eq . ( [ ( 29 ) ] ) for the depolarization amplitude @xmath213 with the corresponding result in equations ( 31 - 35 ) of ref . @xcite , where only the case @xmath168 was analyzed . this corresponds to setting @xmath214 in our analysis . apart from this difference , our eq . ( [ ( 29 ) ] ) can be obtained from eq . ( 35 ) of @xcite by multiplying the latter by the factor @xmath215 . the magnitude of this factor is close to unity if @xmath141 is only slightly less than @xmath142 . for fairly high - energy ucns this is the case for most of the path between the turning points , but not near these points . this minor difference appears to be due to the neglect , in ref . @xcite , of @xmath216 in the derivation of their eq . ( 31 ) . the main difference between the results of @xcite and our numbers , obtained below , is due to the restriction of the previous work to @xmath168 . our analysis of the depolarization current and depolarization rate for magnetically stored ucns yields a loss @xmath310 decades larger for a typical velocity @xmath67 up to @xmath217 m / s than the range of values , @xmath218 to @xmath13 given in ref . @xcite for @xmath219 . we can now insert @xmath159 from ( [ ( 29 ) ] ) into ( [ ( 24 ) ] ) to determine the wave function @xmath103 for spin - flipped ucns . for the downward motion , this gives @xmath220 the phase @xmath149 ( with the index @xmath221 ) indicates that this wave for the ( @xmath143 ) spin state propagates , not with wave number @xmath162 , but with the same wave number @xmath163 as the ( + ) spin state , as it should . using @xmath222 we can verify that the function ( [ ( 32 ) ] ) solves equation ( [ ( 23 ) ] ) , starting from a pure ( + ) spin state , as described following eq . ( [ ( 28 ) ] ) . on a more formal basis , eq . ( [ ( 32 ) ] ) represents a particular solution to ( [ ( 23 ) ] ) and we could add to ( [ ( 32 ) ] ) any solution @xmath223 of the homogeneous equation @xmath224 corresponding to ( [ ( 23 ) ] ) . in the wkb framework , these solutions are @xmath225 with arbitrary constants @xmath226 . these functions represent a constant current in the upward ( downward ) direction for the + ( @xmath143 ) sign . thus the same current enters and leaves the storage space , resulting in a zero contribution to the net flux out which corresponds to the depolarization loss as described below . as an example , such a homogeneous term may represent a neutron that has undergone a spin flip on the way up , proceeds past @xmath169 until it reaches its reversal point in the gravitational field and , on its way down , traverses the storage space without contributing to further depolarization . reverting to solution ( [ ( 32 ) ] ) without added terms , we associate the net depolarization over the path from upper turning point @xmath169 to @xmath199 with the current of spin - flipped ucns at the endpoint @xmath199 , which is consistent with the interpretation in ref . this current represents the net flux out of the storage space since no flux enters at @xmath169 . ) multiplied by @xmath227 , as a function of neutron position for drop heights @xmath144 = 450 mm and 100 mm , stabilization field parameter @xmath228 = 0.005 t , and neutron velocity component @xmath67 = 3 m / s or zero . the sharp peak occurs in the region where the gradient of field angle @xmath84 is largest.,width=291 ] at an arbitrary position @xmath20 along the way the current @xmath229 is given by @xcite @xmath230.\ ] ] for function ( [ ( 32 ) ] ) we have @xmath231 and thus the current as a function of position @xmath20 between @xmath199 and @xmath169 becomes @xmath232 the function @xmath233 corresponds to the depolarization probability of ref . it is plotted in fig . [ fig : two ] for ucns with energy for vertical motion determined by the `` drop heights '' @xmath234 cm and 45 cm ( @xmath144 was defined following eq . ( [ ( 17 ) ] ) ) , a bias magnetic field @xmath228 = 0.005 t and @xmath67 = 3 m / s . as in fig . 3 of @xcite we see a sharp peak at the @xmath20-position where @xmath235 is large , and a decrease as the particle drops further down . the third curve in fig . [ fig : two ] is for @xmath144 = 45 cm , @xmath228 = 0.005 t and @xmath67 = 0 . the peak value and the decrease on the upper side are quite similar . below the peak position the curve for @xmath67 = 0 decreases faster than for @xmath67 = 3 m / s . the current leaving the storage space at @xmath236 is @xmath237 @xmath238 where the index @xmath239 refers to the values at @xmath236 and the last form of eq . ( [ ( 35 ) ] ) uses the fact that @xmath163 vanishes at the turning points . in the analysis of ref . @xcite for @xmath168 only this vanishing term ( @xmath240 ) appeared , and a higher order of approximation as well as numerical integration were used to estimate the depolarization probability . the result was a very small value which is negligible compared to the second term given in the last form of ( [ ( 35 ) ] ) , even for values of @xmath67 as small as 0.1 m / s . the dependence of ( [ ( 35 ) ] ) on the primary field variables is established by noting that @xmath241 , @xmath242 and @xmath243 , where @xmath244 is the frequency of the halbach field as seen by the moving ucn . for upward motion from @xmath199 to @xmath169 we get the same result for the current as in ( [ ( 35 ) ] ) except that all indices @xmath239 are replaced by @xmath245 , i.e. , the quantities relevant for the spin - flipped current leaving the system at the upper turning point are determined by the field angle @xmath246 and by @xmath247 at @xmath169 . the combined depolarization loss for one reflection on the magnetic field , i.e. for one complete round trip down and up , is determined by @xmath248 to approximate the actual situation in magnetic ucn storage , where the ucns have positive and negative velocities in any direction and , for a low - energy maxwell spectrum , with uniform probability per unit of @xmath249 and @xmath127 ( since the phase space density is constant ) , we take the mean value of @xmath250 in ( [ ( 36 ) ] ) for the spectral interval @xmath251 with the result @xmath252 as a final step in this analysis of depolarization in the wkb approximation we establish the explicit connection between the loss current ( [ ( 37 ) ] ) and the rate of depolarization , @xmath253 , that is observable as a contribution to the decay rate ( but should be negligible compared to the neutron @xmath4-decay rate in a neutron lifetime measurement ) . for given neutron energy for vertical motion , i.e. fixed turning levels at @xmath199 and @xmath169 , the depolarization rate ( in s@xmath129 ) is determined by the loss current ( [ ( 37 ) ] ) divided by the number of ucns in the field - repelled spin state present in the trap , @xmath254 we have used the square magnitude of the wkb form ( [ ( 16 ) ] ) for @xmath102 as the density . the factor 2 takes into account that both downward and upward moving ucns are in the trap at the same time . since @xmath255 and @xmath256 , the expression in ( [ ( 38 ) ] ) equals @xmath257 where @xmath258 is the time required for one round trip down and up . thus , the depolarization rate is @xmath259 where we have inserted ( [ ( 37 ) ] ) for the current in the last step . this shows that the loss current ( [ ( 37 ) ] ) of spin - flipped ucns is the loss per round trip , i.e. for one bounce in the magnetic field . this interpretation is consistent with the interpretation in ref . @xcite . to compare to actual experiments storing polarized ( + ) ucns in a broad velocity range in three dimensions , we have to average ( [ ( 39 ) ] ) also over @xmath68 and @xmath260 . averaging over @xmath68 is trivial since ( [ ( 39 ) ] ) does not depend on @xmath68 . averaging over @xmath260 can be achieved as follows . as a measure of @xmath260 for a stored ucn we could take its value at any height within the confinement range , but the most convenient choice of reference plane is the neutral plane at @xmath261 where the gravitational force is compensated by the magnetic force pushing upward , i.e. where @xmath262 . this is the plane where the ucns with the lowest energy for vertical motion reside . in our field model , a ucn with vertical velocity @xmath263 in the neutral plane floats or moves along the plane at constant speed . in actual confinement fields as in @xcite they would follow closed or open paths on the curved neutral surface . for small values of @xmath264 the vertical motion is a classical harmonic oscillation with natural frequency @xmath265 where @xmath266 is the net downward acceleration . this implies that for small oscillations about the neutral plane the time for a round trip becomes @xmath267 . for larger vertical velocities the oscillator potential is strongly anharmonic but the drop height @xmath144 , used originally as a measure of energy for vertical motion , is unambiguously determined by @xmath268 . therefore , if we plot the depolarization rate ( [ ( 39 ) ] ) versus @xmath268 , rather than @xmath144 , the mean height of this curve in the range from @xmath269 to its maximum value for the stored ucn spectrum directly gives the average value of depolarization rate for a maxwell spectrum . ( the trap loading process used in an actual experiment may induce deviations from the maxwell spectrum . ) ) , and neutron @xmath4-decay rate ( for a lifetime of 882s ) , plotted as a function of vertical velocity component @xmath270 in the neutral plane ( where the gravitational and magnetic forces are balanced ) . @xmath270 is normalized with the constant @xmath271 which is determined by the field magnitude in the neutral plane . the curve for @xmath228 = 0.005 t is plotted to scale @xmath272 and the curve for @xmath228 = 0.05 t is plotted with magnification factor @xmath273 @xmath274 . their difference by about two orders of magnitude shows the strong suppression of depolarization by a stabilization field of sufficient strength . for a maxwell spectrum , the mean height of the curves over the range of the abscissa , from 0 to 2.5 for @xmath275 t and from 0 to 4.7 for @xmath276 t , directly determines the average over the full spectrum ( here for @xmath277 m / s @xmath278 @xmath279 m / s and drop heights @xmath144 up to 450 mm).,width=291 ] such a plot is presented in fig . [ fig : three ] where we have normalized @xmath268 to @xmath280 , the @xmath20-velocity for the spin - flipped state on the neutral plane for @xmath281 . @xmath282 is solely determined by the field magnitude @xmath283 on the neutral plane . the parameters used are : @xmath284 cm , @xmath285 t and @xmath286 t and @xmath287 m / s . for these parameters the mean depolarization rate , normalized to the @xmath4-decay rate @xmath288 , is @xmath289 for @xmath276 t and about two orders of magnitude less for @xmath275 t. it might come as a surprise that the largest contribution to the depolarization rate originates from ucns with fairly low energy of vertical motion . they move through the field almost horizontally , with small vertical oscillations about the neutral plane . the result is plausible since these ucns spend the largest fraction of time in the region where the field rotates rapidly in the reference frame of the laterally moving neutron . in section [ sec : iv.b ] we will analyze the same problem of depolarization in magnetic storage of ucns using the semiclassical method and compare the two approaches . next we apply the quantum approximation to analyze the depolarization in reflection of polarized neutrons on a non - magnetic mirror immersed in a non - uniform magnetic field . between the upper and lower turning surfaces at @xmath169 and @xmath199 for a given ucn energy for vertical motion . the reflected spin - flipped wave consists of the two components in eq . ( [ ( 42 ) ] ) : ( a ) the particular solution @xmath290 which is the same as for a wave moving upward from the lower turning point at ii in the absence of the mirror ; ( b ) the homogeneous wave @xmath291 induced at the mirror surface by the wave incident from above , starting from the upper turning level at point i. the position iii of the virtual image of i below the mirror determines a phase angle.,width=291 ] in sec . [ sec : i ] we mentioned two examples where the possible depolarization of ucns in total reflection from a non - magnetic wall , like copper , is of major importance : ( a ) ucn experiments on spin anisotropy parameters in neutron decay , such as the neutron - electron spin correlation coefficient @xmath0 @xcite ; and ( b ) ucn polarizers based on transmission through a high magnetic field . depolarization in mirror reflection has been investigated theoretically in ref . we will study aspects of this problem by imagining an ideal , non - magnetic mirror inserted horizontally into our magnetic model field at various heights @xmath292 , as shown in fig . [ fig : four ] , and comparing the depolarization per bounce on this mirror with that in the field without the mirror . the reflecting mirror surface at @xmath292 lies between the upper and lower turning points @xmath169 and @xmath199 for ucns . we use index @xmath106 to denote the quantities at the reflecting mirror surface and have assumed that the ucns have given lateral velocities @xmath67 and @xmath68 and a fixed energy for vertical motion . an ideal ucn mirror has reflection amplitude @xmath293 and reflectivity @xmath294 . the phase angle @xmath295 depends on the limit @xmath296 of total reflection and , for the system at hand , on the vertical component of incident wave vector , @xmath297 , which is the same for both wave components , @xmath191 and @xmath4 . we will see that the depolarization at the mirror is independent of @xmath298 , i.e. , it should be virtually the same for any low loss ( almost ideal ) mirror material . on physical grounds , no abrupt change of depolarization probability @xmath299 is expected since the dwell time of @xmath300 s inside the wall is much shorter than the larmor precession period ( 1 @xmath17s for @xmath301 mt ) which sets the time scale for any change . as in the previous sections , we assume that the particle started out from an upper turning level @xmath169 in a pure ( + ) spin state . on incidence at @xmath302 its wave function has acquired a depolarized component given by eq . ( [ ( 32 ) ] ) : @xmath303 where @xmath304 has been defined in ( [ ( 30 ) ] ) . the evolution of @xmath103 following reflection is determined by the inhomogeneous differential equation ( [ ( 23 ) ] ) : @xmath305 \alpha(y),\ ] ] where we replaced the @xmath2 sign in ( [ ( 23 ) ] ) by + since the reflected wave is propagating upward . the solution to ( [ ( 41 ) ] ) can be written as a superposition @xmath306 of a particular solution , @xmath307 , and the general solution @xmath308 of the homogeneous equation @xmath309 the amplitude multiplying @xmath310 is adjusted to satisfy initial conditions . a particular solution @xmath290 for downward motion has been given as eq . ( [ ( 32 ) ] ) . for the mirror - reflected beam we need the corresponding solution for upward motion . it follows from the discussion in sec . [ sec : iii.b.1 ] that this particular solution is @xmath311 the phase factor @xmath312 , with @xmath313 , arises as follows : for the mirror reflection problem we choose a time axis with @xmath314 at the upper turning point @xmath169 and increasing as the motion proceeds . however , the corresponding time @xmath315 , chosen in sec . [ sec : iii.b.1 ] for the upward motion , starts from @xmath316 at @xmath199 ( not from @xmath169 ) . the times required to reach a given particle position are related through @xmath317 , where , as before , @xmath258 is the time for a round trip down and up . the phase factor @xmath210 arises due to this difference in the time coordinates . the homogeneous equation ( [ ( 45 ) ] ) has two solutions but only one , @xmath318 , corresponds to upward wave propagation , as required for the reflected wave . the plane wave form is valid only near the mirror surface where @xmath142 is considered constant on the scale of the neutron wavelength . thus , in the framework of the wkb approximation , we use the homogeneous solution @xmath319 where the constant @xmath320 is to be adjusted to match the outgoing wave ( [ ( 42 ) ] ) to the wave @xmath321 excited by the incoming beam . expression ( [ ( 46 ) ] ) is the incoming wave ( [ ( 32 ) ] ) multiplied by the phase factor @xmath322 where @xmath323 is the reflection amplitude . the factor @xmath324 is due to a shift of time scales similar to that defined following eq . ( [ ( 44 ) ] ) but now referring to mirror reflection with start from the upper turning point @xmath169 versus start from its virtual image at @xmath325 below the mirror surface , as shown in fig . [ fig : four ] . now we match the outgoing wave @xmath326 ( from ( [ ( 44 ) ] ) and ( [ ( 45 ) ] ) ) to the wave @xmath327 at the mirror surface @xmath292 ( from ( [ ( 46 ) ] ) ) to determine the constant @xmath320 , with the result @xmath328 @xmath329 \exp(-i\phi_{-m}).\ ] ] thus the mirror reflected wave becomes @xmath330 \exp(-i\phi_{-m } ) \exp\big(i\phi_-(y)\big)\big\}. \nonumber\end{aligned}\ ] ] the first term in the braces represents an outgoing wave with wave number @xmath331 and the second one with wave number @xmath162 . to calculate the outgoing current we also need the derivative @xmath332\exp(-i\phi_{-m } ) \exp\big(i\phi_-(y)\big)\big\}. \nonumber\end{aligned}\ ] ] the current propagating in the upward direction is obtained from @xcite @xmath333,\ ] ] and the result is a sum of slowly varying terms representing the measurable depolarization . there are also fast oscillating terms with phase @xmath334 $ ] which would be averaged to zero by a detector of spin - flipped ucns except within a narrow range of order @xmath335 , i.e. , of a few wavelengths above the mirror . with the same proviso , mixed terms @xmath336 ( or @xmath337 can also be dropped since the phase factors @xmath210 and @xmath338 depend sensitively on the exact position of the mirror and the exact distance between upper and lower turning points . in practice these quantities are blurred by geometrical imperfections as well as the finite spread in ucn energy . thus @xmath210 and @xmath338 can be considered statistically independent of one another . as a result we obtain for the measurable average current @xmath339 to measure the depolarization per one complete bounce on the mirror from upper turning point down and back up to the upper turning point , we insert @xmath340 for the detector position and obtain @xmath341 this corresponds to expression ( [ ( 36 ) ] ) for one bounce in the magnetic field . in ( [ ( 52 ) ] ) we have used @xmath342 and we note that in typical cases the first term in ( [ ( 52 ) ] ) is negligible . we have assumed that the ucns are incident on the mirror from above , i.e. are confined to the space @xmath343 . if , instead , they impinge from below at @xmath344 ( now the lower mirror surface ) and are confined to the space @xmath345 , the expressions ( [ ( 51 ) ] ) and ( [ ( 52 ) ] ) ( now for @xmath346 , not @xmath347 ) remain the same except that in ( [ ( 52 ) ] ) the index @xmath245 is replaced by @xmath239 . we will discuss these results in greater detail in sections [ sec : iv.c ] and [ sec : vi ] . the semi - classical schrdinger equation describes the particle in time @xmath348 , rather than in space coordinates . the particle is assumed to follow a known classical path , so the field variable @xmath109 is considered a known function of @xmath348 . we again choose the quantization axis along the position - dependent direction of @xmath109 , for which the mutually orthogonal basis vectors @xmath77 and @xmath78 are given in ( [ ( 6 ) ] ) and the wave function for the ( + ) and ( @xmath143 ) state is @xmath349 but these quantities are now considered to be functions of @xmath348 , rather than of space variables . as before , the spin flip probability @xmath350 is considered small compared to the probability @xmath351 1 to find the neutron in the storable ( + ) spin state . in this approximation the schrdinger equation reads @xcite @xmath352 where the hamiltonian @xmath79 for spin interaction was given in ( [ ( 3 ) ] ) . we denote time derivatives by a dot and , using the notation and relations of sec . [ sec : iii.a ] and , from @xmath21 , @xmath353 , obtain @xmath354 and @xmath355 inserting into ( 54 ) and keeping only the dominant contributions we get for the terms with @xmath77 @xmath356 and for those with @xmath78 @xmath357 where @xmath358 is the larmor frequency . in practice , @xmath359 . equations ( [ ( 57)]-[(58 ) ] ) correspond to equations ( 14 ) and ( 15 ) of @xcite , but ( [ ( 58 ) ] ) contains the new phase factor @xmath360 ( as the quantum equivalent , eq . ( [ ( 12 ) ] ) , does ) and the term dependent on @xmath67 . equation ( [ ( 57 ) ] ) is solved by @xmath361 where @xmath362 is twice the phase angle accumulated since the start time @xmath363 . as for the quantum case , we assume that the motion starts at the upper or lower turning point with velocity @xmath364 , where @xmath365 const . and the constant @xmath14-component @xmath68 does not induce depolarization for our field model . @xmath366 implies @xmath367 at the start . as in @xcite , eq . ( [ ( 58 ) ] ) is solved using the ansatz @xmath368 , where @xmath369 , the new measure of depolarization amplitude @xcite , satisfies the relation @xmath370 since @xmath371 the factor @xmath372 is time dependent and therefore @xmath373 contributes to the energy . however , as noted before , @xmath374 is much smaller than @xmath375 and can be neglected in the same way small terms are neglected in the wkb approximation . ( [ ( 60 ) ] ) is readily integrated by parts : @xmath376 as for the quantum analog ( [ ( 28 ) ] ) of ( [ ( 68 ) ] ) , the right - hand side of eq . ( [ ( 61 ) ] ) does not include a term for the lower integration limit @xmath363 . as we will see , this leads to results matching those of the quantum approach . we again can argue , as in the discussion of eq . ( [ ( 28 ) ] ) , that in view of the airy - function character of the actual wave solution around the turning region , the initial time @xmath363 is a blurred quantity and the contribution from the lower limit of integration averages to zero . but a rigorous justification may be impossible within this semi - classical mix of ingredients as incongruent as classical and quantum mechanics are . we identify @xmath377 with the probability @xmath378 of finding the neutron in the spin - flipped state @xcite @xmath379 in the last step of eq . ( [ ( 62 ) ] ) we have used @xmath380 , and @xmath381 , since in our field model @xmath84 depends only on @xmath20 . the result ( [ ( 62 ) ] ) agrees with eq . ( [ ( 34 ) ] ) for the function @xmath382 which had also been identified as the depolarization probability . therefore we have full agreement also for the mean depolarization per bounce ( [ ( 37 ) ] ) and for the depolarization rate given in eq . ( [ ( 39 ) ] ) . in the next section we will see that the semi - classical and quantum approaches do not always produce exactly the same results , although a strong correlation between the two will be found also in this case . the semi - classical analysis of the mirror problem follows the same path as for the quantum case analyzed in sec . [ sec : iii.c ] . the crucial step again is matching the superposition of particular plus homogeneous solution for the ascending reflected beam , @xmath383 to the wave @xmath327 induced by the incident beam at the mirror surface at @xmath384 . in eq . ( [ ( 63 ) ] ) , @xmath385 is the general solution of the homogeneous equation @xmath386 , which corresponds to the inhomogeneous equation ( [ ( 58 ) ] ) with the direction of time reversed as described following eq . ( [ ( 44 ) ] ) . using the same sequence of terms as in ( [ ( 63 ) ] ) and the notation of sec . [ sec : iii.c ] , the matching condition @xmath387 reads @xmath388 where we have divided out the common factor @xmath389 . @xmath390 is the larmor frequency at the mirror position and @xmath391 , @xmath392 are statistically independent unitary phase factors similar to their quantum analogs @xmath210 , @xmath338 in ( [ ( 44 ) ] ) and ( [ ( 46 ) ] ) . solving for the constant @xmath320 of the homogeneous term and inserting into ( [ ( 63 ) ] ) gives @xmath393.\nonumber\end{aligned}\ ] ] finally , taking the square magnitude of ( [ ( 65 ) ] ) and performing the same statistical averaging as for ( [ ( 50 ) ] ) , which includes setting @xmath394 and @xmath395 , we find for the probability of depolarization @xmath396\nonumber \\ & + \frac{1}{2\omega_{lm}^2 } \left[\dot\theta_m^2 + ( kv_x\sin\theta_m)^2\right].\end{aligned}\ ] ] to facilitate comparison with the quantum result ( [ ( 51 ) ] ) we convert from time to space dependent variables @xmath397 , as for eq . ( [ ( 62 ) ] ) , and obtain @xmath398}{(k_{-m}^2 - k_{+m}^2)^2}.\end{aligned}\ ] ] this is the same expression as ( [ ( 51 ) ] ) except that the factor @xmath399 for the second term on the right - hand side is missing . except for very low energy ucns hovering in the magnetic field , and for mirror position at a turning point , this factor is close to 1 . for a complete bounce on the mirror , starting from , and ending at , the upper turning point level @xmath169 we obtain @xmath400}{(k_{-m}^2-k_{+m}^2)^2 } .\ ] ] . we use eq . ( [ ( 52 ) ] ) for the quantum treatment ( qm ) and eq . ( [ ( 68 ) ] ) for the semi - classical approach . the plotted values are normalized by dividing ( [ ( 68 ) ] ) by ( [ ( 36 ) ] ) , the depolarization due to the field alone , and by the ratio of duration of one bounce with and without the mirror . the data show a strong enhancement of depolarization due to the mirror . the enhancement factor depends on incident ucn energy ( which increases with larger fall height @xmath144 ) and on the gradient @xmath401 of field angle @xmath84 at the mirror position . @xmath401 is largest in the region where the curves have their peak value which is of order @xmath402 for @xmath67 = 3 m / s . for the examples shown , the turning point levels are : @xmath403 mm , @xmath404 mm for @xmath405 mm and @xmath406 mm , @xmath407 mm for @xmath408 mm.,width=291 ] fig . [ fig : five ] shows the depolarization per bounce on the mirror in our model field as a function of mirror position @xmath292 . for the quantum treatment ( qm ) this probability is given by eq . ( [ ( 52 ) ] ) and for the semi - classical approach ( cl ) by eq . ( [ ( 68 ) ] ) . to separate the role of the non - magnetic mirror in the field from the depolarization due to the field alone we have divided ( [ ( 68 ) ] ) by ( [ ( 36 ) ] ) , the depolarization due to the field alone , and by the ratio of duration of one bounce with and without the mirror . fig . [ fig : five ] shows a strong enhancement due to the mirror . the enhancement factor depends on incident ucn energy ( which increases with larger fall height @xmath144 ) and on the gradient @xmath401 of field angle at the mirror position . @xmath401 is largest in the region where the curves have their peak . the enhancement reaches four orders of magnitude for @xmath409 m and @xmath410 m / s , and six decades for @xmath411 m / s . in the latter comparison , the role of lateral velocity component @xmath67 appears so large because the depolarization in the field alone vanishes for @xmath219 in our approximation ( [ ( 36 ) ] ) ( and is negligibly small in higher - order approximations @xcite ) , while the mirror depolarization remains finite for @xmath219 . the quantum and semi - classical results , compared for identical parameters by the dashed and dash - dotted curves in fig . [ fig : five ] , are quite similar . for the given parameters they differ by a few percent at most if we exclude the range within @xmath3 5 mm from a turning point . as for the quantum approach , the results ( [ ( 67 ) ] ) and ( [ ( 68 ) ] ) remain the same ( except for the index @xmath245 in ( [ ( 68 ) ] ) changing to @xmath239 ) if incidence on the mirror from above is replaced by incidence from below , with start from the lower turning point @xmath412 rather than from @xmath413 . in the next section we will show that for low ucn energies these analytical results can also be obtained by numerical integration of the basic differential equation . numerical integration does not rely on the wkb approximation . however , the lengthy integration over many oscillations of the wave function is plagued with the rounding errors and the error due to the finite step size . reasonable agreement of the two methods would be an indication that the results are reliable . the equation of motion of the wave function @xmath103 for spin flip is given by the second - order inhomogeneous differential equation ( [ ( 23 ) ] ) , @xmath414 \alpha(y ) , \nonumber\ ] ] with ( [ ( 16 ) ] ) @xmath415 and the + ( @xmath143 ) sign refers to motion upward ( downward ) . using the fourth - order runge kutta process , eq . ( [ ( 23 ) ] ) may be integrated numerically , starting from an initial point @xmath416 slightly above ( below ) the turning point @xmath146 . key is the suitable choice of initial values @xmath417 and @xmath418 . we choose the wkb solution which is given by ( [ ( 32 ) ] ) for downward motion . its extension to include also the upward path reads @xmath419 with derivative @xmath420 where @xmath421 and @xmath422 . @xmath208 has been defined in ( [ ( 30 ) ] ) . as discussed following ( [ ( 19 ) ] ) , near a turning point @xmath146 eq . ( [ ( 69 ) ] ) is based on the asymptotic form of the airy function wave solution @xmath423 , where the coefficient @xmath424 is adjusted to match ( [ ( 16 ) ] ) asymptotically . the constants @xmath425 and @xmath426 were defined following eq . ( [ ( 19 ) ] ) . since @xmath427 varies slowly , the constant value @xmath425 is a good approximation over hundreds of oscillations of the airy function , starting from @xmath428 . = 8 cm , @xmath228 = 0.05 t and @xmath67 = 3 m / s . direct numerical integration , represented by the solid curve , coincides with the analytic result shown by the dashed curve , with a maximum deviation of 1% over the entire range from upper to lower turning point ( @xmath429 mm , @xmath430 mm).,width=291 ] using initial values ( [ ( 69 ) ] ) and ( [ ( 70 ) ] ) , eq . ( [ ( 23 ) ] ) may be integrated numerically up to a point slightly before the next turning point , @xmath431 , is reached . the method fails at @xmath431 itself due to the divergence of @xmath432 . we calculate the current @xmath433 $ ] at every point along the integration path and fig . [ fig : six ] shows the result of downward integration for parameters @xmath434 cm , @xmath435 t , @xmath410 m / s and starting point at @xmath436 . we have tested that the solution is stable in a wide range of initial position from @xmath4372 to 20 . [ fig : six ] shows that the numerical result coincides with the analytical solution ( [ ( 69 ) ] ) , with maximum deviations of @xmath438 over the entire range including the far endpoint @xmath199 . this consistency at the lower turning point is vital since the current leaving there from the storage space is identified with the depolarization probability for the move from top to bottom . to be specific , this depolarization probability @xmath439 is obtained by extrapolating the numerical solution over the short distance of order @xmath440 to the next endpoint @xmath431 with the result @xmath441 for the case shown in fig . [ fig : six ] . the extrapolation at @xmath442 mm is straightforward since the current shows a smooth behavior in the entire integration range . the same level of agreement within @xmath443 is obtained for the upward motion which is represented by the same curve of fig . [ fig : six ] , but the extrapolation for the loss current is now made at the upper end @xmath444 mm ( and this contribution is much smaller in the case shown , but it is comparable for smaller drop heights ) . we found agreement within 1% between the various methods also for other parameters @xmath445 within the range of interest . there is one exception . for the strictly vertical mode of motion analyzed in ref . @xcite , where @xmath219 , the analytical solution vanishes and the numerical solution for the intercept is at least four orders of magnitude smaller than for @xmath410 m / s . in this case the numerical precision is insufficient to determine a reliable value of depolarization probability . however , this is inconsequential since extremely small values of @xmath67 make a negligible contribution to the mean depolarization rate in a broad ucn spectrum . we add one remark . if the explicit analytical solutions ( [ ( 69)]-[(70 ) ] ) for @xmath4 and @xmath446 were not known , we could use , for the numerical integration , initial values derived solely from the properties of the airy function solution near the turning points . in these regions the second derivative @xmath170 in eq . ( [ ( 23 ) ] ) is negligible compared with @xmath447 since @xmath448 . therefore , @xmath449 \alpha(y)/k_-^2(y)$ ] is a good approximation . it differs from ( [ ( 69 ) ] ) only by the multiplier @xmath450 $ ] which is close to 1 at either turning point since @xmath163 vanishes there . it turns out that these modified initial values give the same results for @xmath103 at both endpoints , and therefore the same depolarization probability , as the more exact method . however , the function @xmath103 will be somewhat different in the range between the endpoints , where @xmath170 can not be neglected . we may summarize the results on depolarization in magnetic storage as follows : the fact that the three methods used agree ( quantum and semi - classical analysis as well as direct numerical integration ) appears to be a good indication that the approximations made were justified . for ucn reflection from a non - magnetic mirror in a magnetic field the analytical results given in ( [ ( 52 ) ] ) and ( [ ( 68 ) ] ) are similar but not identical . we used numerical integration for the reflected beam , starting from initial conditions ( [ ( 46 ) ] ) at the mirror surface and note that in this case the extrapolation to the next turning point is less straightforward since the curve for current shows , superimposed on a smooth variation , fast oscillations due to beating between two wave components : one propagating with wave number @xmath163 and the other with @xmath162 . this is expected for the superposition of these two waves in eq . ( [ ( 48 ) ] ) . the numerical results are generally consistent with the analytical expressions but not precise enough to decide whether the semi - classical approximation ( [ ( 68 ) ] ) or the quantum approximation ( [ ( 52 ) ] ) is more reliable . depolarization in high precision neutron lifetime experiments using magnetic confinement must either be negligible or else very small and quantitatively understood . using two analytical methods based on ref . @xcite and direct numerical integration we have analyzed the depolarization per bounce and the depolarization rate ( per s ) for ucns stored in a model magnetic field configuration . our magnetic field model is similar to the system envisaged for the `` bathtub '' project @xcite which uses a halbach array of permanent magnets . our model is simplified to a configuration with translational symmetry in both horizontal directions ( @xmath14 and @xmath15 ) and an ideal halbach field whose magnitude depends only on the vertical distance @xmath20 from the horizontal magnet surface . the parameters for the halbach field are the same as those proposed in ref . our analysis shows that depolarization is mainly caused by the rotating halbach field the ucns see as they move through the field with finite lateral velocity component @xmath67 , not by the small field ripple due to imperfections of the halbach system . therefore we do not expect the simplification of the model to affect the depolarization in a significant way . however , the role of the additional horizontal stabilization field @xmath22 perpendicular to the halbach field @xmath451 is critical . it has to be strong enough to suppress depolarization to an acceptable level . the main purpose of this work was to determine tolerance limits for its magnitude @xmath66 . our analysis extends that of ref . @xcite by including arbitrary ucn orbits in 3d space whereas the analysis in @xcite was restricted to purely vertical motion . as a main result of the extension we find that the lateral @xmath15-component of motion in the plane of the halbach field makes the dominant contribution to depolarization while the depolarization due to the vertical motion is insignificant . as a result , some previous estimates of depolarization probability may have been overoptimistic . for the parameters of @xcite ( 0.05 - 0.1 t for @xmath228 ) we estimate on the basis of fig . [ fig : three ] that even a measurement of the neutron lifetime with precision @xmath452 should be possible ( disregarding other potential limitations ) but the safety margin may be smaller than previously expected . systems with smaller stabilization field , as possibly that of ref . @xcite , where @xmath66 has not been specified , may require a separate analysis since they use a cylindrical rather than planar field distribution . however , the main result of the present work is independent of geometrical details : the depolarization loss is determined , not by the largest rotation frequency of @xmath136 as seen by the ucns as they traverse the field , but by the conditions at the turning points where the spin - flipped neutrons can leave the storage system . this is an important point which may appear to contradict the common view , but is also implied by the work of ref . our interpretation of the present results is as follows : the measurable depolarization is not directly caused by critical spots within the storage volume , where the @xmath136-field has a small magnitude @xmath8 and rotates fast in the moving reference frame . all that matters for depolarization are the field conditions at the turning points where the storable ucns are reflected back into the storage space while the spin - flipped fraction leaves the trap . the loss current in eq . ( [ ( 37 ) ] ) is determined by the value of @xmath453 at such a surface . this factor is large for fast field variation ( @xmath454 ) seen by the neutron moving in a horizontal direction ( parallel to the turning surface ) , as well as for small @xmath8 since @xmath455 from ( [ ( 31 ) ] ) is directly proportional to @xmath8 at a turning point where @xmath151 . it also increases quadratically with sin@xmath456 , i.e. with the magnitude @xmath457 of the halbach field at a turning surface . if a broad spectrum of ucns is stored the critical points will be of importance in the sense that some , usually ucns with very low energy of vertical motion , may have a turning surface near such areas and therefore make a large contribution . in fact , the low - energy ucns make the largest contribution to depolarization seen in the peaks in fig . [ fig : three ] , and they are the reason why a larger stabilization field @xmath66 suppresses the net depolarization very effectively . fig . [ fig : three ] shows @xmath458 , the depolarization rate divided by the @xmath4-decay rate , as a function of normalized vertical velocity @xmath268 in the neutral plane , where gravity and the magnetic force are balanced . both curves , for @xmath228 = 0.05 t and for @xmath228 = 0.005 t , are peaked at small values of @xmath459 . thus the ucns most in danger of suffering depolarization are confined to a narrow space of @xmath310 cm about the neutral plane , moving laterally as they float in the field ( for @xmath460 or oscillate up and down about the neutral plane with small amplitude . the magnetic field configurations and neutron orbits in actual or projected 3d magneto - gravitational ucn confinement systems @xcite are more complex than in our field model . however , for some of these concepts the 1d approximation of our model appears to be justified . for the bathtub " system @xcite , the neutral plane of our model , about which low - energy ucns oscillate , corresponds to a strongly anisotropic oscillator potential in the vicinity of the minimum of potential @xmath461 for the low - field seeking spin state . for @xmath276 t the minimum is located about 1.8 cm up from the lowest point of the double - curved surface of permanent magnets of ref . @xcite . for this anisotropic oscillator , the frequency for oscillations in the vertical @xmath20-direction , @xmath462 s@xmath129 , is 22 times larger than for the @xmath14-direction , @xmath463 s@xmath129 where the radius of curvature is @xmath464 m. for the @xmath15-direction , with its asymmetry , there are two curvatures ( 0.5 and 1.0 m ) and , therefore , two ratios replacing 22 : 13 on one side ( @xmath465 ) and 18 for @xmath466 . since all these factors are large , the neutrons move almost freely , on a relative scale , in the peripheral @xmath15- and @xmath14-directions and , therefore , our 1d model should be a good approximation . as a result , we expect the peripheral velocity in the plane of the halbach field , which corresponds to @xmath467 in the model , to be the main source of depolarization . in the cylindrical field geometries of references @xcite , the field magnitude @xmath8 varies more slowly in space . thus , the oscillator is less anisotropic , the degrees of freedom of motion in different coordinate directions are less decoupled and a more complex analysis may be required . to the extent that qualitative features of our model may still apply we expect that the main source of depolarization would be a large peripheral ucn velocity perpendicular to the cylinder axis . in this case , the peripheral velocity corresponds to the component @xmath467 of the model . besides depolarization of magnetically confined ucns we also studied depolarization in ucn reflection on a non - magnetic mirror immersed in a magnetic field . the field was our model field into which we conceptually inserted an ideal neutron mirror horizontally at a variable height . the problem of possible depolarization in mirror reflection is of paramount importance in ucn experiments on the asymmetry parameter @xmath0 in neutron decay @xcite and it is also encountered in high - field ucn polarizers . depolarization is expected since the adiabaticity condition may be violated due to the abrupt change of flight direction at the reflection point , thus @xmath468 changes abruptly . this problem has first been studied in ref . @xcite by adapting the majorana semi - classical approach @xcite to the mirror geometry . in our analysis the three methods used ( quantum approximation , semi - classical and numerical approach ) gave identical results for depolarization in magnetic ucn storage . for the mirror reflection problem the quantum result ( [ ( 52 ) ] ) and the semi - classical result ( [ ( 68 ) ] ) are very similar but not identical , as shown in fig . [ fig : five ] . the numerical method is not accurate enough in this case to distinguish between the two . the semiclassical result ( [ ( 68 ) ] ) lacks the factor @xmath399 which would cause a divergence if the mirror is placed at a turning point height ( since @xmath469 in this region ) . between the turning points the difference is minor and the common result is as follows : depolarization in mirror reflection , averaged over field directions as in our model field , is determined mainly by the second term of eq . ( [ ( 68 ) ] ) . it increases with the frequency of field variation ( @xmath470 ) seen by the neutron moving along the in - plane @xmath15-direction . it also increases quadratically with the sine of the field angle @xmath84 , which is a measure of halbach field strength @xmath471 , and with its gradient , @xmath472 , at the mirror location . depolarization strongly decreases with increasing field strength @xmath8 at the mirror ( @xmath473 ) . [ fig : five ] shows that , for @xmath67 = 3 m / s , the magnitude of depolarization per one bounce on the mirror is up to @xmath119 times larger than the depolarization per bounce in the magnetic field without the mirror . the depolarization on the mirror has its peak value at the vertical location where the depolarization probability plotted in fig . [ fig : two ] also has its peak . in fact , comparing expression ( [ ( 68 ) ] ) for depolarization at the mirror with the result ( [ ( 34 ) ] ) for the field alone ( and neglecting the small first term on the right - hand side of ( [ ( 68 ) ] ) ) we realize that the mirror acts like a polarization analyzer inserted into the particle beam moving through the * b*-field . this interpretation also holds for a non - horizontal or curved mirror since the second term on the right - hand side of ( [ ( 68 ) ] ) is independent of the orientation of the reflecting surface element . in this case , the second term should be averaged over the mirror extension . in cases where many successive wall reflections take place in a weak , non - uniform magnetic field , the depolarization may become significant . comparing our result ( [ ( 66 ) ] ) with equations ( 10 - 11 ) of ref . @xcite we note that both results have the square of the larmor frequency @xmath9 at the mirror position in the denominator . however , a quantitative comparison is difficult since the magnetic field variations assumed in the two approaches are different . in either analysis , no depolarization is expected for a uniform magnetic field . our field model may be too specific to allow a quantitative comparison with the data @xcite on depolarization in ucn reflection from various mirror materials ( like beryllium or samples with diamond - like carbon coating ) . expression ( [ ( 68 ) ] ) does not depend on specific properties of the mirror , as long as it is a good , nearly loss - free ucn reflector . therefore , ( [ ( 68 ) ] ) could explain , without having to invoke any anomalies @xcite , the remarkable similarity and temperature independence of depolarization probabilities measured for different wall materials . such independence would be expected if the samples were exposed to the same non - uniform magnetic field . our results , which were obtained as straightforward solutions to the spin - dependent schrdinger equation , may also provide an alternative to the discussion of new short - range , spin - dependent forces as a possible pathway to explaining the depolarization data for stored ucns . for a recent comprehensive review of fundamental physics with neutrons see @xcite .
the basics of the density matrix renormalization group were developed by s. white in 1992@xcite and since then dmrg has proved to be a very powerful method for low dimensional interacting systems . its remarkable accuracy can be seen for example in the spin-1 heisenberg chain : for a system of hundreds of sites a precision of @xmath0 for the ground state energy can be achieved . since then it has been applied to a great variety of systems and problems including , among others , spin chains and ladders , fermionic and bosonic systems , disordered models , impurities and molecules and 2d electrons in high magnetic fields . it has also been improved substantially in several directions like two ( and three ) dimensional ( 2d ) classical systems , stochastic models , the presence of phonons , quantum chemistry , field theory , the inclusion of temperature and the calculation of dynamical and time - dependent properties . some calculations have also been performed in 2d quantum systems . all these topics are treated in detail and in a pedagogical way in the book @xcite , where the reader can find an extensive review on dmrg . in this article we will attempt to cover the different areas where it has been applied without entering into details but in a few cases , where we have chosen some representative contributions . we suggest the interested reader to look for further information in the referenced work . our aim here is to give the reader a general overview on the subject . one of the most important limitations of numerical calculations in finite systems is the great amount of states that have to be considered and its exponential growth with system size . several methods have been introduced in order to reduce the size of the hilbert space to be able to reach larger systems , such as monte carlo , renormalization group ( rg ) and dmrg . each method considers a particular criterion of keeping the relevant information . the dmrg was originally developed to overcome the problems that arise in interacting systems in 1d when standard rg procedures were applied . consider a block b ( a block is a collection of sites ) where the hamiltonian @xmath1 and end - operators are defined . these traditional methods consist in putting together two or more blocks ( e.g. b - b , which we will call the superblock ) , connected using end - operators , in a basis that is a direct product of the basis of each block , forming @xmath2 . this hamiltonian is then diagonalized , the superblock is replaced by a new effective block @xmath3 formed by a certain number @xmath4 of lowest - lying eigenstates of @xmath2 and the iteration is continued ( see ref . although it has been used successfully in certain cases , this procedure , or similar versions of it , has been applied to several interacting systems with poor performance . for example , it has been applied to the 1d hubbard model keeping @xmath5 states . for 16 sites , an error of 5 - 10% was obtained @xcite . other results@xcite were also discouraging . a better performance was obtained @xcite by adding a single site at a time rather than doubling the block size . however , there is one case where a similar version of this method applies very well : the kondo ( and anderson ) model . wilson@xcite mapped the one - impurity problem onto a one - dimensional lattice with exponentially descreasing hoppings . the difference with the method explained above is that in this case , one site ( equivalent to an onion shell " ) is added at each step and , due to the exponential decrease of the hopping , very accurate results can be obtained . returning to the problem of putting several blocks together , the main source of error comes from the election of eigenstates of @xmath2 as representative states of a superblock . since @xmath2 has no connection to the rest of the lattice , its eigenstates may have unwanted features ( like nodes ) at the ends of the block and this ca nt be improved by increasing the number of states kept . based on this consideration , noack and white@xcite tried including different boundary conditions and boundary strengths . this turned out to work well for single particle and anderson localization problems but , however , it did not improve the results significantly for interacting systems . these considerations led to the idea of taking a larger superblock that includes the blocks @xmath6 , diagonalize the hamiltonian in this large superblock and then somehow project the most favorable states onto @xmath6 . then @xmath6 is replaced by @xmath3 . in this way , awkward features in the boundary would vanish and a better representation of the states in the infinite system would be achieved . white@xcite proposed the density matrix as the optimal way of projecting the best states onto part of the system and this will be discussed in the next section . some considerations concerning the effect of boundary conditions in the nature of the states kept ( and an analogy to the physics of black holes ) is given in @xcite . the justification of using the density matrix is given in detail in ref.@xcite . a very easy and pedagogical way of understanding the basic functioning of dmrg is applying it to the calculation of simple quantum problems like one particle in a tight binding chain @xcite . in the following section we will briefly describe the standard method ; in sect . 3 we will mention some of the most important applications ; in sect . 4 we review the most relevant extensions to the method and finally in sect . 5 we concentrate on the way dynamical calculations can be performed within dmrg . the dmrg allows for a systematic truncation of the hilbert space by keeping the most probable states describing a wave function ( _ e.g . _ the ground state ) instead of the lowest energy states usually kept in previous real space renormalization techniques . the basic idea consists in starting from a small system ( _ e.g _ with @xmath7 sites ) and then gradually increase its size ( to @xmath8 , @xmath9 , ... ) until the desired length is reached . let us call the collection of @xmath7 sites the _ universe _ and divide it into two parts : the _ system _ and the _ environment _ ( see fig . [ figsuperblock ] ) . the hamiltonian is constructed in the _ universe _ and its ground state @xmath10 is obtained . this is considered as the state of the _ universe _ and called the _ target state_. it has components on the _ system _ and the _ environment_. we want to obtain the most relevant states of the _ system _ , i.e. , the states of the _ system _ that have largest weight in @xmath11 . to obtain this , the _ environment _ is considered as a statistical bath and the density matrix@xcite is used to obtain the desired information on the _ system_. so instead of keeping eigenstates of the hamiltonian in the block ( _ system _ ) , we keep eigenstates of the density matrix . we will be more explicit below . = 2.0 in = 1.0 in [ figsuperblock ] let s define block [ * b * ] as a finite chain with @xmath12 sites , having an associated hilbert space with , @xmath4 states where operators are defined ( in particular the hamiltonian in this finite chain , @xmath1 and the operators at the ends of the block , useful for linking it to other chains or added sites ) . except for the first iteration , the basis in this block is nt explicitly known due to previous basis rotations and reductions . the operators in this basis are matrices and the basis states are characterized by quantum numbers ( like @xmath13 , charge or number of particles , etc ) . we also define an added block or site as [ * a * ] having @xmath14 states . a general iteration of the method is described below : \i ) define the hamiltonian @xmath2 for the superblock ( the _ universe _ ) formed by putting together two blocks [ * b * ] and [ * b * ] and two added sites [ * a * ] and [ * a * ] in this way : [ * b a a b * ] ( the primes are only to indicate additional blocks , but the primed blocks have the same structure as the non - primed ones ; this can vary , see the finite - size algorithm below ) . in general , blocks [ * b * ] and [ * b * ] come from the previous iteration . the total hilbert space of this superblock is the direct product of the individual spaces corresponding to each block and the added sites . in practice a quantum number of the superblock can be fixed ( in a spin chain for example one can look at the total @xmath15 subspace ) , so the total number of states in the superblock is much smaller than @xmath16 . in some cases , as the quantum number of the superblock consists of the sum of the quantum numbers of the individual blocks , each block must contain several subspaces ( several values of @xmath13 for example ) . here periodic boundary conditions can be attached to the ends and a different block layout should be considered ( e.g. [ * b a b a * ] ) to avoid connecting blocks [ * b * ] and [ * b * ] which takes longer to converge . the boundary conditions are between [ * a * ] and [ * b * ] . for closed chains the performance is poorer than for open boundary conditions @xcite . \ii ) diagonalize the hamiltonian @xmath2 to obtain the ground state @xmath11 ( target state ) using lanczos@xcite or davidson@xcite algorithms . other states could also be kept , such as the first excited ones : they are all called _ target states_. \iii ) construct the density matrix : @xmath17 on block [ * b a * ] , where @xmath18 , the states @xmath19 belonging to the hilbert space of the block [ * b a * ] and the states @xmath20 to the block [ * b a * ] . the density matrix considers the part [ * b a * ] as a system and [ * b a * ] , as a statistical bath . the eigenstates of @xmath21 with the highest eigenvalues correspond to the most probable states ( or equivalently the states with highest weight ) of block [ * b a * ] in the ground state of the whole superblock . these states are kept up to a certain cutoff , keeping a total of @xmath4 states per block . the density matrix eigenvalues sum up to unity and the truncation error , defined as the sum of the density matrix eigenvalues corresponding to discarded eigenvectors , gives a qualitative indication of the accuracy of the calculation . \iv ) with these @xmath4 states a rectangular matrix @xmath22 is formed and it is used to change basis and reduce all operators defined in [ * b a * ] . this block [ * b a * ] is then renamed as block [ * b@xmath23 * ] or simply [ * b * ] ( for example , the hamiltonian in block [ * b a * ] , @xmath24 , is transformed into @xmath25 as @xmath26 ) . \v ) a new block [ * a * ] is added ( one site in our case ) and the new superblock [ * b a a b * ] is formed as the direct product of the states of all the blocks . \vi ) this iteration continues until the desired length is achieved . at each step the length is @xmath27 ( if [ * a * ] consists of one site ) . when more than one target state is used , _ i.e _ more than one state is wished to be well described , the density matrix is defined as : @xmath28 where @xmath29 defines the probability of finding the system in the target state @xmath30 ( not necessarily eigenstates of the hamiltonian ) . the method described above is usually called the _ infinite - system algorithm _ since the system size increases at each iteration . there is a way to increase precision at each length @xmath7 called the _ finite - system algorithm_. it consists of fixing the lattice size and zipping a couple of times until convergence is reached . in this case and for the block configuration [ * b a a b * ] , @xmath31 where @xmath12 and @xmath32 are the number of sites in @xmath33 and @xmath34 respectively . in this step the density matrix is used to project onto the left @xmath35 sites . in order to keep @xmath7 fixed , in the next block configuration , the right block @xmath34 should be defined in @xmath36 sites such that @xmath37 . the operators in this smaller block should be kept from previous iterations ( in some cases from the iterations for the system size with @xmath38)@xcite . the calculation of static properties like correlation functions is easily done by keeping the operators in question at each step and performing the corresponding basis change and reduction , in a similar manner as done with the hamiltonian in each block@xcite . the energy and measurements are calculated in the superblock . in ref.@xcite an interpretation of the correlation functions of systems at criticality is given in terms of wave function entanglement , conjecturing a modification of dmrg for these cases that preserves the entanglement . a faster convergence of lanczos or davidson algorithm is achieved by choosing a good trial vector@xcite . an interesting analysis on dmrg accuracy is done in ref . fixed points of the dmrg and their relation to matrix product wave functions were studied in @xcite and an analytic formulation combining the block renormalization group with variational and fokker - planck methods in @xcite . the connection of the method with quantum groups and conformal field theory is treated in @xcite . there are also interesting connections between the density matrix spectra and integrable models@xcite via corner transfer matrices . these articles give a deep insight into the essence of the dmrg method . since its development , the number of papers using dmrg has grown enormously and other improvements to the method have been performed . we would like to mention some applications where this method has proved to be useful . other applications related to further developments of the dmrg will be mentioned in sect . a very impressive result with unprecedented accuracy was obtained by white and huse @xcite when calculating the spin gap in a @xmath39 heisenberg chain obtaining @xmath40 . they also calculated very accurate spin correlation functions and excitation energies for one and several magnon states and performed a very detailed analysis of the excitations for different momenta . they obtained a spin correlation length of 6.03 lattice spacings . simultaneously srensen and affleck@xcite also calculated the structure factor and spin gap for this system up to length 100 with very high accuracy , comparing their results with the nonlinear @xmath41 model . in a subsequent paper@xcite they applied the dmrg to the anisotropic @xmath39 chain , obtaining the values for the haldane gap . they also performed a detailed study of the @xmath42 end excitations in an open chain . thermodynamical properties in open @xmath39 chains such as specific heat , electron paramagnetic resonance ( epr ) and magnetic susceptibility calculated using dmrg gave an excellent fit to experimental data , confirming the existence of free spins 1/2 at the boundaries@xcite . a related problem , _ i.e. _ the effect of non - magnetic impurities in spin systems ( dimerized , ladders and 2d ) was studied in @xcite . in addition , the study of magnon interactions and magnetization of @xmath39 chains was done in @xcite , supersymmetric spin chains modelling plateau transitions in the integer quantum hall effect in @xcite and esr studies in these systems was considered in @xcite . for larger integer spins there have also been some studies . nishiyama and coworkers@xcite calculated the low energy spectrum and correlation functions of the @xmath43 antiferromagnetic heisenberg open chain . they found @xmath39 end excitations ( in agreement with the valence bond theory ) . edge excitations for other values of @xmath44 have been studied in ref . @xcite . almost at the same time schollwck and jolicoeur@xcite calculated the spin gap in the same system , up to 350 sites , ( @xmath45 ) , correlation functions that showed topological order and a spin correlation length of 49 lattice spacings . more recent accurate studies of @xmath43 chains are found in @xcite and of @xmath39 chains in staggered magnetic fields @xcite including a detalied comparison to the non - linear sigma model in @xcite . in ref . @xcite the dispersion of the single magnon band and other properties of the @xmath43 antiferromagnetic heisenberg chains were calculated . concerning @xmath42 systems , dmrg has been crucial for obtaining the logarithmic corrections to the @xmath46 dependence of the spin - spin correlation functions in the isotropic heisenberg model @xcite . for this , very accurate values for the energy and correlation functions were needed . for @xmath47 sites an error of @xmath48 was achieved keeping @xmath49 states per block , comparing with the exact finite - size bethe ansatz results . for this model it was found that the data for the correlation function has a very accurate scaling behaviour and advantage of this was taken to obtain the logarithmic corrections in the thermodynamic limit . other calculations of the spin correlations have been performed for the isotropic @xcite and anisotropic cases @xcite . luttinger liquid behaviour with magnetic fields have been studied in @xcite , field - induced gaps in @xcite , anisotropic systems in @xcite and the heisenberg model with a weak link in @xcite . an analysis of quantum critical points and critical behaviour in spin chains by combining dmrg with finite - size scaling was done in @xcite . similar calculations have been performed for the @xmath50 heisenberg chain @xcite . in this case a stronger logarithmic correction to the spin correlation function was found . for this model there was interest in obtaining the central charge @xmath51 to elucidate whether this model corresponds to the same universality class as the @xmath42 case , where the central charge can be obtained from the finite - size scaling of the energy . although there have been previous attempts@xcite , these calculations presented difficulties since they involved also a term @xmath52 . with the dmrg the value @xmath53 was clearly obtained . in ref . @xcite , dmrg was applied to an effective spin hamiltonian obtained from an su(4 ) spin - orbit critical state in 1d . other applications were done to enlarged symmetry cases with su(4 ) symmetry in order to study coherence in arrays of quantum dots@xcite , to obtain the phase diagram for 1d spin orbital models@xcite and dynamical properties in a magnetic field@xcite . dimerization and frustration have been considered in refs . @xcite and alternating spin chains in @xcite . the case of several coupled spin chains ( ladder models ) have been investigated in @xcite , spin ladders with cyclic four - spin exchanges in @xcite and kagome antiferromagnets in @xcite . zigzag spin chains have been considered in @xcite and spin chains of coupled triangles in @xcite . as the dmrg s performance is optimal in open systems , an interesting analysis of the boundary effect on correlation functions is done in @xcite . magnetization properties and plateaus for quantum spin ladder systems@xcite have also been studied . an interesting review on the applications to some exact and analytical techniques for quantum magnetism in low dimension , including dmrg , is presented in @xcite . there has been a great amount of applications to fermionic systems such as 1d hubbard and t - j models @xcite , luttinger liquids with boundaries@xcite , the falicov - kimball model @xcite , the quasiperiodic aubry - andre chain@xcite and fibonacci - hubbard models@xcite . it has also been applied to field theory@xcite . the method has been very successful for several band hubbard models@xcite , hubbard ladders @xcite and t - j ladders@xcite . also several coupled chains at different dopings have been considered @xcite as well as flux phases in these systems @xcite . time reversal symmetry - broken fermionic ladders have been studied in @xcite and power laws in spinless - fermion ladders in @xcite . long - range coulomb interactions in the 1d electron gas and the formation of a wigner crystal was studied in @xcite . several phases including the wigner crystal , incompressible and compressible liquid states , stripe and pairing phases , have been found using dmrg for 2d electrons in high magnetic fields considering different landau levels@xcite . persistent currents in mesoscopic systems have been considered in @xcite . quite large quasi-2d systems can be reached , for example in @xcite where a 4x20 lattice was considered to study ferromagnetism in the infinite - u hubbard model ; the ground state of a 4-leg t - j ladder in @xcite ; the one and two hole ground state in 9x9 and 10x7 t - j lattices in @xcite ; a doped 3-leg t - j ladder in @xcite ; the study of striped phases in @xcite ; domain walls in 19x8 t - j systems in @xcite ; the 2d t - j model in @xcite and the magnetic polaron in a 9x9 t - j lattice in @xcite . also big @xmath54 spin-1/2 lattices reaching 24x11 sites@xcite have been studied . there have been some recent attempts to implement dmrg in two and higher dimensions @xcite but the performance is still poorer than in 1d . a recent extension using a two - step dmrg algorithm for highly anisotropic spin systems has shown promising results@xcite . impurity problems have been studied for example in one- @xcite and two - impurity @xcite kondo systems , in spin chains @xcite and in luttinger liquids @xcite . there have also been applications to kondo and anderson lattices @xcite , kondo lattices with localized @xmath55 configurations @xcite , the two - channel kondo lattice on a ladder @xcite , a t - j chain coupled to localized kondo spins@xcite and ferromagnetic kondo models for manganites @xcite . there have been several extensions to dmrg like the inclusion of symmetries to the method such as spin and parity@xcite . total spin conservation and continuous symmetries have been treated in @xcite and in interaction - round a face hamiltonians@xcite , a formulation that can be applied to rotational - invariant sytems like @xmath39 and 2 chains@xcite . a momentum representation of this technique @xcite that allows for a diagonalization in a fixed momentum subspace has been developed as well as applications in dimension higher than one@xcite and bethe lattices@xcite . the inclusion of symmetries is essential to the method since it allows to consider a smaller number of states , enhance precision and obtain eigenstates with definite quantum numbers . other recent applications have been in nuclear shell model calculations where a two level pairing model has been considered@xcite and in the study of ultrasmall superconducting grains , in this case , using the particle ( hole ) states around the fermi level as the system ( environment ) block@xcite . a very interesting and successful application is a recent work in high energy physics@xcite . here the dmrg is used in an asymptotically free model with bound states , a toy model for quantum chromodynamics , namely the two dimensional delta - function potential . for this case an algorithm similar to the momentum space dmrg@xcite was used where the block and environment consist of low and high energy states respectively . the results obtained here are much more accurate than with the similarity renormalization group@xcite and a generalization to field - theoretical models is proposed based on the discreet light - cone quantization in momentum space@xcite . below we briefly mention other important extensions , leaving the calculation of dynamical properties for the next section . the dmrg has been very successfully extended to study classical systems . for a detailed description we refer the reader to ref . @xcite . since 1d quantum systems are related to 2d classical systems@xcite , it is natural to adapt dmrg to the classical 2d case . this method is based on the renormalization group transformation for the transfer matrix @xmath56 ( tmrg ) . it is a variational method that maximizes the partition function using a limited number of degrees of freedom , where the variational state is written as a product of local matrices@xcite . for 2d classical systems , this algorithm is superior to the classical monte carlo method in accuracy , speed and in the possibility of treating much larger systems . a recent improvement of this method considering periodic boundary conditions is given in @xcite and a detailed comparison between symmetric and asymmetric targetting is done in @xcite . tmrg has also been successfully used to renormalize stochastic transfer matrices in a study of cellular automatons@xcite . the calculation of thermodynamical properties of 3d classical statistical systems has been proposed@xcite where the eigenstate of the transfer matrix with maximum eigenvalue is represented by the product of local tensors optimized using dmrg . a further improvement to this method is based on the corner transfer matrix@xcite , the ctmrg@xcite and can be generalized to any dimension@xcite . it was first applied to the ising model@xcite and also to the potts model@xcite , where very accurate density profiles and critical indices were calculated . further applications have included non - hermitian problems in equilibrium and non - equilibrium physics . in the first case , transfer matrices may be non - hermitian and several situations have been considered : a model for the quantum hall effect@xcite , the @xmath57-symmetric heisenberg chain related to the conformal series of critical models@xcite and the anisotropic triangular nearest and next - nearest neighbour ising models@xcite . in the second case , the adaptation of the dmrg to non - equilibrium physics like the asymmetric exclusion problem@xcite and reaction - diffusion problems @xcite has shown to be very successful . it has also been applied to stochastic lattice models like in @xcite and to the 2d xy model @xcite . the adaptation of the dmrg method for classical systems paved the way for the study of 1d quantum systems at non zero temperature , by using the trotter - suzuki method @xcite . in this case the system is infinite and the finiteness is in the level of the trotter approximation . standard dmrg usually produces its best results for the ground state energy and less accurate results for higher excitations . a different situation occurs here : the lower the temperature , the less accurate the results . very nice results have been obtained for the dimerized , @xmath42 , xy model , where the specific heat was calculated involving an extremely small basis set@xcite ( @xmath58 ) , the agreement with the exact solution being much better in the case where the system has a substantial gap . it has also been used to calculate thermodynamical properties of the anisotropic @xmath42 heisenberg model , with relative errors for the spin susceptibility of less than @xmath59 down to temperatures of the order of @xmath60 keeping @xmath61 states@xcite . a complete study of thermodynamical properties like magnetization , susceptibility , specific heat and temperature dependent correlation functions for the @xmath42 and 3/2 heisenberg models was done in @xcite . other applications have been the calculation of the temperature dependence of the charge and spin gap in the kondo insulator@xcite , the calculation of thermodynamical properties of ferrimagnetic chains@xcite and spin ladders@xcite , the study of impurity properties in spin chains@xcite , frustrated quantum spin chains@xcite , t - j@xcite and spin ladders@xcite and dimerized frustrated heisenberg chains@xcite . an alternative way of incorporating temperature into the dmrg procedure was developed by moukouri and caron@xcite . they considered the standard dmrg taking into account several low - lying target states ( see eq . [ eq : pl ] ) to construct the density matrix , weighted with the boltzmann factor ( @xmath62 is the inverse temperature ) : @xmath63 with this method they performed reliable calculations of the magnetic susceptibility of quantum spin chains with @xmath42 and @xmath64 , showing excellent agreement with bethe ansatz exact results . they also calculated low temperature thermodynamical properties of the 1d kondo lattice model@xcite and of organic conductors @xcite . zhang et al.@xcite applied the same method in the study of a magnetic impurity embedded in a quantum spin chain . a significant limitation to the dmrg method is that it requires a finite basis and calculations in problems with infinite degrees of freedom per site require a large truncation of the basis states@xcite . however , jeckelmann and white developed a way of including phonons in dmrg calculations by transforming each boson site into several artificial interacting two - state pseudo - sites and then applying dmrg to this interacting system@xcite ( called the pseudo - site system " ) . the idea is based on the fact that dmrg is much better able to handle several few - states sites than few many - state sites@xcite . the key idea is to substitute each boson site with @xmath65 states into @xmath7 pseudo - sites with 2 states@xcite . they applied this method to the holstein model for several hundred sites ( keeping more than a hundred states per phonon mode ) obtaining negligible error . in addition , up to date , this method is the most accurate one to determine the ground state energy of the polaron problem ( holstein model with a single electron ) . an alternative method ( the optimal phonon basis")@xcite is a procedure for generating a controlled truncation of a large hilbert space , which allows the use of a very small optimal basis without significant loss of accuracy . the system here consists of only one site and the environment has several sites , both having electronic and phononic degrees of freedom . the density matrix is used to trace out the degrees of freedom of the environment and extract the most relevant states of the site in question . in following steps , more bare phonons are included to the optimal basis obtained in this way . this method was successfully applied to study the interactions induced by quantum fluctuations in quantum strings , as an application to cuprate stripes@xcite and the dissipative two - state system @xcite . a variant of this scheme is the four block method " , as described in @xcite . they obtain very accurately the luttinger liquid - cdw insulator transition in the 1d holstein model for spinless fermions . the method has also been applied to pure bosonic systems such as the disordered bosonic hubbard model@xcite , where gaps , correlation functions and superfluid density are obtained . the phase diagram for the non - disordered bose - hubbard model , showing a reentrance of the superfluid phase into the insulating phase was calculated in ref . it has also been used to study a chain of oscillators with optical phonon spectrum @xcite and optical phonons in the quarter - filled hubbard model for organic conductors@xcite . the dmrg has also been generalized to 1d random and disordered systems , and applied to the random antiferromagnetic and ferromagnetic heisenberg chains@xcite , including quasiperiodic exchange modulation@xcite and a detailed study of the haldane@xcite and griffiths phase@xcite in these systems . strongly disordered spin ladders have been considered in @xcite . it has also been used in disordered fermi systems such as the spinless model@xcite . in particular , the transition from the fermi glass to the mott insulator and the strong enhancement of persistent currents in the transition was studied in correlated one - dimensional disordered rings@xcite . disorder - induced crossover effects at quantum critical points were studied in @xcite . there have been several applications to molecules and polymers , such as the pariser - parr - pople ( ppp ) hamiltonian for a cyclic polyene@xcite ( where long - range interactions are included ) , magnetic keplerate molecules @xcite , molecular iron rings@xcite and polyacenes considering long range interactions @xcite . it has also been applied to conjugated organic systems ( polymers ) , adapting the dmrg to take into account the most important symmetries in order to obtain the desired excited states@xcite . also conjugated one - dimensional semiconductors @xcite have been studied , in which the standard approach can be extended to complex 1d oligomers where the fundamental repeat is not just one or two atoms , but a complex molecular building block . relatively new fields of application are the calculation of dynamical properties in the rubinstein - duke model for reptons @xcite and excitons in dendrimer molecules@xcite . recent attempts to apply dmrg to the _ ab initio _ calculation of electronic states in molecules have been successful@xcite . here , dmrg is applied within the conventional quantum chemical framework of a finite basis set with non - orthogonal basis functions centered on each atom . after the standard hartree - fock ( hf ) calculation in which a hamiltonian is produced within the orthogonal hf basis , dmrg is used to include correlations beyond hf , where each orbital is treated as a site " in a 1d lattice . one important difference with standard dmrg is that , as the interactions are long - ranged , several operators must be kept , making the calculation somewhat cumbersome . however , very accurate results have been obtained in a check performed in a water molecule ( keeping up to 25 orbitals and @xmath66 states per block ) , obtaining an offset of 0.00024hartrees with respect to the exact ground state energy@xcite , a better performance than any other approximate method@xcite . in order to avoid the non - locality introduced in the treatment explained above , white introduced the concept of _ orthlets _ , local , orthogonal and compact wave functions that allow prior knowledge about singularities to be incorporated into the basis and an adequate resolution for the cores@xcite . the most relevant functions in this basis are chosen via the density matrix . an application based on the combination with the momentum version of dmrg is used in @xcite to calculate the ground state of several molecules . the dmrg was originally developed to calculate static ground state properties and low - lying energies . however , it can also be useful to calculate dynamical response functions . these are of great interest in condensed matter physics in connection with experiments such as nuclear magnetic resonance ( nmr ) , neutron scattering , optical absorption , photoemission , etc . we will describe three different methods in this section . a recent development for calculating response functions in single impurity systems in the presence of a magnetic field was done in @xcite by using the dmrg within wilson s nrg to obtain the green s function . an interesting extension of dmrg to tackle time - dependent quantum many - body systems out of equilibrium was considered by cazalilla and marston@xcite . an effective way of extending the basic ideas of this method to the calculation of dynamical quantities is described in ref.@xcite . it is important to notice here that due to the particular real - space construction , it is not possible to fix the momentum as a quantum number . however , we will show that by keeping the appropriate target states , a good value of momentum can be obtained . we want to calculate the following dynamical correlation function at @xmath67 : @xmath68 where @xmath69 is the hermitean conjugate of the operator @xmath70 , @xmath71 is the heisenberg representation of @xmath70 , and @xmath72 is the ground state of the system . its fourier transform is : @xmath73 where the summation is taken over all the eigenstates @xmath74 of the hamiltonian @xmath75 with energy @xmath76 , and @xmath77 is the ground state energy . defining the green s function @xmath78 the correlation function @xmath79 can be obtained as @xmath80 the function @xmath81 can be written in the form of a continued fraction : @xmath82 the coefficients @xmath83 and @xmath84 can be obtained using the following recursion equations @xcite : @xmath85 where @xmath86 for finite systems the green s function @xmath87 has a finite number of poles so only a certain number of coefficients @xmath83 and @xmath84 have to be calculated . the dmrg technique presents a good framework to calculate such quantities . with it , the ground state , hamiltonian and the operator @xmath70 required for the evaluation of @xmath79 are obtained . an important requirement is that the reduced hilbert space should also describe with great precision the relevant excited states @xmath88 . this is achieved by choosing the appropriate target states . for most systems it is enough to consider as target states the ground state @xmath11 and the first few @xmath89 with @xmath90 and @xmath91 as described above . in doing so , states in the reduced hilbert space relevant to the excited states connected to the ground state via the operator of interest @xmath70 are included . the fact that @xmath92 is an excellent trial state , in particular , for the lowest triplet excitations of the two - dimensional antiferromagnet was shown in ref . of course , if the number @xmath4 of states kept per block is fixed , the more target states considered , the less precisely each one of them is described . an optimal number of target states and @xmath4 have to be found for each case . due to this reduction , the algorithm can be applied up to certain lengths , depending on the states involved . for longer chains , the higher energy excitations will become inaccurate . proper sum rules have to be calculated to determine the errors in each case . as an application of the method we calculate @xmath93 for the 1d isotropic heisenberg model with spin @xmath42 . the spin dynamics of this model has been extensively studied . the lowest excited states in the thermodynamic limit are the des cloiseaux - pearson triplets @xcite , having total spin @xmath94 . the dispersion of this spin - wave branch is @xmath95 . above this lower boundary there exists a two - parameter continuum of excited triplet states that have been calculated using the bethe ansatz approach @xcite with an upper boundary given by @xmath96 . it has been shown @xcite , however , that there are excitations above this upper boundary due to higher order scattering processes , with a weight that is at least one order of magnitude lower than the spin - wave continuum . in fig . 2 we show the spectrum for @xmath97 and @xmath98 for different values of @xmath4 , where exact results are available for comparison . the delta peaks of eq . ( [ eq : szzn ] ) are broadened by a lorentzian for visualizing purposes . as expected , increasing @xmath4 gives more precise results for the higher excitations . this spectra has been obtained using the infinite - system method and more precise results are expected using the finite - system method , as described later . = 3.3 in = 2.75 in in fig . 3 we show the spectrum for two systems lengths and @xmath97 and @xmath99 keeping @xmath100 states and periodic boundary conditions . for this case it was enough to take 3 target states , _ i. e. _ @xmath11 , @xmath101 and @xmath102 . here we have used @xmath103 pairs of coefficients @xmath83 and @xmath84 , but we noticed that if we considered only the first ( @xmath104 ) coefficients @xmath83 and @xmath84 , the spectrum at low energies remains essentially unchanged . minor differences arise at @xmath105 . this is another indication that only the first @xmath106 are relevant for the low energy dynamical properties for finite systems . in the inset of fig . 3 the spectrum for @xmath99 and @xmath107 is shown . for this case we considered 5 target states _ i. e. _ @xmath11 , @xmath108 , @xmath109 and @xmath100 . here , and for all the cases considered , we have verified that the results are very weakly dependent on the weights @xmath29 of the target states ( see eq.([eq : pl ] ) ) as long as the appropriate target states are chosen . for lengths where this value of @xmath57 is not defined we took the nearest value . = 3.3 in = 2.75 in even though we are including states with a given momentum as target states , due to the particular real - space construction of the reduced hilbert space , this translational symmetry is not fulfilled and the momentum is not fixed . to check how the reduction on the hilbert space influences the momentum @xmath57 of the target state @xmath110 , we calculated the expectation values @xmath111 for all @xmath112 . if the momenta of the states were well defined , this value is proportional to @xmath113 if @xmath114 . for @xmath115 , @xmath116 . the momentum distribution for @xmath97 is shown in fig . 4 in a semilogarithmic scale where the @xmath117-axis has been shifted by .003 so as to have well - defined logarithms . we can see here that the momentum is better defined , even for much larger systems , but , as expected , more weight on other @xmath112 values arises for larger @xmath7 . as a check of the approximation we calculated the sum rule @xmath118 for @xmath107 , 5 target states and @xmath100 . we obtain a relative error of 0.86% . recently , important improvements to this method have been published @xcite : by considering the finite system method in open chains , khner and white obtained a higher precision in dynamical responses of spin chains . in order to define a momentum in an open chain and to avoid end effects , they introduce a filter function with weight centered in the middle of the chain and zero at the boundaries . recent applications of this method include the calculation of excitations in spin - orbital models ( su(4 ) ) in a magnetic field@xcite , spin dynamics in models for cuprate spin ladders including cyclic spin exchange@xcite , optical conductivity of the ionic hubbard model@xcite , excitations in the one - dimensional bose - hubbard model@xcite and the optical response in 1d mott insulators@xcite . in this section we have presented a method of calculating dynamical responses with dmrg . although the basis truncation is big , this method keeps only the most relevant states and , for example , even by considering a @xmath119 of the total hilbert space ( for @xmath107 only @xmath120 40000 states are kept ) a reasonable description of the low energy excitations is obtained . we show that it is also possible to obtain states with well defined momenta if the appropriate target states are used . introduced in ref . @xcite in the dmrg context and improved in ref . @xcite , this method focuses on a particular energy or energy window , allowing a more precise description in that range and the possibility of calculating spectra for higher energies . instead of using the tridiagonalization of the hamiltonian , but in a similar spirit regarding the important target states to be kept , the spectrum can be calculated for a given @xmath121 by using a correction vector ( related to the operator @xmath70 that can depend on momentum @xmath57 ) . following ( [ eq : din ] ) , the ( complex ) correction vector @xmath122 can be defined as : @xmath123 so the green s function can be calculated as @xmath124 separating the correction vector in real and imaginary parts @xmath125 we obtain @xmath126 and @xmath127 the former equation is solved using the conjugate gradient method . in order to keep the information of the excitations at this particular energy the following states are targeted in the dmrg iterations : the ground state @xmath128 , the first lanczos vector @xmath129 and the correction vector @xmath122 . even though only a certain energy is focused on , dmrg gives the correct excitations for an energy range surrounding this particular point so that by running several times for nearby frequencies , an approximate spectrum can be obtained for a wider region @xcite . a variational formulation of the correction vector technique which leads to more accurate excited energies and spectral weights has been developed in @xcite . it has been successfully applied to calculate the optical conductivity of mott insulators@xcite . this method@xcite relies on a moment expansion of the dynamical correlations using sum rules that depend only on static correlation functions which can be calculated with dmrg . with these moments , the green s functions can be calculated using the maximum entropy method . the first step is the calculation of sum rules . as an example , and following @xcite , the spin - spin correlation function @xmath130 of the heisenberg model is calculated where the operator @xmath70 of eq . ( [ eq : ca ] ) is @xmath131 and the sum rules are@xcite : @xmath132,s^z(-q ) ] \rangle \nonumber \\ & = & 2[1-\cos(q)]\sum_i\langle s^+_{i}s^-_{i+1}+s^-_{i}s^+_{i+1}\rangle\end{aligned}\ ] ] where @xmath133 is the static susceptibility . these sum rules can be easily generalized to higher moments : @xmath134 ... ],s^z(-q)]\rangle \end{aligned}\ ] ] for @xmath12 odd . a similar expression is obtained for @xmath12 even , where the outer square bracket is replaced by an anticommutator and the total sign is changed . here @xmath75 appears in the commutator @xmath135 times . apart from the first moment which is given by the static susceptibility , all the other moments can be expressed as equal time correlations ( using a symbolic manipulator ) . the static susceptibility @xmath136 is calculated by applying a small field @xmath137 and calculating the density response @xmath138 with dmrg . then @xmath139 for @xmath140 . these moments are calculated for several chain lengths and extrapolated to the infinite system . once the moments are calculated , the final spectra is constructed via the maximum entropy method ( me ) , which has become a standard way to extract maximum information from incomplete data ( for details see ref . @xcite and references therein ) . reasonable spectra are obtained for the xy and isotropic models , although information about the exact position of the gaps has to be included . otherwise , the spectra are only qualitatively correct . this method requires the calculation of a large amount of moments in order to get good results : the more information given to the me equations , the better the result . in order to include temperature in the calculation of dynamical quantities , the transfer matrix rg described above ( tmrg@xcite ) was extended to obtain imaginary time correlation functions@xcite . after fourier transformation in the imaginary time axis , analytic continuation from imaginary to real frequencies is done using maximum entropy ( me ) . the combination of the tmrg and me is free from statistical errors and the negative sign problem of monte carlo methods . since we are dealing with the transfer matrix , the thermodynamic limit can be discussed directly without extrapolations . however , in the present scheme , only local quantities can be calculated . a systematic investigation of local spectral functions is done in ref . @xcite for the anisotropic heisenberg antiferromagnetic chain . the authors obtain good qualitative results especially for high temperatures but a quantitative description of peaks and gaps are beyond the method , due to the severe intrinsic limitation of the analytic continuation . this method was also applied with great success to the 1d kondo insulator@xcite . the temperature dependence of the local density of states and local dynamic spin and charge correlation functions were calculated . we have presented here a very brief description of the density matrix renormalization group technique , its applications and extensions . the aim of this article is to give the unexperienced reader an idea of the possibilities and scope of this powerful , though relatively simple method . the experienced reader can find here an extensive ( however incomplete ) list of references covering most applications to date using dmrg in a great variety of fields such as condensed matter , statistical mechanics and high energy physics . the author acknowledges hospitality at the centre de recherches mathematiques , university of montreal and at the physics department of the university of buenos aires , argentina , where this work has been performed . we thank s. white for a critical reading of the manuscript and all those authors that have updated references and sent instructive comments . k. h. is a fellow of conicet , argentina . grants : pict 03 - 00121 - 02153 and pict 03 - 00000 - 00651 . 999 s. white , phys . lett . * 69 * , 2863 ( 1992 ) , edited by i. peschel , x. wang , m. kaulke and k. hallberg ( series : lecture notes in physics , springer , berlin , 1999 ) s. white , phys . b * 48 * , 10345 ( 1993 ) j. bray and s. chui , phys . rev . b * 19 * , 4876 ( 1979 ) c. pan and x. chen , phys . b * 36 * , 8600 ( 1987 ) ; m. kovarik , phys . rev . b * 41 * , 6889 ( 1990 ) t. xiang and g. gehring , phys . b * 48 * , 303 ( 1993 ) k. wilson , _ rev . mod . phys . _ * 47 * , 773 ( 1975 ) s. white and r. noack , phys . . lett . * 68 * , 3487 , ( 1992 ) ; r. noack and s. white , phys . b * 47 * , 9243 ( 1993 ) j. gaite , mod . a*16 * , 1109 ( 2001 ) r. noack and s. white in ref . @xcite , chap . 2(i ) m. martn - delgado , g. sierra and r. noack , j. phys . a : math . gen . * 32 * , 6079 ( 1999 ) r. feynman , _ statistical mechanics : a set of lectures _ , ( benjamin , reading , ma , 1972 ) s. r. white , i. affleck and d. scalapino , phys . b * 65 * , 165122 ( 2002 ) see e. dagotto , rev . mod . phys . * 66 * , 763 ( 1994 ) e. r. davidson , j. comput . phys . * 17 * , 87 ( 1975 ) ; e.r . davidson , _ computers in physics _ * 7 * , no . 5 , 519 ( 1993 ) . t. osborne and m. nielsen , quantum information processing , volume 1 , issue 1 - 2 , 45 ( 2002 ) ( available on - line : http://www.kluweronline.com/issn/1570-0755/ ) s. white , phys . lett . * 77 * , 3633 ( 1996 ) t. nishino and k. okunishi , phys . 64 * , 4084 ( 1995 ) ; u. schollwck , phys . rev . b * 58 * , 8194 ( 1998 ) and phys . rev . b * 59 * , 3917 ( 1999 ) ( erratum ) . . legeza and g. fth , phys . b * 53 * , 14349 ( 1996 ) ; m - b lepetit and g. pastor , phys . rev . b * 58 * , 12691 ( 1998 ) s. stlund and s. rommer , phys . rev . lett . * 75 * , 3537 ( 1995 ) ; phys . b * 55 * , 2164 ( 1997 ) ; m. andersson , m. boman and s. stlund , phys . b * 59 * , 10493 ( 1999 ) ; h. takasaki , t. hikihara and t. nishino , j. phys . soc * 68 * , 1537 ( 1999 ) ; k. okunishi , y. hieida and y.akutsu , phys . e * 59 * , r6227 ( 1999 ) m. a. martn - delgado and g. sierra , int a , * 11 * , 3145 ( 1996 ) g. sierra and m. a. martn - delgado , in `` the exact renormalization group '' by a krasnitz , r potting , y a kubyshin and p. s. de sa ( eds . ) world scientific pub co ; isbn : 9810239394 , ( 1999 ) , ( cond - mat/9811170 ) i. peschel , m. kaulke and . legeza , annalen der physik * 8 * , 153 ( 1999),(cond - mat/9810174 ) s. r. white and d. huse , phys . rev . b * 48 * , 3844 ( 1993 ) e. s. srensen and i. affleck , phys . b * 49 * , 13235 ( 1994 ) e. s. srensen and i. affleck , phys . rev . b * 49 * , 15771 ( 1994 ) ; e. polizzi , f. mila and e. srensen , phys . rev . b * 58 * , 2407 ( 1998 ) c. batista , k. hallberg and a. aligia , phys . rev . b * 58 * , 9248 ( 1998 ) ; phys . rev . b * 60 * , 12553 ( 1999 ) ; physica b , * 259 * , 1017 ( 1999 ) ; e. jannod , c. payen , k. schoumacker , c. batista , k. hallberg and a. aligia , phys . rev b , * 62 * , 2998 ( 2000 ) m. laukamp et al . b * 5 * , 10755 ( 1998 ) t - k . ng , j. lou and z. su , phys . b * 61 * , 11487 ( 2000 ) j. lou , s. qin , t - k . ng , z. su and i. affleck , phys . b * 62 * , 3786 ( 2000 ) s - w tsai and j. b. marston , ann . phys . ( leipzig ) 8 , special issue , 261 ( 1999 ) m. sieling , u. lw , b. wolf , s. schmidt , s. zvyagin and b.lthi , phys . b * 61 * , 88 ( 2000 ) y. nishiyama , k. totsuka , n. hatano and m. suzuki , j. phys . soc . jpn * 64 * , 414 ( 1995 ) s. qin , t. ng and z - b su , phys . b * 52 * , 12844 ( 1995 ) u. schollwck and t. jolicoeur , europhys . lett . * 30 * , 493 ( 1995 ) x. wang , s. qin and lu yu , phys . b. * 60 * , 14529 ( 1999 ) w. tatsuaki , phys . e * 61 * , 3199 ( 2000 ) ; t. wada and t. nishino , cond - mat/0103508 ( proceedings of the conference on computational physics 2000 ( ccp2000 ) , gold coast , queensland , australia , 3 - 8 december 2000 ) m. capone and s. caprara , phys . b * 64 * , 184418 ( 2001 ) j. lou , x. dai , s. qin , z. su and l. yu , phys . rev . b * 60 * , 52 ( 1999 ) e. ercolessi , g. morandi , p. pieri and m. roncaglia , europhys . lett . * 49 * , 434 ( 2000 ) s. qin , x. wang and lu yu , phys . b * 56 * , r14251 ( 1997 ) k. hallberg , p. horsch and g. martnez , phys . rev . b , * 52 * , r719 ( 1995 ) h. e. boos , v. e. korepin , y. nishiyama and m. shiroishi , j. phys . a*35 * , 4443 ( 2002 ) ; v. e. korepin , s. lukyanov , y. nishiyama and m. shiroishi , cond - mat/0210140 m. shiroishi , m. takahashi and y. nishiyama , j. phys . * 70 * , 3535 ( 2001 ) t. hikihara and a. furusaki , phys . b * 58 * , r583 ( 1998 ) g. fath , cond - mat/0208580 j. lou , s. qin , c. chen , z. su and l. yu , phys . rev . b * 65 * , 064420 ( 2002 ) f. capraro and c. gros , cond - mat/0207279 y. hieida , k. okunishi and y. akutsu , phys . rev . b * 64 * , 224422 ( 2001 ) t. byrnes , r. bursill , h - p . eckle , c. hamer and a. sandvik , cond - mat/0205140 s - w . tsai and j. b. marston , phys . b * 62 * , 5546 ( 2000 ) k. hallberg , x. wang , p. horsch and a. moreo , phys . rev . lett . * 76 * , 4955 ( 1996 ) a. moreo , phys . rev . b * 35 * , 8562 ( 1987 ) ; t. ziman and h. schulz , phys . lett * 59 * , 140 ( 1987 ) y. yamashita , n. shibata and k. ueda , phys . b * 58 * , 9114 ( 1998 ) ; j. phys . vol.69 , 242 - 247 ( 2000 ) ; phys . b * 61 * , 4012 ( 2000 ) a. onufriev and b. marston , phys . rev . b * 59 * , 12573 ( 1999 ) c. itoi , s. qin and i. affleck , phys . rev . b * 61 * , 6747 ( 2000 ) w. yu and s. haas , phys . rev . b * 63 * , 024423 ( 2001 ) r. j. bursill , t. xiang and g. a. gehring , j.phys . a * 28 * 2109 ( 1994 ) r. j. bursill , g. a. gehring , d. j. j. farnell , j. b. parkinson , t. xiang and c. zeng , j. phys . c * 7 * 8605 ( 1995 ) u. schollwck , th . jolicoeur and t. garel , phys . b * 53 * , 3304 ( 1996 ) r. chitra , s. pati , h. r. krishnamurthy , d. sen and s. ramasesha , phys . b * 52 * , 6581 ( 1995 ) ; s. pati , r. chitra , d. sen , h. r. krishnamurthy and s. ramasesha , europhys . * 33 * , 707 ( 1996 ) ; j. malek , s. drechsler , g. paasch and k. hallberg , phys . b * 56 * , r8467 ( 1997 ) ; e. srensen et al in ref.@xcite , chap.1.2 ( part ii ) and references therein ; d. augier , e. srensen , j. riera and d. poilblanc , phys . rev . b * 60 * , 1075 ( 1999 ) y. kato and a. tanaka , j. phys . jpn . * 63 * , 1277 ( 1994 ) s. r. white and i. affleck , phys . b * 54 * , 9862 ( 1996 ) g. bouzerar , a. kampf and g. japaridze , phys . rev . b * 58 * , 3117 ( 1998 ) ; g. bouzerar , a. kampf and f. schnfeld , cond - mat/9701176 , unpublished ; m - b . lepetit and g. pastor , phys . b * 56 * , 4447 ( 1997 ) m. kaburagi , h. kawamura and t. hikihara , j. phys . soc . jpn . * 68*. 3185 ( 1999 ) t. hikihara , j. phys . . jpn . * 71 * , 319 ( 2002 ) ; t. hikihara , m. kaburagi , h. kawamura and t. tonegawa , phys . b * 63 * , 174430 ( 2001 ) s. pati , s. ramasesha and d. sen , phys . rev . b * 55 * , 8894 ( 1996 ) ; j. phys . cond * 9 * , 8707 ( 1997 ) ; t. tonegawa _ et al _ , j. mag . * 177 - 181 * , 647 ( 1998 ) , ( cond - mat/9712298 ) m. azzouz , l. chen and s. moukouri , phys . b * 50 * 6223 ( 1994 ) ; s. r. white , r. m. noack and d. j. scalapino , phys . lett . * 73 * 886 ( 1994 ) ; k. hida , j. phys . jpn . * 64 * 4896 ( 1995 ) ; t. narushima , t. nakamura and s. takada , j. phys . * 64 * 4322 ( 1995 ) ; u. schollwck and d. ko , phys . b * 53 * 240 ( 1996 ) ; g. sierra , m. a. martn - delgado , s. white and j. dukelsky , phys . b * 59 * , 7973 ( 1999 ) ; s. white , phys . rev . b * 53 * , 52 ( 1996 ) a. kawaguchi , a. koga , k. okunishi and n. kawakami , cond - mat/0205635 a. trumper and c. gazza , phys . rev . b * 64 * , 134408 ( 2001 ) m. roger , phys . rev . b * 63 * , 144433 ( 2001 ) t. hikihara and a. furusaki , phys . rev . b * 63 * , 134438 ( 2001 ) a. laeuchli , g. schmida and m. troyer , cond - mat/0206153 t. hikihara , t. momoi and x. hu , cond - mat/0206102 t. nunner , p. brune , t. kopp , m. windt and m. grninger , cond - mat/0203472 y. honda and t. horiguchi , cond - mat/0106426 s. k. pati and r. r. p. singh , phys . rev . b * 60 * , 7695 ( 1999 ) ; s. r. white and r. singh , phys . lett . * 85 * , 3330 ( 2000 ) n. maeshima , m. hagiwara , y. narumi , k. kindo , t. c. kobayashi and k. okunishi , cond - mat/0208373 ; n. maeshima and k. okunishi , phys . b * 62 * , 934 - 939 ( 2000 ) c. itoi and s. qin , phys . b * 63 * , 224423 ( 2001 ) j. lou , j. dai , s. qin , z. su and l. yu , phys . rev . b * 62 * , 8600 ( 2000 ) c. raghu , i. rudra , s. ramasesha and d. sen , phys . rev . b * 62 * , 9484 ( 2000 ) a. honecker , m. kaulke and k.d . schotte , eur . j. b * 15 * , 423 ( 2000 ) a. gendiar and a. surda , phys . b * 62 * , 3960 ( 2000 ) k. tandon et al , phys . rev . b * 59 * , 396 ( 1999 ) ; r. citro , e. orignac , n. andrei , c. itoi and s. qin , j. phys.:cond . mat . * 12 * , 3041 ( 2000 ) j. lou , c. chen and s. qin , phys . rev . b * 64 * , 144403 ( 2001 ) x. wang and l. yu , phys lett . * 84 * , 5399 ( 2000 ) s. k. pati , s. ramasesha and d. sen , in _ magnetism : molecules to materials iv _ , j.s . miller and m. drillon ( wiley - vch weinheim , 2002 ) , chapter 4 , ( cond - mat/0106621 ) l. chen and s. moukouri , phys . b * 53 * 1866 ( 1996 ) ; s. j. qin , s. d. liang , z. b. su and l. yu , phys . b * 52 * r5475 ( 1995 ) ; r. noack in ref.@xcite , chap.1.3 ( part ii ) ; s. daul and r. noack , phys . rev . b * 61 * , 12361 ( 2000 ) ; m. vojta , r. hetzel and r. noack , phys . b * 60 * , r8417 ( 1999 ) k. penc , k. hallberg , f. mila and h. shiba , phys . * 77 * , 1390 ( 1996 ) ; phys . b , * 55 * , 15475 ( 1997 ) n. bulut , cond - mat/0207186 , to appear in advances in physics , * 51 * , no . 6 ( 2002 ) a. malvezzi , t. paiva and r. dos santos , cond - mat/0205266 e. jeckelmann , cond - mat/0204244 y. zhang , c. wu and h. q.lin , phys . b * 65 * , 115101 ( 2002 ) c. aebischer , d. baeriswyl and r. m. noack , phys . rev . lett . * 86 * , 468 ( 2001 ) r. arita and h. aoki , phys . b * 61 * , 12261 ( 2000 ) ; r. arita , y. shimoi , k. kuroki and h. aoki , phys . rev . b * 57 * , 10609 ( 1998 ) s. daul , cond - mat/9911361 s. qin , j. lou , t. xiang , g - s . tian and z. su , cond - mat/0004162 a. aligia , k. hallberg , c. batista and g. ortiz , phys . b * 61 * , 7883 ( 2000 ) v. meden , w. metzner , u. schollwoeck , o. schneider , t. stauber and k. schoenhammer , , europhys . j b * 16 * , 631 ( 2000 ) p. farkasovsky , phys . rev . b * 65 * , 081102 ( 2002 ) c. schuster , r. roemer and m. schreiber , phys . rev . b * 65 * , 115114 ( 2002 ) k. hida , phys . lett . * 86 * , 1331 ( 2001 ) w. lay and j. rudnick , phys . lett . * 88 * , 057203 ( 2002 ) ; y. nishiyama , j.phys . a*34 * , 11215 ( 2001 ) ; s. g. chung , phys . e * 62 * , 3262 ( 2000 ) b. srinivasan and m. b. lepetit , phys . b * 66 * , 024421 ( 2002 ) ; h. sakamoto , t. momoi and k. kubo , phys . b * 65 * , 224403 ( 2002 ) k. hamacher , c. gros and w. wenzel , phys . rev . lett . * 88 * , 217203 ( 2002 ) m. vojta , a. huebsch and r. m. noack , phys . b * 63 * , 045105 ( 2001 ) ; z. weihong , j. oitmaa , c.j . hamer and r.j . bursill , j. phys . c * 13 * , 433 ( 2001 ) youngho park , s. liang and t. k. lee , phys . b * 59 * , 2587 ( 1999 ) s. rommer , s. r. white and d. j. scalapino , phys . rev . b * 61 * , 13424 ( 2000 ) r. m. noack , s. r. white and d. j. scalapino , phys lett . * 73 * , 882 ( 1994 ) ; s. r. white , r. m. noack and d. j. scalapino , j. low temp . phys . * 99 * , 593 ( 1995 ) ; r. m. noack , s. r. white and d. j. scalapino , europhys . lett . * 30 * , 163 ( 1995 ) ; c. a. hayward , d. poilblanc , r. m. noack , d. j. scalapino and w. hanke , phys . lett . * 75 * , 926 ( 1995 ) ; s. white and d. scalapino , phys . rev . lett . * 81 * , 3227 ( 1998 ) ; e. jeckelmann , d. scalapino and s. white , phys . rev . b * 58 * , 9492 ( 1998 ) s. nishimoto , e. jeckelmann and d. scalapino , cond - mat/0208189 j. b. marston , j. o. fjaerestad and a. sudbo , phys . rev . lett . * 89 * , 056404 ( 2002 ) u.schollwoeck , sudip chakravarty , j. o. fjaerestad , j. b. marston and m. troyer , cond - mat/0209444 l. g. caron and c. bourbonnais , phys . rev . b * 66 * , 045101 ( 2002 ) g. fano , f. ortolani , a. parola and l. ziosi , phys . rev . b * 60 * , 15654 ( 1999 ) n. shibata and d. yoshioka , phys . 86 * , 5755 ( 2001 ) ; physica e * 12 * , 43 ( 2002 ) d. yoshioka and n. shibata , j. phys . . jpn . * 70 * , 3690 ( 2001 ) v. meden and u. schollwck , cond - mat/0209588 s. liang and h. pang , europhys . lett . * 32 * , 173 ( 1995 ) s. white and d. scalapino , phys . rev . b * 55 * , 14701 ( 1997 ) s. white and d. scalapino , phys . rev . b * 55 * , 6504 ( 1997 ) s. white and d. scalapino , phys . rev . b * 57 * , 3031 ( 1998 ) e. arrigoni , a. p. harju , w. hanke , b. brendel and s.a . kivelson , phys . b * 65 * , 134503 ( 2002 ) ; j. bonca , j. e. gubernatis , m. guerrero , e. jeckelmann and s. r. white , phys . rev . b * 61 * , 3251 ( 2000 ) ; a. l. chernyshev , s. white and a. h. castro neto , phys . rev . b , * 65 * , 214527 ( 2002 ) ; s. white and d. scalapino , phys . rev . b * 61 * , 6320 ( 2000 ) s. white and d. scalapino , phys . rev . lett . * 80 * , 1272 ( 1998 ) ( see also cond - mat/9907375 ) i. p. mcculloch , a. r. bishop and m. gulacsi , phil . mag . b * 81 * , 1603 ( 2001 ) s. r. white and i. affleck , phys . rev . b * 64 * , 024411 ( 2001 ) t. xiang , j. lou and z. su , phys . rev . b * 64 * , 104414 ( 2001 ) ; p. henelius , phys . rev . b * 60 * , 9561 ( 1999 ) s. nishimoto , e. jeckelmann , f. gebhard and r. noack , phys . b * 65 * , 165114 ( 2002 ) n. maeshima , y. hieida , y. akutsu , t. nishino and k. okunishi , phys . rev . e * 64 * ( 2001 ) 016705 m.a . martin - delgado , j. rodriguez - laguna and g. sierra , nucl . b * 601 * , 569 ( 2001 ) p. henelius , phys . b * 60 * , 9561 ( 1999 ) s. moukouri and l. g. caron , cond - mat/0210668 t. a. costi , p. schmitteckert , j. kroha and p. wlfle , phys . lett . * 73 * , 1275 ( 1994 ) ; s. eggert and i. affleck , phys . rev . lett . * 75 * , 934 ( 1995 ) ; e. s. srensen and i. affleck , phys . b * 51*. 16115 ( 1995 ) ; x. q. wang and s. mallwitz , phys . rev . b * 53 * , r492 ( 1996 ) ; w. wang , s. j. qin , z. y. lu , l. yu and z. b. su , phys . b * 53 * , 40 ( 1996 ) ; c. c. yu and m. guerrero , phys . rev . b * 54 * , 15917 ( 1996 ) ; a. furusaki and t. hikihara , phys . rev . b * 58 * , 5529 ( 1998 ) k. hallberg and r. egger , phys . rev . b * 55 * , 8646 ( 1997 ) c. schuster and u. eckern , ann . phys . ( leipzig ) , * 11 * , 901 ( 2002 ) ( cond - mat/0201390 ) ; w. zhang , j. igarashi and p. fulde , phys . b * 56 * , 654 ( 1997 ) v. meden , w. metzner , u. schollwoeck and k. schoenhammer , phys . rev . b * 65 * , 045318 ( 2002 ) c. c. yu and s. r. white , phys . rev . lett . * 71 * , 3866 ( 1993 ) ; c. c. yu and s. r. white , physica b * 199 * , 454 ( 1994 ) s. moukouri and l. g. caron , phys . rev . b * 52 * , 15723 ( 1995 ) ; n. shibata , t. nishino , k. ueda and c. ishii , phys . b * 53 * , r8828 ( 1996 ) ; m. guerrero and r. m. noack , phys . b * 53 * , 3707 ( 1996 ) h. otsuka and t. nishino , phys . b * 52 * , 15066 ( 1995 ) ; s. moukouri , l. g. caron , c. bourbonnais and l. hubert , phys . b * 51 * , 15920 ( 1995 ) j. c. xavier , e. novais and e. miranda , phys . b * 65 * , 214406 ( 2002 ) t. yamamoto , r. manago and y. mori , cond - mat/0204588 i. p. mcculloch , a. juozapavicius , a. rosengren and m. gulacsi , phys . rev . b * 65 * , 052410 ( 2002 ) m. guerrero and r.m . noack , phys . rev . b * 63 * , 144423 ( 2001 ) s. watanabe , j. phys . . jpn . * 69 * , 2947 ( 2000 ) i. p. mcculloch , m. gulacsi , s. caprara and a. juozapavicius , j. low temp . 117 , 323 ( 1999 ) j. xavier , r. pereira , e. miranda and i. affleck , cond - mat/0209623 s. watanabe , y. kuramoto , t. nishino and n. shibata , j. phys . soc . jpn * 68 * , 159 ( 1999 ) j. moreno , s. qin , p. coleman and l. yu , phys . b * 64 * , 085116 ( 2001 ) s. moukouri , l. chen and l. g. caron , phys . rev . b * 53 * , r488 ( 1996 ) j. riera , k. hallberg and e. dagotto , phys . rev . lett . * 79 * , 713 ( 1997 ) ; e. dagotto et al . , b * 58 * , 6414 ( 1998 ) d. garca , k. hallberg , c. batista , m. avignon and b. alascio , phys . rev lett . , * 85 * , 3720 ( 2000 ) ; d. j. garca , k. hallberg , c. d. batista , s. capponi , d. poilblanc , m. avignon and b. alascio , phys . rev . b. * 65 * , 134444 ( 2002 ) b. ammon and m. imada , j. phys . . jpn . * 70 * , 547 ( 2001 ) s. ramasesha et al , phys . b * 54 * , 7598 ( 1996 ) ; synth . metals * 85 * , 1019 ( 1997 ) e. s. srensen and i. affleck , phys . b * 51 * , 16115 ( 1995 ) i. mcculloch and m. gulacsi , europhys . lett . * 57 * , 852 ( 2002 ) g. sierra and t. nishino , nucl . b * 495 * , 505 ( 1997 ) ( cond - mat/9610221 ) t. xiang , phys . b * 53 * , r10445 ( 1996 ) m.s.l . du croo de jongh and j.m.j . van leeuwen , phys . b * 57 * , 8494 ( 1998 ) m - b . lepetit , m. cousy and g. pastor , eur . j. b * 13 * , 421 ( 2000 ) ; h. otsuka , phys . rev . b * 53 * , 14004 ( 1996 ) j. dukelsky and g. dussel , phys . rev . b * 59 * , r3005 ( 1999 ) and references therein . j. dukelsky and g. sierra , phys . lett . * 83 * , 172 ( 1999 ) m. a. martn - delgado and g. sierra , phys . lett . * 83 * , 1514 ( 1999 ) s. glazek and k. wilson , phys . d * 48 * , 5863 ( 1993 ) ; * 49 * , 4214 ( 1994 ) t. eller , h .- c . pauli and s. brodsky , phys . d * 35 * , 1493 ( 1987 ) t. nishino , j. phys . . jpn . * 64 * , 3598 ( 1995 ) ; see also t. nishino in ref . @xcite , chap . 5(i ) ; t. nishino and k. okunishi in _ strongly correlated magnetic and superconducting systems _ , ed . g. sierra and m. a. martn - delgado ( springer , berlin , 1997 ) h. trotter , proc . * 10 * , 545 ( 1959 ) ; m. suzuki , prog . . phys . * 56 * , 1454 ( 1976 ) ; r. feynman and a. hibbs _ quantum mechanics and path integrals _ ( mcgraw - hill , 1965 ) a. gendiar and a. surda , phys . b * 63 * , 014401 ( 2001 ) t. nishino and n. shibata , j. phys . soc . jpn . * 68 * , 3501 ( 1999 ) ; t. enss and u. schollwoeck , j. phys . a * 34 * , 7769 ( 2001 ) a. kemper , a. schadschneider and j. zittartz , j. phys . a : math . gen . * 34 * , l279 ( 2001 ) ; t. enss and u. schollwck , j phys a : math . gen . * 34 * , 7769 ( 2001 ) r. baxter , j. math . phys . * 9 * , 650 ( 1968 ) ; j. stat phys . * 19 * , 461 ( 1978 ) t. nishino and k. okunishi , j. phys . soc . jpn . * 65*,891 ( 1996 ) ; ibid . * 66 * , 3040 ( 1997 ) ; t. nishino , k. okunishi and m. kikuchi , phys . a * 213 * , 69 ( 1996 ) h. takasaki , t. nishino and y. hieida , j. phys . . jpn . * 70 * , 1429 ( 2001 ) t. nishino , y. hieida , k. okunishi , n. maeshima , y. akutsu and a. gendiar , prog . theor . phys . * 105 * , 409 ( 2001 ) c. ritter and g. von gehlen , in `` quantization , gauge theory and strings '' , ed . et al . , vol.i , p.563 - 578 , scientific world publ.co . ( 2001 ) ( cond - mat/0009255 ) t. nishino and k. okunishi , j. phys . . jpn . * 67 * , 3066 ( 1998 ) ; t. nishino , k. okunishi , y. hieida , n. maeshima and y. akutsu , nucl . phys . b * 575 * , 504 ( 2000 ) e. carlon and a. drzewiski , phys . rev . lett . * 79 * , 1591 ( 1997 ) ; phys . e * 57 * , 2626 ( 1998 ) ; e. carlon , a. drzewiski and j. rogiers , phys . b * 58 * , 5070 ( 1998 ) ; a. drzewiski , a. ciach and a. maciolek , eur . j. b * 5 * , 825 ( 1998 ) ; phys . e * 60 * , 2887 ( 1999 ) a. drzewinski , a. maciolek and r. evans , phys . lett . * 85 * , 3079 ( 2000 ) ; a. drzewinski , phys . rev . e * 62 * , 4378 ( 2000 ) m .- c . chung , m. kaulke , i. peschel , m. pleimling and w. selke , eur . j. b * 18 * , 655 ( 2000 ) e. carlon and f. igli , phys . b * 57 * , 7877 ( 1998 ) ; phys . rev . b * 59 * , 3783 ( 1999 ) ; e. carlon , c. chatelain and b. berche , phys . rev . b * 60 * , 12974 ( 1999 ) j. kondev and j. marston , nucl . b * 497 * , 639 ( 1997 ) ; t. senthil , b. marston and m. fisher , phys . b * 60 * , 4245 ( 1999 ) ; j. marston and s. tsai , phys . lett . * 82 * , 4906 ( 1999 ) ; s. tsai and j. marston , ann . ( leipzig ) * 8 * , special issue , 261(1999 ) m. kaulke and i. peschel , eur . j. b * 5 * , 727 ( 1998 ) y. hieida , j. phys . . jpn . * 67 * , 369 ( 1998 ) i. peschel and m. kaulke , in ref . @xcite , chap . 3.1(ii ) e. carlon , m. henkel and u. schollwck , eur . phys . j. b * 12 * , 99(1999 ) ; e. carlon , m. henkel and u. schollwoeck , phys . e * 63 * , 036101 ( 2001 ) j. hooyberghs , e. carlon and c. vanderzande , phys . e * 62 * , 036124 ( 2001 ) s.g . chung , phys . b * 60 * , 11761(1999 ) r. bursill , t. xiang and g. gehring , j. phys . c * 8 * , l583 ( 1996 ) h. trotter , proc . * 10 * , 545 ( 1959 ) ; m. suzuki , prog . . phys . * 56 * , 1454 ( 1976 ) x. wang and t. xiang , phys . b * 56 * , 5061 ( 1997 ) n. shibata and k. ueda , j. phys . soc vol.70 , 3690 ( 2001 ) n. shibata , j. phys . . jpn . * 66 * , 2221 ( 1997 ) t. xiang , phys . b * 58 * , 9142 ( 1998 ) n. shibata , b. ammon , t. troyer , m. sigrist and k. ueda , j. phys . . jpn . * 67 * , 1086 ( 1998 ) k. maisinger , u. schollwck , s. brehmer , h - j . mikeska and s. yamamoto , phys . b * 58 * , r5908 ( 1998 ) s. rommer and s. eggert , phys . rev . b * 59 * , 6301 ( 1999 ) i. maruyama , n. shibata and k. ueda , phys . rev . b * 65 * , 174421 ( 2002 ) k. maisinger and u. schollwck , phys . rev . lett . * 81 * , 445 ( 1999 ) b. ammon , m. troyer , t. rice and n. shibata , phys . lett . * 82 * , 3855 ( 1999 ) ; n. shibata and h. tsunetsugu , j. phys . . jpn . * 68 * , 3138 ( 1999 ) f. naef and x. wang , phys . lett . * 84 * , 1320 ( 2000 ) a. klmper , r. raupach and f. schnfeld , phys . b * 59 * , 3612 ( 1999 ) s. moukouri and l. caron , phys . rev . lett . * 77 * , 4640 ( 1996 ) . s. moukouri and l. caron , see ref.@xcite , chap . 4.5(ii ) s. moukouri , cond - mat/0011169 w. zhang , j. igarashi and p. fulde , j. phys . . jpn . * 66 * , 1912 ( 1997 ) l. caron and s. moukouri , phys . lett . * 76 * , 4050 ( 1996 ) ; phys . b * 56 * , r8471 ( 1997 ) e. jeckelmann and s. white , phys . rev . b * 57 * , 6376 ( 1998 ) r. noack , s. white and d. scalapino in _ computer simulations in condensed matter physics vii _ , edited by d. landau , k .- k . mon and h .- b . schttler ( springer verlag , heidelberg and berlin , 1994 ) e. jeckelmann , c. zhang and s. white in ref . @xcite , chap . 5.1 ( ii ) c. zhang , e. jeckelmann and s. white , phys . lett . * 80 * , 2661 ( 1998 ) ; e. jeckelmann , c. zhang and s. r. white , physical review b * 60 * , 7950 ( 1999 ) y. nishiyama , cond - mat/0102123 eur . j. b * 12 * , 547 ( 1999 ) r. bursill , y. mckenzie and c. hammer , phys . lett . * 80 * , 5607 ( 1998 ) ; * 83 * , 408 ( 1999 ) ; r. bursill , phys . rev . b * 60 * , 1643 ( 1999 ) r. pai , r. pandit , h. krishnamurthy and s. ramasesha , phys . lett . * 76 * , 2937 ( 1996 ) ( see also the comment by n. v. prokofev and b. v. svistunov , phys . rev . lett . * 80 * , 4355 ( 1998 ) ) ; s. rapsch , u. schollwck and w. zwerger , europhys . lett . * 46 * , 559 ( 1999 ) t. khner and h. monien , phys . b * 58 * , r14741 ( 1998 ) i. peschel and m - c . chung , j. phys . gen . * 32 * , 8419 ( 1999 ) p. maurel and m - b . lepetit , phys . b * 62 * , 10744 ( 2000 ) ; p. maurel , m - b . lepetit and d. poilblanc , eur . j. * b 21 * , 481 ( 2001 ) k. hida , j. phys . . jpn . * 65 * , 895 ( 1996 ) and 3412 ( 1996 ) ( erratum ) ; j. phys . . jpn . * 66 * , 330 ( 1997 ) ; j. phys . . jpn . * 66 * , 3237 ( 1997 ) ; prog . . suppl . * 145 * , 320 ( 2002 ) k. hida , j. phys . . jpn . * 68 * , 3177 ( 1999 ) k. hida , phys . lett . * 83 * , 3297 ( 1999 ) f. igloi , r. juhasz and p. lajko , phys . lett . * 86 * , 1343 ( 2001 ) r. mlin , y - c . lin , p. lajk , h. rieger and f. igli , phys . b * 65 * , 104415 ( 2002 ) p. schmitteckert , t. schulze , c. schuster , p. schwab and u. eckern , phys . rev . lett . * 80 * , 560 ( 1998 ) ; p. schmitteckert and u. eckern , phys . b * 53 * , 15397 ( 1996 ) d. weinmann , p. schmitteckert , r. jalabert and j. pichard , eur j. b * 19 * , 139 - 156 ( 2001 ) p. schmitteckert , r. jalabert , d. weinmann and j. l. pichard , phys . lett . * 81 * , 2308 ( 1998 ) e. carlon , p. lajko and f. igli , phys . lett . * 87 * , 277201 ( 2001 ) g. fano , f. ortolani and l. ziosi , j. chem . phys . * 108 * , 9246 ( 1998),(cond - mat/9803071 ) ; r. bursill and w. barford , phys . rev . lett . * 82 * , 1514 ( 1999 ) m. exler and j. schnack , cond - mat/0205068 b. normand , x. wang , x. zotos and daniel loss , phys . b * 63 * , 184409 ( 2001 ) c. raghu , y. anusooya pati and s. ramasesha , journal - ref : j. phys . a * 34 * , 11215 ( 2001 ) w. barford and r. bursill , chem . . lett . * 268 * , 535(1997 ) ; w. barford , r. bursill and m. lavrentiev , j. phys : cond . matt , * 10 * , 6429 ( 1998 ) ; w. barford in ref.@xcite , chap.2.3 ( part ii ) and references therein ; m. lavrentiev , w. barford , s. martin , h. daly , r. bursill , physical review b * 59 * , 9987 ( 1999 ) e. carlon , a. drzewinski and j. van leeuwen , j. chem . phys . * 117 * , 2425 ( 2002 ) ; phys . rev . e * 64 * , 010801(r ) ( 2001 ) ; m. paessens and g. schtz , phys . e * 66 * , 021806 ( 2002 ) ; m. a. martn - delgado , j. rodriguez - laguna and g. sierra , phys . b * 65 * , 155116 ( 2002 ) s. white and r. martin , j. chem . phys . * 110 * , 4127 ( 1999 ) ; see also s. white in @xcite , chap . s. white in ref . @xcite , chap . 2.1 . . legeza , j. rder and b.a . hess , cond - mat/0208187 s. daul , i. ciofini , c. daul and s. r. white , int . j. of quantum chem . , * 79 * , 331(2000 ) ( cond - mat/9912348 ) c. bauschlicher and p. taylor , j. chem . phys . * 85 * , 2779 ( 1986 ) . legeza , j. roder and b. a. hess , cond - mat/0204602 w. hofstetter , phys . lett . * 85 * , 1508 ( 2000 ) m. a. cazalilla and j. b. marston , phys . rev . lett . * 88 * , 256403 ( 2002 ) k. hallberg , phys . b * 52 * , 9827 ( 1995 ) . e. r. gagliano and c. a. balseiro , phys . lett . * 59 * , 2999 ( 1987 ) . g. grosso and g. partori parravicini , in _ memory function approaches to stochastic problems in condensed matter _ , adv . in chemical physics , * 62 * , 133 ( wiley , n. y. , 1985 ) p. horsch and w. von der linden , z. phys . b * 72 * 181 ( 1981 ) j. des cloiseaux and j. j. pearson , phys . rev . * 128 * , 2131 ( 1962 ) t. yamada , prog . . jpn . * 41 * , 880 ( 1969 ) ; l. d. fadeev and l. a. takhtajan , phys . lett . * 85 a * , 375 ( 1981 ) g. mller , h. thomas , h. beck and j. bonner , phys . b * 24 * , 1429 ( 1981 ) and references therein . s. haas , j. riera and e. dagotto , phys . b * 48 * , 3281 ( 1993 ) t. khner and s. white , phys . rev . b * 60 * , 335(1999 ) p. brune , g. japaridze and a. p. kampf , cond - mat/0106007 ( unpublished ) t. d. khner , s. r. white and h. monien , phys . b * 61 * , 12474 ( 2000 ) s. s. kancharla and c. j. bolech , phys . rev . b * 64 * , 085119 ( 2001 ) y. anusooya , s. pati and s. ramasesha , j. chem . phys . * 106 * , 1 ( 1997 ) ; s. ramasesha , k. tandon , y. anusooya and s. pati , proc . of spie , * 3145 * , 282 ( 1997 ) ; s. ramasesha , z. shuai and j. brdas , chem . . lett . * 245 * , 224 ( 1995 ) e. jeckelmann , phys . b * 66 * , 045114 ( 2002 ) e. jeckelmann and f. gebhard , phys . rev . lett . * 85 * , 3910 ( 2000 ) ; f. essler , f. gebhard and e. jeckelmann , phys . rev . b * 64 * , 125119 ( 2001 ) ; e. jeckelmann , cond - mat/0208480 h. b. pang , h. akhlaghpour and m. jarrell , phys.rev . b * 53 * 5086 ( 1996 ) p. hohenberg and w. brinkman , phys . b * 10 * , 128 ( 1974 ) x. wang , k. hallberg and f. naef in ref.@xcite , chap.7(i ) t. mutou , n. shibata and k. ueda , phys . rev . lett * 81 * , 4939 ( 1998 ) ( erratum : phys . lett . * 82 * , 3727 ( 1999 ) ) f. naef , x. wang , x. zotos and w. von der linden , phys . b * 60 * , 359 ( 1999 )
aerosol particles have a direct and indirect impact on the earth s climate . one of the most important physical properties of aerosol particles is their size , and the concentration of particles in terms of their size is referred to as the particle size distribution . an important characteristic of these data is that because aerosol particles are governed by formation and transformation processes they tend to form well distinguished modal features . investigating these features provides an understanding of the dynamic behaviour of aerosol particles , their effect on the climate and their association with adverse health effects . this type of data is increasingly being measured on a regular basis , with the potential to provide more detailed information than , for example , measurements of particle mass concentrations such as pm10 or pm2.5 , traditionally used for regulatory purposes [ @xcite ] . the most common approach for representing particle size distributions is by treating the size distribution at any time point as a set of individual typically normal distributions or modes [ @xcite , @xcite ] . in this formulation the estimation of particle size distributions is then analagous to a finite parametric mixture model problem at each time point . while interest is in the representation of the particle size distribution as a mixture at each time point , it is also of interest to describe how this distribution evolves over time . to better understand aerosol dynamic processes , a feature of the measurements of particle size distributions is that they are often collected at regular points in time , and often at quite small time intervals ( e.g. , every 10 minutes ) . in this setting , parameters of the mixture model at each time point are likely to be correlated with neighboring time points and useful information about the parameters may be gained by incorporating this information in estimation . the standard setting in which mixture models have been applied has largely been for independent random samples [ @xcite ] , but literature is developing for situations in which the data are spatially and/or temporally structured [ @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ] . the development has largely been driven by the increasing availability of information in a wide variety of applications . an example includes analysing images ( cat scan ) of sheep over time in which interest is in changes to the composition of fat , bone and tissue [ @xcite ] . in @xcite interest is in cell fluorescent imaging tracking modelled using a dynamic spatial point process and a mixture representation for the different intensity functions observed . for the air pollution example considered in this paper , we are interested in a mixture representation using a missing data approach , in which the components themselves can be interpreted as potential substrata of the data and for which further interest is in their behaviour over time . in particular , we are interested in the evolution of the parameters of the components over time , which in the case of the particle size distribution data is able to reveal important information about the number and change in size of particles for particular modes along with a measure of their variation . this information can then be used to better understand the potential variables affecting the dynamic behaviour of each mode ( e.g. , from local effects such as combustion from petrol and diesel vehicle engines , construction activity , wind speed , temperature , etc . , and regional effects ) , which are likely to vary substantially between modes , and provide for a more accurate risk assessment of potential effects on adverse health outcomes ( e.g. , respiratory illnesses ) . popular recent approaches that allow for the correlated nature of the parameters in a mixture setting , both within and across epochs , include dependent dirichlet process mixture models ( ddpm ) and ( spatial ) dynamic factor models ( sdfm ) [ @xcite , @xcite , @xcite , @xcite , @xcite ] . while these approaches are appealing in the context of our case study , they do have some drawbacks . importantly , interpretation of component parameters is less straightforward under the ddpm , and the sdfm typically requires relatively long time series [ @xcite ] . an alternative that we consider here is the use of informative priors at each epoch in a finite parametric mixture setting , where the information required at each epoch is obtained from neighbouring epochs . this has appeal both in terms of a general bayesian learning framework and in terms of interpretability of the mixture components and weights , which is important in our application . moreover , while some of the methods developed for mixture models in the spatial setting [ e.g. , @xcite ] can potentially be adapted for use in a time series setting , the influence or choice of informative priors in a time series framework and the implications in different data environments have largely not been examined . in this paper we explore three different informative priors for estimation of mixtures where the data are highly correlated , and all parameters in the mixture are allowed to vary . different simulated data sets , with features similar to actual particle size distribution data , are used to highlight the influence of using informative priors and to identify situations where placing informative priors may not be beneficial . the paper is structured as follows . in section [ secpsddata ] we briefly describe particle size distributions and provide an illustration with actual data . in section [ secmixture ] we outline the finite parametric mixture model setup for a single time point and then outline the three approaches to estimation of a mixture model at multiple time points . section [ secres ] presents results on the performance of the approaches on several simulated data sets and actual data , and we conclude in section [ secdiscussion ] with some discussion and possibilities for further work . -axis : @xmath0 ) ] . the black ( overall ) and red ( components ) lines show the inferred density from estimation using . ] figure [ figplotexamplefit ] shows an example of particle size distribution data for one measurement or time point . the histogram shows the number of particles @xmath1 per cubic centimeter binned by particle size , with the horizontal axis representing the natural logarithm of the particle diameter in nm [ @xmath2 . the histogram is normalised , so that its total area equals 1 . -axis and ( bottom ) @xmath3-axis is on the @xmath4 scale [ i.e. , @xmath5 . ] because aerosol particles are charged , their size can be determined from their electrical mobility [ @xcite ] and a common instrument that utilises this principle is the differential mobility particle sizer ( dmps ) [ @xcite ] . in this study we present , as an example , the aerosol particle evolution before , during and after a new particle formation event at a boreal forest in southern finland ( figure [ figplotoneday ] ) . this data set was selected as it provides a wide ranging representation of modes for particle size distributions [ @xcite ] . because aerosol particles are governed by formation and transformation processes , they tend to form well distinguishable modal features . for example , during background conditions in the boreal forest the particle number size distribution of fine aerosols [ diameter @xmath62500 nm ( @xmath4 , 7.82 ) ] is bimodal : an aitken mode [ below 100 nm ( @xmath4 , 4.60 ) ] and an accumulation mode ( over 100 nm ) . during a new particle formation event a new particle mode , which is commonly known as a nucleation mode , is formed in the atmosphere with geometric mean diameter below 25 nm ( @xmath4 , 3.22 ) . however , in the urban atmosphere aerosol particles are more dynamic because of the different types and properties of sources of aerosol particles and may show more than three modes . typically the number concentrations of aerosol particles in the urban background can be as high as @xmath7 @xmath8 and close to a major road they often exceed @xmath9 @xmath8 [ @xcite ] . in this section we briefly describe the mixture model , outline a two stage approach to estimation of parameters over time , and describe three types of priors for temporal evolution of the parameters . the density of data ( @xmath10 ) at a given time period may be represented by a finite parametric mixture model @xmath11 where @xmath12 is the number of components in the mixture , @xmath13 represents the probability of membership of the @xmath14th component ( @xmath15 ) , and @xmath16 ) is the density function of component @xmath14 which has parameters @xmath17 . let @xmath18 indicate the observed data index . as component membership of the data is unknown , a computationally convenient method of estimation for mixture models is to use a hidden allocation process and introduce a latent indicator variable @xmath19 , which is used along the lines of a missing variable approach to allocate observations @xmath20 to each component [ @xcite ] . in this paper we adopt the common assumption of fitting normal distributions to aerosol particle size distribution data [ @xcite ] . as psd data are often measured with a definite lower and upper bound for the size of the particles , we introduce a slight modification and assume that the data follow a truncated normal distribution . as is commonly assumed , we take the data ( @xmath10 ) to be the @xmath4 of particle diameters ( nm ) , and the parameters ( @xmath17 ) for each component are the mean ( @xmath21 ) and variance ( @xmath22 ) . the number of components @xmath12 is also assumed to be unknown . in the first stage of the temporal analysis , for each time period we implemented a reversible jump markov chain monte carlo ( rjmcmc ) algorithm [ @xcite ] . although this algorithm is easily fit at a single time point , the use of rjmcmc for mixture models with temporal data requires significant preprocessing with respect to mixing coverage and convergence , as well as postprocessing to provide adequate summary statistics and between time component mapping . as an alternative , we considered a two - stage approach . in the first stage , the number of components was estimated at each time point using rjmcmc . in the second stage , we fixed the number of components ( @xmath12 ) to the maximum observed at any time point and independently estimated the parameters of the mixture model ( @xmath23 , @xmath24 and @xmath25 ) for each time point using a mcmc sampler algorithm . details of the mcmc scheme used for the different cases are given below . as we do not observe all of the components in every time point , we allow component weights to be `` effectively '' zero [ @xmath26 if required [ for details on the asymptotic behaviour of the posterior distribution using this approach see @xcite ] . estimation of parameters of the components that are effectively `` empty '' under this criterion will then essentially be governed by their respective prior information . for the results to follow in section [ secres ] we thus only plot the parameters of components which are not `` empty '' . priors for the first stage of the analysis were @xmath27 where @xmath28 , @xmath29 , @xmath30 , @xmath31 , @xmath32 , @xmath33 , @xmath34 are hyperparameters . for the second stage , priors were @xmath35 where again @xmath36 and @xmath37 are hyperparameters , detailed below . the prior for @xmath23 and @xmath38 could alternatively be decoupled and expressed as in stage 1 , but we did not see a noticeable difference in the results ( section [ secres ] ) using either form . for the independent prior case , we use uninformative priors for @xmath23 , @xmath24 and @xmath25 . in the second stage , four priors were considered for linking parameter values ( @xmath39 ) over time . the first of these was the independent prior , in which the correlated nature of the data was ignored completely and parameters were independently estimated at each time point . the second , third and fourth were termed the `` informed prior '' , `` penalised prior '' and `` hierarchical informed prior '' , as described below . in this approach we use the information provided from the previous time period as prior information for the current period . for the main results we focus on a simple case where posterior estimates from the previous period are used as prior information for the current period . we do this to illustrate the influence of a simple prior specification on the posterior estimates of parameters ( @xmath40 ) . in the case of a mixture model using gaussian distributions , we have three parameters ( @xmath23 , @xmath24 and @xmath25 ) for which we could utilise available prior information to aid in estimation . preliminary investigation indicates that all three parameters are likely to show strong evidence of autocorrelation , so here we examine the effect of smoothing on each of these parameters . for @xmath41 , we allow @xmath42 in equation ( [ eqnprior ] ) to reflect prior information about @xmath43 . thus , we set @xmath44 , where @xmath45 is the mean of the number of observations allocated to component @xmath14 in the previous time period , and @xmath17 is fixed at some value . an alternative is to impose a distribution on @xmath40 , say , @xmath46 [ or @xmath47 ) ] , but we do not present the results for this approach in this paper . for the specification of prior information for @xmath23 and @xmath24 , we set @xmath48 , @xmath49 and @xmath50 [ to ensure @xmath51 is centred on @xmath52 and increase the value of @xmath53 from the value set for the independent case to reflect the degree of dependency for these parameters from the previous period . in this approach we base the priors at time @xmath54 on the aggregated information at all other time periods . this can be achieved by employing a reparameterisation of the prior to reflect the degree of dependency between parameters . @xcite proposed a prior for @xmath25 ( in a different context ) which can be used in our setting to downweight large changes in probabilities in successive time periods . let @xmath55 , then @xmath56 is defined as @xmath57 where smaller values of @xmath58 imply greater smoothing . a potential advantage of using information about estimates both forwards and backwards in time is the additional information this may provide to guide parameter estimates in the current period . this may be most useful if large changes in the parameter estimates occur for single periods of time . for the purposes of comparison with section [ secsimulateddata ] , we compare the results of using a similar formulation for @xmath25 in the informed prior approach ( without smoothing on @xmath23 and @xmath24 ) . thus , prior distributions @xmath59 and @xmath60 are set as for the independent approach [ equation ( [ eqnprior ] ) ] . for this formulation , we sampled from the posterior distribution of @xmath25 using a rejection sampling approach outlined in appendix [ app1 ] . in this approach an informative prior is placed at two different levels . the aim of allowing for different levels is to provide flexibility to the form in which prior information is given in the model . this flexibility may be needed in cases where the correlation structure can vary greatly over time : instead of imposing a smoothing structure directly on strongly varying parameters , we can provide a less restrictive smoothing through the hyperparameters . for the hierarchical approach , we will focus on parameters @xmath23 and @xmath25 as they are the main parameters of interest for the psd data ( see section [ secrealdata ] and the ) . the hierarchical approach for @xmath23 is specified as @xmath61\\[-9pt ] \phi_{jt } & \sim&\mathcal{n } \bigl(\phi_{j , t-1 } , \varepsilon^{(s)}_{\phi } \bigr),\nonumber\end{aligned}\ ] ] where @xmath62 and @xmath63 are scalars , reflecting the variability of @xmath64 and @xmath65 , respectively . under this formulation , @xmath23 is used to estimate the mixture distribution at the level of the data , and @xmath58 represents the underlying correlation of @xmath23 over time [ assuming in this case an ar(1 ) process ] . in this setting , we can interpret the ratio @xmath66 as reflecting the amount of information we have about the underlying behaviour ( signal ) of @xmath23 in comparison to estimates at the level of the data ( noise ) . for the first time period ( @xmath67 ) , we set @xmath68 . for estimation of @xmath23 and @xmath58 , we use a gibbs sampling scheme . for details see appendix [ app2 ] . for the parameter @xmath25 , the dirichlet prior used in equation ( [ eqnprior ] ) for the independent model and the informed prior approach is a very common prior for discrete probabilities . a natural extension to the dirichlet prior with a temporal component is to use its representation in terms of a gamma distribution . however , the inflexibility of the gamma distribution makes it difficult to construct a temporal structure to the dirichlet prior . an alternative formulation of the dirichlet in terms of the beta distribution does not appear to provide greater flexibility . another alternative is to use a logistic - normal prior for @xmath25 , where @xmath69\\[-8pt ] \lambda_{jt } & = & \frac{\exp(w_{jt})}{\sum_{j=1}^{k-1}{\exp(w_{jt})}}\nonumber\end{aligned}\ ] ] and where @xmath70 is the mean value ( number of particles ) at time period @xmath54 . using this functional form , the parameterisation of @xmath25 in terms of a multivariate normal distribution allows for a suitably flexible form in which to explore a hierarchical structure for this parameter . such flexibility , in comparison to the dirichlet distribution , has been investigated in a hierarchical approach for pooling of estimates across different sampling units [ @xcite ] . in a hierarchical setting and similar to the model used for @xmath23 , we can further say that @xmath71\\[-8pt ] \gamma_{jt } & = & \frac{\exp(x_{jt})}{\sum_{j=1}^{k-1}{\exp(x_{jt})}},\nonumber\end{aligned}\ ] ] where @xmath72 and @xmath73 reflect the variability of @xmath74 and @xmath70 , respectively . analogous to the above discussion for @xmath23 , under this formulation the parameter @xmath25 is used to estimate the mixture model at the level of the data , and @xmath75 represents the underlying or smoothed behaviour of @xmath25 over time , which may be prone to large fluctuations from the data . for the simulation results and actual data to follow , we specify the diagonal entries of @xmath72 and @xmath73 , and fix off - diagonal entries to be zero . for comparability with the hierarchical approach for @xmath23 , and using similar notation for the smoothing parameters , we specify @xmath76 and @xmath77 . the interpretation of @xmath78 and @xmath79 is then the same as before , but this time in terms of @xmath25 . for estimation of @xmath25 and @xmath75 we use a gibbs sampling scheme with a metropolis hastings step . for details see appendix [ app2 ] . for identifiability both @xmath74 and @xmath70 are @xmath80 dimensional , and @xmath81 ( with the same identification used for @xmath75 ) . in practice , it may be difficult to specify _ a priori _ the parameter values for @xmath78 and @xmath79 , as little information about the variability of the parameters for the mixture components may be known . estimation of these parameters also requires a choice to be made about the degree of smoothing required . for the purposes of this paper we focus on specifying the parameter values for @xmath78 and @xmath79 , and explore briefly the effect on the results of varying these values . in the discussion we talk about this issue further . for now , one approach to specifying @xmath78 may be to use the results from a rjmcmc approach used in the first stage , estimate or explore the variability of @xmath23 or @xmath25 over time , and then use this information to set the parameter values for @xmath78 . the parameter value for @xmath79 could be set as a smaller multiple of @xmath78 , and be varied to assess the influence of the results . using the information from the results of the independent approach , for a fixed upper bound @xmath12 could also be used . although there is extra computational time involved in running either approach in the first stage , such a strategy may prove useful in order to assess the influence of different prior information on the results . where component parameters are themselves the subject of the analysis , an important and commonly encountered issue in bayesian mixture modelling relates to the labelling of these parameters during the mcmc run . as the likelihood of a mixture is by definition multimodal , using exchangeable or noninformative priors can ( and should ) result in parameters moving freely over the parameter and component labelling space during sampling . ( in theory , for a @xmath12 component mixture , @xmath82 permutations of the labelling of the parameters are possible . ) estimation of functionals of these parameters ( conditional on labelling ) at the end of sampling is thus then problematic . several empirical approaches to deal with this issue ( commonly called `` label switching '' ) have been proposed in the literature [ @xcite ( @xcite ) , @xcite , @xcite ( @xcite ) , @xcite , @xcite , @xcite , @xcite ] , generally by relabelling parameters in proximity to one of the @xmath82 modal regions during the run [ e.g. , @xcite ] or at the end of sampling [ e.g. , @xcite ] . under exchangeable or noninformative priors the main reason for label switching relates to a lack of identifiability of the mixture model ( particularly with respect to enforcing a unique labelling ) . given two component parameter sets @xmath83 and @xmath84 , a finite mixture model is weakly identifiable if at least one element of @xmath83 and @xmath84 differs . for practical purposes then , the element that differs can be used to enforce a unique labelling . for exchangeable priors , in the sense that priors for @xmath83 and @xmath84 are the same , it is clear that there is a lack of identifiability . in the case of using informative priors on @xmath83 and @xmath84 , the issue of model identifiability is less clear . although the use of different informative priors can help to separate @xmath83 and @xmath84 , in practice , it will depend on the strength of the informative prior to separate at least one element of each parameter set . to our knowledge , there has been very little theoretical investigation of this issue . in practice , one can assess whether parameters are well separated by analysing the path of parameters over the sampling run and/or from plots of marginal densities . however , only part of the picture may be revealed by doing this . first , a relatively long sampling run is needed to allow the parameter to fully explore the space ( perhaps many thousands of iterations ) . in the case of gibbs sampling , the sampler may also become trapped in one of the modal regions [ @xcite ] . apart from the case of using a fully exchangeable prior , this second case is more difficult to identify in practice , as it can be unclear whether the gibbs sampler is truly trapped or has found a uniquely labelled parameter space . a pragmatic solution could be to start the sampler from different values , although good starting locations can be difficult to determine in high - dimensional space . alternative ( and often more involved ) solutions could be to : reparameterise and change the conditioning [ @xcite ] ; use tempering to facilitate more exploration [ @xcite ] ; and/or modify the gibbs sampling proposal and acceptance [ @xcite , @xcite ] . for both the independent and informed prior approaches outlined previously , it is possible to relabel the output using existing empirical approaches . in these cases , the posterior is updated sequentially across time and relabelling can take place during or at the end of each time period . in the results to be presented , we used the _ maximum a posteriori _ ( map ) estimate to select one of the @xmath82 modal regions and chose either a distance - based measure on the space of parameters [ @xcite ] , or on the space of allocation probabilities [ @xcite , @xcite ] to relabel parameters in proximity to this region . in the case of the penalised and hierarchical informed priors , a joint posterior that includes all components and all parameters over time is used and sampling updates occur globally over time . relabelling of sampling output thus necessitates a permutation of labels for all time points together . such a joint approach is similar in spirit to the viterbi algorithm approach used in state space models [ @xcite ] , but the joint posterior in our case is different and would require further work outside the scope of this paper . using informative priors that are incompatible with the data can also force too many distinct components and lead to overfitting , which can in some cases lead to label switching . the potential for label switching could arise if two components with similar parameters are fitted where one component would suffice under the true model . as before , the similarity of the parameters could result in a lack of identifiability of the model . in using an informative prior there is also a model choice issue in terms of selecting the best model for the data , where `` best '' is defined , for example , as the most parsimonious model in terms of the smallest value of @xmath12 . in practice , discrimination between the choice and use of informative priors can be made by fitting several mixture models to the data . point - process representations of the estimated parameters from using a noninformative prior ( such as figure [ figrjplotmeans ] ) can also offer a good visual guide as to the potential range and scope of the parameter space . aerosol particle measurements are commonly recorded in the form of a number of distinct particle size ranges , or channels , the size and number of the channels being governed by the type and setup of the measurement instrument . for example , in the sampled data from hyytil ( see figure [ figplotexamplefit ] and section [ secrealdata ] ) , we observed 32 distinct size partitions ( bins ) covering the range from 3 nm to 650 nm . such coarsening of the data created by binning has an impact on density estimates and in a mixture context the number of components required to adequately model the data [ @xcite ] . to address this , we add another step in the gibbs sampler in which we simulate a new latent variable ( say @xmath3 ) which is drawn from the believed underlying density of the data ( @xmath10 ) , in this case the fitted mixture model at current estimates of the parameters , at each iteration of the gibbs sampler . as the sampling takes place within each bin , the simulation of the latent variable essentially involves sampling from a truncated normal distribution for which there are a number of proposed approaches . for computational efficiency we used the slice - sampling approach of @xcite . for details of the approach and of the gibbs sampler see @xcite . as the number of observations within each bin ( and hence the latent sample size ) is quite large , an extra step can be added after the latent variables are simulated in which the samples within each bin are divided into a number of sub - bins and computations in the gibbs sampler proceed based on the new binned data . this can greatly speed up computations compared to using the full latent sample whilst reducing the coarsening of the data created by the original bins . in general , we found comparable results to the full latent sample by using 3 additional sub - bins for each original bin ( in total 97 bins are used compared to the original 32 bins ) . in this section we present and assess the results using simulated data and then present the results of applying the approaches to particle size distribution data from hyytil , finland . we use the simulated data to test the impact of the different prior representations and the degree of smoothing . we first use an informative and penalised prior only on the weights ( @xmath25 ) , and then assess the influence of using an informative prior on @xmath23 and @xmath24 in order to assess the influence of using prior information for each parameter separately . for the independent , informed and penalised prior approaches the results are based on 50,000 iterations with a burn - in period of 20,000 ( i.e. , the first 20,000 samples are discarded ) . results using rjmcmc ( used in the case study ) are based on 200,000 iterations with a burn - in period of 100,000 . convergence was assessed by visual inspection and using the gelman rubin statistic [ @xcite ] . we simulated data sets indicative of the type of behaviour of aerosol particle size distribution data observed at hyytil , a boreal forest site in southern finland ( smear ii ) [ @xcite ] . a particular feature of these particle size distribution data is both a growth in the mean and weight for some of the modes ( components ) and a decline in weight for others . changes can also occur to the variance of the modes and at times they can follow a similar pattern to the weights over time . we simulated data from two different cases . in the first case ( d1 ) , we simulated data which are highly correlated across time , a feature of particle size distribution data observed in practice for most time periods where measurements are commonly taken at small time intervals . this data set was also simulated with parameter estimates where at times the mixture is not well identified ( component means and weights are not well separated ) . of interest in this setting is the effect of using either the informed prior or penalised prior approach compared to the independent approach . in practice , it is quite common to observe sudden large changes in the number of particles measured which may persist for a number of time periods . this is more often observed when there are relatively few particles for a particular size group , and more so for the smaller sized particles ( an example of this type of data is examined in section [ secrealdata ] ) . thus , for the second data set ( d2 ) we simulated data for the first component where the weight for the smaller sized particles is quite volatile . for this data set the mixture is well identified . further details and results are available in the supplementary material [ @xcite ] . for both cases ( d1 ) and ( d2 ) , we simulated data using three components on 32 distinct size partitions ( bins ) equally spaced ( on the @xmath4 scale ) covering the range from 3 nm to 20 nm in particle diameter ( on the @xmath4 scale 1 to 3 ) . the sample size for each time period is 1000 and the total number of time periods was 100 . further details of the sampling process for each case are provided in section [ secd1 ] and the supplementary material [ @xcite , second data set ( d2 ) ] . for the results to follow , except as specified otherwise , for the independent , informed prior and penalised prior approaches , we set the hyperparameters to be @xmath85 ; @xmath86 ; @xmath87 ; and @xmath88 , which were chosen to be weakly uninformative considering the range and size of the data . for the independent and informed prior approaches , the original gibbs sampling output has been relabelled using a distance - based measure on the space of parameters [ @xcite ] and also ( as a check ) on the allocation space [ @xcite , @xcite ] . as shown in figure [ figsimoactualind ] ( black line ) , for the first data set we simulated data for the first component with a mean value increasing from 1.5 to 3.0 , and weight increasing from 0.1 to 0.6 and then decreasing to 0.3 , over time . often a consequence of the growth in the first component is a decline in size and weight for the larger sized particles and this is reflected in the weight for the second component following an opposite pattern to the first component . for the third component , the weight increases from 0.1 to 0.3 over time . the parameters @xmath23 and @xmath25 are simulated with some noise around the parameter values , and the sample size is 1000 . ( top ) , @xmath38 ( middle ) and @xmath25 ( bottom ) ] over time ( @xmath3-axis ) for independent approach : simulated data ( black ) ; independent ( red ) ; and 95% credible interval ( dotted line ) . the columns represent the components ( components 1 to 3 ) . ] figure [ figsimoactualind ] also shows the results of using the independent approach . we see that at times the parameter estimates for the independent approach deviate from the actual data . figures [ figsimoactualip ] and [ figsimoactualpp ] show the results for the informed prior and penalised prior compared to the actual data , respectively . in figure [ figsimoactualip ] , the results show the effect of varying the degree of smoothing on @xmath25 for the informed prior using @xmath89 . for the results of the penalised prior , we vary the degree of smoothing on @xmath25 using @xmath90 . ( top ) , @xmath38 ( middle ) and @xmath25 ( bottom ) ] over time ( @xmath3-axis ) for informed prior approach : simulated data ( black ) ; theta@xmath91 ( green ) ; theta@xmath920.8 ( blue ) ; theta@xmath93 ( purple ) ; and 95% credible interval ( dotted line ) . the columns represent the components ( components 1 to 3 ) . ] ( top ) , @xmath38 ( middle ) and @xmath25 ( bottom ) ] over time ( @xmath3-axis ) for penalised prior approach : simulated data ( black ) ; @xmath94 ( brown ) ; @xmath95 ( light blue ) ; @xmath96 ( dark green ) ; and 95% credible interval ( dotted line ) . the columns represent the components ( components 1 to 3 ) . ] in figure [ figsimoactualip ] , we can see that the parameter estimates for @xmath25 for all three values of @xmath40 appear to closely follow the actual data , with the closest estimates to the actual data being for @xmath97 and 1.3 . as we are only using an informed prior on the weights , the parameter estimates for @xmath23 and @xmath24 appear to be quite variable over time compared to the actual data . however , the variability appears to be slightly less for these variables than for the independent approach ( figure [ figsimoactualind ] ) and closer to the actual data over time . of interest is the closeness of the parameter estimates of @xmath23 and @xmath24 for components 1 and 2 , which more clearly follow the true growth occurring in component 1 and the stability over time for component 2 compared to that observed for the independent approach . in figure [ figsimoactualpp ] , the parameter estimates for the penalised prior approach appear to deviate slightly from the actual data for components 1 and 2 . for the third component , the parameter estimates for the penalised prior approach follow the actual data with some noise . overall , the results from the penalised prior approach are similar to the independent approach but with less variability over time . we turn now to an assessment of the impact of using an informative prior for @xmath23 or @xmath24 over time . we present results for the highly correlated data set , since this is the most sensitive of the simulated data as discussed above . here we set @xmath98 , @xmath48 , @xmath99 and . ( top ) , @xmath100 ( middle ) and @xmath25 ( bottom ) ] over time ( @xmath3-axis ) for informed prior approach : simulated data ( black ) ; smoothing on @xmath23 ( orange ) ; smoothing on @xmath24 ( dark green ) ; and 95% credible interval ( dotted line ) . the columns represent the components ( components 1 to 3 ) . ] in figure [ figsimoactualtpmusigma ] , the parameter estimates for the informative prior for @xmath23 appear to more closely follow the actual data than using an informative prior for @xmath24 . although the parameter estimates for both approaches appear to be further away from the actual data than using an informative prior for @xmath25 , they do appear to be closer than under the independent approach . figure [ figcaseiisimosmlambdamu ] shows the results of using an informative prior on both @xmath23 and @xmath25 . in this example , the results are similar to using an informative prior only on @xmath25 . thus , depending on the objectives of the analysis , using an informative prior on both parameters may not be needed . ( top ) , @xmath38 ( middle ) and @xmath25 ( bottom ) ] over time ( @xmath3-axis ) for approaches : simulated data ( black ) ; independent ( red ) ; smoothing on @xmath23 and @xmath25 ( green ) ; and 95% credible interval ( dotted line ) . the columns represent the components ( components 1 to 3 ) . ] for the second simulated data set , where the weight for the smaller sized particles is quite volatile , the results of smoothing on @xmath23 , @xmath24 and @xmath25 for the informed prior and penalised prior suggests that large adjustments to one parameter ( e.g. , from volatility in some time periods ) are not supported unless compensatory measures can be taken by the other parameters . in contrast to these results , we do not see any large compensatory adjustments being made to parameters by using a hierarchically based informative prior for @xmath75 . details are available in the supplementary material [ @xcite ] . the data set studied here was taken from a measurement site at hyytil , finland ; a plot of the measurements for the day selected is shown in figure [ figplotoneday ] . this particular day was selected as it shows a new particle formation event occurring , whereby a new mode of aerosol particles appears with a significant influx of particles ( as high as @xmath101 per @xmath102 ) with a geometric mean diameter ( @xmath610 nm ) , growing later into the aitken ( 2590 nm ) or accumulation modes ( 100@xmath103 nm ) . in terms of a temporal mixture model setting , we will be able to assess the performance of the four prior specifications outlined previously as new components are introduced and both a growth in the mean and weight for those components are observed . the data from hyytil consists of measurements which were taken every 10 minutes ( 144 time periods ) and for each time period in the form of 32 distinct size partitions ( bins ) equally scaled ( on the @xmath4 scale ) covering the range of 3 nm to 650 nm ( on the @xmath4 scale 1 to 6.5 ) . as outlined in section [ secmixture ] , the first stage of our approach is to apply rjmcmc to each time period . these results are then used to guide the choice of the number of components and initial parameter estimates for the second stage analysis , in which temporally correlated priors are used to model the evolution of the mixture parameters over time . figure [ figrjplotmeans ] shows the results of the first stage of the algorithm , with a plot of the posterior mean estimates for @xmath64 at each time point @xmath54 ( bottom panel ) , with the size of the circles indicating the corresponding weight @xmath104 . the average number of components estimated with the highest probability over the day was four ; the minimum number of components was one , and the maximum number of components was five ( see top panel of figure [ figrjplotmeans ] ) . ( posterior median ) from rjmcmc algorithm over time ( every 10 minutes ) ( hyytil ) . ( bottom ) : plot of posterior mean estimates for @xmath105 [ @xmath0 ) ] from algorithm over the same time period . stage 1 of analysis for temporal evolution of parameters . the size of the circles is proportional to the weight ( @xmath104 ) corresponding to @xmath64 . ] for the second stage , we fixed the number of components to be five with the initial mean values equally spaced across the range of possible diameter values [ @xmath106 and @xmath107 . figure [ figdm512hmind ] shows the results of using the independent approach ( original output has been relabelled using a distance - based measure on the space of parameters [ @xcite ] ; similar results were obtained by relabelling on the allocation space [ @xcite , @xcite ] ) . [ ( top ) @xmath0 ) ] , @xmath25 ( middle ) and particle count ( per @xmath108 ) ( bottom ) . stage 2 of the analysis for the evolution of parameters . measurements taken every 10 minutes . colours indicate the components to which parameter estimates belong . ( the parameter estimates for the first component are black , parameters for the second component are red , for the third component they are green , etc . ) note : only components where @xmath109 are plotted . ] due to the large number of particles ( per @xmath108 ) observed over the day ( see figure [ figplotoneday ] ) , the results of estimation using the informed prior approach for @xmath23 and/or @xmath25 are very similar to the results of the independent approach , so are not shown here . as we are interested in the effect of using an informative prior in this context , we rescale the number of particles by a factor of @xmath110 and assess the results . the median number of particles over the course of the day is then 6893 , reaching a maximum of 17,740 . using this rescaling of the data , the results of the independent approach are the same as shown in figure [ figdm512hmind ] . figure [ figdm512hmlambda ] shows the results of estimation using the informed prior approach for @xmath23 and @xmath25 . for these results , @xmath111 , @xmath112 and @xmath48 , which provides for a moderately informative prior across time . similar results were obtained using the hierarchical model with similar strength of prior information . of interest to note is that in the original output we did not see any evidence of label switching within the gibbs sampling runs . although there is some evidence of instability between 12:00 and 13:00 for the weight ( @xmath13 ) of two of the components ( newly formed and background particles ) , this appears to be due to instability in the data ( see figure [ figrjplotmeans ] ) , with the means of these components remaining reasonably well separated during this period . and @xmath25 . posterior mean estimates for @xmath23 [ ( top ) @xmath0 ) ] , @xmath25 ( middle ) and particle count ( per @xmath108 ) ( bottom ) and 95% credible interval ( dotted line ) . stage 2 of the analysis for the evolution of parameters . measurements taken every 10 minutes . colours indicate the components to which parameter estimates belong . ( the parameter estimates for the first component are black , parameters for the second component are red , for the third component they are green , etc . ) note : only components where @xmath109 are plotted . ] compared to the results from the independent approach ( figure [ figdm512hmind ] ) , the results from the informed prior approach for @xmath23 and @xmath25 suggest a clear pattern for both over the course of the day . in particular , and of interest to aerosol physicists , is the clear growth pattern shown for the mean of the first component representing the smaller sized particles . the results indicate that these particles grow from approximately 1.40 ( 4 nm ) to 3.0 ( 20 nm ) in diameter on the natural @xmath4 scale ( or in nanometers ) . the path of the parameters for the first component ( black ) is less clear using the independent approach , principally as a result of the newly formed smaller particles merging into the component representing the background particles ( green ) . there is also some evidence of instability in the labelling _ across time _ , which is natural given the absence of temporal information in the prior . we note that different representations of this data are possible depending on prior assumptions . if we relax the assumed moderate correlation of the new and background particles , then there is less separation between the component associated to the newly formed particles and to that of the background particles , similar to the results we see for the independent approach . ideally , more information is needed to model these data ; this could take the form of greater assumptions about the underlying stochastic process and the possible inclusion of external factors ( e.g. , meteorological data ) impacting / influencing the observed process . we discuss this issue more in the following section . in this paper we explored the problem of estimating bayesian mixture models at multiple time points . under different situations , approaches that employ information about neighbouring time points compared favourably to results based on an independent approach . by including additional temporal information about parameters for correlated time periods , we may be able to better identify individual components at each time point . as an aid for inference , we may also be able to obtain smoother parameter estimates over time and from this be able to clearly establish patterns or identify anomalies from the data . the results highlight a number of observations about mixture representations at multiple time points . first , analysis of the evolution of parameters of a mixture over multiple time points highlights the large degree of dependency that exists between component parameters . changes to a parameter in one component may flow on to the parameter in a nearby component . depending on the context of the study , we can anticipate this dependency to be more readily apparent for the weight parameters , but we found similar dependencies to exist for other parameters . the second is the need to be mindful that the same parameter in one component may have a different correlation structure over time to the same parameter in another component . in the context of particle size distribution data , we often observed greater volatility in estimates for the smaller particles compared to the larger sized particles and so at times the correlation structure of the parameters between these respective components appeared to be quite different . a possible effect of using informative priors in this context is to impose a prior not supported by the data or to impose a temporal correlation structure where such a structure does not exist , and thereby cause unnecessary adjustments to other parameters . we observed this most clearly in the results from the simulated data where at times the data was quite noisy . for this data set , using an informative prior for a parameter , which supported large adjustments away from the actual data , resulted in large compensatory adjustments being made not only by other parameters within the same component , but also to parameters in neighbouring components . the easy solution may be to use an appropriate correlation structure for components , but of course this may not always be known _ a priori_. a further result of the dependency that can exist between parameters of components and within component parameters is that the inclusion of correlation information to aid in the identifiability of the mixture may not be required for all or , alternatively , all components . in the context of a mixture with a small number of components , we may only need to provide more information about one parameter for an influential component in order to separate out the influence of competing components . this result will also be useful if the correlation structure for one parameter or parameters for one component are more readily known . in the context of a mixture of gaussians , we generally found that an informative prior was only needed on @xmath23 or @xmath25 or possibly both . this result could well be context specific and influenced by any reliance on the means for defining ( in terms of size ) and ordering of components . the choice of which parameter to use more information may also be guided by whether it is a parameter of interest for inference as demonstrated in analysis of the case study where most interest was in the behaviour of both @xmath23 and @xmath25 over time . in this case , and in general , one must be careful in the analysis of selected parameters , as it can largely be a conditional analysis in view of the behaviour of other possible cross - correlated parameters within the same component and between components . in the hierarchical informed prior approach the influence of the informative prior at the two levels was specified by parameters @xmath78 ( low level ) and @xmath79 ( high level ) , and the values assigned to these parameters are critical in carrying information about the correlation structure of the parameter of interest . in this paper , we decided to choose parameter values based on prior belief in the correlation structure of the data ; alternatively , these parameters could be estimated . to this effect , a number of approaches are available for estimation [ @xcite , @xcite ] . however , in order to estimate @xmath78 and @xmath79 , we still face a choice as to the degree of penalisation or smoothing of the parameter in light of the apparent variability in the data . this is a common issue in temporal and spatial modelling in general . while many of the above difficulties may seem to be avoided if smoothing approaches are applied retrospectively on parameter estimates from an independent mixture model , this type of analysis may largely ignore the true mapping of components or the path of parameters over time . from the results of the simulated data , the large degree of dependency that we observe between the parameters of a mixture over time suggests that including temporal information to better identify one of the parameters at a single time point can flow on to affect other parameters . this could change inference about both the mixture representation at a point in time and also the behaviour of mixture parameters over time . in general , one of the potential difficulties in using an informative prior approach to smooth parameter estimates over time is the variable degree of influence the prior may have in the posterior . if the primary objective is to obtain smoothed parameter estimates over time , larger sample sizes and noisiness of the data at times may warrant increasingly restrictive priors . in such cases where the objective might be to downplay the influence of the data , a number of alternative approaches to increase the influence of prior information can be used [ @xcite ] . in all cases , it is valuable to undertake a sensitivity analysis in order to assess the effect of the prior . such an analysis should include the independent prior as a baseline comparison . a further limitation of the approach outlined is that it is computationally expensive . most of this expense is experienced in the first stage of the analysis ( which can be skipped in the presence of good prior knowledge of the parameter space ) . for estimation of psd data over one day using 144 time points , the running time of the rj approach with 200,000 iterations was approximately 3 hours using an intel centrino 2 processor 2.80 ghz . in comparison , the second stage approach using 50,000 iterations took approximately one hour . such computational expense quickly becomes burdensome if analyses is required for several days or , indeed , several weeks . of course , the use of the first stage for subsequent days may not be required , considerably reducing the computational time involved . although we have focussed on developing a hierarchical approach for parameters @xmath23 and @xmath25 , we could equally apply the same approach to consider estimation of @xmath24 . such an approach may be to consider a half-@xmath54 distribution which has previously been used in similar hierarchical settings [ @xcite ] . the hierarchical approach considered here can be readily generalised to include covariates . moreover , through the flexibility of assuming a logistic normal distribution on the weights , we can better explore and estimate transitory movements between components . in some situations it may be of interest to combine components and allow components to share a common grouping . this could be of interest where some components are only needed to account for the skewness of a larger component or to allow an analysis based on a mixture representation with fewer components of most interest . although this grouping of components could be undertaken retrospectively , it may also be interesting to see the effects such a grouping has upon estimation of the parameters and their evolution over time . for estimation of aerosol particle size distributions , the dynamics of the aerosol process and the complexity of the influences on particle concentration and size demand the use of approaches which utilise as much information from the data as possible . to this end , the inclusion of temporal information may be helpful . in this section we outline the rejection sampling algorithm for @xmath25 proposed by @xcite for the penalised prior approach : prior @xmath113 posterior @xmath114\\[-16pt ] & & \qquad \propto\prod_{j=1}^{k } \biggl\ { \prod_{t=1}^{t } \operatorname{dirichlet}(m_{jt}+1)\mathrm{i}(\lambda_{jt } ) \biggr\ } \exp\biggl(-\frac{1}{\phi}\sum_{t=2}^{t } \|\underline{\lambda}_{t}-\underline{\lambda}_{t-1 } \|^{2 } \biggr).\nonumber\end{aligned}\ ] ] @xcite suggest sampling @xmath115 from a @xmath116 , @xmath117 distribution and accepting when @xmath118 ( @xmath119 ) , where @xmath120\bigr)\nonumber\end{aligned}\ ] ] and @xmath121\bigr),\nonumber\end{aligned}\ ] ] where @xmath122 @xmath123 and @xmath124 . here @xmath125 is an indicator function equal to 1 when @xmath126 ^ 2 $ ] and 0 otherwise . _ hierarchical model for @xmath23 _ update @xmath127 , @xmath128 , @xmath25 , @xmath38 as in the independent approach . update @xmath58 and @xmath23 by sampling from the conditionals , @xmath129 _ hierarchical model for @xmath25 _ update @xmath127 , @xmath128 , @xmath23 , @xmath38 as in the independent approach . update @xmath130 by sampling from the conditional , @xmath131 where @xmath132 . update @xmath133 using a metropolis hastings step . sample from @xmath134 , where @xmath135 and @xmath136 is the variance of the proposal . let @xmath137 and for @xmath138 , @xmath139 , where @xmath140 is the mean number of observations allocated to component @xmath14 in the previous time period ( under the independent approach ) . the authors gratefully acknowledge funding from the australian research council ( arc ) as part of two arc discovery projects and the arc centre of excellence in complex dynamic systems and control . the authors are also grateful for helpful discussions with christian p. robert and tony pettitt in early stages of this work , and to comments from the editor , associate editor and referees .
the origin of cosmic rays still remains an unanswered fundamental research question . cosmic ray distributions of various aspects have been measured , most notably the steeply falling spectrum up to the ultra - high energy regime with cosmic ray energies even exceeding @xmath4 eev @xcite . for ultra - high energy cosmic rays , deflections in magnetic fields should diminish with increasing energy , such that directional correlations should lead to a straight - forward identification of accelerating sites . however , even at the highest energies the arrival distributions of cosmic rays appear to be rather isotropic . only hints for departures from isotropic distributions have been reported , e.g. , a so - called hot spot @xcite , and a dipole signal @xcite . at least with the apparent isotropy , limits on the density of extragalactic sources were derived which depend on the cosmic ray energy @xcite . a recent determination of ultra - high energy cosmic ray composition from measurements of the shower depth in the atmosphere revealed contributions of heavy nuclei above @xmath5 eev @xcite . this observation may explain the seemingly isotropic arrival distribution as deflections of nuclei in magnetic fields scale with their nuclear charges @xmath1 . obviously , when searching for cosmic ray sources , a key role is therefore attributed to magnetic fields . the galactic field in particular is strong enough to displace original arrival directions of protons with energy @xmath6 eev by several degrees from their original arrival directions outside the galaxy @xcite . the displacement angles for nuclei even reach tens of degrees . the knowledge on the extragalactic magnetic fields is much less certain , but is likely to be less important than the galactic field and is not studied in this contribution . to identify sources of cosmic rays , rather precise corrections for the propagation within the galactic magnetic field are needed . in previous directional correlation analyses , only the overall magnitude of deflections was taken into account , e.g. @xcite , or corrections for cosmic ray deflections were applied using analytic magnetic field expressions reflecting the spiral structure of our galaxy @xcite . recently , parameterizations of the galactic magnetic field have been developed which are based on numerous measurements of faraday rotation @xcite , and in addition polarized synchrotron radiation for the second reference . based on directional characteristics and the field strength of the parameterizations , deflections of cosmic rays are predicted to depend strongly on their arrival direction , charge and energy . in the following we will refer to the regular field with the bisymmetric disk model of the first reference as the pt11 field parameterization , and to the regular field of the latter as the jf12 field parameterization , respectively . angular distributions of cosmic rays in these galactic field parameterizations have been studied before , e.g. , with respect to general properties of the jf12 parameterization @xcite , specific source candidates @xcite , general properties of deflections and magnifications @xcite , and to the potential of revealing correlations between cosmic rays and their sources @xcite . in this work we investigate whether cosmic ray deflections in the galactic magnetic field can be reliably corrected for , given the current knowledge of the field . to simplify discussions of energy and nuclear dependencies we will define rigidity as the ratio of the cosmic ray energy and number @xmath1 of elementary charges @xmath7 @xmath8 in our investigations we use galactic coordinates as our reference system , with longitude @xmath9 and latitude @xmath10 . for a number of visualizations we use cartesian coordinates alternatively with height @xmath11 above the galactic plane , with the earth being located at @xmath12 kpc . based on the two field parameterizations pt11 and jf12 we initially discuss key distributions of cosmic ray deflection , dispersion effects in arrival distributions , directional variance in field transparency , and the influence of random field components . from the rigidity dependencies of these distributions , we recommend a minimum rigidity threshold above which cosmic ray deflection may be controlled in terms of probability distributions . furthermore , we take the different results of the two galactic field parameterizations as a measure of our current knowledge of the galactic field . we compare their cosmic ray angular deflections and study differences in the dispersion of arrival distributions . finally , we study the practical consequences of galactic field corrections and their uncertainties by performing simulated point source searches and by quantifying the field impact in terms of discovery potential . the two field parameterizations pt11 and jf12 each follow a different ansatz . both take into account about @xmath13 faraday rotation measurements . the pt11 field has been fitted to two large sets of faraday rotation measurements . the jf12 field has been adapted to several large sets of faraday rotation measurements and to synchrotron polarization measurements , thereby increasing the information per analysed direction by two additional complementary measurements @xcite . both use the electron density model ne2001 @xcite with an enlarged vertical scale for weighting the line - of - sight integrals of the magnetic field . a ) b ) fig . [ fig : parameterization ] shows the field strength as a function of the radial distance from the galactic center along the solar system line - of - sight and the distance perpendicular to the galactic plane . the fields exhibit different shapes and magnitudes ; especially notable in fig . [ fig : parameterization]a is the field extent of the jf12 parametrization above and below the galactic plane with non - negligible field strengths even at a distance of @xmath14 kpc . the pt11 field ( fig . [ fig : parameterization]b ) , on the other hand , exhibits a rather concentrated halo field , which is centered around a distance of @xmath15 kpc to the galactic plane . when studying the magnitude of angular deflections of cosmic rays resulting from these parameterizations , we take the angle @xmath16 between the incoming direction to the galaxy and the arrival direction on earth as a measure of the directional change ( fig . [ fig : beta ] ) . between the direction on earth and the direction outside the galaxy ( pt11).,scaledwidth=50.0% ] to get a first impression of the different deflections resulting from the two field parameterizations we use backward tracking techniques of antiprotons through the galactic field . with this technique we obtain individual trajectories for matter particles entering from outside the galaxy and then following the reverse path . the method ensures that every trajectory leads to observation on earth . in fig . [ fig : angularsky ] we show the magnitudes of the angular deflections @xmath16 of cosmic rays with rigidity @xmath17 ev . the position in the map denotes the initial direction on earth in galactic coordinates for the backtracked antiprotons . the color code refers to the magnitude of angular deflections which reach up to @xmath18 deg . for the jf12 parameterization ( fig . [ fig : angularsky]a ) , deflections are largest near directions of the galactic center which is expected from the magnitude of the field shown in fig . [ fig : parameterization]a . with the pt11 parameterization ( fig . [ fig : angularsky]b ) , deflections are largest in any direction near the galactic plane which is attributed to the strong disk field ( fig . [ fig : parameterization]b ) . in the galactic magnetic field for antimatter with rigidity @xmath17 ev starting from earth in the direction presented in galactic coordinates , a ) jf12 , b ) pt11 . , title="fig:",scaledwidth=45.0%]a ) in the galactic magnetic field for antimatter with rigidity @xmath17 ev starting from earth in the direction presented in galactic coordinates , a ) jf12 , b ) pt11 . , title="fig:",scaledwidth=45.0%]b ) as expected , the differences in the field parameterizations relate directly to a different impact on cosmic ray deflections . in the following section we study a number of aspects related to the directional changes of cosmic rays when traversing the galactic field . the goal of this section is to determine a kinematic regime where information on cosmic ray arrival directions can be obtained by transformation of probability distributions . for this purpose we first analyze cosmic ray angular deflections as a function of rigidity . as our primary criterion we require angular deflections to be below @xmath19 deg in order to distinguish ballistic deflections from diffusive - type random walk . beyond this we investigate the dispersion of arrival probability distributions by the galactic field , and the splitting of arrival distributions into several images . furthermore we show directional dependencies of the field transparency for cosmic matter and antimatter particles . we also study the influence of random components of the field which imposes uncertainties on the arrival directions . the two magnetic field parameterizations exhibit different field strengths above and below the galactic plane , and differ substantially in their field characteristics near the plane ( fig.[fig : parameterization ] ) . therefore , we divide the sky into three regions of equal solid angles , and study angular deflections for each region separately . the boundaries of these regions are fixed at galactic latitudes of @xmath20 deg . to ensure that every cosmic ray trajectory leads to observation on earth we use the backward tracking method explained in the previous section . we use the term `` northern region '' to refer to antiparticles originating on earth in the direction of positive latitudes above @xmath21 deg . negative latitudes below @xmath22 deg , on the other hand , are referred to as the `` southern region '' , while for latitudes in - between we use the term `` disk region '' . a ) b ) c ) d ) in fig . [ fig : angular]a we show the average directional change @xmath23 between the direction on earth and the direction outside the galaxy as a function of the cosmic ray rigidity @xmath24 using the jf12 field . the distribution was derived from @xmath25 million simulated cosmic rays per rigidity interval . in fig . [ fig : angular]c we show the corresponding spread @xmath26 in terms of standard deviations . for low rigidities @xmath27 ev , cosmic ray confinement owing to the size of our galaxy and its magnetic field leads to large directional changes @xmath28 deg and large average variations in @xmath16 ( @xmath29 deg ) . at rigidity @xmath30 ev , the largest average deflection of @xmath31 deg is found in the southern region ( fig . [ fig : angular]a , downward - pointing triangles ) . the corresponding spread amounts to @xmath32 deg ( fig . [ fig : angular]c ) , such that for @xmath33 of the cosmic rays the deflection angle remains below @xmath34 deg . the average deflection in the northern region is substantially smaller with only @xmath35 deg , however , the spread of @xmath36 deg is larger ( upward - pointing triangles ) . also here most of the cosmic ray deflections are below @xmath34 deg . a similar conclusion holds for the disk region ( square symbols ) . the pt11 parameterization exhibits very similar tendencies for rigidity @xmath30 ev as can be seen in figs . [ fig : angular]b , d . in the disk region the deflections exceed those of the other regions for rigidities above @xmath37 ev ( square symbols ) . here the deflections also exceed that of the jf12 parameterization as was already visualized in fig . [ fig : angularsky ] above . overall , at rigidities @xmath38 ev the deflections are consistently reduced and correspondingly enhance the control over cosmic ray deflection . in the following studies we will therefore use the rigidity of @xmath30 ev as a benchmark . cosmic rays originating from a point source may arrive slightly dispersed after their propagation through extragalactic fields . when traversing the galactic field the extent of the arrival distribution may even be enlarged . in order to obtain information on the arrival direction and arrival probability on earth of a cosmic ray that enters the galaxy in any direction , we use a lensing technique @xcite . the lenses consist of matrices based on the healpix format @xcite , where we divide the sphere into @xmath39 equally sized pixels of approximately @xmath40 in size . for each rigidity interval a separate matrix is produced by backtracking a set of @xmath41 antiparticles , which are distributed uniformly in each pixel . the matrices thus contain the probability of a cosmic ray entering the galaxy with rigidity @xmath24 at pixel direction @xmath42 to be observed in pixel direction @xmath43 . as defined above , @xmath44 refer to the galactic longitudes , and @xmath45 to the latitudes , respectively . by design , the lenses project an extragalactic isotropic distribution onto an isotropic distribution on earth . note that some incoming directions have more simulated trajectories leading to observation on earth , while other directions have less , such that there are directionally dependent variations in the transparency of the field . we normalize the lenses to ensure that the lenses return relative arrival probabilities , and that an isotropic cosmic ray flux is preserved . however , for cosmic rays arriving from individual sources the flux varies depending on the source directions which we will show below . the technical details of the lenses and their production are outlined in @xcite . the lenses used in this contribution were calculated with the crpropa v@xmath46 program @xcite using the pt11 and jf12 parameterizations . in fig . [ fig : images ] we show examples of arrival directions on earth for simulated cosmic rays with rigidity @xmath47 ev together with their original source directions using the jf12 parameterization of the regular field . the incoming cosmic rays followed a fisher probability distribution @xcite @xmath48 with a gaussian width of @xmath49 deg . their arrival directions were calculated with the lensing techniques described above . the color code indicates the relation between sources and their cosmic rays . for all scenarios , the fraction of arriving cosmic rays is shown on the right side of the figure , normalized to the source with the highest arrival probability . different images of the cosmic rays appear depending on their incoming direction . for example , a source direction which coincides on average with the cosmic rays after traversing the galactic field is denoted by the green symbols . only a widening of the directional distribution is observed . another example is a source direction where the cosmic ray distribution is displaced without a strong spread ( purple symbols ) . examples of source directions where the cosmic rays are substantially deflected and exhibit a wide - spread distribution of arrival directions are denoted by the light blue and dark blue symbols . for some source directions , small variations in the cosmic ray incoming direction lead to largely different paths , and therefore to several distinct images of the arrival directions ( red symbols ) . to investigate dispersion effects in the galactic field we again choose initial cosmic rays to follow a fisher probability distribution with a gaussian width of @xmath46 deg . in principle this value could be related to dispersion effects caused by extragalactic fields , which implies a dependency on cosmic ray rigidity . however , to ensure clarity of our galactic field investigations we will use a fixed gaussian spread for all cosmic rays incoming to our galaxy throughout this work . ev originating from arbitrarily chosen sources ( star symbols ) , and the relative arrival probability in percent . relations of cosmic rays with their sources and corresponding arrival probabilities are indicated by the color code ( jf12).,scaledwidth=45.0% ] after the cosmic rays traversed the galactic field we quantify the direction and extent of the resulting arrival distribution by calculating around every healpix pixel @xmath43 with non - zero probability a circular curve which includes @xmath50 of all arrival probabilities . we then use the radius @xmath51 of the smallest circle as a measure of the extent of the probability distribution @xcite . in fig.[fig : extent ] we show the extent of the arrival probability distributions in terms of the smallest average radius @xmath52 as a function of cosmic ray rigidity @xmath24 . again we show separately the three galactic regions defined above . as we start with extragalactic directions we use the term `` northern region '' to refer to initial directions with galactic latitudes above @xmath21 deg , etc . of arrival probability distributions on earth resulting from cosmic rays incoming to our galaxy with a fixed gaussian width of @xmath46 deg in three regions separated by galactic latitudes @xmath20 deg as a function of rigidity @xmath24 , a ) jf12 , b ) pt11 . for the exact definition of @xmath51 refer to the text . , title="fig:",scaledwidth=45.0%]a ) of arrival probability distributions on earth resulting from cosmic rays incoming to our galaxy with a fixed gaussian width of @xmath46 deg in three regions separated by galactic latitudes @xmath20 deg as a function of rigidity @xmath24 , a ) jf12 , b ) pt11 . for the exact definition of @xmath51 refer to the text . , title="fig:",scaledwidth=45.0%]b ) the projection of the incoming distribution causes a dispersion or a focusing effect , depending on the original cosmic ray direction and its rigidity @xmath24 . in the southern region ( downward - pointing triangles ) a focusing effect is visible for rigidities around @xmath53 ev . in contrast , the disk region ( square symbols ) exhibits dispersion effects up to the largest rigidities . below @xmath54 ev the average extent @xmath52 is large for both fields , such that it appears difficult to identify arrival directions on earth from a given extragalactic direction . the jf12 field exhibits several pronounced features which appear to be specific to the jf12 field parametrization . such effects are not visible in the pt11 parametrization where @xmath52 appears to increase continuously with reduced rigidity up to @xmath55 deg . above @xmath30 ev both parametrizations lead to similar results with a dispersion effect in the disk region ( square symbols ) , and dispersion and focusing effects in the southern region ( downward - pointing triangles ) . the largest extent of the probability region is found for the pt11 field corresponding to a dispersion of the initial probability distribution by a factor of @xmath46 at @xmath30 ev , which may still be acceptable for an analysis of cosmic ray arrival directions . extending the above study of dispersion we investigate multiple images arising in cosmic ray arrival distributions on earth . again cosmic rays from an incoming direction are fisher - distributed with a gaussian width of @xmath46 deg . for some incoming directions , the small angular deviations within this distribution are sufficient to change the arrival direction on earth substantially , leading to distinct maxima in the arrival distributions . examples for such multiple images are denoted in fig . [ fig : images ] by the red symbols . we count the number of arrival images by searching for connected areas of arbitrary shape in - between which the probability falls below a pre - defined threshold . for this we coarsen the healpix resolution in the arrival distribution from originally @xmath56 deg to @xmath57 deg and require each pixel to carry at least @xmath58 of the pixel with the maximum arrival probability . the image multiplicity then arises from counting connected areas that are separated from one another by at least one pixel below the pre - defined threshold . in fig . [ fig : jets ] we show the average multiplicity @xmath59 of arrival images as a function of cosmic ray rigidity @xmath24 in the three regions of the galactic sphere separated by galactic latitudes @xmath20 deg . overall , the image multiplicity decreases with increasing rigidity . the multiplicity arising from the pt11 parametrization ( fig . [ fig : jets]b ) appears to exceed that of the jf12 parameterization ( fig . [ fig : jets]a ) . of images in the arrival probability distributions on earth resulting from cosmic rays incoming to our galaxy with a gaussian width of @xmath46 deg in three regions separated by galactic latitudes @xmath20 deg as a function of rigidity @xmath24 , a ) jf12 , b ) pt11 . for the image definition refer to the text . , title="fig:",scaledwidth=44.0%]a ) of images in the arrival probability distributions on earth resulting from cosmic rays incoming to our galaxy with a gaussian width of @xmath46 deg in three regions separated by galactic latitudes @xmath20 deg as a function of rigidity @xmath24 , a ) jf12 , b ) pt11 . for the image definition refer to the text . , title="fig:",scaledwidth=44.0%]b ) for directions in the northern and southern regions , typically @xmath56 image of the arrival direction arises . in the disk region , however , multiple images appear even for cosmic rays with large rigidity . here the image multiplicity is especially large for the pt11 parameterization ( fig . [ fig : jets]b , square symbols ) which is related to the pronounced halo field visualized in fig . [ fig : parameterization]b . multiple images reduce the predictive power of cosmic ray arrival directions and may require additional selection depending on the individual analysis . as criteria to reduce image multiplicity both cosmic ray rigidity and a selection of incoming directions away from the galactic disk region are relevant . related to the above study on dispersion we investigate extragalactic directions causing a relatively enhanced flux of cosmic rays on earth @xcite , and directions for which the arrival probability disappears as no simulated trajectory leads to earth . these effects have a direct impact on the visibility of a source by cosmic ray messengers , and the luminosity required for observation on earth . examples of varying transparency of the field depending on the incoming directions are shown in fig . [ fig : images ] . to demonstrate the enhanced flux of a few extragalactic directions we show in fig . [ fig : focusing]a for incoming cosmic rays with rigidity @xmath30 ev the probability @xmath60 of arriving on earth coded in color . the lightly colored regions indicate incoming cosmic ray directions with a high probability of observation on earth . in order to quantify flux enhancement we organize the incoming directions ( binned in @xmath39 pixels of @xmath56 deg ) according to their arrival probabilities @xmath61 on earth and select the leading @xmath62 directions . these incoming directions cover a solid angular region of @xmath63 and provide a relative flux contribution of @xmath64 . of observing a cosmic ray with rigidity @xmath30 ev on earth as a function of the incoming direction to our galaxy ( jf12 ) . b ) flux @xmath65 on earth originating from the @xmath66-percentile of the directions with the largest arrival probabilities in fig . a ) . , title="fig:",scaledwidth=45.0%]a ) of observing a cosmic ray with rigidity @xmath30 ev on earth as a function of the incoming direction to our galaxy ( jf12 ) . b ) flux @xmath65 on earth originating from the @xmath66-percentile of the directions with the largest arrival probabilities in fig . a ) . , title="fig:",scaledwidth=45.0%]b ) in fig . [ fig : focusing]b we show for the above example the observed flux @xmath65 of cosmic rays originating from the @xmath66 incoming directions with the highest probabilities indicated in fig . [ fig : focusing]a , when assuming an isotropic extragalactic flux . we find that these few incoming directions cause a wide spread distribution and contribute @xmath67 to the observed flux . in a more general approach we show in fig . [ fig : int_flux ] the relative flux @xmath68 of observed cosmic rays as a function of the solid angular region @xmath69 covered by the incoming directions with the highest arrival probabilities . at low rigidity @xmath30 ev in the jf12 parameterization @xmath33 of the cosmic ray flux on earth is caused by about @xmath50 of the extragalactic directions ( fig . [ fig : int_flux]a ) . with increasing rigidity all extragalactic directions contribute equally to the flux on earth . the pt11 parameterization yields similar results as shown in fig . [ fig : int_flux]b , however , with less inhomogeneity at low rigidity @xmath24 . of observed cosmic rays originating from the incoming directions with the highest arrival probabilities covering @xmath69 of the sky , a ) jf12 , b ) pt11 . , title="fig:",scaledwidth=45.0%]a ) of observed cosmic rays originating from the incoming directions with the highest arrival probabilities covering @xmath69 of the sky , a ) jf12 , b ) pt11 . , title="fig:",scaledwidth=45.0%]b ) in contrast , the flux from certain extragalactic directions is not only suppressed but can even disappear , as no path leads to observation on earth . as an example we show in fig . [ fig : antimatter]a example trajectories of cosmic antimatter with rigidity @xmath17 ev from the same extragalactic direction traversing a thin slice of @xmath70 pc around the @xmath71-@xmath11-plane in the galactic coordinate system . the trajectories are expected to miss the solar environment ( marked by the yellow star ) . to enable this demonstration for antimatter we produced a separate set of lenses by backtracking matter particles in the jf12 field . ev incoming to the jf12 field from the same direction that miss the solar system ( yellow star ) . the trajectories are shown in a thin slice of @xmath70 pc around the @xmath71-@xmath11-plane . b ) directions with unobserved cosmic matter ( full curves ) and antimatter ( dashed curve ) in terms of covered solid angle @xmath72 as a function of rigidity @xmath24 ( blue curve jf12 , red curve pt11 ) . , title="fig:",scaledwidth=45.0%]a ) ev incoming to the jf12 field from the same direction that miss the solar system ( yellow star ) . the trajectories are shown in a thin slice of @xmath70 pc around the @xmath71-@xmath11-plane . b ) directions with unobserved cosmic matter ( full curves ) and antimatter ( dashed curve ) in terms of covered solid angle @xmath72 as a function of rigidity @xmath24 ( blue curve jf12 , red curve pt11 ) . , title="fig:",scaledwidth=44.0%]b ) in general , both matter and antimatter particles entering the galaxy exhibit directions with a negligible arrival probability on earth . in fig . [ fig : antimatter]b we show the fraction of the sky in terms of solid angles @xmath72 , where none of @xmath73 simulated cosmic rays reached earth as a function of their rigidity @xmath24 . for antimatter trajectories in the jf12 field , invisible directions appear at small and large rigidities ( dashed curve ) . for matter particles of rigidity @xmath30 ev traversing the jf12 field , the invisible sky fraction is about @xmath58 and disappears above @xmath74 ev ( blue curve ) . in contrast , for the pt11 field the invisible fraction of the sky for matter particles appears to be generally small ( red curve ) . at cosmic ray rigidities of @xmath30 ev , variations in the field transparency from specific extragalactic directions lead to relative suppression and enhancement effects which may eventually require corrections in individual arrival direction analyses . the variations are strongly reduced with increasing cosmic ray rigidity . beyond the regular large - scale component , recent galactic magnetic field models also contain small - scale random structures which are motivated e.g. by supernovae @xcite . such local disturbances are expected to cause a randomly oriented field component which introduces uncertainties in the predicted arrival directions of extragalactic cosmic rays . we expect this impact to be small , since the direction of the random component changes on a scale that is substantially smaller than the gyroradius of cosmic rays constrained in the galaxy . to investigate the influence of such random fields we compare two different realizations of the so - called striated and turbulent random components as described in @xcite with a coherence length of @xmath75 pc . cosmic rays are then deflected in both the regular jf12 and the first random field realization , and in the regular jf12 and the second random field , respectively . as we aim to investigate the influence of the random fields on arrival directions we use the lensing technique . to obtain the most probable arrival direction on earth we use the same techniques described above when studying the dispersion of the probability distribution ( section [ sec : dispersion ] ) . we calculate the radius @xmath51 containing @xmath50 of the arrival probabilities , and use the center of the pixel with the smallest radius @xmath51 as the expected arrival direction . in fig.[fig : delta]a we sketch the angular distance @xmath76 between the arrival directions resulting from the two random field realizations . here the incoming cosmic ray direction to the galaxy is indicated by the star symbol , and the two alternative arrival directions are denoted by the circular symbols . in fig.[fig : delta]b we show the angular distances @xmath76 between the arrival directions on earth using the two realizations of the random fields as a function of cosmic ray rigidity @xmath24 . the red curve indicates the median values . for rigidity @xmath30 ev the uncertainty in the arrival directions is below @xmath14 deg for @xmath50 of the cosmic rays , and rarely extends to more than @xmath19 deg . between two cosmic ray arrival directions ( circular symbols ) from an incoming direction ( star symbol ) , and directional difference @xmath77 of the deflections ( tangential to the sphere ) resulting from different magnetic field orientations . b ) angular distances @xmath76 between the arrival directions using two different realizations of the striated and turbulent random components in addition to the regular jf12 field parametrization as a function of rigidity @xmath24 . the median is depicted by the curve.,title="fig:",scaledwidth=40.0%]a ) between two cosmic ray arrival directions ( circular symbols ) from an incoming direction ( star symbol ) , and directional difference @xmath77 of the deflections ( tangential to the sphere ) resulting from different magnetic field orientations . b ) angular distances @xmath76 between the arrival directions using two different realizations of the striated and turbulent random components in addition to the regular jf12 field parametrization as a function of rigidity @xmath24 . the median is depicted by the curve.,title="fig:",scaledwidth=45.0%]b ) as expected , the impact of the random field is significantly smaller compared to the deflections of the cosmic rays arising from the regular galactic field ( fig.[fig : angular ] ) . however , the uncertainties generated by such random deflections are sufficiently sizable to consider optimization of the rigidity threshold for individual arrival direction analyses . in order to work in a phase space region in which cosmic ray deflections can be controlled at least in a probabilistic way a minimum cosmic ray rigidity of @xmath30 ev is recommended . this value results from avoiding deflections leading to a bend of @xmath19 deg which overlaps with cosmic ray diffusion . all other distributions presented above on the dispersion of arrival probability distributions and multiple images , variance in field transparency , and uncertainties due to random field components are in accordance with this minimal rigidity value . for a typical large - scale analysis the @xmath30 ev rigidity threshold may be sufficient . however , depending on individual analysis requirements , the above key distributions may help to determine whether the rigidity threshold needs to be adjusted to larger values , or whether restrictions to incoming directions aside the galactic disk region need to be introduced . for a typical point source search , the reliability of galactic field corrections are of utmost importance . as a first step we compare directly the cosmic ray deflections resulting from the two field parameterizations pt11 and jf12 . in order to exemplify the impact of these differences on point source searches we simulate a typical corresponding analysis . we determine the discovery potential when the true galactic field is known , and quantify the reduced discovery potential when taking into account the different deflections of the two field parameterizations . uncertainties in the current knowledge of the galactic field can be obtained to some extent from the different arrival distributions of the two field parameterizations pt11 and jf12 . note that their fields are not completely independent regarding the overlap in the measurements constraining their fits , and usage of similar electron density distributions . however , the two parameterizations follow different ansatzes and include disjoint measurements , such that a direct comparison of cosmic ray deflections at least gives an idea of limited knowledge in the magnetic field . for cosmic rays originating from the identical extragalactic direction @xmath78 we investigate the different deflections resulting from the two field parameterizations . we use the above - mentioned lensing technique to deliver arrival probability distributions . as outlined in section [ sec : turbulent ] , we calculate the most probable arrival direction on earth by using the center of the pixel with the smallest radius @xmath51 containing @xmath50 of the arrival probabilities . this gives the expected arrival directions @xmath79 for the jf12 parametrization and for the pt11 parametrization , respectively . we first study the different directions of cosmic ray deflections by their azimuthal angular distance @xmath77 which is measured tangentially to the sphere ( see fig.[fig : delta]a ) . these directional differences reflect different field orientations of the two parameterizations . in the directions of the deflections between jf12 and pt11 , a ) for rigidity @xmath17 ev in the three regions separated by galactic latitudes @xmath20 deg , b ) as a function of rigidity . the curve depicts the median values . , title="fig:",scaledwidth=44.0%]a ) in the directions of the deflections between jf12 and pt11 , a ) for rigidity @xmath17 ev in the three regions separated by galactic latitudes @xmath20 deg , b ) as a function of rigidity . the curve depicts the median values . , title="fig:",scaledwidth=45.0%]b ) in fig.[fig : psi]a we show the directional difference @xmath77 for cosmic ray rigidity @xmath17 ev for the three regions separated by galactic latitudes @xmath20 deg . for the southern region almost all directions of the deflections are within @xmath19 deg ( red histogram ) . in the northern region most of the incoming directions show @xmath80 deg ( blue histogram ) . about @xmath81 of the incoming directions exhibit @xmath82 deg , i.e. here the directions are nearly opposite . in the disk region the differences between the two fields are large as half of the incoming directions are within @xmath19 deg , and the other half has directional differences above @xmath19 deg ( green histogram ) . of cosmic rays after deflections in jf12 and pt11 , a ) for rigidity @xmath17 ev in the three regions separated by galactic latitudes @xmath20 deg , b ) as a function of rigidity . the curve depicts the median values . , title="fig:",scaledwidth=44.0%]a ) of cosmic rays after deflections in jf12 and pt11 , a ) for rigidity @xmath17 ev in the three regions separated by galactic latitudes @xmath20 deg , b ) as a function of rigidity . the curve depicts the median values . , title="fig:",scaledwidth=45.0%]b ) in fig.[fig : psi]b we show the rigidity dependence of the directional difference @xmath77 for all regions where the curve represents the median values . for all rigidities above @xmath30 ev , the most likely differences in the directions of the deflections are below @xmath14 deg , and for @xmath50 of the cosmic rays the field directions are within @xmath83 deg . the second important aspect of the field differences is the absolute angular distance in the arrival directions resulting from the pt11 and jf12 parameterizations which are denoted by @xmath76 ( see fig.[fig : delta]a ) . in fig.[fig : comparison]a we show the angular distance @xmath76 for cosmic rays with rigidity @xmath17 ev for the three regions separated by galactic latitudes @xmath20 deg . for the northern and southern regions the angular distance between the two parametrizations is below @xmath84 deg for @xmath85 of the incoming directions ( blue , red histograms ) . in the disk region , only @xmath86 of the arrival directions show angular distances below @xmath84 deg , while the majority of incoming directions have larger angular distances up to @xmath87 deg . in fig.[fig : comparison]b we show the angular distance @xmath76 for all regions as a function of cosmic ray rigidity . the curve represents the median values . for cosmic rays with small rigidity @xmath30 ev , half of them result at an angular difference below @xmath88 deg . however , there is a long tail towards large angular distances resulting from the two field parameterizations . although sizable differences in the directional characteristics and magnitudes of the two field parameterizations exist , their influence is sufficiently reduced at large cosmic ray rigidity . for example , at @xmath17 ev , the absolute deflection angles @xmath16 as well as the angular distances @xmath76 arising from the two fields become consistently small for most incoming directions . in fig . [ fig : source_arrivals ] we show example arrival distributions of cosmic rays with rigidity a ) @xmath89 ev , b ) @xmath17 ev originating from ten sources . the directions of the sources are denoted by the star symbols . the initial cosmic ray distributions followed the fisher distribution with a gaussian width of @xmath46 deg . and @xmath33 levels . the incoming cosmic ray distributions were centered at the directions denoted by the star symbols and fisher distributed with a gaussian width of @xmath46 deg ; rigidity a ) @xmath89 ev , b ) @xmath17 ev . , title="fig:",scaledwidth=44.0%]a ) and @xmath33 levels . the incoming cosmic ray distributions were centered at the directions denoted by the star symbols and fisher distributed with a gaussian width of @xmath46 deg ; rigidity a ) @xmath89 ev , b ) @xmath17 ev . , title="fig:",scaledwidth=44.0%]b ) indicated by the dark ( light ) red regions are the @xmath90 ( @xmath33 ) arrival probability distributions of the cosmic ray after deflections by the pt11 field . the blue regions give the corresponding arrival probability distributions from the jf12 field . at cosmic ray rigidity @xmath89 ev ( fig . [ fig : source_arrivals]a ) at least half of the arrival probability distributions exhibit substantial overlap for the pt11 and jf12 fields . with increasing rigidity ( fig . [ fig : source_arrivals]b , @xmath17 ev ) the overlap increases as expected . the number of images is also reduced at larger rigidity . this implies that , at large rigidities , the agreement of the two field parameterizations is sufficiently large to investigate the impact of the field uncertainties on a point source search which we present in the following section . for our simulated search for origins of cosmic rays we study separately sources in the three regions of the galaxy ( latitude @xmath20 deg ) . in each region we repeatedly simulate ten sources and demand the cosmic rays to follow a fisher probability distribution with a gaussian width of @xmath46 deg . we also simulate isotropically distributed cosmic rays with full sky coverage as a background contribution . in the following we perform multiple analyses with sets of @xmath91 cosmic rays for which we vary the contribution of signal cosmic rays , i.e. cosmic rays arriving from the ten sources , between signal fraction @xmath92 and @xmath93 . to quantify the analysis sensitivity we use the log - likelihood function @xmath94 } \;. \label{eq : loglikeli}\ ] ] the sum refers to all simulated @xmath95 cosmic rays . parameter @xmath96 denotes the anticipated fraction of signal cosmic rays from the sources when analyzing the data , and the isotropically distributed cosmic rays are assumed to contribute with @xmath97 correspondingly . the probabilities @xmath98 represent the anticipated arrival probability distributions for cosmic rays with rigidity @xmath24 which originate from the sources and are expected to be observed in directions ( @xmath99 ) on earth . they were obtained using the lensing techniques . the background probability @xmath100 corresponds to the inverse number of pixels for which we used the above @xmath39 pixels of approximately @xmath40 . as the test statistics we use the likelihood ratio @xmath101 which approximately follows a @xmath102 distribution with @xmath56 degree of freedom @xcite for each anticipated signal fraction @xmath96 we repeat the simulation of cosmic ray sets @xmath103 times and determine the average maximum @xmath104 . the significance by which isotropic arrival distributions can be excluded is then estimated by converting the integral @xmath105 above @xmath104 to gaussian standard deviations @xmath106 . in the analysis we use as the simulated scenario the jf12 arrival probability distributions @xmath98 to describe cosmic ray deflections . to obtain a benchmark for a best - case scenario , where the field and the cosmic ray rigidities are perfectly known , we first analyze cosmic rays with rigidity @xmath89 ev by using the jf12 field , here representing the true field . note that this scenario returns optimistic results as we neglect deflections in the small - scale random field and demand sources to be located in one of the three galactic regions exclusively . in fig.[fig : sensitivity]a we show the significance @xmath106 as a function of the signal fraction @xmath107 of the simulated sample . with perfect knowledge of the galactic field a signal fraction of @xmath108 is sufficient for a @xmath109 discovery ( full curves ) . to take into account uncertainties in the galactic field as encoded in the two different parameterizations , we then perform the analysis with the pt11 arrival probability distributions instead of the true jf12 probability distributions . as the results are slightly dependent on the exact directions of the sources , we repeat the analyses nine times in each region and present the average resulting values . in fig.[fig : sensitivity]a we show the significance @xmath106 of a deviation from isotropic arrival distributions as a function of the average signal fraction @xmath107 for the three regions ( dashed curves ) . a signal fraction of @xmath110 is sufficient for a @xmath111 discovery for the northern and southern regions , and slightly larger for the disk region ( @xmath112 ) . when compared to the above best - case scenario , the field uncertainties require the signal fraction for discovery to increase substantially by a factor of @xmath113 . in fig . [ fig : sensitivity]b we present the required signal fraction @xmath114 for discovery as a function of cosmic ray rigidity @xmath24 . the full curves represent the required signal fraction for the benchmark scenario having perfect knowledge of the galactic magnetic field . the required signal fraction @xmath114 when including field uncertainties are shown by the dashed curves . it is interesting to note that , although deflections are on average twice as large in the southern region compared to the northern region , their sensitivities are of similar value above rigidity @xmath89 ev . in the disk region the different characteristics of the two parameterizations have a large impact on the point source search . here the required signal fraction almost doubles irrespective of rigidity . deg and sets of @xmath91 cosmic rays with fraction @xmath107 originating from the sources and @xmath115 from isotropic background with full sky coverage . a ) significances for deviation from isotropic arrival distributions at @xmath89 ev using the true galactic field ( full curve ) for the anticipated arrival probability @xmath116 in eq . ( [ eq : loglikeli ] ) , and the other field alternatively ( dashed curves ) . b ) required signal fraction for a @xmath111 deviation from isotropic arrival directions using the true field ( full curve ) compared to using the other field ( dashed curves ) as a function of rigidity @xmath24 . , title="fig:",scaledwidth=45.0%]a ) deg and sets of @xmath91 cosmic rays with fraction @xmath107 originating from the sources and @xmath115 from isotropic background with full sky coverage . a ) significances for deviation from isotropic arrival distributions at @xmath89 ev using the true galactic field ( full curve ) for the anticipated arrival probability @xmath116 in eq . ( [ eq : loglikeli ] ) , and the other field alternatively ( dashed curves ) . b ) required signal fraction for a @xmath111 deviation from isotropic arrival directions using the true field ( full curve ) compared to using the other field ( dashed curves ) as a function of rigidity @xmath24 . , title="fig:",scaledwidth=45.0%]b ) as expected , the sensitivity of the point source search improves consistently with increasing cosmic ray rigidity . again comparing with the above best - case scenario , cosmic rays with @xmath17 ev entering the galaxy in the northern or southern regions require a moderate @xmath58 increase in the required signal fraction for discovery owing to field uncertainties . in contrast , incoming directions at the disk region need a @xmath117-fold larger signal fraction . in view of our current knowledge of the galactic field , for analyses which aim at selected source directions and require corresponding galactic field corrections we recommend a cosmic ray rigidity of at least @xmath89 ev . in our simulated search the required signal fraction for a @xmath111 discovery remains below @xmath58 . our direct comparisons of cosmic ray deflections in the two field parameterizations provide some information on the actual knowledge of the entire field map . for cosmic ray rigidity @xmath30 ev , the angular distance after deflections in the two fields is within @xmath118 deg for @xmath50 of the directions incoming to our galaxy . this value is much smaller at @xmath17 ev rigidity , where the differences are below @xmath119 deg for @xmath120 of the directions incoming to our galaxy . to study the influence of uncertainties in the field on searches for cosmic ray origins , we analyzed simulated astrophysical scenarios using a log - likelihood method . the method includes the anticipated probability distributions for cosmic ray arrival after traversing the galactic field , and quantifies deviations from isotropic arrival distributions . we estimated the influence of field uncertainties by using one field parameterization in the simulation of the scenario , and by applying the other field parameterization in the log - likelihood analysis . for cosmic rays with rigidity @xmath17 ev arriving from sources within galactic latitudes @xmath20 deg ( disk region ) we found that the field uncertainties increase the required signal fraction for a @xmath121 discovery substantially by more than a factor of two . however , for sources in the northern and southern regions emitting cosmic rays with @xmath17 ev , the field uncertainties are relatively small and increase the required signal fraction for a @xmath121 discovery by @xmath58 only . corrections for deflections in the galactic magnetic field using the two parameterizations pt11 and jf12 can be meaningfully considered for cosmic ray rigidities above @xmath38 ev . above this rigidity , deflections can be distinguished from diffusive random walk . this has strong implications for analyses using cosmic ray data with mixed composition . for protons this rigidity corresponds to energies above @xmath122 eev . however , when analyzing , e.g. , neon nuclei with charge @xmath123 , meaningful corrections can be performed for energies above @xmath6 eev only . when quantifying uncertainties in the galactic field from comparisons of the two field parameterizations pt11 and jf12 , the rigidity threshold needs to be raised substantially . then both fields give similar predictions for cosmic ray deflections in the northern and southern regions with galactic latitudes @xmath124 deg . in the disk region @xmath125 deg , however , the differences in the predictions remain large . consequently , in our simulated search for cosmic ray origins the arising uncertainties are substantial for sources near the galactic disk , and may be considered acceptable for sources aside the disk emitting cosmic rays with rigidities @xmath126 ev . we wish to thank very much g. farrar for fruitful discussions , and g. farrar , p. tinyakov and m. sutherland for valuable comments on the manuscript . this work is supported by the ministerium fr wissenschaft und forschung , nordrhein - westfalen , the bundesministerium fr bildung und forschung ( bmbf ) , and the helmholtz alliance for astroparticle physics . abraham , j. , abreu , p. , aglietta , m. , et al . ( the pierre auger collaboration ) , 2010 , phys . b 685 , 239 abu - zayyad , t. , aida , r. , allen , m. , et al . , ( the telescope array collaboration ) , 2013 , astrophys . j. , 768 , l1 abbasi r. u. , abe , m. , abu - zayyad , t. , et al . ( the telescope array collaboration ) , 2014 , astrophys . j. , 790 , l21 aab , a. , abreu , p. , aglietta , m. , et al . ( the pierre auger collaboration ) , 2015 , astrophys . j. , 802 , no . 2 , 111 abreu , p. , aglietta , m. , ahlers , m. , et al . , ( the pierre auger collaboration ) , 2013 , jcap 1305 , no . 05 , 009 aab , a. , abreu , p. , aglietta , m. , et al . ( the pierre auger collaboration ) , 2014 , phrvd , 90 , 122005 aab , a. , abreu , p. , aglietta , m. , et al . ( the pierre auger collaboration ) , 2014 , phrvd , 90 , no . 12 , 122006 stanev , t. , 1997 , astrophys . j. , 479 , 290 aartsen , m. g. , abraham , k. , ackermann , m. , et al . ( the pierre auger collaboration , telescope array collaboration , icecube collaboration ) , 2016 , jcap 01 , 037 tinyakov , p. g. and tkachev , i. i. , 2002 , astropart . phys . , 18 , 165 pshirkov , m. s. , tinyakov , p. g. , kronberg , p. p. , & newton - mcgee , k. j. , 2011 , astrophys . j. , 738 , 192 jansson , r. , & farrar , g. r. , 2012 , astrophys . j. , 757 , 14 farrar , g. r. , 2014 , comptes rendus physique , 15 , 339 , arxiv:1405.3680 keivani , a. , farrar , g. r. , & m. sutherland , m. , 2014 , astropart . 61 , 47 farrar , g. r. , awal , n. , khurana , d. , & sutherland , m. , 2015 , 34th int . cosmic ray conf . , the hague , the netherlands , pos ( icrc2015 ) 560 , arxiv:1508.04530 farrar , g. r. , et al . , 2016 , in preparation erdmann , m. , mller , g. , & urban , m. , 2015 , 34th int . cosmic ray conf . , the hague , the netherlands , pos ( icrc2015 ) 557 , arxiv:1508.03737 cordes , j. m. , & lazio , t. j. w. , 2002 , arxiv : astro - ph/0207156 harari , d. , mollerach , s. , & roulet , e. , 2000 , jhep , 0002 , 035 bretz , h .- p . , erdmann , m. , schiffer , p. , walz , d. , & winchen , t. , 2014 , astropart . c54 , 110 gorski , k. , hivon , e. , banday , a. , et al . , 2005 , j. , 622 , 759 , 771 winchen , t. , 2013 , phd thesis , rwth aachen university batista , r. a. , dundovic , a. , erdmann , m. , et al . , 2016 , jcap 05 , 38 fisher , r. , 1953 , proc . a , 217 , 295 mller , g. , 2016 , phd thesis , rwth aachen university pshirkov , m. s. , tinyakov , p. g. , & urban , f. r. , 2013 , mnras , 436 , 2326 jansson , r. , & farrar , g. r. , 2012 , astrophys . j. , 761 , l11 beck , m. c. , beck , a. m. , beck , r. , et al . , 2014 , arxiv:1409.5120 wilks , s. s. , 1938 , ann . math . number 1 , 60
electrostatic interactions of charged macromolecules and colloids are often governed by small oppositely charged ions ( counterions ) that maintain global electroneutrality . these counterion - mediated interactions play a fundamental role in classical charged ( coulomb ) fluids that are abundant in biological and soft matter context @xcite and include many charged macromolecules ( such as nucleic acids dna and rna , actin filaments , microtubules and globular proteins ) , affecting their functional , structural and dynamical behavior . in spite of the importance of electrostatic interactions , there is no general method that would allow for an accurate prediction of electrostatic effects in all regions of the parameter space , defined by the surface charge density of macroions , charge valency of counterions , dielectric mismatches between the often hydrophobic core of the macromolecule and the surrounding aqueous medium , etc . often the electrostatic interactions are treated on the poisson - boltzmann ( pb ) level leading to effective interactions which turn out to be always repulsive between like - charged macromolecules . conceptually , the pb approach corresponds to a mean - field treatment of electrostatic interactions and is asymptotically valid only for sufficiently large separations between macromolecules , low enough surface charge densities and low counterion valency @xcite . it characterizes a situation where electrostatic fluctuations and correlations due to the counterions are negligible . there are other regions in the parameter space of charged macromolecules where one expects the mean - field framework to break down leading to the emergence of a completely different non - pb - type physics . a notorious example is the phenomenon of like - charge attraction , which emerges between highly charged macroions or in the presence of high valency counterions and has been at the focus of both experimental @xcite and theoretical investigations in recent years ( see refs . @xcite for an extensive reference list ) . it appears to us that among the most important recent advances in this field has been the systematization of non - pb effects based on the notions of _ weak coupling _ ( wc ) and _ strong coupling _ ( sc ) approximations . these terms refer to the strength of electrostatic coupling in the system and may be understood conceptually in terms of the electrostatic interactions of mobile counterions with fixed external charges ( macroions ) in the system when compared with direct electrostatic interactions between the counterions themselves . this latter contribution may be characterized in terms of the _ bjerrum length _ , @xmath0 which corresponds to the separation at which two unit charges , @xmath1 , interact with thermal energy @xmath2 in a medium of dielectric constant @xmath3 ( in water and at room temperature , the value is @xmath4 nm ) . if the charge of the counterions is @xmath5 then the bjerrum length scales as @xmath6 . the interaction of counterions with macroion charges ( of surface charge density @xmath7 ) can be characterized in terms of the so - called _ gouy - chapman length _ , @xmath8 which gives the separation at which the counterion - surface interaction energy equals @xmath2 . the ratio of these two fundamental length scales introduces a dimensionless parameter @xmath9 which is known as the ( netz - moreira ) electrostatic coupling parameter @xmath10 @xcite and quantifies the strength of electrostatic coupling in the system . this parameter is closely related to the plasma parameter of ionic systems @xcite and may be written also in terms of the typical lateral spacing , @xmath11 , between counterions in the proximity of a charged surface , i.e. , @xmath12 as set by the local electroneutrality condition @xmath13 . it then follows that in the weak coupling regime , defined by @xmath14 , the width of the gouy - chapman layer @xmath15 is much larger than the separation between two neighbouring counterions in the counterion layer and thus it behaves basically as a three - dimensional gas . each counterion in this case interacts with many others and the collective mean - field approach of the poisson - boltzmann type is thoroughly justified . on the other hand , in the strong coupling regime , defined by @xmath16 , the mean distance between counterions , @xmath11 , exceeds the layer width and thus the counterion layer behaves essentially as a two - dimensional layer @xcite . in this case the mean - field approach breaks down as each counterion is isolated laterally in a relatively large correlation hole of size @xmath11 . in fact , as each counterion can move almost independently from the others along the direction perpendicular to the charged surface , the properties of the system are dominated by single - particle contributions on the leading order , which is in stark contrast with the collective mean - field picture and emerges as a direct consequence of strong electrostatic correlations in the system . the two dychotomous limits are characterised by a low ( high ) valency of the counterions , a small ( large ) value of the surface charge density and/or large ( small ) medium dielectric constant . conceptually the study of the sc regime has been pioneered in several recent works @xcite using various analytical methods . it was shown @xcite that both the wc and the sc limits may be described analytically as two exact _ limiting laws _ from a single general theory for classical coulomb fluids : while the pb theory follows in the limit of @xmath17 , a limiting single - particle sc theory follows in the limit @xmath18 , which thus allow for an exact statistical mechanical treatment of charged systems at two opposed limiting conditions . the parameter space in between can be analyzed by approximate methods @xcite , being accessible effectively only _ via _ computer simulations @xcite . exact solutions for the whole range of coupling parameters are available only in one dimension @xcite . the wc - sc paradigm has been tested extensively @xcite and was found to describe computer simulations quantitatively correctly in the respective regimes of validity , thus providing a unifying view of the behavior of coulomb fluids . the main facets of these results are retained even when the model system is generalized in order to include more realistic features describing the bathing solution or the nature of the fixed or mobile charges in the system . though , for instance , multipolar charge distribution of mobile counterions @xcite or statistically disordered distribution of fixed charges @xcite unavoidably introduce novel features in the counterion - mediated electrostatic interaction , the application of the same general philosophy embodied in the weak and strong coupling limits remains sound and valid . the only case where it needs to be amended in an essential manner is when the bathing solution contains a mixture of univalent as well as polyvalent salts , which incidentally are also the most common experimental conditions . in that case a more sophisticated mixed weak - strong coupling analysis is in order leading to qualitatively different results @xcite . though originally formulated in the context of planar macroions , these advances have transpired also in the dna - like models of polyelectrolytes which deal with electrostatic interactions in the context of macroions with cylindrical or indeed helical fixed charge distribution . indeed this particular variant of counterion - mediated interactions has a venerable history @xcite . during the last two decades several distinct analytical approaches aimed in different directions improved the classical pb results for simple dna - like models and revealed the importance of correlation effects as well as several other factors including the discrete or helical charge distributions , chain flexibility , finite molecular size , and dielectric inhomogeneities @xcite . guided by these developments we set ourselves the goal of systematically analyzing the interactions between cylindrical macroions mediated by mobile counterions in the presence of dielectric inhomogeneities . contrary to the case of planar macroions which can be completely characterized by a single dimensionless coupling parameter , cylindrical macroions require in general two dimensionless coupling parameters that consistently describe the range of validity of the strong and weak coupling approximations . the existence of two coupling parameters is due to the simple fact that , if compared to the planar case , the cylinder has a finite radius that needs to enter the fundamental description of the problem . this other dimensionless parameter brought fourth by cylindrical geometry of the macroion is nothing but the so - called _ manning parameter _ @xcite . for a cylinder of radius @xmath19 and linear charge density @xmath20 , it is given by @xmath21 the manning parameter thus represents the dimensionless linear charge density or , on the other hand , also the rescaled radius of the cylindrical charge distribution . note that the two parameters ( i.e. , @xmath10 and @xmath22 ) in the cylindrical geometry are independent and can be set separately . for double - stranded dna ( @xmath23 ) , the manning parameter and the ( netz - moreira ) electrostatic coupling parameter are given by @xmath24 and @xmath25 ( at room temperature in water ) in terms of the counterion valency @xmath26 . in this paper we will analyze systematically the interactions between two cylindrical macroions characterized by a fixed surface charge density as well as a finite dielectric jump between the external dielectric background ( corresponding to a continuum solvent ) and the hydrophobic cores of cylindrical macroions . analytically derived results in both limits of strong and weak coupling will be compared directly with monte - carlo simulation results . the organization of this paper is as follows : we first introduce our model and study the image - charge effects within the problem of a single charged cylinder with neutralizing counterions in section [ sec : one_cyl ] . we then focus on the interaction between two identical charged dielectric cylinders within the wc and the sc theory as well as mc simulations in section [ sec : two_cylinder ] . the results in the presence and absence of dielectric discontinuity at the cylindrical boundaries are analyzed in detail in order to bring up the effects due to the image charges in the two - cylinder geometry . let us focus first on the problem of a single infinitely long uniformly charged cylindrical macroion of radius @xmath19 . the charge of the cylinder is assumed to be distributed uniformly on its surface according to the charge distribution function @xmath27 with @xmath20 being the absolute _ linear _ charge density and @xmath28 the radial coordinate from the cylinder axis . the cylinder is arbitrarily chosen to be negatively charged and its electrical charge is thus neutralized by positively charged mobile counterions of valency @xmath29 , which are present in the region @xmath30 only . in general , the interior of the cylindrical macromolecule can be characterized by a different dielectric constant @xmath31 than the surrounding continuum solvent medium , @xmath3 . this is certainly the case for dna that has a hydrophobic inner core composed mostly of stacked nitrogen bases that have a vastly different dielectric response from an aqueous solution @xcite . other charged polymers that usually do not share structural features with dna can nevertheless also posses hydrophobic inner cores with a local dielectric response that differs from the one of the solvent . the hydrophobic apolar core of the macromolecular backbone would then have a static dielectric constant @xmath32 ( hydrocarbon ) , while the surrounding continuum dielectric medium which is usually water , a polar associated liquid , would be characterized by @xmath33 . one can thus introduce the dielectric discontinuity parameter @xmath34 which measures the relative dielectric mismatch at the interface of the two materials . in a dielectrically homogeneous system , we have @xmath35 and no image charge effects are present , while , for instance , in water - hydrocarbon systems , one has @xmath36 , which suggests strong image charge effects ( note that @xmath37 and the largest value for @xmath38 is obtained when one medium is ideally polarizable , i.e. , is an ideal metal ) . therefore , a single cylinder can be described by three different dimensionless parameters , namely , the electrostatic coupling parameter @xmath10 , the manning parameter @xmath22 and the dielectric discontinuity parameter @xmath38 as defined in eqs . ( [ eq : xicouple ] ) , ( [ eq : xi_m ] ) and ( [ eq : jump ] ) , respectively . the presence of a dielectric inhomogeneity across the boundary of the cylinder , see fig . [ fig : model_1cyln ] , influences the electrostatic potential that can be thus described by the green s function connecting two points @xmath39 outside the cylinder as @xmath40 where @xmath41 is the direct standard coulomb kernel in the absence of dielectric inhomogeneities , @xmath42 and @xmath43 is the image correction " due to the dielectric jump just as in the case of a planar discontinuity . unfortunately in cylindrical geometry the concept of kelvin image charges , that plays such a fundamental role for planar dielectric discontinuities , is a bit diluted since the image correction can not be in general formulated in a way that would entail a summation of properly positioned point image charges . we nevertheless consistently refer to the effects of dielectric inhomogeneities in this system as dielectric image effects . if explicitly stated we believe this inconsistency in nomenclature can not be a source of confusion . instead of using the concept of point image charges one must in fact explicitly solve the poisson equation in cylindrical geometry specified by cylindrical coordinates @xmath44 with appropriate boundary conditions at the surface of the cylinder , which requires the normal component of the electric field to fulfill the following relation @xmath45 where @xmath46 is the cylinder surface charge density . solving the poisson equation with the proper eigenfunction expansion @xcite one can express the image part of the green s function as @xmath47 with the following definition @xmath48 where @xmath49 and @xmath50 are the angle and height differences between the position vectors @xmath51 and @xmath52 , respectively . @xmath53 and @xmath54 are the modified bessel functions of the first and the second kind that enter the eigenfunction expansion @xcite . having derived the appropriate green s function for the cylindrical geometry , we now investigate the behavior of counterions in proximity to a charged dielectric cylinder in the two limits specified by the weak and strong coupling approximations that can be treated analytically . the behavior at intermediate coupling strengths will be analyzed by using extensive mc simulations . as noted before , the wc regime is characterized by a small coupling parameter @xmath14 , which is adequate for low valency counterions , small surface charge , high temperature or high solvent dielectric constant . in the strict limit of @xmath17 , ignoring the usually small fluctuations around the mean field configuration , the system is described exactly by the mean - field pb theory @xcite . formally the pb equation corresponds to the saddle - point condition imposed on the action field - functional in this limit @xcite . for the present system , the pb equation governing the mean electrostatic potential takes the form @xmath55 where the right - hand side is obviously the number density of the counterions , i.e. @xmath56 note that due to axial symmetry and translational invariance in the @xmath57 direction , the problem is reduced to a one - dimensional formulation with @xmath58 and the solution of the pb equation depends only on the radial coordinate @xmath28 . the analytical solution of the pb equation in this case is well known and was first discussed in refs . @xcite within the so - called _ cell model _ where the cylindrical macroion is confined ( coaxially ) in an outer cylindrical boundary of radius @xmath59 . the cell model guarantees a finite solution for the counterion density despite the extremely slowly varying ( logarithmic ) electrostatic potential in two dimensions . the solution of the pb equation for the counterion density takes different forms depending on whether the manning parameter is larger or smaller than a threshold value given by @xmath60 $ ] @xcite . here we are interested mainly in the situations with sufficiently large manning parameters @xmath61 , where the normalized counterion density can be written as @xmath62 , \label{npb}\ ] ] with @xmath63 determined from the transcendental equation @xmath64}.\ ] ] note that the density profile is renormalized such that we have @xmath65 the above density distribution is a monotonically decaying function with very slowly convergent asymptotics @xcite . it is also independent of the dielectric jump parameter , @xmath38 , due to the axial symmetry which implies that the electrostatic field vanishes inside the cylinder . the absence of dielectric discontinuity effects is specific to the mean - field limit where no fluctuations are taken into account and can be derived also in the case of planar slabs @xcite . when the manning parameter is decreased , the system exhibits a continuous counterion condensation transition @xcite at a critical manning parameter @xmath66 . the nature of this transition has been analyzed throughly by means of analytical and numerical methods @xcite . it was in particular shown that the behavior close to the transition point is described by the mean - field theory and fluctuation and correlation effects ( that will be important eventually at sufficiently large manning parameters ) play no role at the transition point . here we shall not consider the behavior of counterions close to the transition point but do note that since the image - charge effects are absent in the mean - field limit , such effects are expected to have no influence on the counterion condensation transition itself . in what follows , we shall also take into account the hard - core repulsion between counterions and the cylinder by assuming that the counterions have a finite radius @xmath67 . within the pb theory , this can not be done explicitly unless full corrections due to excluded - volume repulsions between counterions are taken into account @xcite . here we shall consider a simplified version of the ion size effects by taking an effective hard - core radius for the cylinder , i.e. , by setting @xmath68 . though approximate on the wc level , this procedure turns out to be exact in the sc limit @xcite . ( a ) 0.25 cm ( b ) 0.25 cm ( c ) 0.25 cm in the regime when @xmath69 , the mean - field approximation breaks down , and a different kind of approach is needed . we shall focus on the limit of @xmath18 , where the system can be treated via a systematic strong coupling virial expansion @xcite leading to an exact analytical theory in the leading order in @xmath10 . by construction the sc theory is a single particle theory and contains contributions only from single - particle interactions . we do not elaborate on the derivation of the sc theory here as further details can be found in previous publications @xcite . in this section , we shall be interested in the counterion density profile around the cylinder in the sc limit . the general form of the counterion distribution within the sc theory turns out to be given in terms of a boltzmann factor containing the interaction energy of an individual counterion with a fixed charged macroion @xcite . in a dielectrically inhomogeneous system , such as a charged cylindrical core considered here , the sc counterion density may be written as @xcite @xmath70 , \label{eq : density_ci}\ ] ] where @xmath71 is the dielectric image _ self - energy _ of a single counterion , i.e. , the electrostatic energy of a charged point particle in the vicinity of a neutral dielectric cylinder . this energy of course represents the contribution from the interaction of a counterion with its image charge and thus depends crucially on the value of the dielectric discontinuity at the macroion surface . furthermore , @xmath72 where @xmath73 is the bare electrostatic interaction energy of a single counterion located at a given position @xmath74 with the cylinder charge ( up to an irrelevant additive constant ) . it goes to zero when the charge density on the macroion surface goes to zero . the image - dependent part of the interaction energy between a point charge and the surface charge distribution on the dielectric cylinder , @xmath75 , can be shown to be nil due to the axial symmetry , i.e. @xmath76 if the surface charge distribution on the macroion is non - uniform , such as in the case of charged helical stripes @xcite , the image dependent part of the interaction energy is non - zero and should be considered explicitly . note that physically the dielectric image - dependent part of the interaction energy corresponds to the interaction between the dielectrically induced image charge of the surface charge on the cylinder with the counterion as well as the image charge of the counterion with the surface charge on the cylinder , as follows from the general definition of the green s function ( [ eq : u_0u_im ] ) . using the above equations , we obtain the sc counterion number density profile ( @xmath30 ) as @xmath77 , \label{nsc}\ ] ] where we have introduced the dimensionless integral @xmath78 and the numerical prefactor @xmath79 is determined from the normalization condition , eq . ( [ normrho ] ) . we can extract two limiting behaviors for the function @xmath80 , viz . @xmath81 the limiting form of @xmath80 for @xmath82 implies that at very small separations between a single counterion and the cylindrical surface , the dielectric self - energy has the same form as in the case of a counterion next to a planar wall , where the polarization effects can be described by an image charge inside the wall @xcite . the sc density for @xmath35 reduces to the well - known form for a homogeneous system @xmath83 @xcite . if the dielectric core has a smaller dielectric constant than the medium ( @xmath84 ) , as one typically encounters in the case of biological macromolecules , the image charges have the same sign as counterions and thus lead to a depletion of counterions in the vicinity of the cylinder . this behavior can be seen from the sc density profiles plotted in fig . [ fig : dens]a as the dielectric mismatch parameter @xmath38 or the coupling parameter @xmath10 is increased . they show a reduced density close to the cylinder ( depletion zone ) and a peak some distance away from the cylinder surface . one can estimate the location of the density peak by setting the density ( [ nsc ] ) derivative to zero , @xmath85 . using the approximation ( [ eqi ] ) for @xmath86 , where @xmath87 is the peak distance from the cylinder surface , we obtain to the leading order for small @xmath87 @xmath88 which shows good agreement with the peak location values of the sc density profile as seen in fig . [ fig : dens]a . as noted before , effects of a finite counterion radius @xmath67 within the sc approximation can be taken into account exactly ( due to the single particle nature of the theory ) by increasing the hard - core radius of the cylinder to @xmath89 . the counterion - counterion excluded - volume repulsion effects are absent within the leading - order sc theory and enter only in the subleading terms . we have performed extensive mc simulations in order to verify the validities of the weak- and strong - coupling approaches . the detailed description of mc simulation is given in appendix [ app : mc ] . in fig . [ fig : dens]b we show the radial density profile of counterions around the cylinder without dielectric image effect , @xmath35 . apparently in this case the simulation data are always bracketed by the two analytical results obtained from the wc and sc theories , eqs . ( [ npb ] ) and ( [ nsc ] ) , respectively . the wc theory is found to be valid at sufficiently small coupling parameter or sufficiently large radial distances from the cylinder . the sc theory is valid in the opposite regime , i.e. , for sufficiently large coupling parameters or sufficiently small radial distances . in fact , in agreement with the general trend obtained for planar surfaces @xcite , the validity regime of the sc ( wc ) theory expands to large ( smaller ) separations as the coupling becomes larger ( smaller ) . in the case of a planar charged wall , the validity regime of the sc theory can be estimated systematically as @xmath90 , where @xmath57 is the distance from the charged plane @xcite . in the case of a charged cylinder , a similar criterion is yet to be obtained . in fig . [ fig : dens]c , we show the results for the case where the dielectric constant of the cylinder is different from that of the medium such that @xmath91 . as already discussed the dielectric discontinuities have no effects in the wc limit as confirmed also by the agreement obtained between the simulation data and the wc ( mean - field ) result for small @xmath10 . by contrast , the image effects become quite significant at large coupling parameters as the density profile deviates qualitatively from those obtained with @xmath35 ( fig . [ fig : dens]b ) . the simulation data again show good agreement with the sc approximation at high enough couplings and small enough radial distances and in particular exhibit a depletion zone and a peak at small distances from the cylinder surface as predicted within the sc theory . note that in a dielectrically inhomogeneous system ( @xmath92 ) the sc theory has an explicit dependence on the coupling parameter @xmath10 because of the self - interaction of counterions with their image charges . we now turn our attention to the problem of the interaction between two like - charged cylinders , which is commonly used as a model for the interaction between rigid polyelectrolyte chains @xcite . we consider two identical parallel cylinders of radius @xmath19 oriented along the @xmath57 axis at an interaxial separation of @xmath93 ( see fig . [ fig : model_2cyln ] ) . the electric charge is assumed to be uniformly distributed on the surface for both cylinders with equal linear charge density @xmath20 . the cylinders are assumed to be infinitely long of length @xmath94 and are standardly confined within a confining square box of lateral size @xmath95 ( we shall assume @xmath96 in section [ subsec : wc_2cyln ] and @xmath97 in sections [ subsec : sc_2cyln]-[subsec : sc_2cyln_im ] ) , but we should emphasize that in the regime of manning parameters considered here , the lateral box size plays no significant role @xcite . ( a ) ( b ) we follow the same approach as in the case of a single cylinder and thus first analyze the problem in the weak - coupling limit and then in the strong - coupling limit . we then compare the results from these two analytical theories with the mc simulations . the weak coupling regime ( @xmath98 ) is again characterized by the mean - field pb equation . in this case however , no closed - form solution is available for the nonlinear pb theory . we thus take full recourse to the numerical methods appropriate to the problem . in the two - cylinder geometry , the axial symmetry is broken but the translational symmetry along the @xmath57 axis remains intact ; hence one has to deal with a two - dimensional pb equation in a finite bounding square @xmath99 , which is then solved numerically @xcite . we shall focus on the interaction force between the cylinders , which can be evaluated via the electrostatic stress tensor in a standard manner @xcite . note that the wc scheme used here based on the numerical solution of the pb equation allows for a full analysis of the dielectric discontinuity effects . ( later on in our sc analysis , which will be based on analytical methods , we shall restrict our discussion to only the first - order image approximation , as a full analysis of dielectric discontinuity effects in the sc limit is not yet available . ) here again the counterion size effect will be taken into account approximately by setting the distance of closest approach to each cylinder as @xmath89 . fluctuation contribution around the mean - field solution in cylindrical geometry is non - trivial to calculate @xcite but is by construction small and we thus skip its detailed analysis . it does however depend crucially on the dielectric discontinuity at the cylinder surface . in general , the wc force obtained between two like - charged cylinders is repulsive and decays monotonically with their interaxial separation @xmath93 . the mean - field prediction for the rescaled force per unit length defined via @xmath100 is shown in fig . [ fig : forcedelta_wc ] as a function of the interaxial separation @xmath93 and for different values of the dielectric discontinuity parameter @xmath38 ( see eq . ( [ eq : jump ] ) ) . note that in contrast to the case of a single cylinder studied in the previous section , the dielectric discontinuity is expected to matter in the two - cylinder geometry even within the mean - field approximation . this is due to the axial symmetry breaking which leads to electric field penetrating into the cylindrical cores . as seen in the figure , the wc mean - field results compare very well with the mc data ( to be discussed later ) at sufficiently large interaxial separations . the dielectric discontinuity has a relatively small effect on the interaction force . it leads to an increased repulsion ( and also larger deviations from the mean - field prediction ) at small interaxial separations but its effects diminish at larger separations . as noted before , the sc theory follows systematically from the leading order contribution to the partition function in the limit of infinite coupling parameter @xmath18 @xcite . the corresponding sc free energy involves only interactions between single counterions and the charged cylinders as well as the direct electrostatic interaction between the cylinders themselves ( see refs . @xcite ) . in the presence of the dielectric discontinuity effects , the sc free energy per counterion can be written as @xmath101\rmd{{\mathbf r}}. \label{f1}\ ] ] here @xmath102 is the total number of counterions and @xmath103 is the electrostatic energy due to the interaction between the two cylinders in the absence of any counterions , which contains both the direct coulomb interaction between their surface charges as well as the contribution due to their images that results from the polarization of their dielectric cores . furthermore , @xmath104 is the energy due to the interaction between a single counterion and the surface charges on both cylinders and includes contributions from the corresponding image charges as well . finally , @xmath105 is the image self - energy of counterions , i.e. , the contribution from the interaction of an individual counterion with its own image charges in both cylindrical cores . note that the volume integration should be again taken over the space available to counterions , i.e. , inside the confining square box of lateral size @xmath95 excluding the two cylindrical cores . note also that on the sc level , the counterion excluded - volume repulsions are absent ( as implied by the fact that strongly coupled counterions are highly isolated within large correlation holes in the sc limit ) and only the excluded - volume interaction between the counterions and the cylinders will be present . in fact , the counterion size effects can be accounted for exactly within the sc theory via a hard - core repulsion which simply amounts to setting the effective ( hard core ) cylinder radius as @xmath106 @xcite . as in the single cylinder case all charges are coupled with the interaction kernel , composed of direct and image part , but with the difference that now the image part of the kernel @xmath107 corresponds to two dielectric cylinders separated by the distance @xmath93 . this kernel should satisfy the electrostatic boundary conditions on both cylindrical surfaces ( [ bc ] ) , which leads to very complicated numerical expressions and is unfortunately not available in a closed analytical form . our aim is nevertheless to give a simple approximate analytical expression for the final results , so we need to make an approximation at this step . in what follows we focus on the first - order - image approximation , where we neglect higher orders of inter - cylindrical image interaction . this approximation implies that the image green s functions of the two cylinders can be written as the sum of the green s functions of the isolated cylinders , eq . ( [ vim ] ) , @xmath108 where @xmath109 is the one - cylinder image kernel , eq . ( [ vim ] ) . here , @xmath110 , @xmath111 are distances centered at the first and @xmath112 , @xmath113 at the second cylinder , for instance , @xmath114 with @xmath115 a unit vector pointing in the @xmath116 direction , i.e. , horizontally on fig . [ fig : model_2cyln ] . recall that the surface charge is uniformly distributed on both cylinders so that the fixed charge distribution is composed of two single cylinder distributions @xmath117 where @xmath118 is the charge distribution of a single cylinder , eq . ( [ n0 ] ) . with these assumptions one can then evaluate all interaction terms in eq . ( [ f1 ] ) analytically as follows . the cylinder - cylinder interaction energy per counterion can be obtained as @xmath119 furthermore the image self - energy of a single counterion interacting with the dielectric cores of both cylinders can be derived in the form @xmath120 , \label{wself}\end{aligned}\ ] ] where @xmath80 is defined in eq . ( [ eqw ] ) . finally , the single particle cylinder - counterion contribution can be written as the sum of the interaction with each cylinder separately and assumes the form @xmath121 the cylinder - counterion energy contribution from the @xmath15-th cylinder ( with @xmath122 ) is itself composed of three parts , @xmath123 which correspond respectively to three parts of the green s function , i.e. , @xmath124 , within the approximation implied by eq . ( [ vim2 ] ) and will be evaluated as follows . the first contribution in eq . ( [ eq : v_contrib ] ) , @xmath125 , is the direct interaction of the bare surface charge of each cylinder with a single counterion , i.e. , for the the @xmath15-th cylinder , @xmath126 where @xmath127 is the radial distance from the center of the @xmath15-th cylinder to the counterion , and @xmath22 is the manning parameter as defined in eq . ( [ eq : xi_m ] ) . the second contribution , @xmath128 corresponds to the interaction of a counterion image charge with the surface charge on the _ same _ cylinder , which is thus _ zero _ by symmetry reasons as already discussed in the one - cylinder case , eq . ( [ vself ] ) . this contribution can be also considered as the interaction between the image of the surface charge on the same cylinder with the counterion . finally , we have the cross contributions due to the interaction of a counterion image in one cylinder with the surface charge on the _ other _ cylinder . formally , it follows for each cylinder as ( see appendix [ app : cross ] ) @xmath129 for @xmath130 , where @xmath131 is the radial distance from an axis shifted by @xmath132 from the center of the @xmath15-th cylinder toward the axis of the other cylinder , see fig . [ fig : model_2cyln]b , i.e. , @xmath133 or in cartesian coordinates , @xmath134 the sc free energy thus follows by inserting the above expressions into eq . ( [ f1 ] ) in the form @xmath135- \nonumber \\ & -&\xi(1-\delta)\ln\,\rho_1 ^ 2\rho_2 ^ 2- \xi \delta\,\ln\,{\rho_1^{*}}^2{\rho_2^{*}}^2\biggr\}\rmd{{\mathbf r}}. \label{free}\end{aligned}\ ] ] the corresponding sc force between the two cylinders is obtained simply by differentiating the free energy with respect to @xmath93 @xmath136 which may be expressed in dimensionless units according to eq . ( [ forcedef_wc ] ) . the sc density profile follows from the general expression ( [ eq : density_ci ] ) , which is the same as the integrand in ( [ free ] ) , viz . @xmath137- \label{dens2}\\ & -&\,\xi(1-\delta)\ln\,\rho_1 ^ 2\rho_2 ^ 2- \xi \delta\,\ln\,{\rho_1^{*}}^2{\rho_2^{*}}^2\biggr\}\nonumber.\end{aligned}\ ] ] where the normalization prefactor is determined from the electroneutrality condition . note that the forms of the sc free energy , eq . ( [ free ] ) , and the sc counterion density profile , eq . ( [ dens2 ] ) , reflect the fact that , within the first - order image approximation ( and using the appropriate volume constraints for counterions ) , the two charged dielectric cylindrical cores can be equivalently replaced by four parallel lines of charges , two of which are located along the central axis for each of the cylinders with ( renormalized ) linear charge density @xmath138 and the two other lines of charges with linear charge density @xmath139 are placed along two axes shifted by @xmath132 from the central axis for each of the cylinders , see fig . [ fig : model_2cyln ] . note that the image charge language is applicable for two charged dielectric cylinders in the absence of pointlike charges ( counterions ) @xcite , and can be exactly solved by applying higher orders of images . but in order to be consistent with the first - order image counterion - cylinder approximation , we have used the same image kernel ( [ vim2 ] ) , which accounts for the first - order image approximation between the two cylinders as well ( [ f1 ] ) . at interaxial separtaion @xmath140 and in the absence of image charges ( @xmath35 ) . the density is shown across an arbitrary plane perpendicular to the two cylinders in a color - coded fashion as indicated on the graph . the small depletion zone around each cylinder is due to the counterion volume exclusion as counterions are assumed to have a finite radius of @xmath141 . ] 0.25 cm at interaxial separtaion @xmath140 and in the absence of image charges ( @xmath35 ) . the density is shown across an arbitrary plane perpendicular to the two cylinders in a color - coded fashion as indicated on the graph . the small depletion zone around each cylinder is due to the counterion volume exclusion as counterions are assumed to have a finite radius of @xmath141 . ] 0.25 cm ( a ) 0.25 cm ( b ) 0.25 cm ( c ) 0.25 cm in order to proceed , let us first consider the case where there is no dielectric mismatch between the cylinders and the surrounding medium , i.e. , @xmath35 ; the image charges will thus be absent . the sc interaction in this case has been considered also in previous works @xcite . we shall reproduce some of these previous results for the sake of completeness , but will also provide several new results , including a direct comparison with mc simulations for the force dependence as a function of separation at different values of the coupling parameter as well as a global interaction phase diagram , which have not been considered previously . in the absence of image charges , the sc free energy ( [ free ] ) reduces to a simple form as @xmath142 where the first term is again the bare repulsion between the two cylinders and the second term contains energetic and entropic contributions from counterions . the results for the sc force that follow from eq . ( [ free0 ] ) are shown in fig . [ fig : force0 ] for a few different manning parameters , where we have accounted for the finite counterion radius by choosing @xmath143 . the sc force is repulsive at small separations and becomes attractive beyond the equilibrium interaxial distance @xmath144 where the force vanishes . the sc attraction is mediated by counterions that are strongly coupled to both cylinders in this limit and are thus accumulated mainly in the region between the two cylinders as can be seen from the counterion distribution ( [ dens2 ] ) in fig . [ fig : densdelta0 ] . this is a direct consequence of the energy contributions included in the second term in eq . ( [ free0 ] ) @xcite . by decreasing the separation between the cylinders , the entropic osmotic pressure from counterions sandwiched between the two cylinders becomes increasingly important and the effective interaction becomes repulsive . this repulsion is reduced as the surface - surface separation , @xmath145 , becomes smaller than the counterion diameter , @xmath146 as the counterions are depleted from the intervening region due to excluded - volume effects . this behavior due to the presence of the counterion depletion interaction has been investigated thoroughly in the case of cylinders with helical charge pattern and we shall not delve on it any further here @xcite . as seen in fig . [ fig : force0 ] the entropic repulsion effects at small separations decrease when the manning parameter @xmath22 is increased . for very large manning parameters , the free energy is dominated by purely energetic contributions and does not contain any temperature effects . in fact the limit @xmath147 formally corresponds to the zero temperature limit as we have already taken the limit @xmath18 appropriate within the sc approximation . the connection between the sc limit and the zero temperature limit has been analyzed in detail in refs . @xcite . the asymptotic form of the sc force in the limit of large @xmath22 can be obtained as @xmath148 which is shown as a black solid line in fig . [ fig : force0 ] . the equilibrium hard - core surface - surface separation @xmath149 tends to zero as @xmath150 as @xmath147 @xcite . note that even for moderate values of the manning parameters , @xmath151 , the sc theory predicts a closely packed _ bound state _ with a relatively small surface - surface separation @xmath152 . let us now consider the situation where the coupling parameter has a finite value . in this case we shall study the system using mc simulations as detailed in appendices [ app : mc ] and [ app : mc_force ] . the mc results for the interaction force between like - charged cylinders ( with no dielectric mismatch ) are shown in fig . [ fig : force0mc ] along with the wc and sc predictions . as seen in fig . [ fig : force0mc]a , the wc results agree with the simulation data for a sufficiently small coupling parameters @xmath153 as expected . furthermore , by increasing the coupling parameter , the mc results start deviating strongly from the wc prediction and tend to the sc prediction , where a reasonable agreement is obtained already at coupling parameters of the order @xmath154 . note that the wc and sc predictions bracket the simulation data and thus establish upper and lower bounds for the interaction force at any realistic value of the coupling parameter . ( a ) 0.25 cm ( b ) ( c ) in figs . [ fig : force0mc]b and c we show the results for larger manning parameters of @xmath155 and 100 , respectively . as manning parameter becomes larger , a larger @xmath10 is required in order to achieve the same level of agreement with the sc prediction at a given separation and thus the convergence to the sc limit becomes weaker . this effect is more pronounced at larger interaxial separations . in fact , as is generally known @xcite , the sc effects become more dominant and the sc theory becomes increasingly more accurate at smaller surface - surface separations . in general , the applicability regime of the sc theory ( obtained in the limit of @xmath18 ) to the situations where the coupling parameter @xmath10 is finite can be specified via simple validity criteria as applied successfully in several previous studies @xcite . in what follows we shall derive the particular form of the sc validity criterion for the case of two like - charged cylinders . it should be noted first that the sc theory ( obtained as an exact asymptotic theory in the limit of @xmath18 ) contains only contributions stemming from the interaction of individual counterions with charged objects and thus the counterion - counterion repulsions and other higher order many - body effects are absent . these effects however become increasingly more important as the coupling parameter is decreased , depending crucially on the precise value of all the system parameters . in the case of two cylinders , as noted before , counterions are mostly accumulated in the intervening region between the two cylinders ( fig . [ fig : densdelta0 ] ) where the typical counterion spacing can be estimated by stipulating the local electroneutrality condition as @xmath156 or in rescaled units as @xmath157 . it is thus evident that as the manning parameter becomes larger at a fixed coupling parameter the spacing between counterions tends to become small . hence , a larger effect due to counterion - counterion repulsions and thus larger deviations from the sc theory would be expected to be observed at larger @xmath22 . in other words , the sc theory is expected to overestimate the counterion density in the intervening region at any _ finite _ value of the coupling parameter , which is consistent with the general result that the sc theory gives the upper bound for the density profile of the counterions @xcite . one can thus identify the validity regime of the sc theory by estimating the counterion - counterion contributions and comparing them with the sinlge - particle counterion - cylinder contribution at a given interaxial separation . this can be done by noting that the ground state configuration ( obtained for @xmath158 and @xmath18 ) is predicted within the sc theory to be the configuration where all counterions are localized between the two cylinders and are lined - up along the @xmath57 axis as obviously favored by the electrostatic interaction energy . when the coupling parameter is reduced to a finite value , repulsions between counterions become more important and some counterions tend to escape from the intervening region to the external side of the cylinders ( shown by an arrow in fig . [ fig : diagrams0]a ) . in order to prevent this from happening , the coupling parameter @xmath10 has to be large enough . by comparing the energy of the ground state with the energy of a single defect obtained by allowing one counterion to move to the external side of one of the cylinders , it follows analytically that for a large @xmath22 and in the absence of image charges , one should have @xmath159 these criteria indeed capture the trends observed in fig . [ fig : force0mc ] when viewed in terms of the three main parameters describing the system , i.e. , @xmath10 , @xmath22 and @xmath93 . they automatically cover the one proposed before based on the surface - surface separation between the cylinders , where one requires that the surface - surface separation between the cylinders must be smaller than the spacing between counterions , @xmath160 , or in rescaled units @xcite @xmath161 this latter criterion is inspired by the observations in the case of charged planar walls where it can indeed be derived from systematic analysis of higher - order corrections to the sc theory @xcite . it can describe the validity regime of the sc predictions for the equilibrium separation between strongly coupled cylinders that form a closely packed bound state @xcite , but when the applicability of the sc theory for the interaction force is considered , the criteria ( [ sccriterion ] ) should be used instead . the difference between the new criteria ( [ sccriterion ] ) and the one in eq . ( [ eq : crit_old ] ) is in fact related to the fundamental difference between the topology of the space available to counterions in the case of two cylinders as compared to two planar walls : in the latter case counterions can not escape " from the intersurface gap and the ground state of the system has a universal two - dimensional configuration @xcite , while in the former the ground state corresponds to a one dimensional arrangement of counterions and the whole space is available for thermal excitations from the ground state . the criteria ( [ sccriterion ] ) are more stringent and cover also the situation where the cylinders are placed at large interaxial separations . the ground state of neutralizing counterions between two charged cylinders ( without the effects of the dielectric mismatch ) has been investigated extensively by arnold and holm in ref . @xcite , where a validity criterion for the sc theory has been proposed as @xmath162 ( when expressed in rescaled units ) based on computer simulations . this agrees with our estimate in eq . ( [ sccriterion ] ) within 16% of the numerical prefactor . let us now turn our attention to the behavior of the equilibrium interaxial separation or the so - called bound - state separation , @xmath144 , where the interaction force between the two cylinders vanishes . as noted before , the sc theory is generally more accurate at smaller separations . a closer inspection of fig . [ fig : force0mc ] shows that the sc prediction for the bound - state separation agrees with the mc data even outside the regime set by the sc criteria ( [ sccriterion ] ) , which reiterates the point discussed above that a less stringent criterion such as eq . ( [ eq : crit_old ] ) would be sufficient to describe the validity of the sc prediction for the bound - state separation . in fig . [ fig : diagrams0]b the bound - state interaxial separation is plotted as a function of the manning parameter , which shows that a reasonable agreement between mc data and the sc theory for this quantity can be achieved for a coupling parameter as small as @xmath163 , in agreement with previous results in ref . as seen in the figure , the bound - state separation diverges ( i.e. , the cylinders unbind ) for manning parameter approaching a minimum value of @xmath164 @xcite below which the two cylinders merely repel each other . figure [ fig : diagrams0]b can be also thought of as a phase diagram identifying the attraction ( the region above the lines ) and repulsion ( the region below the lines ) regimes between two like - charged cylinders in terms of the parameters @xmath144 and @xmath22 for a given coupling parameter @xmath10 . in fig . [ fig : diagrams0]c we show a global phase diagram in terms of the parameters @xmath10 and @xmath22 . the boundary lines are determined from mc simulations ( open symbols ) and separate the region ( above the lines ) where the interaction force becomes attractive at some finite interaxial separation between the cylinders and the region ( under the lines ) where the force never becomes attractive at _ any _ interaxial separation . in other words , the boundary lines themselves correspond to the parameter values where the force - distance diagrams exhibit a global minimum at @xmath165 , i.e. the force curve only touches the abscissa . they were determined by bisection procedure for a series of mc runs . if the cylinder parameter values are chosen to describe dna , we have @xmath166 and @xmath167 for @xmath26 valency counterions , which are shown as solid circles in fig . [ fig : diagrams0]c . this suggests that dna - dna interaction in the case of monovalent counterions falls well within the repulsive region , while with divalent counterions and beyond one can expect to observe an attractive force , in qualitative agreement with recent experiments @xcite . and the dielectric discontinuity parameter @xmath36 at interaxial separtaion @xmath140 . the density is shown across an arbitrary plane perpendicular to the two cylinders in a color - coded fashion as indicated on the graph . the small depletion zone around each cylinder is due to the counterion volume exclusion as counterions are assumed to have a finite radius of @xmath141 . a large depletion zone develops due to the influence of image charges as the coupling parameter increases from @xmath168 to 5000 ( a - d ) . , title="fig : " ] ( a ) @xmath168 and the dielectric discontinuity parameter @xmath36 at interaxial separtaion @xmath140 . the density is shown across an arbitrary plane perpendicular to the two cylinders in a color - coded fashion as indicated on the graph . the small depletion zone around each cylinder is due to the counterion volume exclusion as counterions are assumed to have a finite radius of @xmath141 . a large depletion zone develops due to the influence of image charges as the coupling parameter increases from @xmath168 to 5000 ( a - d ) . , title="fig : " ] ( c ) @xmath169 0.25 cm and the dielectric discontinuity parameter @xmath36 at interaxial separtaion @xmath140 . the density is shown across an arbitrary plane perpendicular to the two cylinders in a color - coded fashion as indicated on the graph . the small depletion zone around each cylinder is due to the counterion volume exclusion as counterions are assumed to have a finite radius of @xmath141 . a large depletion zone develops due to the influence of image charges as the coupling parameter increases from @xmath168 to 5000 ( a - d ) . , title="fig : " ] ( b ) @xmath170 and the dielectric discontinuity parameter @xmath36 at interaxial separtaion @xmath140 . the density is shown across an arbitrary plane perpendicular to the two cylinders in a color - coded fashion as indicated on the graph . the small depletion zone around each cylinder is due to the counterion volume exclusion as counterions are assumed to have a finite radius of @xmath141 . a large depletion zone develops due to the influence of image charges as the coupling parameter increases from @xmath168 to 5000 ( a - d ) . , title="fig : " ] ( d ) @xmath171 ( a ) 0.25 cm ( b ) ( c ) we now consider the case of two charged cylinders with a dielectric core whose dielectric constant may be in general different from that of the surrounding medium . in this case , the induced image charges ( due to the polarization of the dielectric cores ) can play an important role as well . however , as we discussed before ( fig . [ fig : forcedelta_wc ] ) , the image - charge effects turn out to be small in the wc regime where the coupling parameter is small ( e.g. , when counterions are monovalent ) . this may be the reason why such dielectric effects have not yet been throughly investigated in the particular case of two charged cylinders . the situation turns out to be quite different in the opposite limit of strong coupling as we will see in this section . this is tied in with a fundamental difference between counterion distribution in the sc limit as opposed to the wc limit : while in the former limit one deals with highly isolated single counterion close to a charged surface , in the latter limit individual nature of counterions fades in the wake of dominant collective many - body effects ( as counterions form a diffuse and uncorrelated ionic cloud around the cylinders ) and thus the polarization effects are highly reduced . if the dielectric cores have a smaller dielectric constant than the medium ( @xmath84 ) , as relevant to most macromolecules in water , the induced image charges will have the same sign as the counterions and thus tend to cause depletion of counterions from the vicinity of the dielectric cores . this behavior is shown in fig . [ fig : densdelta ] , where the sc density is plotted across an arbitrary plane perpendicular to the two cylinders in a color - coded fashion . as can be seen , the depletion effect is enhanced as the coupling parameter , @xmath10 , is increased such that a large depletion zone " develops that eventually encloses both cylinders . ( note that the sc free energy in the presence of a dielectric discontinuity , eq . ( [ free ] ) , has an explicit dependence on the coupling parameter @xmath10 as already known from the case of planar charged dielectric slabs @xcite ) . in fact , the image depletion at larger couplings turns out to have a similar effect qualitatively as the counterion volume exclusion , i.e. , as if the counterion radius @xmath67 is renormalized to a large effective value . note that the region with highest concentration ( shown in red color ) is squeezed " and eventually splits in two regions shifted away from the common @xmath172 plane that passes through the axes of the two cylinders . the splitting appears when the second derivative of the density ( [ dens2 ] ) with respect to the @xmath173 coordinate becomes positive at the midpoint , @xmath174 . using the small - distance approximation @xmath175 from eq . ( [ eqi ] ) , we can estimate the distance @xmath93 at which the density splitting happens as @xmath176 thus for the manning parameter @xmath155 and at separation @xmath140 , the splitting is estimated to occur at @xmath177 as is the case in fig . [ fig : densdelta ] . in figs . [ fig : forceqdelta]a and b , we show the sc force per unit length ( solid lines ) as obtained from eqs . ( [ free ] ) and ( [ free_to_force ] ) along with the mc simulation results ( symbols ) for two different manning parameters and several different values of the coupling parameter . as seen upon increasing the coupling parameter , the repulsive peak at small separations shifts towards larger values of the interaxial separation @xmath93 , which reflects the emergence of a depletion zone due to repulsions between counterions and their image charges as discussed before . as a result , the main contribution to the inter - cylinder interaction at small separations comes from the bare interaction between the two cylinders and the force is approximately given by the bare repulsion force @xmath178 as given by eq . ( [ f00 ] ) in the appendix ( dashed line , fig . [ fig : forceqdelta]b ) . at larger separations , we still find strong sc - type attraction even in the presence of image charges . one should thus note that in general the repulsive forces are enhanced in the effective interaction in the sc limit especially at small separations as compared to a homogeneous system with no image charges . therefore , unlike in the wc limit ( see fig . [ fig : forcedelta_wc ] ) , the introduction of image charges may lead to qualitative changes in the behavior for a strongly coupled system . in general , the sc results for the interaction force show better agreement with the simulations for larger @xmath10 and smaller separation distances @xmath93 . especially the results for the bound - state interaxial separation show very good agreement between the sc results ( solid lines ) and the mc data ( symbols ) as shown in fig . [ fig : forceqdelta]c ( compare with fig . [ fig : diagrams0]b ) . in plan - parallel geometry the repulsive self - image interaction compresses counterions toward the mid - plane and hence expands the range of the sc regime @xcite . here in the cylindrical geometry such mechanism can not work , since counterions can escape from inter - cylindrical region . therefore , it seems that dielectric image effects do not have a noticable influence on the range of validity of the sc approximation . note that just as in the case with @xmath35 ( blue line in fig . [ fig : diagrams0]c ) , the simulation results show no attraction between the cylinders at too low electrostatic coupling parameters . the attraction can only appear once the coupling parameter exceeds a threshold value . this threshold value is shown as a red line in the global phase diagram in fig . [ fig : diagrams0]c for the dielectric discontinuity parameter @xmath179 . note that bare cylinder - cylinder repulsion is enhanced in the case of images ( @xmath92 ) , eq.([f1 ] ) . therefore , larger coupling parameter @xmath10 is needed to get attraction at smaller manning parameters compared to no - image case ( red line stays above blue for @xmath180 and @xmath36 in fig . [ fig : diagrams0]c ) . for larger manning parameters , the attractive interaction between the counterion image and the cylindrical charges ( represented by the term @xmath181 , eq . ( [ vcross ] ) ) becomes more important and , hence , attraction can appear even at lower coupling parameters ( red line is below the blue line for @xmath182 on fig . [ fig : diagrams0]c ) . we have analyzed the electrostatic interaction between two like - charged cylindrical macromolecules surrounded by counterions . we have in particular examined the role of image charges when the cylinders have a dielectric constant which is different from that of the surrounding medium within the the weak- and strong - coupling frameworks elaborated first by netz and coworkers @xcite . the two limits are defined in terms of a single coupling parameter @xmath10 . the weak coupling limit , or equivalently the mean - field theory , is relevant when the valency of counterions and/or the macromolecule surface charge density is small and/or when the dielectric constant of the medium is large enough . it is based on the poisson - boltzmann equation which turns out to give purely repulsive interactions between two - like charged cylinders . the image - charge effects in this case turn out to be small even at small separations and quickly diminish as the inter - cylinder separation is increased . on the contrary , the strong coupling limit is relevant when the valency of counterions is large and/or the macromolecule surface charge density is large and/or when the dielectric constant of the medium is small enough . the sc approach is based on a single - particle description which is obtained as an exact limiting result for large coupling parameters @xmath18 @xcite . while some aspects of the sc theory for cylindrical macromolecules were studied in previous works @xcite , other aspects have remained unadressed . our work is aimed specifically at addressing the effects due to image charges , which indeed turn out to be quite significant in the sc limit . using a generalized sc theory and extensive mc simulations we have shown that the counterions are strongly depleted away from the cylindrical cores due to repulsion from image charges leading to stronger effective repulsions between the two cylinders at smaller separations . the counterion - mediated attraction will be present at intermediate to large separation when the coupling parameter is sufficiently large . these features are in marked contrast with those obtained with the wc theory . we have mapped a global phase diagram where attractive and repulsive interactions emerge between two like - charged cylinders with or without the image - charge effects . here we have employed a first - order image approximation to deal with the image - charge effects within the sc theory ( note , however , that the numerical scheme used within the wc limit allows to account for the full effect of the dielectric discontinuity ) . generally , the treatment of the image charges in nontrivial geometries becomes very complicated and the two - cylinder model is no exception . the first - order image approximation ( [ vim2 ] ) has the obvious advantage that it allows for an analytical treatment of the system . the study of the full effect due to the dielectric discontinuity in the two - cylinder model requires more advanced numerical and analytical developments that might become available in the future . another interesting effect which can be investigated in the present context is that of the additional salt that may be present besides the neutralizing counterions . recent generalizations @xcite should allow for a systematic study of salt screening effects within the wc - sc framework . r.p . would like to acknowledge the financial support by the slovenian research agency under contract nr . p1 - 0055 ( biophysics of polymers , membranes , gels , colloids and cells ) . m.k . would like to acknowledge the financial support by the slovenian research agency under the young researcher grant . is a newton international fellow . in this section , we explain how the cross energy term ( [ vcross ] ) may be evaluated in the case of two charged cylinders ( section [ subsec : sc_2cyln ] ) . this requires employing some mathematical identities as we shall explain below . note that here the position vectors centered at two different cylinders appear simultaneously , @xmath130 , @xmath184 inserting the image kernel @xmath109 , eq . ( [ vim ] ) , and the cylinder charge density @xmath118 , eq . ( [ n0 ] ) , we first transform the coordinates @xmath127 and @xmath185 centered at the @xmath15-th cylinder to the those centered at the @xmath186-th cylinder using the graf s addition theorems @xcite , @xmath187 with @xmath188 and @xmath189 being integers and @xmath190i_n(k\rho_\nu)\hspace{3ex}\\ c_{mn}(k\rho_\nu)&=&[k_{m+n}(kr)+k_{m - n}(kr)]i_n(k\rho_\nu)\nonumber\\ & & \times\ > ( 1-\frac{1}{2}\delta_{n0}).\end{aligned}\ ] ] we can then carry out the integration in ( [ app : vcross_def ] ) over @xmath191 , @xmath192 and @xmath193 . integration over @xmath193 produces a @xmath194 term , where @xmath195 is the upper limit and thus we obtain @xmath196 in the next step we integrate over the wave - vector @xmath197 , where the integrand contains the rapidly oscillating factor @xmath198 , which supresses contributions of the integrand in whole the range , except at @xmath199 . formally , we use the following mathematical identity @xmath200 where @xmath201 is the residue of function @xmath202 at @xmath203 . this integral is finite only if @xmath202 behaves as @xmath204 for @xmath205 and goes to 0 for @xmath206 . this statement can be proven by considering the split of function @xmath202 into a @xmath204 term and a remainder , i.e. , @xmath207 . the first part can be integrated straightforwardly , whearas the remainder yields 0 according to the riemann - lebesgue lemma @xcite . this finally leads us to @xmath208 which may be simplified further using the identity @xmath209\nonumber\end{aligned}\ ] ] and yields the result given in eq . ( [ vcross ] ) . we performed monte - carlo simulations in order to study the system of one and two charged dielectric cylinders at arbitrary coupling parameters , which thus go beyond the weak and strong coupling theories . all simultations were performed in the canonical ensemble ( @xmath210 ) using standard metropolis algorithm @xcite . in the case of one charged cylinder the system ( including the neutralizing counterions ) is enclosed in a cylindrical simulation box of outer radius @xmath211 whereas in the case of two cylinders the system is enclosed in a square box of lateral size @xmath96 ( wc ) and 100 ( sc ) ( note that in this work we focus on the regime of large manning parameters where the lateral size and shape of the confining box becomes unimportant @xcite ) . the simulation box height is assumed to have a finite value @xmath212 , which in terms of other physical parameters by invoking the global electroneutrality condition : in the case of a single cylinder it is given by @xmath213 , and in the case of two cylinders @xmath214 . we use periodic boundary conditions in the @xmath57 direction by replicating the main simulation box infinitely many times in that direction . the energy of the system for a given configuration of the cylinder(s ) is composed of two main parts @xmath215 the first term is the counterion - cylinder interaction energy as given by eqs . ( [ eq_w0c ] ) and ( [ eq : w0c_two ] ) in the text . the second term in ( [ total_en ] ) is counterion - counterion interaction energy which includes also the contribution from interactions with counterion image charges . due to the periodic boundary conditions used in the simulations , this latter term involves infinite summation series @xcite . this is because the counterions and image charges in the main simulation box interact also with their periodic copies " as obtained by the replication of the main simulation box to an infinite number of simulation box copies . these summations can be evaluated as explained in the forthcoming sections . the interaction energy between two given counterions @xmath216 and @xmath217 can be written as @xmath218 the first equation corresponds to the hard - core repulsion between two counterions overlapping counterions . the second equation gives the image self - energy of the @xmath216-th counterion , i.e. , it takes into account the interaction of a given counterion with its dielectrically induced images in the polarizable dielectric core . the third equation in ( [ ionion ] ) gives the interaction between two different counterions @xmath216 and @xmath217 . it is written as the sum of two different contributions which will be calculated below : i ) the coulomb interaction , @xmath219 , between the @xmath216-th ( located at @xmath220 ) and the @xmath217-th counterions and all its periodic copies ( located at @xmath221 for integer @xmath222 ) , @xmath223 and ii ) the contribution involving interactions with the dielectrically induced image charges @xmath224 that will be defined later in this appendix . the first contribution @xmath219 can be written as @xmath225 where we have introduced two different summation schemes depending on the radial separation @xmath226 between the two position vectors @xmath220 and @xmath227 in order to increase the convergence of the sum over the long - range electrostatic interactions . if radial distance @xmath228 is smaller than a threshold @xmath229 , we first sum up @xmath230 terms directly and then use an approximation to estimate the remaining part of the series ( from @xmath231 to infinity ) . we refer to this as the direct " summation scheme . this method is different from the so - called sperb @xcite summation scheme which involves calculating various special functions . for our purposes of relative precision up to @xmath232 we found it more efficient and can be made in principle as accurate as necessary . it follows as @xmath233 where the last sum over @xmath234 is equal to @xmath235 , where @xmath236 is digamma function , and needs to be calculated only once in the course of the simulations . the term @xmath237 is defined as @xmath238 this scheme gives an error of the order @xmath239 . for larger radial separations ( @xmath240 ) , we use the so - called lekner summation scheme @xcite as it converges more rapidly . it is given by @xmath241 where @xmath242 accounts for the offset generated at the threshold , i.e. , @xmath243 we choose the threshold value as @xmath244 and use a cut off of @xmath245 in the direct scheme and sum up to @xmath246 $ ] terms in the lekner scheme , which give a relative error smaller than @xmath232 . let us now consider the second contribution that enters eq . ( [ ionion ] ) , i.e. , @xmath247 , which is the interaction energy obtained by summing up all contributions that involve dielectrically induced image charges throughout different periodic copies of the main simulation box along the @xmath57 axis . it can be written in the case of one cylinder as @xmath248 where @xmath249 is a unit vector pointing in @xmath250 direction . inserting the image kernel ( [ vim ] ) into the above equation , one ends up with an infinite summation of terms including @xmath251 , which can be expressed in terms of @xmath252 functions , i.e. , @xmath253 this enables us to simply integrate over the wave - vector variable @xmath197 and thus obtain @xmath254 where @xmath255 . the @xmath256 term needs to be handled carefully by taking the limit @xmath257 as @xmath258 using the summation identity as used in eq . ( [ app : vcross_2 ] ) , we find the final expression @xmath259+\nonumber\\ & & + \frac{4}{l_z}\sum_{m=0}^\infty\sum_{n=1}^\infty\xi_{m}(k_n a)k_m(k_n\rho_i)k_m(k_n\rho_j)\nonumber\\ & & \times\>\cos\ , k_n\delta z\ , \cos\ , m\delta \varphi . \label{app : imagesum}\end{aligned}\ ] ] this expression is the most computationally expensive part of the simulations , and its efficiency decreases for @xmath260 , i.e. , for counterions near the cylindrical core . in the simulations we truncate these summations in such a way that the relative cut - off error is less than @xmath261 . in the case of two cylinders , we use the first - order image approximation in the simulations as well , and thus the image - charge contributions are simply the sum of the contributions for each cylinder separately , i.e. , @xmath262 where indices 1 and 2 represent coordinates centered at the first and the second cylinder , respectively . in this section , we explain the method used to directly evaluate the force between two charged cylinders within mc simulations . the force acting on a given cylinder ( say cylinder 1 ) is composed of three contributions , @xmath263 where @xmath178 is the bare electrostatic force due to the interaction with the other cylinder , @xmath264 , is the electrostatic force due to the interaction of the cylinder with the @xmath216-th counterion ( summed over all counterions @xmath265 ) , and @xmath266 is the force due to the osmotic pressure from the @xmath216-th counterion . the first contribution follows from the interaction energy @xmath103 , eq . ( [ w00_delta ] ) , as @xmath267 which gives @xmath268 the second contribution follows by differentiating @xmath104 , eq . ( [ eq : w0c_two ] ) , with respect to @xmath93 by assuming that the cylinder 1 and the @xmath269-th counterion are fixed , i.e. , @xmath270 which after some algebra gives @xmath271.\qquad\ ] ] here @xmath272 and @xmath273 are the radial distance and the azimuthal angle of the @xmath216-th counterion position with respect to cylinder 1 , and @xmath274 and @xmath275 are the corresponding coordinates with respect to the shifted axis by @xmath132 from the axis of cylinder 1 toward the axis of the cylinder 2 , fig . [ fig : model_2cyln ] . similar definitions apply to @xmath276 , @xmath277 , @xmath278 and @xmath279 . the osmotic ( third ) term in eq . ( [ force ] ) results from the collisions of counterions with the cylinder . this contribution amounts to a pressure of counterions exerted on the cylinder surface , which is proportional to the contact density of counterions at the cylinder surface @xmath280 . thus the infinitesimal force acting on the cylinder is given by @xmath281 . note that the contact counterion density @xmath282 is given by the number of counterions inside a small box of dimensions @xmath283 next to the cylindrical surface . the total osmotic force of all ions in simulation cell is then @xmath284 where @xmath285 is the heaviside step function which is @xmath286 only if the @xmath216-th counterion falls inside a shell thickness @xmath287 from the cylinder surface . the above procedure in principle gives the exact value of the osmotic force when @xmath288 . it also gives a convenient method to calculate the osmotic contribution within mc simulations by evaluating the force for different values of the shell thickness and estimating the limiting value by extrapolation . r. r. netz , eur . j. e * 5 * , 557 ( 2001 ) ; a. g. moreira and r. r. netz , _ ibid _ * 8 * , 33 ( 2002 ) ; phys . lett * 87 * , 078301 ( 2001 ) ; europhys . lett . * 52 * , 705 ( 2000 ) ; _ ibid _ * 57 * , 911 ( 2002 ) . r. h. french , v. a. parsegian , r. podgornik _ et al . ( 2010)in press . l. guldbrand , b. jnsson , h. wennerstrm , and p. linse , j. chem . phys . * 80 * , 2221 ( 1984 ) ; d. bratko , b. jnsson , and h. wennerstrm , chem . lett . * 128 * , 449 ( 1986 ) ; j. p. valleau , r. ivkov , and g. m. torrie , j. chem . phys . * 95 * , 520 ( 1991 ) ; r. kjellander , t. kesson , b. jnsson , and s. marelja , j. chem . phys . * 97 * , 1424 ( 1992 ) . a. naji and r. podgornik , phys . e , 2005 , * 72 * 041402 ; r. podgornik and a. naji , europhys . letts . , 2006 , * 74 * , 712 ; y. s. mamasakhlisov , a. naji , and r. podgornik , j. stat . phys . * 133 * , 659 ( 2008 ) ; a. naji _ et al . _ , phys . rev . * 104 * , 060601 ( 2010 ) . stevens , m.o . robbins , europhys . lett . * 12 * , 81 ( 1990 ) . a. diehl , m.n . tamashiro , m.c . barbosa , y. levin , physica a * 274 * , 433 ( 1999 ) ; m.c . barbosa , m. deserno , c. holm , europhys . * 52 * , 80 ( 2000 ) . kornyshev , s. leikin , j. chem . phys . * 107 * , 3656 ( 1997 ) ; phys . lett . * 82 * , 4138 ( 1999 ) . arenzon , j.f . stilck , y. levin , eur . j. b * 12 * , 79 ( 1999 ) ; y. levin , j.j . arenzon , j.f . stilck , phys . . lett . * 83 * , 2680 ( 1999 ) ; a. diehl , h.a . carmona , y. levin , phys . e * 64 * , 011804 ( 2001 ) . henle , c.d . santangelo , d.m . patel , p.a . pincus , europhys . lett . * 66 * , 284 ( 2004 ) . m. manghi and r. r. netz , eur . j. e * 14 * , 67 ( 2004 ) s. liu , m. muthukumar , j. chem . phys . * 116 * , 9975 ( 2002 ) ; s. liu , k. ghosh , m. muthukumar , j. chem . phys . * 119 * , 1813 ( 2003 ) . m. muthukumar , j. chem . phys . * 120 * , 9343 ( 2004 ) . d. chapot , l. bocquet , e.trizac , j. chem . phys . * 120 * , 3969 ( 2004 ) a. g. cherstvy , j. phys . chem . b * 111 * , 12933 ( 2007 ) ; j. phys . : condens . matter * 17 * , 1363 ( 2005 ) . we make use of the commercial software comsol multiphysics which employes a finite element mesh approach in order to solve the nonlinear pb equation in the two - cylinder geometry ( http://www.comsol.com ) .
giant molecular clouds , both galactic and extragalactic , are observed to have velocity gradients that many authors have interpreted as being caused by rotation ( e.g. , kutner et al . 1977 , phillips 1999 ; rosolowsky et al . 2003 ) . if we start with the premise that these clouds are rotating because they have inherited the angular momentum of the rotating galactic disk out of which they formed , conservation of angular momentum should provide clues that give us insight to the origin of their formation . however , simple formation theories that assume giant molecular clouds ( gmcs ) form by condensing out of the galactic disk are at odds with some of the observations . for instance , they do not adequately explain why the directions of gmc velocity gradients are not typically aligned with the direction of galactic rotation , counter to the expectation of conservation of angular momentum . furthermore , simple formation scenarios tend to overpredict the observed specific angular momentum o f gmcs ( e.g. , blitz 1990 ; rosolowsky et al . this is the so - called `` angular momentum problem . '' provided there is no transfer of angular momentum , the angular momentum of a gmc should be equal to that of the gas out of which it formed . but blitz ( 1990 ) , working under the assumption that the velocity gradients in molecular clouds are due to rotation , showed that the angular momentum due to galactic differential rotation in the solar neighborhood interstellar medium is consistently greater than that contained within molecular clouds . even molecular clouds with the largest observed velocity gradients such as the rosette and orion a molecular clouds have less specific angular momentum compared to the ism from which they presumably formed . blitz ( 1990 ) also pointed out that because the molecular clouds in his sample are rotating in a sense _ opposite _ to that of galactic rotation , they could not have conserved angular momentum from the initial states he calculated , unless the local galactic rotation curve is falling , or unless the clouds always collapsed azimuthall y. the angular momentum problem also extends to extragalactic molecular clouds . roso - lowsky et al . ( 2003 ) showed that simple gmc formation theories consistently overestimate the magnitude of the observed angular momentum of molecular clouds in the galaxy m33 . on average , they found that simple theory overpredicts the observed magnitudes of specific angular momentum by more than a factor of 5 . furthermore , they found that 40% of the gmcs in m33 are counterrotating with respect to the galactic plane . in this paper , we shed light on the angular momentum problem by doing a detailed analysis of the kinematics in gmcs _ and _ the surrounding ism . whereas previous studies estimated the initial angular momentum imparted to gmcs from the galactic rotation curve ( e.g. , blitz 1990 ; blitz 1993 ) , we compare the velocity fields of gmcs to those of the ism with which they are associated directly from observation . in light of the observation that gmcs have a spatial and kinematic correlation with high - surface density atomic gas ( [ sec : analysis ] ) , we pose the question : does the rotation of the large - scale @xmath0 associated with gmcs mirror that of the gmcs themselves ? our primary goal is to determine whether or not rotation is the cause of the velocity gradients in gmcs . to that end , we create first - moment maps of the molecular clouds in our sample and of the atomic gas surrounding them are for comparison . in the following section , we describe the and 21-cm data used to conduct this study . in 3 , we create intensity - weighted first moment and @xmath0 maps of five galactic clouds : perseus , orion a , ngc 2264 , monoceros r2 ( monr2 ) , and the rosette . the results from these measurements are given in 4 , and a summary is provided in 5 . to measure velocity gradients and other properties across the molecular clouds , we use high - resolution , high - sensitivity published observations . because emission is nearly always optically thin in galactic gmcs , we have the advantage of getting a detailed view of the kinematic structure of the molecular clouds in our sample . and because has narrower line widths than the optically thick , the former permits finer separation of velocity components than the latter . nevertheless , the large scale velocity gradient of a gmc measured using is generally consistent with that measured using , since the gradient is being measured across several parsecs and small scale variations in the velocity field tend to get averaged out . the data for perseus , orion a , ngc 2264 , and the monr2 molecular clouds were generously provided by j. bally ( see bally et al . observations of these clouds were taken at the at&t bell laboratories 7 m telescope and have a beam size of 100 . perseus , orion a , and ngc 2264 data were resampled onto @xmath2 grids . the monr2 data were resampled onto a @xmath3 grid . the spectral resolution of 128 channels at 100 khz corresponds to a velocity resolution of 0.27 . on the @xmath4 scale , the cubes have rms noise levels of 0.17 k ( perseus ) , 0.32 k ( orion a ) , 0.42 ( ngc 2264 ) and 0.29 k ( monr2 ) . j. williams and m. heyer graciously provided the fcrao data of the rosette molecular cloud ( see heyer , williams , & brunt 2006 ) . the beam size is @xmath5 and the data were interpolated onto a @xmath6 grid . the spectral resolution is 59 khz per channel , and the velocity resolution is 0.133 . the data have an rms noise of 0.21 k in @xmath4 , similar to that of the bell labs data . the @xmath0 data are obtained from the leiden / argentine / bonn ( lab ) galactic @xmath0 survey ( kalberla et al . 2005 ) , which spans velocities from @xmath7 to @xmath8 . ( the lsr velocities of the molecular clouds in our sample range from 6 to 14 with velocity dispersions of a few . ) the survey has a half power beam width of @xmath9 , velocity resolution of 1.3 , and an rms noise level of 0.07 k. the high sensitivity and resolution of the lab data set enables a detailed study of the atomic gas from which the gmcs formed . observations in the milky way indicate the molecular clouds are typically associated with high - density atomic gas with column densities around @xmath10 ( e.g. , mckee & ostriker 2007 ) . the rosette molecular cloud is a prototypical example : williams et al . ( 1995 ) measured the mean column of the @xmath0 associated with the gmc to be @xmath11 . furthermore , in external galaxies , gmcs are often observed to be located on or near bright @xmath0 peaks . in the large magellanic cloud , mizuno et al . ( 2001 ) observed that most gmcs are associated with @xmath0 having column densities greater than @xmath12 . rosolowsky et al . ( 2003 ) found that every gmc they identified in m33 lies on an over - dense @xmath0 filament , though every over - dense @xmath0 region does not contain a gmc . this seems to imply that high column density atomic gas is necessary but not sufficient for gmc formation . it is our goal to do a detailed comparison of the velocity fields in the gmcs and local atomic gas . to that end , we first describe how the physical properties including column densities , masses , and velocity gradients in both the atomic and molecular gas are determined . we then provide our criteria for choosing spatial and kinematic regions of atomic gas associated with the molecular clouds . lastly , in this section , we discuss how we estimate the specific angular momentum . these parameters are summarized in tables 1 and 2 . the @xmath0 column density , @xmath13 , is calculated along each line of sight by integrating the atomic hydrogen emission above a certain background value and over the selected velocity range , ( see 3.1 and table 1 ) , @xmath14 where @xmath15 is the brightness temperature of the @xmath0 observations and @xmath16 is the channel velocity width . assuming that the @xmath0 is optically thin , equation [ eq : nhi ] provides a lower limit to measured column density , which we convert into units of surface density in figures [ fig : oriona0 ] [ fig : rosette0 ] . the total @xmath0 mass is then determined by summing over all pixels in the map where emission is detected , that is , where the emission is at least three times the root - mean - square ( rms ) noise level : @xmath17 where @xmath18 is the correction for helium , @xmath19 is the distance to a given molecular cloud , and @xmath20 is the area of one pixel , which corresponds to one resolution element . the column density , @xmath21 , is derived assuming that the emission is optically thin and in local thermodynamic equilibrium . following frerking et al . ( 1982 ) , @xmath22^{-1 } \nonumber \\ \times \int_{v_{\mathrm{min}}}^{v_{\mathrm{max}}}{\frac { t_{{\mathrm{b , co}}}~{\ensuremath{\mbox{d}{v}}}}{{\mbox{k km s$^{-1}$ } } } } ~{\mbox{cm$^{-2}$ } } , \end{aligned}\ ] ] where @xmath23 is the integrated intensity and @xmath24 is the excitation temperature . normally , @xmath24 is determined by measuring the radiation temperature toward peaks . since we lack observations at the same resolution as the data , we use a constant excitation temperature in the calculation of @xmath21 for each of the five gmcs . based on the following arguments , we adopt a value of 20 k for each of the gmcs . if the actual excitation temperature in a given region is between 10 and 30 k , the derived column density will be in error by less than a factor of 2 . castets et al . ( 1990 ) showed that the emission in orion a mainly arises from regions where @xmath25 k , and dense cloud cores have temperatures of @xmath26 k. nagahama et al . ( 1998 ) showed that , with the exception of two peaks at @xmath27 and @xmath28 associated with embedded young stellar groups , @xmath24 rises slowly and monotonically in galactic longitude , with an average ranging from @xmath29 k. in his study of molecular clouds , including perseus , orion a , and monr2 , carpenter ( 2000 ) adopts a constant value of @xmath30 k , though the coefficient in his formula for @xmath21 yields slightly higher values than ours in equation [ eq : nco13 ] . thus , for orion a , perseus , ngc 2264 , and monoceros r2 , we use @xmath31 k in our calculation of @xmath21 . williams et al . ( 1995 ) showed that @xmath24 in the rosette decreases slowly with increasing distance from the rosette nebula ( centered at @xmath32 ) from @xmath33 to @xmath34 k. in our analysis , we adopt a uniform value of @xmath31 k for the rosette . our estimate of the cloud s mass using this value is slightly lower than that estimated by williams et al . ( 1995 ) , who measured the mass over a larger surface area ( see below ) . next , the @xmath35 column density is evaluated assuming a ratio of @xmath36 ( frerking et al . pixels having values at least three times the rms noise level are counted as detected . finally , the molecular mass @xmath37 is calculated over the areas where emission is detected ( that is , higher than the 3-@xmath38 level ) using @xmath39 where @xmath40 is the mass of an @xmath35 molecule . we note that the areas over which emission is detected and , subsequently , the masses we calculate will be smaller than cited in previous studies in which these quantities were measured using emission . this is because the stronger line is observed over larger areas in gmcs than the line . in the rosette , for instance , we measure a projected area of 1500 @xmath41 and a mass of @xmath42 @xmath43 , while williams et al . ( 1995 ) measure 2200 @xmath41 and @xmath44 @xmath43 . in table 1 , both the masses and previously measured masses are listed for the gmcs . velocity gradients are measured from first moment maps of the atomic and molecular gas . first , the intensity - weighted average velocity along each line of sight is determined using @xmath45 where @xmath46 and @xmath47 are the velocity and temperature at location @xmath48 . following goodman et al . ( 1993 ) , the uncertainty of a given measurement is @xmath49 where @xmath50 is the spectrum noise , @xmath51 is the maximum temperature along the line of sight , and @xmath52 is the fwhm linewidth of the spectrum along the line of sight . a plane is then fitted to the first moment map of velocity centroids , as in goodman et al . ( 1993 ) , assuming a linear velocity gradient : @xmath53 where @xmath54 is the mean cloud velocity , and @xmath55 is an arbitrary reference position , which we take to be the center of our maps , and the coefficients are @xmath56 the gradient magnitude and direction , @xmath57 and @xmath58 , are derived from the coefficients to the fit , @xmath59 @xmath60 where @xmath58 , measured in degrees east from north , points in the direction of increasing velocity . note that since we have no information regarding the inclination of a given cloud , @xmath48 , to our line of sight , gradient measurements are underestimates of the actual values , @xmath61 . the uncertainties in these values are calculated by propagating the errors in the coefficients . to check whether planes are good fits to the velocity centroid maps , we make plots of the central velocity at a given location in the cloud versus the perpendicular offset from the cloud s rotation axis . this is done by taking the average velocity along lines parallel to the rotation axis at various distances . in most cases , as will be discussed in [ sec : results ] , these plots show that planes are good fits to the velocity centroid maps . we select @xmath0 regions in the position - velocity lab data cube that are centered , spatially and kinematically , on the five gmcs in our sample . we do not know _ a priori _ the kinematic or spatial extent of the @xmath0 regions that are associated with each gmc nor the extent of the @xmath0 velocity gradients . molecular clouds have well - defined boundaries at which the molecules are dissociated by uv radiation and where there is a distinct transition from primarily molecular to primarily atomic gas ( e.g. savage et al . 1977 ; blitz & thaddeus 1980 ) . the atomic gas associated with gmcs does not have such distinct boundaries , however , making it difficult to distinguish @xmath0 that may be related to gmcs from background emission . thus , we begin by examining regions in position - velocity space that are far from the center of the gmc . while we want to capture the full extent of any linear velocity gradient we may measure in a given region of atomic gas , we do not want to make our aperture so large that we end up including in our measurements too much atomic gas that is unrelated to the molecular clouds . we start by varying the size of the region ( that is , the subcube extracted from the lab data ) and examine how the velocity gradient magnitude and direction change . we fix the velocity range ( see below ) and vary the spatial size of the region centered on the gmc from about 10 to several tens of parsecs , in increments of 10 pc . we find that the gradient direction remains roughly constant until the radius of the region over which it is measured reaches @xmath62 pc , independent of the size of the gmc . beyond this , the measurements start to fluctuate , as the gradient in the vicinity of the molecular cloud becomes washed out by unrelated features . thus for each cloud , we end by selecting a spatial boundary of atomic gas which extends roughly 30 50 pc from the center of the co emission . because the peaks of the @xmath0 regions are included in the maps we generate , varying the size of the region within @xmath62 pc does not change the gradient direction by more than a few degrees . the 1-@xmath63 uncertainty level of @xmath58 ranges from 3 to 6 degrees for the clouds in our sample . our criterion is supported by previous studies such as that of andersson et al . ( 1991 ) who found that the spatial extent of high - intensity @xmath0 halos , measured from the edge of the molecular cloud , ranges 5 10 pc ( see also [ sec : discussion ] ) . figures [ fig : oriona0 ] through [ fig : rosette0 ] show the @xmath0 surface density maps derived from the zeroth - moment intensity maps , with the outline of the molecular clouds overlaid . also overplotted in each figure is a dashed circle indicating the @xmath0 region selected for the analysis . keep in mind , we are showing the emission of the molecular clouds , and so the maps in figures [ fig : oriona0 ] [ fig : rosette0 ] do not show the full extent of the molecular emission in the gmcs . to choose relevant velocities of the atomic gas associated each gmc , we begin by examining the @xmath0 emission in the velocity range @xmath64 centered about the mean lsr velocity of the emission . again , this is because we want to be sure that our measurements include as much as possible of the associated atomic gas . in the direction of the rosette , for instance , the @xmath0 has been observed to extend several beyond the co emission ( williams et al . studies of both milky way molecular clouds ( e.g. , wannier et al . 1983 , williams et al . 1995 ) and extragalactic clouds ( e.g. engargiola et al . 2003 ) have shown that the @xmath0 emission line profile tends to peak in the direction of gmcs . in effect , we are using the @xmath0 velocity as a proxy for distance in order to associate the atomic gas with the gmcs . as figures [ fig : oriona0 ] [ fig : rosette0 ] show , although the @xmath0 emission line is broader than the co line , the velocity difference between the peaks in the respective lines never exceeds @xmath65 , where @xmath65 is the velocity dispersion of the @xmath0 profile . the bottom panel in figures [ fig : oriona0 ] through [ fig : rosette0 ] show the average spectrum through each gmc with the @xmath0 spectrum in the same direction ( within the dashed circle ) overplotted . because the @xmath0 peaks are nearly coincident in each case , this indicates that most of the @xmath0 in the direction of a given cloud is associated with that cloud within the selected velocity range . in the cases of perseus , orion a , and monr2 , the @xmath0 emission drops abruptly beyond @xmath66 of the emission . ngc 2264 and the rosette have more complicated @xmath0 spectra , each showing double peaks that may be indicative expansion or of a blended , possibly unrelated component . the latter explanation would not be surprising since , of the five clouds in the sample , ngc 2246 and the rosette are located closest to the galactic plane where line - of - sight blending is more of a problem . based on figures [ fig : oriona0 ] [ fig : rosette0 ] , we determine the boundaries of the @xmath0 emission we will use for the subsequent analysis . for perseus , orion a , and monr2 , the @xmath0 line profiles are approximated as gaussians and we assume that @xmath0 emission having velocities within @xmath67 is associated with a given molecular cloud . since both ngc 2264 and the rosette each have a second peak in their @xmath0 temperature profiles ( @xmath68 ) at higher velocities and because we want to be careful to exclude as much unrelated emission as possible , we set a slightly more stringent criterion on the velocities we select . for each cloud , a maximum velocity is identified at the local minimum in @xmath15 where the gaussians overlap . in the ngc 2264 spectrum , for instance , @xmath15 drops to 29 k at 15 and then peaks again at around 20 ( see figure [ fig : ngc22640 ] ) . thus , we eliminate all emission having velocities above 15 , the locat ion of the local minimum . all of these selections are listed in the third to last column of table 1 . once the magnitudes of the velocity gradients in the molecular clouds and the surrounding @xmath0 are measured , we may calculate and compare their specific angular momenta , under the assumption that the linear gradients are due to solid body rotation . the specific angular momentum , @xmath69 , is simply the total angular momentum of a body divided by its mass , @xmath70 where @xmath71 is the radius of the region , and the constant @xmath72 takes into account the moment of inertia of a rotating body . for roughly spherical gmcs having constant surface mass density distributions , @xmath73 . unless otherwise stated , we take the size of a given molecular cloud to be its effective radius , as defined by its projected area : @xmath74 . we would like to estimate the expected specific angular momentum initially imparted to a gmc by the ism from which it forms . this depends on the process of gmc formation , and we assume here that gmcs form via a `` top - down '' formation mechanism . for instance , it has been suggested that molecular cloud formation occurs when an instability triggers collapse or condensation from the galactic disk ( e.g. , mouschovias et al . 1974 ; blitz & shu 1980 ; elmegreen 1982 ; kim et al . blitz & shu ( 1980 ) show that `` bottom - up '' formation of gmcs via the random agglomeration of pre - existing low - mass clouds is unlikely because of the long timescales for this process . gravitational instabilities and magneto - gravitational instabilities , however , tend to proceed more quickly . if the @xmath0 surrounding the gmcs is reflective of the ism out of which the gmcs initially formed , the quantity @xmath75 provides an estimate of the initial specific angular momentum of the gmcs , assuming they are rotating . henceforth , we will often refer to this as the _ expected _ specific angular momentum . note that this disregards the possible effects of magnetic fields ; that is , we are assuming that the magnetic field strength of the forming cloud is 0 . the initial angular velocity imparted to a given gmc is the local value of @xmath57 , which we calculate from the first - moment maps of the @xmath0 using equation [ eq : omega ] . in principle , the size of the region from which a forming cloud gathers material , the `` accumulation radius '' @xmath76 , could have a range of values because it depends on details of the formation mechanism , as well as on assumptions regarding the initial surface density and geometry of the gas from which a gmc formed . nevertheless , we can set an effective lower limit on @xmath76 by calculating the size of the region from which a gmc must contract to obtain its present mass , @xmath77 . following blitz ( 1993 ) , let us assume that the initial geometry of the collapsing region is a cylinder with a diameter equal to its height . the size of the cylinder is determined by requiring that the mass contained within it is equal to the present mass of the gmc : @xmath78 where @xmath79 is the mean volume density of the atomic gas from which the gmc formed . blitz ( 1993 ) estimated @xmath76 using the mean value of @xmath80 in the galactic plane near the sun , 5 ( henderson et al . 1982 ) . this corresponds to a mass density of @xmath81 ( or a number density of 0.5 @xmath82 ) , using an effective scale height of atomic gas of 200 pc in the solar vicinity ( falgarone & lequeux 1973 ) . however , using larger values of @xmath80 ( leading to lower estimates of @xmath76 ) might be more appropriate , given the observation that gmcs tend to form in regions of @xmath0 with densities higher than global galactic values ( e.g. , engargiola et al . 2003 ; imara et al . we estimate the accumulation radii using @xmath83 , a value more in keeping observations of the atomic gas associated with gmcs . for instance , around the rosette , williams et al . ( 1995 ) measure an @xmath0 column density of @xmath84 , which corresponds to a @xmath85 , twice the mean galactic value . from the sancisi ( 1974 ) study of atomic gas near perseus , the inferred surface density is 11 . our estimates of @xmath76 are listed in table 2 , as well as the effective radii of the gmcs . finally , we note that the predicted specific angular momentum in the atomic gas has a marked dependence on the gmc mass . substituting @xmath86 from equation 2.12 for the radius in equation 2.11 yields @xmath87 . the masses we calculate using the observations underestimate the total molecular mass of the gmcs . for this reason , we use the larger gmc masses , as previously measured using observations , to predict the initial specific angular momentum imparted to forming molecular clouds . typically , @xmath88 . figures [ fig : oriona ] through [ fig : rosette ] show the @xmath0 velocity maps ( grayscale ) overlaid with velocity maps of ( color ) for each gmc . overplotted on these maps are axes of rotation : the lines perpendicular to the gradient directions , @xmath89 and @xmath90 , as calculated from equation [ eq : theta ] . these lines are the position angles of the rotation axes , @xmath91 , ( where @xmath92 ) , of the molecular and atomic gas _ if _ the gradients are in fact due to rotation . in order to check whether planes are good fits to the first - moment velocity maps , we make position - velocity cuts parallel to the maximum gradient directions and plot the results ( bottom panels of figures [ fig : oriona ] [ fig : rosette ] ) . we are essentially plotting the central velocity at a given location in the cloud versus displacement along the gradient on a pixel by pixel basis . figures [ fig : oriona ] , [ fig : monr2 ] , and [ fig : rosette ] show that planes are good fits to the first - moment maps of @xmath0 surrounding orion a , monr2 , and the rosette because there is a clear linear trend in the gradient in the atomic hydrogen . the velocity fields of the @xmath0 associated with perseus and ngc 2264 appear to have more complex structure . in the case of the ngc 2264 and monr2 molecular clouds ( figures [ fig : ngc2264 ] and [ fig : monr2 ] ) , however , the position - velocity plots do not have monotonically increasing or decreasing slopes , indicating a more complex velocity structure in the _ molecular _ gas ( red points ) . in these two cases , we nevertheless overplot the lines perpendicular to the gradient direction calculated from equation [ eq : theta ] . the specific angular momenta in the atomic and molecular gas , listed in table 3 , are compared in figure [ fig : jplot ] . there appears to be a reasonable correlation@xmath93 and @xmath94 increasing together as @xmath95although small number statistics prevent us from making a firm conclusion . in each case , the initially expected specific angular momentum , @xmath93 , is always greater than @xmath94 . these measurements alone are consistent with a picture whereby gmcs form via some top - down mechanism , such as a gravitational instability , and somehow shed angular momentum in the process . but since angular momentum is a vector quantity , this scenario is difficult to reconcile with the observation that the gradient position angles in the molecular and atomic gas differ and appear to be uncorrelated ( figure [ fig : theta ] ) . the key finding of our analysis is that the regions of atomic gas associated with molecular clouds have linear velocity gradients , yet the directions of these gradients are with one exception unaligned with the direction of the gradients in the associated gmcs . under the hypothesis that the gradients are caused by solid body rotation , this would imply that gmcs are _ not _ corotating with the surrounding ism . the second key result is that the magnitudes of the velocity gradients in the gmcs are larger than the gradient magnitudes in the atomic gas . thirdly , if the gradients in the molecular and atomic gas are from rotation , the angular momenta in the molecular clouds is less than predicted from calculations of the angular momenta in the associated atomic gas . below , we describe the results in detail for each gmc . the gradient directions , magnitudes , and specific angular momenta of both the molecular and atomic gas are given in table 3 . * * figure [ fig : oriona ] shows that the orion a molecular cloud has a large - scale gradient whose velocity decreases from about 12 to 3 with increasing galactic longitude . of the gmcs in this study , observations of the kinematics and morphology of the orion a molecular cloud seem to make the best case for a top - down picture of gmc formation . the direction of the gradient and the long axis of orion a are parallel to each other and to the galactic plane . if the gradient were due to rotation , this would imply that orion a is rotating in a sense nearly opposite to that of the galactic disk . this result is in agreement with previous studies ( e.g. , kutner et al . 1977 , blitz 1993 ) . the velocity gradient of the @xmath0 in the immediate surroundings of orion a points in nearly the same direction as that in the molecular cloud . figure [ fig : oriona ] shows that the gradient in the atomic hydrogen , integrated over the velocity range from about @xmath96 to 22 , differs from th e gradient in the molecular cloud by only @xmath97 . the origin of the velocity gradient in orion a has been debated ; it has previously been explained by cloud rotation ( kutner et al . 1977 ) , expansion driven by the orion ob association ( bally et al . 1987 ) , and expansion driven by stellar winds from newborn stars ( heyer et al . if due to rotation , the 0.22 gradient in the molecular cloud implies a specific angular momentum of 42.2 . however , considering that orion a is filamentary and much more closely resembles a cylinder rotating about its minor axis than a sphere , this estimate of @xmath69 is likely to be low limit , since the configuration of the former has a higher moment of inertia than the latter . for a cylinder rotating about its minor axis , @xmath98 , where @xmath71 is the radius of the cylinder and @xmath99 is the length . the cylindrical radius and length of orion a are approximately @xmath100 and @xmath97 , corresponding to 11 pc and 65 pc . consequently , @xmath101 , much c loser to , but still a bit less than the estimated angular momentum of the associated atomic gas ( 107 ) . these two results that the gradients of the molecular and local atomic gas are in near alignment , and the magnitudes of the specific angular momenta are within range of each other suggest that orion a may be a case in which the molecular cloud and surrounding @xmath0 are corotating and in which conservation of angular momentum is demonstrated . as we will see below , it appears to be an exceptional case . * * figure [ fig : perseus ] shows that a strong linear gradient exists in the perseus molecular cloud . the velocities in the gmc range from about 1 to 11 with increasing galactic latitude . the gradient direction is tilted about @xmath102 to the north - west . of the molecular clouds in the sample , perseus has the largest velocity gradient at 0.23 . if the gradient is caused by rotation , from equation [ eq : j ] , the specific angular momentum in perseus is 24.2 , as indicated in table 3 . examination of the black points in the position velocity - plot shown in figure [ fig : perseus ] indicates that the velocity gradient of the @xmath0 region centered on perseus is not linear . as with each cloud in this study , we take the center of the @xmath0 field to be located at the position centroid of the gmc in galactic coordinates , ( @xmath103 ) . this affects the appearance of the position - velocity plot , which shows the velocity of points in the field as a function of distance from the gradient judged from @xmath103 . the @xmath0 surface density map in figure [ fig : perseus0 ] shows a high - surface density ( @xmath104 @xmath43 ) @xmath0 filament extending vertically near a galactic longitude of @xmath105 . whereas the other gmcs in this sample have some portion of their molecular material laying directly on top of an @xmath0 peak , this is not strictly the case with perseus , which is tilted toward the east from the bright @xmath0 filament . thus , we generate another first - moment map of the @xmath0 ce ntered on the filament at @xmath106 , as well as another position - velocity plot centered here . figure [ fig : perseus2 ] shows that a linear fit is a much more suitable fit to the gradient in the @xmath0 when we shift the reference point . when measured this way , the magnitude of the gradient measured using equation [ eq : omega ] shifts from 0.038 to 0.067 , the value we cite in table 3 . the gradient direction is nearly @xmath107 , very close to the sense of galactic rotation , in which velocities increase from west to east . but the directions of the gradients in the molecular cloud and the atomic gas surrounding the gmc differ by nearly @xmath108 . of the clouds in this analysis , perseus has the largest velocity gradients in the molecular gas . the magnitude of the gradient in the molecular gas ( 0.23 ) is 3.4 times that in the @xmath0 ( @xmath109 ) . the estimated specific angular momentum in the atomic gas is @xmath110 , 2.5 times @xmath69 measured in the molecular cloud . although this is consistent with a scenario in which the perseus molecular cloud formed by collapsing out of the atomic gas , its angular momentum being redistributed somehow in the process , based on the observation that there is such a large difference between the gradient directions in the molecular and atomic gas , it is difficult to see how the gmc could have formed in such a simple way . * * the first - moment maps and position - velocity plot in figure [ fig : ngc2264 ] show that the ngc 2264 molecular cloud and surrounding @xmath0 both have complex kinematic features . though there is no significant linear gradient across the entire gmc , there is a weak gradient in the atomic gas . the gradient in the atomic gas is stronger to the south of the rotation axis , where the velocity decreases from about 8 to 5 as the galactic latitude increases from @xmath111 to @xmath112 . as seen in figure [ fig : ngc22640 ] , the ngc 2264 molecular cloud is composed of two main structures . the structure in the north has a larger range of velocities , with @xmath113 10 , than does the structure in the south , which averages around 6 , close to the mean velocity of the molecular cloud as a whole . this is also the average velocity of the @xmath0 peak located at near @xmath114 , @xmath115 ( figure [ fig : ngc22640 ] ) , which is associated with the southern segment of t he gmc . the magnitude of the gradient in the atomic gas near ngc 2264 is 0.019 , which is less than @xmath57(@xmath116 ) in the solar vicinity assuming a flat rotation curve . the magnitude of the gradient fitted across the entire gmc , using equation [ eq : omega ] , is 0.046 . but the northern segment of the cloud in isolation has a gradient closer to 0.08 . it seems unlikely that rotation is the origin of the velocity field in the ngc 2264 molecular cloud . table 3 records the velocity magnitude and direction calculated by fitting a plane to the field , even though the gradient is not linear . using these values , the specific angular momentum of the gmc , @xmath117 , is smaller than the expected value by at least a factor of 6 . the gradient direction in the gmc , pointing nearly @xmath118 east from north , differs from @xmath90 by @xmath119 . an alternative explanation of the velocity field in ngc 2264 may be , at least in part , the internal stellar activity . it is noteworthy that the @xmath120 region associated with ngc 2264 ( the cone nebula ) is located in the north , near the part of the gmc that has a larger velocity dispersion than the southern part . the nebula , at @xmath121 , @xmath122 ( kharchenko et al . 2005 ) is located between two high velocity regions of the gmc that are receding at speeds near 10 . this morpholog y is suggestive of an expanding ring seen from an edge - on perspective . taken all together , it appears that these features could be causally connected : an @xmath120 region causes the high - speed expansion of the surrounding molecular gas , which subsequently sweeps up atomic gas into a high - density ridge . it is perhaps significant that we see a similar pattern in the rosette . yet another explanation of the kinematics of ngc 2264 is put forth by fursz et al . ( 2006 ) , based on their finding that the stars and emission in ngc 2264 are well - correlated in position - velocity space . they suggest that the velocity field of the gmc is explained by the models of burkert & hartmann ( 2004 ) , in which molecular clouds form from supersonic collisions of gas . this argument seems to be corroborated by the overall pattern of the cloud s velocity structure : higher velocities at the outskirts of the cloud and lower velocities proceeding toward the center of the cloud . this pattern could be explained by gravitationally driven infall at the interface of the colliding flows . in this scenario , star formation occurs preferentially in the condensations that develop during the collapse of the molecular cloud , thus explaining the position - velocity correlation observed between stars and high - density molecular gas . the colliding flows hypothesis may also explain the velocity structure of the atomic gas . figure [ fig : ngc2264 ] shows that the @xmath0 has slightly higher velocities at positions in the field far from the gradient axis and lower velocities close to the axis again , suggestive of infall accelerated by gravity . if other molecular clouds form by this mechanism , however , it is unclear why we do not see this pattern in the atomic gas surrounding the rest of the gmcs in our sample . possibly , the ngc 2264 system still retains signs of its early formation history due to the relatively young age of the gmc ( @xmath123 3 myr ; flaccomio et al . 2000 ; ramrez et al . it is difficult to make a definitive conclusion since we are observing @xmath0 so close to the galactic plane where the problem of line - of - sight blending is exacerbated . * * figures [ fig : monr20 ] and [ fig : monr2 ] show that the monr2 molecular cloud sits right on top of a high - density cloud of @xmath0 that has a strong linear velocity gradient . the gradient , which extends for over 80 pc , has a magnitude of 0.035 and increases from about 7 to 12 from south to north - west . if the gradient is due to rotation , @xmath124 180 , depending on the size of the region considered . from conservation of angular momentum , one might expect monr2 to have a gradient somewhere between @xmath125 0.84 , pointed about @xmath126 west of north like the gradient in the atomic gas . this is not what we see in figure [ fig : monr2 ] , however . the velocity field of the monr2 molecular cloud has a much more complex structure than does the atomic gas . the outer northern , western , and southern edges of the cloud reach velocities up to 12 13 . the inner portion of the cloud is moving at slower speeds around 9 11 , which is also the velocity range at which the @xmath0 peak located at @xmath127 , @xmath128 dominates . considering the apparently random nature of the velocity field , it is unlikely that the gmc is undergoing large - scale , coherent rotation . the position - velocity plot for the molecular cloud in figure [ fig : monr2 ] ( red points ) has a zero slope , as does ngc 2264 , indicating that it has no significant linear gradient . it is difficult to say whether or not the present velocity field originated during the cloud s formation or during some later stage in its evolution . if the gmc originally had a more organized velocity field and did _ not _ inherent its present field during formation , perhaps a series of interactions with external forces or internal events such as turbulence and star formation activity have washed out any signature of a systematic velocity gradient which was previously present in the molecular cloud . * * the separation between the gradient directions in the rosette molecular cloud and the surrounding atomic gas is approximately @xmath1 . figure [ fig : rosette ] shows that in the velocities of the molecular cloud tend to increase from roughly 7 14 in the north - east to as high as @xmath129 in the south and south - west . the gradient in the @xmath0 is directed perpendicular to the galactic plane . we note that , of the clouds in our sample , the @xmath0 maps near the rosette probably suffer the greatest degree of line - of - sight blending . the cloud is very close to the galactic plane , and at the distance of the rosette , 1600 pc , we have the lowest spatial resolution ( @xmath129 pc ) in the atomic gas . therefore the surface density and velocity maps displayed in figures [ fig : rosette0 ] and [ fig : rosette ] almost certainly fail to capture many of the local , small - scale variations in the structure of the atomic gas . we measure a gradient in the molecular cloud of 0.09 , consistent with the 0.08 measured by williams et al . if the gradient is due to rotation , the specific angular momentum of the cloud is approximately 26 . based on the estimate of the minimum value of the accumulation radius , the specific angular momentum in the surrounding ism is nearly 3 times larger . figure [ fig : rosette0 ] shows the outline of the molecular cloud overlaid on a surface density map of the atomic gas , which is integrated over the range @xmath130 27 . a `` shell '' of @xmath0 containing two high - density peaks is associated with the molecular cloud . the brightest peak of the shell sits at the southern edge of the molecular cloud near a latitude of @xmath131 , and another peak in the surface density occurs near @xmath132 . the southern portion of the @xmath0 shell appears to mimic the ring - like structure in the molecular cloud sitting just to the north of it . when the location of the structures in both the molecular and atomic gas are compared with the first - moment map in figure [ fig : rosette ] , we see that the southern portion of the @xmath0 shell is moving near the same velocity , @xmath133 - 14 , as the south - east segment of the ring in the gmc . the western half of the ring in the gmc is moving at higher velocities . also note how the @xmath0 peak locate d near @xmath132 is moving slightly faster ( @xmath134 ) than the molecular gas in that region ( @xmath133 ) . all of this suggests that both the @xmath0 shell and the ring - like structure in the gmc are expanding . kuchar & bania ( 1993 ) demonstrated that the @xmath120 region ngc 2244 ( the rosette nebula ) could have given rise to the expansion in the atomic gas . and williams et al . ( 1995 ) found that certain properties of clumps within the gmc vary with distance from the nebula , centered at @xmath135 , @xmath136 ( celnik 1983 ) . taken all together , the evidence leads us to suggest that the gradient in the rosette molecular cloud is _ not _ caused by rotation , but by the high - luminosity @xmath120 region , ngc 2244 . high - energy winds of stars in ngc 2244 may have excavated a hole in the gmc , causing @xmath0 to be swept up into a high - density ridge by the expanding molecular gas ( e.g. , kuchar et al . 1991 , kuchar & bania 1993 ) . in this picture , because the nebula has had less of an impact on distant regions of the cloud , these distant regions are moving at lower velocities compared to molecular gas near the @xmath120 region . our key findings are that the angular momentum in the gmcs is less than that in the surrounding atomic gas , and the velocity gradient position angles in the molecular and atomic gas are widely divergent with orion a being the one exception . this leads us to suggest that rotation may _ not _ be the best explanation of the velocity fields observed in the gmcs . traditionally , at least three possible solutions have been invoked to resolve the angular momentum problem : \1 . one or more of the assumptions in the theory are inappropriate . for instance , if the average surface density of the precursor atomic gas is underestimated , this will lead to an overestimate in the accumulation radius and , consequently , an overestimate of the angular momentum initially imparted to a gmc . in our analysis , we used 10 for the mean surface density of the precursor gas , twice the average value in the solar vicinity . assuming that the initial gas had an even higher density , say 20 , this would not change the main result , namely that @xmath93 is consistently greater than @xmath94 . this is because the accumulation radius depends weakly on the initial surface density of the gas ( @xmath137 ) . on the other hand , if collapsing molecular clouds do not gather material far from the galactic plane , using larger values of @xmath76 might be appropriate . yet this would only exacerbate the discrepancy between the predicted and obser ved angular momenta . it might also be argued that using the effective radius of a gmc leads to underestimates in @xmath138 if the gmc is filamentary . in the previous section we took this into consideration with orion a and recalculated @xmath138 assuming a cylindrical morphology . this raised @xmath138 , which , nevertheless , remained less than @xmath93 . moreover , varying the size of the gmc or the region from which it gathers material does not solve the problem of the gradient directions in the molecular and atomic gas being unaligned . an alternative explanation of the angular momentum problem is that though the assumptions regarding the initial conditions may be valid , there may be some kind of external braking force which rapidly reduces the angular momentum of a gmc during its initial condensation . magnetic braking is often evoked as an angular momentum shedding mechanism ( e.g. , mouschovias 1977 ; fleck & clark 1981 ; mestel & paris 1984 ; rosolowsky et al zeeman splitting of the oh 18-cm line and the 21-cm line of neutral hydrogen have been used to measure the magnetic field strengths of gmcs . since gmcs are magnetized and mhd effects are expected to play a significant role in their evolution , magnetic braking in which magnetic field lines anchoring a gmc to the ambient ism provide the tension necessary to slow down rotation is a possible solution to the angular momentum problem . heiles & troland ( 2005 ) measured the mean magnetic field strength in the cold neutral medium of the milky way to be @xmath139 . the braking time is set by the time it takes for alfvn waves to travel across a region of gas having a moment of inertia comparable to the gmc ( mestel & paris 1984 ) . in order for magnetic braking to be efficient at slowing down cloud rotation , it would have to occur on timescales no greater than the timescale for cloud collapse . for gas having an initial density of 1 @xmath82 ( = 0.025 @xmath43 @xmath140 ) , the alfvn speed is @xmath141 . in a region having an accumulation radius of @xmath142 pc , this corresponds to a braking time of roughly @xmath143 myr . by comparison , the timescale for self - gravitational cloud formation in a region having the same density , assuming that this process occurs on a timescale close to the dynamical free - fall time ( mouschovias & paleologou 1979 ; mestel & paris 1984 ; elmegreen 2007 ) , is @xmath144 myr . however , the effectiveness of magnetic braking largely depends on how the mass of the forming cloud compares to its magnetic critical mass . mestel & paris ( 1984 ) argue that braking will efficiently slow down rotation only if the mass , @xmath145 , of the forming cloud is much less than its magnetic critical mass , @xmath146 . using crutcher s ( 1999 ) magnetic field strength measurements of molecular clouds , mckee & ostriker ( 2007 ) infer that gmcs have approximately @xmath147 , that is , they are magnetically supercritical . elmegreen ( 2007 ) argues that the ism out of which gmcs form has a magnetic field that is `` near - critical , '' i.e. , @xmath148 . furthermore , magnetic braking slows rotation most efficiently when the cloud s angular momentum vector is perpendicular to the magnetic field ( mouschovias & paleologou 1979 ) . due to the complexity of making precise measurements of the magnetic field in the ism , it is not established whether or not this is the case . clearly , more observatio ns are needed in order to determine the importance of magnetic fields in gmc evolution . finally , perhaps gmcs are not rotating after all , and the `` problem '' is unfounded . as some authors have pointed out , the interpretation that velocity gradients indicate rotation is not unique . expansion and shear , for instance , also produce velocity gradients . the numerical simulations of hennebelle et al . ( 2008 ) suggest that the converging flows of atomic gas could produce gmcs . if shearing occurs at the interface of colliding flows , might this result in an excess of shear , that is , values of @xmath57 that are higher than the shear arising from the local galactic rotation curve ? assuming a flat rotation curve , the local value of @xmath57 in the solar neighborhood is @xmath149 . this is slightly less than , but basically comparable to the gradient magnitudes we measure in perseus , orion a , and monr2 ( table 3 ) . as discussed previously , we are underestimating @xmath57 in every case , since we have no information regarding the inclination of clouds . also , since bl ending of emission along the line - of - sight is most extreme in the cases of ngc 2264 and the rosette , which are situated close to the galactic plane , it is likely that @xmath150 is underestimated to an even greater degree in these cases . nevertheless , we can not conclude whether or not shear is the cause of the velocity gradients based on these observations . it is an issue we would like to further investigate with a larger sample of clouds . burhert & bodenheimer ( 2000 ) show that turbulence may also cause linear velocity gradients . they demonstrated that the gradient magnitude of turbulent cores scales with size as @xmath151 . figure [ fig : r_vgrad ] shows @xmath57 as a function of @xmath71 for both the molecular clouds and @xmath0 . neither the gmcs nor the @xmath0 appear to follow the @xmath151 relationship , though we do not have enough data points to make a definite conslusion . studying the properties of gmc kinematics in other galaxies will also help to develop our picture of gmc formation . in their analysis of 45 gmcs in m33 , rosolowsky et al . ( 2003 ) did the first systematic , extragalactic study of gmc angular momentum properties . they showed that simple gmc formation theories consistently overestimate the magnitude of the observed angular momentum . they measured the velocity gradients from high - resolution ( @xmath152 ) data , finding that the gradients of m33 clouds are similar in magnitude to galactic clouds . they then tested several formation models by calculating the accumulation radii of the cataloged clouds which are predicted by the various models , including the toomre and parker instabilities . on average , the theories , which do not include the effects of magnetic fields on rotation , overpredict observed magnitudes of velocity gradients by more than a factor of 5 . and 40% of the gmcs in m33 are counterrotating with respe ct to the sense of galactic rotation . we extend this study by performing the analysis established here on a much larger sample of gmcs in m33 cataloged by rosolowsky et al . since m33 has a relatively low inclination of @xmath153 ( corbelli & salucci 2000 ) , such a study has the advantage of bypassing the problem of source confusion along lines of sight that arises when doing comparable surveys of molecular clouds in the milky way . we have presented a detailed comparison between the kinematics of five galactic molecular clouds and the atomic gas that surrounds them . we developed a method for selecting regions of @xmath0 that are associated with the gmcs and found that each gmc was nearby high - density @xmath0 which peaked above the mean galactic value . first - moment maps were created using the emission of the gmcs and the @xmath0 21-cm emission , and then a plane was fit to each map of velocity centroids . we determined the magnitudes and directions of the velocity gradients from the coefficients to the fits . from these observations and measurements , we arrived at the following conclusions : \1 . orion a , perseus , and the rosette each have a significant linear velocity gradient across the face of the molecular clouds , while ngc 2264 and monr2 have complex , non - linear velocity fields . the perseus molecular cloud has the strongest linear gradient in the sample , with a magnitude of @xmath154 . the atomic gas associated with orion a , monr2 and the rosette has significant linear velocity gradients , regardless of whether the molecular cloud has one . the @xmath0 gradients range from 0.019 to 0.067 , or 0.76 to 2.7 times the shear in the solar vicinity as measured by the galactic rotation curve . if the gradients in the molecular and atomic gas were due to rotation , the specific angular momentum in the gmcs is less than that predicted by the formation scenario in which a gmc preserves angular momentum while undergoing top - down collapse from the surrounding ism by a factor of 1 to 6 . the discrepancy can be narrowed if different assumptions are made regarding the initial density and geometry of the gas . but the direction of the trend that the observed angular momentum is less than the predicted remains the same . we observe large differences between the velocity gradient directions in the gmcs and the atomic gas , with orion a being the one exception . at more than @xmath1 , two of the most extreme angle separations can be seen in the monr2 and the rosette systems ( figure [ fig : monr2 ] and [ fig : rosette ] ) . furthermore , the gradient directions in neither the molecular nor the atomic gas are in alignment with the overall direction of galactic rotation . if the gradients were due to rotation , this indicates that some gmcs are counterrotating with respect to the galaxy . \5 . that the velocity gradient position angles in the atomic and molecular gas are divergent indicates that the gmcs in our sample probably did not inherit their present velocity fields from the atomic gas from which they formed . finally , in at least two cases , ngc 2264 and the rosette , a good explanation of the morphology and kinematics observed in the gas is that they are caused by stellar winds from o stars in the @xmath120 regions located in the gmcs , not rotation . lcccccccccc + + & & & @xmath37 & @xmath77 & & @xmath155 & @xmath156 & + cloud & @xmath157 & @xmath158 & [ pc ] & [ @xmath159 & [ @xmath159 & [ pc@xmath160 & [ ] & [ ] & & + perseus ....... & 159 & -20 & 320 & 1.4 & 1.2 & 825 & 7.5 & @xmath161 & 1 & 6 + orion a ....... & 210 & -19 & 414 & 2.4 & 6.9 & 1510 & 9.6 & @xmath162 & 2 & 7 + ngc 2264 ... & 202 & 1 & 800 & 2.5 & 2.2 & 674 & 6.6 & @xmath163 & 3 & 8 + monr2 ........ & 213 & -13 & 830 & 3.0 & 9.0 & 2940 & 10.8 & @xmath164 & 4 & 9 + rosette ........ & 207 & -2 & 1600 & 4.0 & 7.7 & 2200 & 13.1 & @xmath165 & 5 & 10 + lcccccc + + cloud & @xmath167 & @xmath150 & @xmath89 & @xmath90 & @xmath168 & @xmath93 + & [ @xmath169 & [ @xmath169 & [ deg ] & [ deg ] & [ ] & [ ] + perseus ....... & 23.1 & 6.70 & -20.2 & 99.4 & 24.2 & 60.3 + orion a ....... & 22.0 & 3.71 & -111 & -102 & 42.2 [ 84.1 ] & 107 + ngc 2264 ... & 4.60 & 1.87 & 47.8 & -173 & 3.92 & 25.2 + monr2 ........ & 6.68 & 3.48 & -72.4 & -11.2 & 25.5 & 120 + rosette ........ & 9.18 & 2.36 & -132 & 1.12 & 25.8 & 73.4 +
phonons are important in metallic systems . they affect the electronic behaviour in diverse ways causing the dominant temperature dependent contribution to the resistivity , an enhancement of the specific heat , and an attractive effective electron - electron interaction , which may induce superconductivity . one can expect strong electron - phonon interactions in strongly correlated systems , such as heavy fermion compounds@xcite , where the radius of the rare earth ions is very sensitive to the @xmath0-electron occupation . the coupling of electronic states to the lattice also plays an important role in the anomalous electronic behaviour of the manganites@xcite . the superconductivity in the fullerides@xcite appears to be induced by electron - phonon interactions within a strongly correlated electron band . however , the fully quantum mechanical treatment of lattice effects in these compounds has so far received comparatively little attention . the reason for that lies in the difficulty in handling strong electron - phonon coupling together with strong electron - electron repulsion . recent advances in non - perturbative techniques , most notably the development of the dynamical mean field theory ( dmft)@xcite , enable one to investigate the interplay of electron and phonon interactions in these systems . one of the most prominent models in the field of strongly correlated electron systems is the hubbard model@xcite which been used extensively to study the effects of strong local electron - electron interactions . especially within the framework of dmft@xcite the nature of the mott transition could be clarified@xcite . the holstein model@xcite has been used to study polaronic effects in the absence of electron - electron repulsion , mainly in the limit of low electron density . systems with finite electron density have received less attention , but have been studied within the dmft using monte carlo@xcite , exact diagonalization ( ed)@xcite , perturbative approaches@xcite and the numerical renormalization group ( nrg)@xcite . even less studied has been the interplay of electron - electron interactions and the electron - phonon coupling@xcite . the most natural model incorporating both these terms is the holstein - hubbard model defined by the hamiltonian @xmath1 here @xmath2 describes the electron electron interaction within a band of dispersion @xmath3 . the electron density @xmath4 at site @xmath5 couples linearly to the local displacement operator @xmath6 with an electron - phonon coupling @xmath7 . the phonons are assumed to be dispersionless ( local einstein phonons ) with energy @xmath8 and @xmath9 is the mass of the vibrating ions . and a semielliptic band of width @xmath10 as calculated in ref . . the dashed line is the locus of @xmath11 . points on the dot - dashed line fulfil @xmath12.,scaledwidth=46.0% ] recently , the @xmath13 phase diagram of the half filled holstein - hubbard model has been calculated@xcite using various local approximations , and is shown in fig . [ phases.eps ] . the results are for a semi - elliptic band of width @xmath10 and phonon frequency @xmath14 ( @xmath15 ) . all types of long - range order are excluded . we can distinguish three different phases : metallic , bipolaronic and mott insulating phase . the metallic phase is always found to be a fermi liquid . the latter two phases display a gap in the one - electron spectra and will be referred to as gapped phases . the parameters of fig . [ phases.eps ] , namely @xmath16 and @xmath17 are taken throughout this paper , unless mentioned otherwise . one way of simplifying the discussion of the holstein - hubbard model would be to integrate out the phonons in order to derive an effective model for the electrons . the effective action is then governed by the @xmath18-dependent potential@xcite @xmath19 this reduces to the static quantity @xmath20 in the limits @xmath21 and @xmath22 . therefore , on small energy scales , we expect an effective attractive or repulsive interaction depending on the value of @xmath23 . for small phonon energies @xmath8 , however , there is no reason to expect the static quantity @xmath23 to be sufficient to characterize the behaviour of the holstein - hubbard model . the properties of this model are expected to depend on both @xmath2 and @xmath7 individually . this is also apparent from the phase diagram in fig . [ phases.eps ] . the dot - dashed line , where the double occupancy takes the value of a free system @xmath12 is quite distinct from the line @xmath24 . the aim of this study is to clarify the interplay of these different types of interactions . an interesting question about this model is whether the suppression of local charge fluctuations by a significant hubbard @xmath2 will effectively decouple electrons and phonons on all energy scales . it is possible , however , that for the behaviour on low energy scales the effective onsite-@xmath2 is strongly renormalized and hence the electron phonon coupling will have a comparatively stronger effect on energy scales @xmath25 ( see ref ) . in this paper , we present results on dynamical electronic and phonon response functions for the holstein - hubbard model as calculated within the dynamical mean field theory@xcite ( dmft ) . the effective impurity problem is solved using wilson s numerical renormalization group ( nrg ) method@xcite extended to treat the electron phonon coupling@xcite . the combination of dmft and nrg has proven to be a reliable and effective method for calculations of both metallic and insulating phases at @xmath13 . it has been applied to the mott transition in the pure hubbard model@xcite and to the pure holstein model@xcite . in our calculations we use the discretization parameter @xmath26 and retain up to @xmath27 states . the value of @xmath26 gives rise to a bandwidth correction factor @xmath28 ( see eq . ( 5.42 ) in ) which we take into account in the nrg procedure . for the calculation of dynamical response functions we use the method described in ref . . apart from the usual local one - electron green s function @xmath29 we will be interested in two different phonon green s functions , @xmath30 since we will treat only homogeneous phases here and exclude all types of long - range order , we have dropped the site indices in all of the definitions . the spectral weight of @xmath31 will be denoted by @xmath32 , and similar for all other green s functions . the dynamical charge and spin susceptibility are given by @xmath33 respectively . the operator @xmath34 represents the electronic term that couples to the phonons , @xmath35 it should be noted that our definition of @xmath36 differs from the usual one by a factor of four . we also calculate the quasiparticle weight @xmath37 given by @xmath38 within the local approximation of the dmft . the different regimes of the phase diagram of fig . [ phases.eps ] give rise to the following structure of the paper : in sec . [ sec : g_dependence ] we choose a fixed value of @xmath2 and study dynamical properties as a function of the phonon coupling @xmath7 . we find continuous transitions from a metallic to a bipolaronic state for small values of @xmath39 and @xmath40 ( sec . [ ssec : weaku ] ) . for a larger value of @xmath41 , as discussed in sec . [ ssec : strongu ] , the transition to the bipolaronic state becomes discontinuous . complementarily , we look at the @xmath2-dependence of dynamical properties in sec . [ sec : u_dependence ] keeping @xmath7 fixed to @xmath42 . for this value of @xmath7 the system can be in all three phases , depending on @xmath2 . in sec [ sec : polaronic ] we study the properties of the model on the two polaronic lines as given by @xmath24 and @xmath43 ( dashed and dot - dashed lines in fig . [ phases.eps ] ) . we also show spectra for @xmath24 for different values of @xmath8 . in sec . [ sec : overview ] we give an overview of our results , and we present our conclusions in sec . [ sec : conclusions ] . in this section , we fix the value of the hubbard interaction @xmath2 and study dynamical properties for a varying electron - phonon coupling @xmath7 . we distinguish between the cases of weak and strong @xmath2 . and various values of @xmath7 . the inset shows the variation of the quasiparticle weight @xmath37 for @xmath40 ( solid line ) and @xmath39 ( dashed line).,scaledwidth=46.0% ] first of all , we look at the one - electron spectra for a fixed @xmath44 as shown in fig . [ g_spec_u10_gx.eps ] . there is a strong similarity of these results with those of the pure holstein model ( @xmath39 ) calculated previously@xcite . at weak coupling a narrow peak emerges at the fermi energy . the top of this feature is rather flat and the imaginary part of the self energy has a shallow @xmath45-dependence . this one would expect from lowest order perturbation theory in @xmath7 , where the imaginary part of the self energy vanishes in the range @xmath46 for @xmath39 ( see eq . ( 19 ) in ref . ) . upon increasing @xmath7 further , there is a rapid narrowing of this feature until it disappears at a critical coupling of @xmath47 and a gap opens . similar to the pure holstein model , there is no preformed gap in this case in contrast to the mott transition in the hubbard model . the metallic regime corresponds to a renormalized fermi liquid . the quasiparticle weight @xmath37 is shown as a function of @xmath7 in the inset together with that for the pure holstein model ( @xmath39 ) . for @xmath40 , the quasiparticles are already slightly renormalized at @xmath48 . however , initially the renormalization with @xmath7 is weaker than in the case @xmath39 but more rapid on the approach to @xmath49@xcite . as with the pure holstein model , there is no evidence of multiple solutions near the phase transition . this is in contrast to the situation near the mott transition for the hubbard model , where there is a finite coexistence region . , top ) and @xmath40 ( bottom ) . the values of @xmath7 are @xmath50 ( bottom ) , @xmath51 ( top ) in the upper panel and @xmath50 to @xmath52 in steps of @xmath53 in the lower panel . the dotted lines are scaled up by a factor of @xmath54.,title="fig:",scaledwidth=23.0% ] , top ) and @xmath40 ( bottom ) . the values of @xmath7 are @xmath50 ( bottom ) , @xmath51 ( top ) in the upper panel and @xmath50 to @xmath52 in steps of @xmath53 in the lower panel . the dotted lines are scaled up by a factor of @xmath54.,title="fig:",scaledwidth=23.0% ] + , top ) and @xmath40 ( bottom ) . the values of @xmath7 are @xmath50 ( bottom ) , @xmath51 ( top ) in the upper panel and @xmath50 to @xmath52 in steps of @xmath53 in the lower panel . the dotted lines are scaled up by a factor of @xmath54.,title="fig:",scaledwidth=23.0% ] , top ) and @xmath40 ( bottom ) . the values of @xmath7 are @xmath50 ( bottom ) , @xmath51 ( top ) in the upper panel and @xmath50 to @xmath52 in steps of @xmath53 in the lower panel . the dotted lines are scaled up by a factor of @xmath54.,title="fig:",scaledwidth=23.0% ] figure [ csss_spec_u10_gx.eps ] shows the spectra of the corresponding charge- and spin susceptibilities at @xmath40 and @xmath39 for comparison . in the free system , @xmath55 as defined in eq . ( [ eq : chi_csss ] ) , which can be seen in the lowest curves of the two upper panels . looking at the case @xmath39 first ( upper panel ) , we see the emergence and buildup of a low - energy peak in @xmath56 and the suppression of weight in @xmath57 for @xmath58 . in the bipolaronic state ( @xmath59 ) , both charge and spin susceptibilities have the same gap , roughly twice that seen in the electronic spectrum . they have an identical peak above this gap . these can be seen in the dotted curves of fig . [ csss_spec_u10_gx.eps ] , which have been scaled up by a factor of @xmath54 to make these features visible . the coincidence of these peaks can be explained by the fact that spin fluctuations necessarily require the breakup of a bipolaronic pair . their weight is very small due to the strong bipolaronic binding . there is an additional very small peak in the charge susceptibility in the gap close to @xmath8 from the residual couplings to the phonon mode . in the absence of charge ordering , there must be charge fluctuations at @xmath60 which , however , can not be observed numerically . for @xmath40 ( lower panel ) we observe the same overall trends . however , starting from a reduced charge susceptibility due to the finite @xmath2 , the peak in @xmath56 develops more slowly . the spin susceptibility , which is enhanced due to the finite @xmath2 , changes very little in the metallic phase . the low - energy spin fluctuations are almost completely suppressed with the emergence of the gap in the bipolaronic state . the charge fluctuations can be directly related to the phonon propagator via the equation ( see appendix [ app : d_chic ] ) @xmath61 where @xmath62 is the free phonon propagator . however , use of this equation with an _ approximate _ form of @xmath63 can lead to severe numerical errors , as discussed in appendix [ app : d_chic ] . in fig . [ bb_spec_u10_gx.eps ] , we give the results for the phonon spectra for @xmath39 and @xmath40 calculated via the self energy@xcite , similar to the procedure we use for the electronic spectra@xcite . the results for @xmath39 differ significantly from those published earlier@xcite based on eq . [ eq : d_chic ] . in contrast to the earlier results , the main phonon peak is found initially to soften significantly in a similar way to the migdal elisahberg result . the two - peak structure develops only upon approaching the transition to the bipolaronic state . in the gapped state , the high - frequency peak narrows and tends towards @xmath8 . the same trend can be seen for @xmath40 ( right - hand plot ) . the slower initital softening correlates with the suppression of charge fluctuations seen in @xmath63 . in the pure holstein model ( @xmath39 ) . in the metallic range ( @xmath64 ) , the direct calculation ( @xmath65 ) agrees well with the integration over the phonon green s function ( @xmath66 ) and charge susceptibility ( @xmath67 ) ) . for details see appendices [ app : d_chic ] and [ app : n_vs_x ] . the dot - dashed line shows the result from the migdal eliashberg calculation.,scaledwidth=46.0% ] on the approach to the transition , significant negative spectral weight for @xmath68 can be seen in the spectra shown in fig . [ bb_spec_u10_gx.eps ] . as a consequence of the spectral theorem , this relates directly to the average number of excited phonons in the ground state . the spectra of the green s function @xmath69 show similar features as @xmath31 , and integrated up to @xmath60 yield the average lattice fluctuations as shown in fig . [ x2_av_u00_gx.eps ] . these results agree well with the direct evaluation of the expectation value @xmath70 in the metallic phase . for low to intermediate values of @xmath7 , the nrg results do not deviate significantly from the midgal eliashberg calculation and are significantly smaller than reported in ref . based on eq . [ eq : d_chic ] . on the approach to the transition , there is a large increase in the lattice fluctuation . in the gapped phase , we observe a large difference between the direct calculation and the integration over the negative part of the phonon spectral density . one way of looking at it is that due to the limited numerical resolution we miss the contribution from the peak at @xmath71 in the spectrum of @xmath69 . an other way of explaining this difference is that there is no dynamics connecting the two degenerate ground states with @xmath72 and @xmath73 in the gapped phase . these two ground states have nonvanishing average displacements @xmath74 , see appendix [ app : n_vs_x ] , and integration over the phonon spectrum yields the fluctuations about @xmath74 , i.e. , @xmath75 . the calculation of @xmath76 in one of the ground states quantitatively explains the different values of the two methods in the gapped phase . and various values of @xmath7 . the central peak broadens with increasing @xmath7 and at @xmath77 disappears in a first order phase transition to the bipolaronic state . the inset shows the quasiparticle weight @xmath37 ( @xmath66 ) and the double occupancy @xmath78 ( @xmath79 ) as functions of @xmath7.,scaledwidth=46.0% ] next we look at the transition to the bipolaronic state for a larger fixed value of @xmath41 . figure [ g_spec_u50_gx.eps ] shows the corresponding one - particle electron spectral functions for various values of @xmath7 . their @xmath7-dependence is in sharp contrast to the small @xmath2 case shown in fig . [ g_spec_u10_gx.eps ] . initially for @xmath48 , we have the three - peak structure of a strongly correlated hubbard model . with increasing @xmath7 , the central resonance broadens slightly as is reflected in the increasing quasiparticle weight @xmath37 ( see inset ) . this can be ascribed to the partial cancellation of the hubbard repulsion by phonon mediated electron - electron attraction , see eq . ( [ eq : ueff_def ] ) . the effect , however , is much weaker than one would deduce from @xmath23 . the position of the high - energy hubbard bands is virtually unaffected by the phonons , as expected from eq . ( [ eq : ueff_def ] ) evaluated at @xmath80 . when approaching the critical coupling @xmath81 , the quasiparticle peak disappears abruptly and @xmath37 jumps to zero , indicating a discontinuous transition to the bipolaronic phase . the double occupancy , shown in the inset to fig . [ g_spec_u50_gx.eps ] , is much less than a quarter in the metallic state but makes a sudden jump to @xmath82 on the point of the phase transition . a coexistence of metallic and gapped solutions is found in the narrow range @xmath83 . . the values of @xmath7 correspond to those in fig . [ bb_spec_u50_gx.eps ] . note the different scales of the @xmath84-axes . dotted lines have been scaled up by a factor of @xmath54.,title="fig:",scaledwidth=23.0% ] . the values of @xmath7 correspond to those in fig . [ bb_spec_u50_gx.eps ] . note the different scales of the @xmath84-axes . dotted lines have been scaled up by a factor of @xmath54.,title="fig:",scaledwidth=23.0% ] the spectra of the charge and spin susceptibilities are shown in fig . [ csss_spec_u50_gx.eps ] . they display similar trends , but more pronounced , than the corresponding results for @xmath40 ( lower panels of fig . [ csss_spec_u10_gx.eps ] ) . the low - energy peak in @xmath63 only becomes clearly visible when @xmath7 reaches the value of @xmath85 . in the bipolaronic phase , the peaks above the gap in the charge and spin susceptibilities are not identical as in the @xmath39 case , but very similar however , the gaps are too large to be visible on the @xmath18-scale of fig . [ csss_spec_u50_gx.eps ] . the only peak visible , when enhanced by a factor of @xmath86 , is the peak in the charge susceptibility within the gap near @xmath8 . and various values of @xmath7.,scaledwidth=46.0% ] in fig . [ bb_spec_u50_gx.eps ] , the spectra of the corresponding phonon propagators are plotted . over a large initial range of @xmath87 there is very little softening , corresponding to a complete suppression of charge fluctuations . when approaching @xmath49 , the phonon mode softens and develops a low - energy shoulder . after the discontinuous transition to the bipolaronic phase , the mode hardens back to @xmath8 and narrows . in this section we will fix the electron - phonon coupling at the value @xmath42 and study the dependence of dynamic response functions on the hubbard repulsion @xmath2 . the value of @xmath42 is chosen because all three phases can be found for it , depending on @xmath2 ( see fig . [ phases.eps ] ) . figure [ g_spec_ux_g45.eps ] shows the one - electron spectral function in the vicinity of both phase transitions . the right - hand plot for large values of @xmath2 is very similar to what is observed in the mott transition of the pure hubbard model . the lower and the upper hubbard peak at @xmath88 are well developed and move to higher energies as @xmath2 increases . at the same time , the central peak narrows and vanishes at the mott transition . as for the pure hubbard model , there is a preformed gap in the one - electron spectrum and a broad region of numerical coexistance of metallic and insulating solutions . the electronic spectra are very similar to those of the pure hubbard model near the mott transition . the phonons do not seem to alter much the picture of the phase transition . the transition from the bipolaronic to the metallic state is completely different , as can be seen in the left - hand plot of fig . [ g_spec_ux_g45.eps ] . the phase transition occurs close to @xmath40 and the spectra show a similar behaviour as discussed in sec . [ ssec : weaku ] , where @xmath7 is varied . in the bipolaronic state , with the initial increase of @xmath2 , the two polaron bands move towards the fermi level . on the metallic side , just after the transition , we see a very sharp resonance which broadens when @xmath2 is increased further . there is no signature of a preformed gap once the system is metallic . again , there is no evidence of a significant coexistence region . the spectra of the charge and spin susceptibilities for @xmath42 are shown in fig . [ csss_spec_ux_g45.eps ] . we clearly see that the low - energy features of the two susceptibilities signal the appearance of two different instabilities of the system , depending on @xmath2 . in the bipolaronic phase , both spectra are gapped and we see only the peak near @xmath8 in the charge susceptibility . upon entering the metallic phase , the zero frequency peak in the charge susceptibility moves to finite @xmath18 . this peak loses weight and slightly disperses to higher energies as the system becomes more metallic . in this small @xmath2 regime , spin fluctuations are suppressed . the converse occurs when approaching the mott transition at @xmath89 . the charge fluctuations are suppressed and peak in the spin susceptibility builds up , signalling instability towards antiferromagnetic ordering . in the mott insulating state , both susceptibilities are gapped . the spin fluctuations above the gap ( beyond the scale of fig . [ csss_spec_ux_g45.eps ] ) must largely be due to the charge fluctuations , as in the pure hubbard case . however these peaks are not identical , in contrast to the bipolaronic gapped state . the difference in the peaks can be interpreted as a result of the spin correlations induced by @xmath2 . however , due to the phonon coupling the spin fluctuation peak is reduced compared to the pure hubbard case . and various values of @xmath2 . the transition from the bipolaronic to the metallic state is clearly visible by the two - peak structure whereas the transition to the mott insulator does not obviously affect the phonon propagator.,scaledwidth=46.0% ] figure [ bb_spec_ux_g45.eps ] displays the phonon propagator for this scan . in the bipolaronic phase at @xmath39 , electrons and phonons are almost decoupled , and the phonon spectrum shows only one peak just below @xmath14 . when increasing @xmath2 , this mode softens and splits into two peaks close to the phase transition . on the metallic side , the low - energy peak survives and continuously hardens back to @xmath8 , as the system approaches the mott transition . there is no signature of the mott transition in the phonon spectrum . the effect of an increasing @xmath2 is just to suppress continuously the charge fluctuations which results in a decoupling of electrons and phonons which drives the hardening of the phonon peak . another feature of the suppression of charge fluctuations is the narrowing of the phonon peak with increase of @xmath2 , as also observed in ref . . in our case , however , the shift of the peak appears to be more marked . the narrowing can also be seen in comparing the curves with the same ( low ) value of @xmath7 in figs . [ bb_spec_u10_gx.eps ] and [ bb_spec_u50_gx.eps ] . . the weight for @xmath90 indicates the degree of double occupancy ; @xmath42 and various values of @xmath2 . the inset shows the double occupancy calculated directly ( @xmath67 ) and via the spectral theorem from @xmath91 ( @xmath79 ) . circles indicate the total spectral weight of @xmath91 which should evaluate to @xmath92 at half filling . the discrepancies are most pronounced near the phase transitions at @xmath93 and @xmath94.,scaledwidth=46.0% ] further insight into the nature of these transitions can be gained by looking at the behaviour of the higher order electron green s function @xmath95 this green s function is needed for the calculation of the self energy in the nrg procedure@xcite . the integrated spectral weight for @xmath96 equals the double occupancy @xmath97@xcite . moreover , the total spectral weight of @xmath98 yields the average density @xmath99 , which should be @xmath100 in the particle - hole symmetric case treated here . figure [ f_spec_ux_g45.eps ] shows the spectral function of @xmath91 for @xmath42 . in the bipolaronic phase , almost all of the weight is located below the fermi energy signalling a high value of the double occupancy ( see inset ) . as we enter the metallic phase , a large part of this weight is transferred to higher energies . in the mott insulator , virtually no weight is left at @xmath68 and the double occupancy is completely suppressed . the inset to fig . [ f_spec_ux_g45.eps ] also shows the total spectral weight of @xmath91 , which should evaluate to @xmath101 . we see that this is indeed more or less the case , except for values of @xmath2 close to the phase transitions @xmath93 and @xmath94 . for these values , the truncation of the hilbert space in the nrg introduces the most significant errors . there we also find the strongest discrepancies between the direct calculation of the double occupancy from the ground state expectation value ( marked with @xmath67 ) and the indirect , less accurate calculation via @xmath91 ( marked with @xmath79 ) . in this section we discuss the dynamical properties along two particular lines in the @xmath102 plane ( see fig . [ phases.eps ] ) . the first of these lines is given by @xmath103 where one might naively expect free quasiparticles close to the fermi surface . the other line is the location of points where the double occupancy @xmath43 , as in the free system . only for very large values of @xmath8 can the hubbard term be completely cancelled by the phonon - mediated attractive term ( see eq . [ eq : ueff_def ] ) , in which case both polaronic lines would coincide . we investigate these lines to understand the degree of compensation of these competing interactions for small phonon frequency @xmath104 . at @xmath14 ( dashed line in fig . [ phases.eps ] ) . the inset shows the variation of the quasiparticle weight @xmath37 and double occupancy @xmath78.,scaledwidth=46.0% ] figure [ g_spec_ueff0_ux.eps ] shows the one - electron spectral functions for various values of @xmath7 and @xmath2 such that @xmath24 . for small values of @xmath2 ( and @xmath7 ) , we see the sharp feature in the electronic spectrum as discussed in sec . [ sec : g_dependence ] . however , upon increasing @xmath2 further , hubbard bands develop and the central feature becomes the central quasiparticle peak similar to the pure hubbard model for large @xmath2(@xmath105 ) . this indicates that phonons play their assumed role of compensating the hubbard repulsion only to a very limited extent . the only remarkable difference to the pure hubbard model is that on the line @xmath24 the phase transition is shifted towards a higher value of @xmath106 , as compared to @xmath107 for @xmath48 . the inset to fig . [ g_spec_ueff0_ux.eps ] shows that the quasiparticle weight decreases steadily from @xmath108 to zero and the double occupancy from @xmath43 to a small but finite value . . the inset shows the expectation value of the lattice fluctuations as a function of @xmath2.,scaledwidth=46.0% ] the phonon spectra , as displayed in fig . [ bb_spec_ueff0_ux.eps ] , are always made up of only one single peak that initially moves towards lower energies as more phonons are excited and lattice fluctuations increase ( see inset ) . at @xmath109 this trend reverses and the phonon peak hardens back when approaching the phase transition to the gapped state . simultaneously , the number of excited phonons decreases to zero and the lattice fluctuations reach the value of the quantum mechanical zero point fluctuations . the reason for this non - monotonic behaviour of the lattice fluctuations as function of @xmath7 is not obvious . the maximum in the lattice fluctuations as shown in the inset correlates with the value of @xmath2 where the transition to the bipolaronic state changes from second order ( for @xmath110 ) to first order ( for @xmath111 ) . this could be a possible explanation as one would expect to see stronger lattice fluctuations as one aproaches a second - order transition to the bipolaronic state . . ( dot - dashed line in fig . [ phases.eps ] ) . the inset shows the variation of the quasiparticle weight @xmath37 along this line.,scaledwidth=46.0% ] the line in the phase diagram where @xmath113 starts at the uncorrelated system @xmath114 and ends at @xmath115 where it merges with the phase boundary to the bipolaronic state . as a consequence , for larger values of @xmath2 , the double occupancy has a jump at the transition from the metallic state ( @xmath116 ) to the bipolaronic state ( @xmath117 ) ( see fig . 3 in ref . and inset of fig . [ g_spec_u50_gx.eps ] ) . therefore this transition has to be first order in this range . figure [ g_spec_d25_ux.eps ] shows the one - electron spectral functions along this line . we observe the appearance of the narrow feature at the fermi energy that is a typical for the electron - phonon coupling . the band develops broad shoulders , but they never separate into distinct subbands . . the inset shows the expectation value of the lattice fluctuations as a function of @xmath2 along this line.,scaledwidth=46.0% ] the phonon propagator , as displayed in fig . [ bb_spec_d25_ux.eps ] shows a remarkable softening of the original peak and the build - up of negative spectral weight . as a consequence , the lattice fluctuations increase almost linearly with @xmath2 and are quite pronounced for @xmath118 . and various values of @xmath8.,scaledwidth=46.0% ] so far we have worked with a fixed value of @xmath14 . however , the degree of compensation of the competing interactions is dependent on @xmath8 . here we return to the case @xmath24 and study the variation of @xmath8 with fixed @xmath41 and @xmath119 . in fig . [ g_spec_ueff0_u50_wx.eps ] we plot the one - electron spectral density for a range of values of @xmath8 . for the largest value of @xmath120 , the spectrum is virtually the same as that of the free system , because we have almost complete compensation . we now examine the behaviour as we progressively reduce @xmath8 . first the high - energy phonon subbands gain weight and move towards the fermi energy . their position is roughly given by @xmath8 . there is a commensurate reduction of the width of the central peak . for @xmath121 the phonon subbands are visible as shoulders of the central peak . upon further reduction of @xmath8 , these shoulders move back towards higher energies again and , for @xmath122 , become the hubbard bands located at @xmath123 . for the smallest value of @xmath14 , the spectrum is identical to the one discussed in sec . [ ssec : ueff0 ] . and @xmath41 for various values of @xmath8.,scaledwidth=46.0% ] in fig . [ bb_spec_ueff0_u50_wx.eps ] we plot the spectra of the phonon propagator as a function of the relative energy scale @xmath124 . for large @xmath125 , where the compensation is almost complete , we see the narrow phonon peak at @xmath8 . in addition we see a small feature near @xmath60 . as @xmath8 is decreased , the low - energy feature gains weight and the mode at @xmath8 broadens and moves to slightly higher energies , as one would expect when @xmath8 approaches but lies above the upper band edge . at @xmath126 , the two features have approximately equal weight . for smaller values of @xmath8 the upper peak has almost zero weight . the remaining peak is the soft mode discussed in the previous sections . ( o ) , the rescaled double occupancy @xmath127 ( @xmath79 ) , and lattice fluctuations @xmath128 ( @xmath67 ) as a function of @xmath8 for @xmath24 . the dot - dashed line represents the number of excited phonons @xmath129.,scaledwidth=46.0% ] in fig . [ zdp_ueff0_u50_wx.eps ] we illustrate how different static quantities reflect the trend of the holstein - hubbard model towards a free system in the limit of large @xmath8 . we plot @xmath37 and @xmath127 as a function of @xmath8 . the double occupation value converges relatively rapidly to that of the free state . the quasiparticle weight , however , converges rather more slowly . in fact , only for @xmath130 is it approaching the uncorrelated limit @xmath108 . in the same fig . [ zdp_ueff0_u50_wx.eps ] we also plot the lattice fluctuations @xmath131 . this curve goes monotonically to zero as @xmath22 asymptotically behaving as @xmath132 for large @xmath8 . however , if we plot the rescaled values @xmath133 , we find a a maximum which coincides with the region of maximum softening of the phonon mode . for larger values of @xmath8 this quantity decreases to the quantum limit of @xmath134 . one would expect the phonons to decouple from the electrons in the limit of large @xmath8 . it is interesting to observe , however , that @xmath135 ( dot - dashed line in fig . [ zdp_ueff0_u50_wx.eps ] ) fits well the high-@xmath8 behaviour of the lattice fluctuations which indicates that the ground state of the phonons in this limit is a coherent state rather than an eigenstate of the free phonon hamiltonian @xmath136 . for an overview we will find it useful to classify the metallic state in the phase diagram into three regions , separated by the two polaronic lines , in each of these we find qualitatively different behaviour . we consider first of all the region below the line @xmath24 . here one interesting question is how the coupling to the phonons affects the transition to the mott insulating state . our results indicate that this transition is always very similar to that found in the pure hubbard model : the appearance of a preformed gap , a rather large coexistence region between insulating and metallic solutions and the quasiparticle weight continuously going to zero . there is only a small increase in the critical value of @xmath2 which is found empirically to depend upon @xmath7 to good approximation as @xmath138 the small influence of the phonon coupling can be readily explained by suppression of charge fluctuations for large @xmath2 . as the electron - phonon coupling in the holstein - hubbard model is to the charge fluctuations , the suppression of these renders the coupling ineffective . in fact , the whole region @xmath137 appears to be dominated by the hubbard term . for @xmath139 the second term in the retarded interaction as defined in eq . [ eq : ueff_def ] is negligible and the effective interaction is essentially given by @xmath2 . also on the lowest energy scales @xmath140 , @xmath141 . next we look at the complementary region defined by @xmath142 which is dominated by the coupling to phonons . here the similarity is with the pure holstein model . the metal to bipolaronic transition takes place without a preformed gap and with the characteristic softening of the phonon mode . the transition in this region appears also to be continuous with no significant coexistence region . the main effect of the hubbard @xmath2 is to push the phase transition and its precursors to somewhat larger values of @xmath7 . finally , in the region bounded by the two curves @xmath24 and @xmath43 there is more complex interplay of the two interactions , as can be seen in our results on the response functions . beyond the point where the line @xmath43 merges into the phase boundary the transition to the bipolaronic state is first order and thus qualitatively different from that discussed in the last paragraph . along the lower boundary line of this region , i.e. , for @xmath24 , the effective interaction is still repulsive for @xmath139 . the question arises whether on the lowest energy scale there is a complete cancellation of the two competing interactions . this question can be investigated by examining the lowest - lying one- and two - particle excitations from the interacting ground state . the calculation of the effective interaction @xmath143 between the renormalized quasiparticles has been considered in detail for impurity models in ref . . a similar analysis is possible using the effective impurity model within the dmft . from ref . , @xmath144 where @xmath145 ( @xmath146 ) is the energy of the lowest - lying single - particle ( particle - particle ) excitation in the @xmath147-th iteration of the nrg procedure and @xmath148 is the nrg discretization parameter . as for the impurity model , we find that , within the numerical accuracy of the nrg , @xmath143 vanishes on the line @xmath24 and is positive ( negative ) below ( above ) this line . therefore , along this line we have a system of renormalized ( @xmath149 ) but non - interacting quasiparticles . on the high - energy scale @xmath150 , the system is still dominated by @xmath2 . this can be seen by the formation of the hubbard bands as shown in fig . [ g_spec_ueff0_ux.eps ] and the suppression of the double occupancy . in this study we have investigated the particle - hole symmetric holstein - hubbard model . we have used the dynamical mean field theory in combination with the numerical renormalization group to calculate dynamical correlation functions for the full hamiltonian with quantum phonons . this non - perturbative approach allows us to access all parameter regimes of the model at zero temperature . a non - perturbative technique is indeed necessary because perturbative methods such as midgal - eliashberg approach for the holstein model are known to break down@xcite in the strong - coupling limit . we find three regions with qualitatively different behaviour . in the region of the mott metal insulator transition we find the phonon effects are largely suppressed by the on - site repulsion @xmath2 . this conclusion may be specific for the non - degenerate holstein - hubbard model where the coupling is purely to the local charge density . a system with orbital degeneracy and jahn - teller phonons shows@xcite strong phonon effects even for strong @xmath2 . in our case of the non - degenerate holstein - hubbard model , the electronic effects on phonons yield only a modest softening of the phonon mode in the metallic regime . this softening disappears in the mott insulator . weak electron - electron interactions seem to have little effect other than to delay the onset of the transition to the bipolaronic phase . for larger values of @xmath2 the transition changes from continuous to discontinuous . for small values of @xmath2 we see a complete softening of the phonon peak when approaching the transition . the softening is much less pronounced for larger @xmath2 . the line @xmath24 in the metallic phase does seem to have some significance in that for @xmath137 the spin susceptibility has a dominant low - energy peak . for @xmath151 the dominant low - energy peak is in the charge susceptibility . the line @xmath24 also appears to be significant for the low - energy fermi liquid behaviour as the on - site quasiparticle interaction changes sign on or close to this line . this is not obvious as one might have expected that on the low energy scale of the electron - phonon interaction , the electron - electron interaction is no longer given by the bare @xmath2 but some renormalized value @xmath152 . this would imply that one should use @xmath152 in eq . [ eq : ueff_def ] on the energy scale @xmath140 , so that the change of sign of the quasiparticle interaction @xmath143 , as defined in eq . ( [ eq : tildeueff ] ) , would occur when @xmath153 . however , our results indicate that this is not the case and that @xmath143 vanishes on or close to the line @xmath154 . a similar situation was found in the case the anderson - holstein impurity model@xcite . the holstein - hubbard model is very rich and shows also diverse forms of behaviour that could not be addressed in this paper . apart from the results presented here , this model is known to exhibit various types of symmetry - breaking phases@xcite . we intend to extend our calculations to investigate the competition between charge order , antiferromagnetism , superconductivity and the effects of doping@xcite . this should enable us to make direct contact with experimental results on such compounds as the fullerides , where this model with the assumption of the coupling of the electron density to the local phonon mode should be directly applicable . for applications to jahn - teller systems , such as in the manganites , the model would have to be extended to include degenerate orbitals . we wish to thank the epsrc ( grant gr / s18571/01 ) for financial support . this work was partially supported by sunnynames llp . we acknowledge fruitful discussions with a. gogolin , d. edwards and y. = ono . here we reconsider the relation between the phonon propagator and the charge susceptibility ( eq . ( [ eq : d_chic ] ) ) for the holstein - hubbard model and comment on pitfalls in its exploitation . [ [ equations - of - motion ] ] equations of motion : + + + + + + + + + + + + + + + + + + + + for the derivation of eq . ( [ eq : d_chic ] ) , we can use the standard equation of motion for the fourier transform of the double - time green s function , @xmath156_\eta \rangle+ { { \big<\!\big < { [ a , h]_{-}\,;b } \big>\!\big > } } _ { \omega}\ ] ] for the boson green s function @xmath31 we take @xmath157 , @xmath158 , and @xmath159 . for @xmath69 we take @xmath160 and @xmath159 . the following procedure works for both @xmath31 and @xmath69 , but we will use the former in what follows . the equation of motion gives @xmath161 where @xmath162 . taking the equation of motion for the right - hand operator , we have @xmath163 hence the result ( @xmath164 is the free phonon propagator ) , @xmath165 which is eq . ( [ eq : d_chic ] ) linking the phonon propagator @xmath31 with the charge susceptibility @xmath166 . as a green s function , the phonon propagator has a series of poles of first order , but none of second order . for this to be the case , eq . ( [ eq : d_chic ] ) tells us that the charge susceptibility has to vanish as @xmath167 in order to ensure that the pole of @xmath31 at @xmath168 is first order . this is also evident if we express @xmath63 in terms of the irreducible particle - hole bubble @xmath169 . it takes the form @xmath170 with the non - interacting phonon propagator @xmath171 . since @xmath172 diverges at @xmath168 , we have @xmath173 . [ [ spectral - density ] ] spectral density : + + + + + + + + + + + + + + + + + we will derive an expression for the spectral density of the phonon propagator , i.e. , for @xmath174 that implicitly includes the fact that @xmath173 . consider the system with a finite basis such that , @xmath175 for @xmath176 , where @xmath177 are the discrete set of excitations in the charge density response function . then @xmath178 the contributions to the spectral density @xmath179 arise solely from the poles at @xmath168 and @xmath180 . these do not coincide because @xmath173 . the contribution to @xmath179 from the poles at @xmath180 are straightforward to evaluate and give @xmath181 the contribution from the pole at @xmath168 is a little trickier to evaluate . the term in @xmath182 contributes @xmath183 where @xmath184 denotes the derivative of the delta function . however , we have @xmath185 where @xmath186 is the derivative of @xmath63 and is given by @xmath187 using @xmath173 and collecting all the terms together we obtain @xmath188 this expression can be used to calculate the spectral density of the phonon propagator @xmath179 from the peaks @xmath189 of the charge susceptibility @xmath63 . in fact , the numerical evaluation of eq . ( [ eq : rho_b ] ) yields the correct result for the phonon propagator and the lattice fluctuations , as shown in fig . [ x2_av_u00_gx.eps ] for the case of the pure holstein model . the condition @xmath173 has , of course , been used explicitly in the derivation of eq . ( [ eq : rho_b ] ) . the problem of the direct evaluation of eq . ( [ eq : d_chic ] ) for the phonon propagator is the numerical error in @xmath63 . in fact , the truncation of the hilbert space in the nrg procedure entails that the condition ( [ eq : chic_w0 ] ) is not exactly met . therefore @xmath63 has no root exactly at @xmath168 . this leads to a double pole of @xmath31 at @xmath168 and to a distorted result for the phonon propagator at energies close to @xmath8 , when derived from eq . ( [ eq : d_chic ] ) . lattice fluctuation , as calculated by integrating @xmath69 from eq . ( [ eq : d_chic ] ) are strongly overestimated for the same reason . in this appendix we derive an expression linking the average local displacement @xmath191 to the local electron density . for this derivation , we modify the hamiltonian by coupling an extra term linear in @xmath192 at each site . the extra term in the hamiltonian reads @xmath193 with the coupling @xmath194 . upon differentiating the expression for the partition function @xmath195 with respect to @xmath196 , we obtain @xmath197 we then apply a canonical transformation to the hamiltonian , @xmath198 , with @xmath199 given by @xmath200 this is a displaced oscillator transformation at each lattice site . the phonon and electron operators are transformed as @xmath201 after some algebra we find that the terms depending on @xmath202 in the transformed hamiltonian @xmath203 read @xmath204 the partition function @xmath195 is , of course , not changed by the canonical transformation of the hamiltonian . upon expressing @xmath195 in terms of @xmath205 , differentiating with respect to @xmath196 and then putting @xmath206 we find the result @xmath207 which relates the average local displacement to the local average electron density
ion pressure anisotropies are ubiquitous in heliospheric and astrophysical plasmas . in the absence of coulomb collisions , the magnetic moment of ions , @xmath4 ( @xmath5 , where @xmath6 is the ion velocity perpendicular to the local magnetic field @xmath7 and @xmath8 ) is an adiabatic invariant . thus , if ion collisions are infrequent , the ion velocity distributions parallel and perpendicular to the magnetic field decouple , making @xmath9 ( where @xmath10 and @xmath11 are the pressure components of ions perpendicular and parallel to @xmath7 ) . examples of systems where ion pressure anisotropies are important are low - luminosity accretion disks around compact objects @xcite , the intracluster medium ( icm ) @xcite , and the heliosphere ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in these systems the ion pressure anisotropy is expected to play a key role in the large scale dynamics of the plasma . for instance , @xcite pointed out that pressure anisotropies can give rise to an anisotropic stress that can contribute to angular momentum transport and plasma heating in low - luminosity accretion disks . this physics is not included in standard mhd models of accretion disks . velocity space instabilities limit the growth of the pressure anisotropy and give rise to small scale fluctuations in the magnetic field . these fluctuations can , in turn , affect the large scale transport properties of the plasma by modifying the mean free path of particles . magnetic field amplification in a collisionless system generically drives @xmath12 . there are two ion - scale instabilities that can be excited in this regime : the mirror and the ion - cyclotron ( ic ) instabilities @xcite . the mirror instability consists of non - propagating , strongly compressional modes . their fastest growing wave vectors @xmath13 are oblique to @xmath7 , with the magnetic variations parallel to @xmath7 being much larger than the perpendicular fluctuations , @xmath14 . for a bi - maxwellian distribution of ions , and assuming cold electrons , the threshold condition for mirror instability growth is given by @xmath15 , where @xmath16 ( @xmath17 ) is the ion temperature perpendicular ( parallel ) to @xmath7 , and @xmath18 @xcite . the ic instability , on the other hand , consists of transverse electromagnetic waves , with the fastest growing @xmath13 being preferentially parallel to @xmath7 ( e.g. , * ? ? ? whether the ic or mirror instability sets in first depends on how fast these instabilities grow for a given set of plasma conditions . @xcite showed that for @xmath19 , the growth rate of the ic instability is @xmath20 , where @xmath21 is the cyclotron frequency of the ions . in addition , for @xmath22 , the threshold anisotropy depends very weakly on @xmath23 . in the linear regime , the dominant instability will be determined by which threshold is reached first as @xmath7 is amplified . these estimates for the threshold conditions imply that , in the linear regime , the ic instability should dominate for @xmath24 , while the mirror instability should dominate for @xmath25 . although both the mirror and ic instabilities have been extensively studied in the linear regime , a complete theory of their nonlinear evolution and saturation is still lacking . in this paper , we are particularly interested in the question of how the mirror and ic modes behave after the initial phase of exponential growth . indeed , in most astrophysical scenarios where the magnetic field is amplified , the growth of the field occurs on time scales much longer than the growth time of the relevant kinetic instabilities . therefore , most of the evolution of the velocity space instabilities happens in a nonlinear regime , where the conditions typically assumed in linear studies , such as a homogeneous plasma or a bi - maxwellian distribution of particle velocities , are not necessarily satisfied . moreover , if the field is amplified by order unity or more , a quasi - linear analysis is not applicable . one of the key questions we are interested in addressing in this case is whether the mirror instability saturates via pitch - angle scattering that violates magnetic moment conservation @xcite or via a nearly @xmath4-conserving rearrangement of the magnetic field , with fluctuations @xmath26 @xcite . we shall see that both of these saturation processes can in fact be important . the nonlinear regime of the ic and mirror instabilities has been studied both theoretically ( e.g. , * ? ? ? * ; * ? ? ? * ) and numerically @xcite . in most numerical studies , however , the pressure anisotropy is treated as an initial condition , which decreases as the instabilities grow and saturate . thus , in these approaches the nonlinear saturation can not be followed for a time much longer than that of the initial exponential growth . moreover , many of the previous numerical studies have focused on one - dimensional simulations . in this paper we study the long term , nonlinear evolution of the mirror and ic instabilities using two - dimensional particle - in - cell ( pic ) simulations in driven systems . in order to self - consistently explore a time much longer than the initial exponential growth , we continuously induce the growth of @xmath27 by amplifying the mean field @xmath28 during the simulation . we concentrate on amplification by incompressible plasma motions . thus , our simulations impose a shear velocity on the plasma , which sustains the growth of the magnetic field and maintains an overall positive pressure anisotropy . this growth is intended to mimic magnetic field fluctuations in a turbulent plasma and mhd - scale instabilities that amplify the magnetic field , like the magnetorotational instability ( mri ; * ? ? ? * ) or convective instabilities driven by anisotropic thermal conduction in dilute plasmas @xcite . we suspect that our results are also relevant to cases in which perpendicular heating or parallel cooling drives a plasma mirror and/or ic unstable , as in , e.g. , the solar wind . @xcite conducted an analogous study to ours using hybrid simulations . in their study , expansion of the simulation box decreased the mean magnetic field , so that on average @xmath29 . this in turn led to the growth and nonlinear saturation of the firehose instability . also , @xcite conducted a similar hybrid simulation study where the simulation box is expanded along the directions parallel and perpendicular to @xmath7 . with this setup , @xmath10 becomes larger than @xmath11 on average , and both the mirror and ic instabilities can grow and reach saturation . our work is complementary to these hybrid simulation studies and focuses on the process in which the condition @xmath12 is achieved by field amplification due to shearing plasmas . we also consider higher values of @xmath30 @xmath31 , relevant to accretion disks , the icm , and the near - earth solar wind . our calculations are less directly applicable to heliospheric measurements of mirror modes driven unstable by rapid heating through the earth s bow shock ( e.g. , * ? ? ? this is because we assume that the pressure anisotropy is generated on a timescale long compared to the ion cyclotron timescale , which is probably not true at collisionless shocks . our paper is organized as follows . in [ sec : numsetup ] we describe the simulation set up , emphasizing the key physical and numerical parameters . [ sec : results ] shows our results and [ sec : conclu ] presents our conclusions . throughout the paper we compare some of our simulation results with linear theory predictions appropriate to our pic simulations . these have artificially low ion to electron mass ratios ( @xmath32 . the linear theory results are based on the linear vlasov solver developed in @xcite . during the completion of this work , @xcite presented calculations of firehose and mirror saturation in shearing plasmas with very similar results to those that we present here . we use the electromagnetic , relativistic pic code tristan - mp @xcite in two dimensions . the simulation box consists of a square box in the @xmath33 plane , containing plasma with a homogeneous magnetic field @xmath34 initially pointing along the @xmath35 axis . since we want to simulate a magnetic field that is being amplified in an incompressible way , we impose a shear motion of the plasma so that the mean particle velocity is @xmath36 , where @xmath37 is a shear parameter with units of frequency and @xmath38 is the distance along @xmath35 . from flux conservation , the @xmath39-component of the mean field evolves as @xmath40 , implying a net growth of @xmath41 . although we present simulations resolving the @xmath33 plane only , we also tried runs where the @xmath42 plane was resolved . in those cases the isotropization efficiency was substantially lower this is because , if the @xmath42 plane is resolved , the growing component of @xmath7 is perpendicular to the plane of the simulation . this way the angle between the relevant wave vectors , @xmath13 , and @xmath7 can not be 0 . this over - constrains the wave vectors that are allowed to exist , artificially reducing the isotropization efficiency . an important physical parameter in our simulations will be the ratio of the initial ion cyclotron frequency to the shear frequency , @xmath43 . we refer to this as the magnetization . in typical astrophysical environments @xmath44 . due to computational constraints , however , we will use values of @xmath43 much smaller than expected in reality , although still satisfying @xmath45 . the dependence of our results on the ratio @xmath43 will be carefully assessed . ccccccccc beta6mag93 & 1 & 6 & - & 93 & 0.05 & 14 & 30 & 12 + beta6mag670a & 1 & 6 & - & 670 & 0.15 & 14 & 10 & 18 + beta6mag670b & 1 & 6 & - & 670 & 0.15 & 14 & 10 & 24 + beta20mag93a & 1 & 20 & - & 93 & 0.05 & 10 & 60 & 35 + beta20mag93b & 1 & 20 & - & 93 & 0.05 & 10 & 30 & 35 + beta20mag670a & 1 & 20 & - & 670 & 0.05 & 7 & 30 & 36 + beta20mag670b & 1 & 20 & - & 670 & 0.05 & 7 & 30 & 48 + beta20mag2000 & 1 & 20 & - & 2000 & 0.05 & 7 & 50 & 55 + beta80mag670 & 1 & 80 & - & 670 & 0.05 & 5 & 30 & 24 + beta80mag1340 & 1 & 80 & - & 1340 & 0.05 & 5 & 30 & 24 + betai20magi240mass10 & 10 & - & 20 & 240 & 0.05 & 5 & 20 & 25 [ table:1d ] in standard mhd simulations where shear plasma motions are imposed , shearing periodic boundary conditions would be used along the @xmath38 direction ( see , e.g. , * ? ? ? in that case , the flow velocities at the @xmath38-boundaries of the box are matched using galilean transformations of the fluid velocity . in the case of relativistic pic simulations this can not be done in a self - consistent way . the reason is that , under a relativistic change of reference frame , the current @xmath46 transforms in a way that is inconsistent with simply transforming the particle velocities . we instead implement _ shearing coordinates _ , in which the grid moves with the shearing velocity @xmath47 . in this new frame the average plasma velocity vanishes everywhere in the box and simple periodic boundary conditions are allowed both in the @xmath38 and @xmath39 axes . in these shearing coordinates , maxwell s equations gain additional terms ( see appendix a of * ? ? ? * ) , becoming @xmath48 @xmath49 where @xmath50 is the speed of light , and @xmath51 is the electric field . in addition to the modifications to maxwell s equations , the forces on the particles also acquire an extra term : @xmath52 where @xmath53 , @xmath54 , and @xmath55 are the particle s momentum , velocity , and charge , respectively . the third term on the right hand side of equation [ eq : ind ] and the fourth term on the right hand side of equation [ eq : amp ] are proportional to time and arise from the motion of the shearing coordinate grid points with respect to those of the non - shearing coordinates . indeed , as time goes on , the @xmath38 axis of the shearing coordinates is gradually tilted ( as seen from the non - shearing frame ) ; therefore the @xmath56derivatives of the fields ( @xmath57 ) in the shearing coordinates must include a time dependent term that accounts for the evolution of the @xmath38 axis . the last terms in equations [ eq : ind ] and [ eq : amp ] , which are proportional to the coordinate @xmath39 , will be neglected for the following reasons . in our simulations the box size will typically be a few times the larmor radius of the ions , @xmath58 . thus , @xmath59 ( where @xmath60 is the thermal velocity of the ions ) , which is much smaller than unity since we are interested in the regime @xmath61 and @xmath62 . as a result , the last term in equation [ eq : ind ] will be much smaller than the term on the left hand side , especially since @xmath63 . the last term of equation [ eq : amp ] , on the other hand , is not necessarily much smaller than the displacement current ( left hand side term ) , mainly because we expect @xmath64 . however , if the characteristic timescale for the mirror and ic modes is close to @xmath65 ( below we will check that this is indeed the case ) , the last term in equation [ eq : amp ] will be much smaller than the first term on the right hand side ( the ratio between these terms scales as @xmath66 ) , so we choose to neglect it in the limit @xmath67 . the fact that the neglected term can still be comparable to the displacement current implies that we may be excluding a relevant contribution to the charge density in gauss law . however , the physics of interest in this paper is non - relativistic and charge separation does not play a role . as a result , @xmath68 ( equivalent to @xmath69 in the _ non - shearing _ coordinate system ) . neglecting the terms proportional to @xmath39 is also required by the periodic boundary conditions in the y - direction . the existence of these terms at all is a consequence of the incompatibility of the galilean invariance of the shearing box and the lorentz invariance of the pic calculations . for consistency we will also neglect the term @xmath70 in equation [ eq : amp ] , which is also comparable to the displacement current ( in the case where @xmath51 evolves on a timescale close to @xmath65 ) can be integrated using simple time and space interpolations of @xmath71 . this way , after this modification is implemented , the numerical algorithm used by tristan - mp continues to be second order accurate in time and space . ] our set of simulations are summarized in table 1 . the physical setup is defined by the magnetization parameter , @xmath43 ( the ratio of the initial cyclotron frequency to the shear rate ) , the ratio between the initial particle and magnetic pressures , @xmath72 , the mass ratio between ions and electrons , @xmath73 , and the initial alfvn velocity of the plasma @xmath74 , relative to the speed of light @xmath50 , where @xmath75 and @xmath76 is the mass density . the numerical parameters of our runs are defined by the spatial resolution ( @xmath77 ) , box size relative to the ion larmor radius ( @xmath78 ) , and number of particles per cell ( @xmath79 ) , where @xmath80 is the plasma frequency of the electrons and @xmath81 is the spacial separation of the grid points . the simulations shown in this paper are described in table 1 . these simulations are a subset of a higher number of runs , where we used different combinations of numerical and physical parameters , which confirmed numerical convergence . we want to explore the regime @xmath82 . in order to do so , most of our runs use the mass ratio @xmath83 . since for @xmath83 both species will be essentially indistinguishable , we initially give ions and electrons maxwellian energy distributions with the same temperature , and use the parameters @xmath84 and @xmath85 to quantify the pressure of the plasma perpendicular and parallel to @xmath7 . simulations with @xmath83 of course do not allow us to study the physics of electron isotropization . however , in [ sec : compmime ] we use an example of our @xmath86 runs to show that @xmath83 calculations reproduce the ion - related phenomena fairly well , with @xmath87 and @xmath88 playing the same role of @xmath89 and @xmath90 in the @xmath86 runs . we defer a detailed study of electron scale pressure isotropization to future work . we divide our runs into cases with three different initial betas : @xmath91 , 20 , and 80 . our analysis focuses on the nonlinear structure of the ic / mirror generated fluctuations , the evolution of the pressure anisotropy @xmath92 , and the conservation of the ion magnetic moment . runs . panels @xmath93 and @xmath94 show the evolution of @xmath95 ( @xmath96 ; solid ) and @xmath97 ( @xmath98 ; dotted ) for runs beta6mag93 and beta6mag670a , respectively ( @xmath99 and @xmath100 correspond to black , red , and green , respectively ) . for the same runs , panels @xmath50 and @xmath101 show @xmath102 ( @xmath103 ; black - solid ) , compared with the linear mirror and ic thresholds ( @xmath104 and @xmath105 , in red and green , respectively ) for pair plasma and growth rates @xmath106 and @xmath107 , respectively . panels @xmath108 and @xmath109 show @xmath110 and @xmath111 , where the subscripts @xmath112 and @xmath113 denote the components parallel and perpendicular to @xmath28 , respectively . after the plasma pressure anisotropy exceeds the mirror and ic thresholds , there is an initial phase of exponential growth followed by a secular phase . the pressure anisotropy initially grows but saturates at the linear threshold for the mirror / ic instabilities.,width=340 ] runs with @xmath114 and @xmath115 ( beta6mag93 and beta6mag670a in table [ table:1d ] ) . we show the evolution of the true average magnetic moment @xmath116 @xmath117 ( black line ) and an effective global magnetic moment , @xmath118 @xmath119 ( red line ) , where the subscript @xmath120 denotes an average over all particles . the ion magnetic moment @xmath116 is conserved until @xmath121 , at which point it decreases on the same timescale as the mean magnetic field ( see fig . [ fig : muderivs ] ) . the close similarity between @xmath116 and @xmath118 shows that there are not strong correlations between the particle velocity @xmath122 and magnetic field @xmath123 ; such correlations are more prominent at higher @xmath124 where the ic instability is subdominant ( see fig . [ fig : mubeta20]).,width=340 ] , for different @xmath124 , compared to the growth rate of the mean magnetic field @xmath125 ( black line ; this quantity is the same for the different @xmath124 runs ) . @xmath126 is shown for our simulations with @xmath127 and 80 ( beta6mag670a , beta20mag670a , and beta80mag1340 , respectively ) . the vertical - dotted lines mark the beginning of the saturated state for each run based on when the growth of the fluctuations becomes roughly secular . in the saturated state , the magnetic moment changes on a timescale comparable to that of the mean magnetic field ( to within @xmath128 ) . at earlier times , however , the magnetic moment is reasonably well conserved.,width=283 ] as discussed above , the expectation is that the ic instability should play an important role in the @xmath91 case , and it should become significantly less important in the @xmath129 and 80 cases . the contributions from the ic and mirror modes can be seen from figure [ fig : fieldsbeta6 ] . this figure shows 2d images of the field fluctuations @xmath130 ( @xmath131 ; @xmath132 stands for volume average " ) and the plasma density fluctuations @xmath133 for run beta6mag670@xmath50 ( @xmath134 " stands for the component @xmath38 , @xmath39 , or @xmath100 of @xmath135 ) at two different times : @xmath136 ( upper panels ) and @xmath137 ( lower panels ) . we can see that at the earlier time ( @xmath136 ) the three components of @xmath135 have about the same amplitude ( see figures [ fig : fieldsbeta6]@xmath93 , [ fig : fieldsbeta6]@xmath94 , and [ fig : fieldsbeta6]@xmath50 ) , but appear to be dominated by different mode orientations , indicating the simultaneous presence of ic and mirror modes . the mirror modes are expected to have @xmath135 mainly in the plane of the simulation ( the plane of @xmath13 and @xmath28 ; * ? ? ? on the other hand , the ic modes have the three components of @xmath135 of comparable magnitude , so their presence can be most clearly revealed by @xmath138 ( figure [ fig : fieldsbeta6]@xmath50 ) . indeed , while @xmath139 and @xmath140 are dominated by a combination of oblique ( mirror ) and quasi - parallel ( ic ) modes , @xmath138 appears to be dominated by only quasi - parallel modes . thus , initially the mirror and ic instabilities contribute comparably to @xmath135 . at the later time ( @xmath137 ) , however , the fluctuations become dominated by an oblique wave vector , with the ic instability playing a subdominant role . the density fluctuations at all times are @xmath141 , and seem to correlate primarily with the mirror modes . the relative contribution of the mirror and ic instabilities can also be seen from figure [ fig : beta6 ] , which shows the time evolution of different volume - averaged quantities for runs with @xmath142 ( first column ; run beta6mag93@xmath93 ) and @xmath143 ( second column ; run beta6mag670@xmath93 ) . in figures [ fig : beta6]@xmath93 and [ fig : beta6]@xmath94 we plot @xmath95 for the two magnetizations . in both cases there is an initially exponential growth of @xmath144 until @xmath145 . both instabilities have about the same growth rates , which are @xmath146 . note that the dominant growth rate of the mirror / ic modes is nearly the same in the two simulations with different values of @xmath21 . the growth rate is thus set by the background shear rate @xmath37 instead of @xmath21 . this can be understood by noting that the smaller growth rate modes have a smaller anisotropy threshold . therefore , as @xmath123 and @xmath147 grow , modes with smaller growth rate will begin to grow first . this implies that the dominant modes will be those that can reach a significant amplitude to stop the growth of @xmath147 on the shear timescale @xmath65 . after the exponential phase , the growth becomes dominated by the mirror modes and is secular ( as predicted by * ? ? ? this transition from exponential to secular is clearly seen at @xmath148 for @xmath149 in figure [ fig : beta6 ] . despite the similar exponential growth rate for the mirror and ic instabilities in the linear regime , the mirror modes are more robust and dominate in the nonlinear regime . this dominance of the mirror modes is remarkable because it occurs even for @xmath24 , where the linear analysis predicts an important contribution of the ic instability ( in the nonlinear regime the mean magnetic energy has been amplified by about one order of magnitude , so that @xmath24 ) . we suggest below ( fig . [ fig : spectrabeta6 ] ) that departure from the bi - maxwellian energy distribution assumed in the linear instability analysis is likely playing an important role in this sub - dominance of the ic instability . run beta6mag670a at two different times . _ top : _ energy spectra as a function of the energies perpendicular and parallel to the magnetic field , @xmath150 ( black - dotted line ) and @xmath151 ( black - solid line ) . _ bottom : _ energy spectra as a function of @xmath152 . bi - maxwellian fits are shown in red . there is a clear deviation from the bi - maxwellian at the highest energies , which grows in time . there is , however , little deviation from a maxwellian at @xmath153 , in contrast to some of the 1d runs discussed in the appendix . comparison with figure [ fig : spectrabeta20 ] suggests that the high energy deviation from maxwellian is due to the ic instability , since it is less prominent at higher @xmath154 when the ic instability is sub - dominant.,width=340 ] one difference in the evolution of @xmath135 for the two different magnetizations is that the end of the exponential phase occurs at somewhat higher amplitude in the @xmath142 case . this difference implies that , for smaller values of @xmath43 , the mirror and ic fluctuations are less efficient in suppressing the pressure anisotropy , requiring larger values of @xmath144 . at the end of this section , we will explain in further detail this dependence of @xmath144 on magnetization by focusing on the mechanism by which the isotropization occurs . one consequence of the difference in @xmath144 is that the lower magnetization runs take longer to reach the point when @xmath102 ( @xmath103 ) saturates . thus in the @xmath142 run the anisotropy @xmath102 has more time to grow , reaching larger values before saturating . this can be seen in the evolution of @xmath102 , shown in black lines in figures [ fig : beta6]@xmath50 and [ fig : beta6]@xmath101 for runs beta6mag93@xmath93 and beta6mag670@xmath93 , respectively . for both magnetizations , there is an initial overshoot in @xmath102 , whose amplitude is larger for the @xmath142 run . after the overshoot , @xmath102 evolves in a similar way for both magnetizations . since in the nonlinear regime @xmath135 is dominated by mirror modes , after the saturation one would expect the pressure anisotropies to behave according to the marginal stability conditions of the mirror instability . since in our simulations both the ic and mirror modes grow at @xmath155 , we calculated the ic and mirror thresholds for equivalent growth rates in the pair plasma case with magnetizations @xmath142 and @xmath143 . thus , figures [ fig : beta6]@xmath50 and [ fig : beta6]@xmath101 show that the threshold for mirror ( red ) and ic instability ( green ) in the cases @xmath156 and @xmath157 , respectively . we see that the mirror threshold coincides fairly well with the saturated @xmath102 for both magnetizations . these results imply that the linear marginal stability condition for the mirror instability is fairly accurate in determining @xmath102 in the nonlinear regime . in realistic astrophysical setups , where @xmath21 can be many orders of magnitude larger than @xmath37 , the expectation is that @xmath102 will follow the @xmath158 mirror threshold , given by @xmath159 . this differs from the results shown in figure [ fig : beta6 ] , which are only appropriate for pair plasmas and modest magnetization . ( red ) for @xmath160 and @xmath149 ( run beta6mag670a ) . the good agreement between the two results implies that the anisotropic stress coupling to the background shear is the primary mechanism for particle heating.,width=283 ] as discussed in the introduction , it is not known whether the mirror instability saturates via pitch - angle scattering ( see , for instance , * ? ? ? * ) or whether it cancels the appearance of a pressure anisotropy by substantially modifying the structure of the magnetic field @xcite . in the latter case @xmath135 should grow secularly until @xmath161 , with the breaking of @xmath162invariance not necessary for the regulation of @xmath163 . figures [ fig : beta6]@xmath108 and [ fig : beta6]@xmath109 show the evolution of @xmath111 and @xmath110 for @xmath114 and @xmath115 , respectively , where @xmath164 and @xmath165 refer to the magnetic field component perpendicular and parallel to the mean field @xmath28 . we see that @xmath166 @xmath167 dominates in the nonlinear regime , as expected from the larger amplitude of the oblique mirror modes , and @xmath168 the saturation amplitude is @xmath169 , which favors the scenario where the non - conservation of @xmath116 is the key mechanism for the isotropization of the plasma pressure , at least if the background field has been amplified significantly . to quantify the time variation of the average ion magnetic moment we define @xmath170 where @xmath171 stands for average over all the particles . @xmath116 corresponds to the actual average of the magnetic moment . thus if a nearly @xmath162conserving process dominates the mirror saturation , @xmath116 should remain fairly constant . @xmath118 , on the other hand , can be thought as an _ effective _ global magnetic moment that ignores the fluctuations in @xmath172 or the correlation between @xmath122 and @xmath123 due to particles collecting in mirrors . @xmath118 necessarily decreases since the mirror and ic instabilities suppress the growth that @xmath173 would have if the particles were only affected by the mean field @xmath28 . thus , if @xmath116 is nearly conserved , @xmath116 and @xmath118 should evolve quite differently , with @xmath174 . on the other hand , if there is significant pitch angle scattering @xmath175 and both should decrease on the same timescale that @xmath176 increases . figures [ fig : beta6b]@xmath93 and [ fig : beta6b]@xmath94 show @xmath116 and @xmath118 for our @xmath160 runs with magnetizations @xmath142 and @xmath143 , respectively . we see that for both magnetizations the difference between @xmath116 and @xmath118 is very small ( only @xmath177 a few @xmath178 difference ) . this implies that there is little spatial correlation between @xmath122 and @xmath123 . we shall see below that @xmath116 and @xmath118 differ somewhat more at higher @xmath154 where the mirror instability dominates over the ic instability . figure [ fig : muderivs ] compares the rate of change of the ion magnetic moment with that of the mean magnetic field for runs with @xmath1796 , 20 , and 80 . this comparison is important since it is the evolution of the mean field that is driving the velocity space instabilities in our calculations . if the total thermal energy varies on a timescale long compared to the mean magnetic field , maintaining marginal stability to the mirror instability via pitch - angle scattering implies @xmath180 $ ] . the term @xmath181 is exactly @xmath182 for the canonical high magnetization mirror instability threshold but is somewhat different in our pair plasma , modest magnetization simulations ( hence the use of @xmath183 ) . this expression for @xmath184 implies that pitch - angle scattering should lead to the magnetic moment varying at a rate slightly less than that of the mean magnetic field . this is consistent with our numerical results in figure [ fig : muderivs ] in the saturated state where the magnetic field grows secularly in time . at earlier times , when the magnetic field fluctuations due to the mirror and ic instability are smaller , @xmath116 is approximately conserved even though the mirror and ic are present and grow exponentially . in particular , there is a temporal lag of @xmath185 between the onset of the exponential growth of the mirror and ic instabilities and the onset of pitch angle scattering that decreases @xmath116 . our interpretation of this result is that the mirror / ic fluctuations must reach a sufficient amplitude , roughly @xmath186 , in order for pitch angle scattering to be effective . for the mirror instability , this violation of @xmath116 conservation by finite amplitude fluctuations is likely due to the stochasticity of particle orbits that sets in for large amplitude ( albeit low frequency ) fluctuations ( e.g. , @xcite ) . for the ion cyclotron instability , the need for finite amplitudes before significant scattering sets in is a consequence of the well - known result that the scattering rate for cyclotron - frequency fluctuations is @xmath187 . the breaking of local @xmath116-invariance for this calculation ( @xmath87 , @xmath188 ) might be affected by the subdominant ( but still significant ) contribution of the ic modes to the magnetic fluctuations . however , figure [ fig : muderivs ] shows that the magnetic moment varies at the same rate as the mean magnetic field for all of the @xmath124 we have simulated . we will discuss the higher @xmath124 results in more detail in the next two subsections . at @xmath189 for @xmath190 and magnetization @xmath114 ( panel @xmath93 ; run beta20mag93b ) and @xmath149 ( panel @xmath94 ; run beta20mag670b ) . panels @xmath50 and @xmath101 show the same quantities at @xmath191 . the black arrows show the direction of @xmath28 . the higher magnetization run produces mirror modes whose wavevectors are more perpendicular to @xmath28 and with longer wavelengths . a migration to larger wavelengths in time can also be seen.,width=340 ] an interesting question is whether the non - linear evolution of the velocity space instabilities makes the energy distribution of the particles substantially different from a bi - maxwellian spectrum . this question is particularly relevant in terms of understanding the suppression of the ic modes in the nonlinear regime as @xmath87 goes from 6 to @xmath192 . indeed , it has been shown@xcite that a significant departure from a bi - maxwellian distribution can increase the anisotropy threshold for the growth of the ic modes and that a change of this kind is expected given the resonant character of the ic instability . figure [ fig : spectrabeta6 ] shows that the particle spectrum does differ somewhat from a bi - maxwellian for our @xmath160 run , mainly due to a significant excess of particles at the highest energies . in future work , we will study whether or not the spectral variation in figure [ fig : spectrabeta6 ] can completely account for the decrease in the amplitude of the ic modes in the saturated state . figure [ fig : spectrabeta6 ] also shows that the ion velocity distribution does not deviate significantly from a bi - maxwellian at low @xmath152 , as has been found in some 1d studies of the saturation of the mirror instability ( e.g. , @xcite ) . we discuss this in more detail in the appendix . finally , figure [ fig : anistress ] shows the evolution of the total heating rate of the particles as a function of time ( black line ) . for comparison , in the red line we plot the particle heating rate by the anisotropic stress , @xmath193 @xcite . we see a reasonable agreement between the two results , implying that the anisotropic stress plays the dominant role in particle heating . other mechanisms , like wave - particle interaction , must play a secondary role in the total particle energization . this is true for all of the @xmath124 we have simulated . we now consider a somewhat weaker magnetic field case with @xmath129 . in this case , the field fluctuations are dominated by the mirror instability at all times . this can be seen in figures [ fig : beta20b2s]@xmath93-@xmath50 which show the evolution of the three components of @xmath194 for @xmath190 runs with magnetizations @xmath43=93 , 670 , and 2000 , respectively ( runs beta20mag93@xmath93 , beta20mag670@xmath93 and beta20mag2000@xmath93 in table 1 ) . in all three cases @xmath195 is small compared to @xmath194 , showing the subdominant role of the ic instability . figures [ fig : beta20b2s]@xmath93-@xmath50 also show the characteristic transition between exponential growth ( with @xmath196 ) and secular growth ( with @xmath197 ) for the mirror modes . as in the @xmath91 case , the transition between the exponential and secular regimes occurs at smaller amplitudes and in a smoother way in the more strongly magnetized runs . figures [ fig : beta20b2s]@xmath101-@xmath109 also show that , as in the case of @xmath91 , after the overshoot the pressure anisotropy follows the marginal stability condition for the mirror modes reasonably well , particularly at higher magnetization . figures [ fig : beta20b2s]@xmath198-@xmath199 show @xmath200 and @xmath201 as a function of time . while @xmath200 is about the same for the three magnetizations , @xmath201 is significantly smaller in the higher magnetization runs . the relative magnitude of @xmath202 and @xmath203 is a measure of the orientation of the mirror s dominant wave vector , @xmath13 , with respect to @xmath28 : a large @xmath204 ratio implies that @xmath13 and @xmath28 are nearly perpendicular . the results of figures [ fig : beta20b2s]@xmath198-@xmath199 imply that for larger magnetization , @xmath13 and @xmath28 are more perpendicular , which is consistent with linear calculations @xcite . and @xmath94 show the evolution of the ion magnetic moment @xmath116 for @xmath190 runs with @xmath114 ( beta20mag93a ; panel @xmath93 ) and @xmath149 and @xmath205 ( beta20mag670a and beta20mag2000 ; panel @xmath94 ) . panels @xmath50 and @xmath101 compare the corresponding rate of change of @xmath116 with that of the mean magnetic field for runs beta20mag93a and beta20mag670a , respectively . all plots show the evolution of the true average magnetic moment @xmath116 @xmath117 ( black line ) and an effective global magnetic moment , @xmath118 @xmath119 ( red line ) , where the subscript @xmath120 denotes an average over all particles . the magnetic moment begins to decrease when the fluctuations reach @xmath206 ( see fig . [ fig : beta20b2s ] ) . the difference between @xmath116 and @xmath118 is more significant than for @xmath160 ( cf fig . [ fig : beta6b ] ) . this is a consequence of particles bunching in mirrors , which leads to a correlation between @xmath122 and @xmath123 . nonetheless , the ion magnetic moment changes at about the same rate as the background magnetic field after saturation at @xmath207 , indicating that pitch angle scattering regulates the nonlinear saturation on sufficiently long timescales.,width=340 ] the orientation of the modes in the nonlinear stage can also be seen directly in figure [ fig : fieldsbeta20 ] , which shows the spatial distribution of @xmath208 at two different times for the runs with @xmath142 and @xmath143 . figures [ fig : fieldsbeta20]@xmath93 and [ fig : fieldsbeta20]@xmath94 correspond to the time @xmath136 , while figures [ fig : fieldsbeta20]@xmath50 and [ fig : fieldsbeta20]@xmath101 correspond to @xmath137 . for each magnetization , the mirror fluctuations are dominated by two modes that are symmetric with respect to the magnetic field , and that are more oblique at higher magnetization . note also that the wavelength of the dominant modes ( in units of the larmor radius of the particles , @xmath58 ) is larger for larger magnetization and that the modes tend to grow in wavelength as time goes on . figure [ fig : mubeta20 ] shows the evolution of the ion magnetic moment @xmath116 and the rate of change of @xmath116 relative to that of the mean magnetic field for our @xmath190 calculations . a comparison of figures [ fig : mubeta20 ] & [ fig : beta20b2s ] shows that the magnetic moment is reasonably conserved until the magnetic fluctuation amplitude is close to ( though somewhat less than ) its saturated value . as in our @xmath160 calculations , this implies that there is a time lag of @xmath209 between the onset of the mirror instability and the onset of significant pitch - angle scattering ( and its associated decrease in @xmath116 ) . note , however , that for @xmath129 the saturated amplitude of the fluctuations in the secular regime , @xmath210 , is larger than in the @xmath91 case , for which @xmath211 . this implies that when _ only _ the mirror instability plays a significant role , a somewhat larger fluctuation amplitude is necessary for efficient pitch - angle scattering . we will see below that @xmath210 continues even for @xmath212 ; the saturated amplitude of the mirror modes is thus fairly independent of the beta of the plasma as long as the ic instability is not significant ( @xmath213 6 ) . figure [ fig : mubeta20 ] also shows that for @xmath190 , the true ion magnetic moment @xmath116 decreases somewhat more slowly than the effective global magnetic moment @xmath118 ( by @xmath214 ) . this was not seen in our @xmath160 calculations ( figs . [ fig : beta6b ] ) . the modest difference between @xmath116 and @xmath118 indicates that the nonlinear saturation of the mirror instability involves correlations between @xmath122 and @xmath123 , i.e. , particles bunching in mirrors . this effect is independent of magnetization , as can be seen from the solid and dotted lines in figure [ fig : mubeta20]@xmath94 ( corresponding to @xmath215 670 and 2000 , respectively ) . run beta20mag670a at two different times during the nonlinear regime . _ top : _ energy spectra as a function of the energies perpendicular and parallel to the magnetic field , @xmath150 ( black - dotted line ) and @xmath151 ( black - solid line ) . _ bottom : _ energy spectra as a function of @xmath152 . in all cases , a bi - maxwellian energy spectrum ( red ) provides a good approximation to the numerical simulations.,width=340 ] figure [ fig : spectrabeta20 ] shows that the particle energy spectrum stays fairly close to bi - maxwellian in our @xmath190 runs , regardless of the magnetization . since these simulations are almost purely mirror dominated , this suggests that the mirror modes provide rather momentum - independent pitch - angle scattering to the particles . as a result , it is likely that ic modes produce the deviations from a bi - maxwellian distribution seen in figure [ fig : spectrabeta6 ] . figure [ fig : spectrabeta20 ] also shows that the parallel velocity spectrum @xmath216 remains nearly maxwellian during most of the saturated regime ( with only a transient @xmath217 decrease for @xmath218 especially at the end of the exponential growth regime ) . the lack of significant flattening of @xmath216 ( which contrasts with the prediction of * ? ? ? * ) is analyzed in further detail in the appendix . for @xmath212 we focus on simulations with @xmath143 and 1340 , respectively ( runs beta80mag670 and beta80mag1340 in table 1 ) . the results are essentially the same as for the @xmath129 case so we do not discuss them in detail . figure [ fig : beta80b2s ] shows that the mirror instability dominates the fluctuations in @xmath7 . the evolution of the pressure anisotropy @xmath102 contains the initial overshoot followed by a rather flat behavior . the amplitude of the overshoot is controlled by the magnetization , while the subsequent evolution is fairly well described by the threshold condition for mirror modes . runs with @xmath149 ( beta80mag670 ; left column ) and @xmath219 ( beta80mag1340 ; right column ) . panels @xmath93 and @xmath94 show the evolution of @xmath95 ( @xmath96 ; solid ) and @xmath97 ( @xmath98 ; dotted ) , respectively ( @xmath99 and @xmath100 correspond to black , red , and green , respectively ) . for the same runs , panels @xmath50 and @xmath101 show @xmath102 ( @xmath103 ; black - solid ) , compared with the linear mirror threshold in red ( @xmath104 ) for a pair plasma and growth rates @xmath107 and @xmath220 , respectively . panels @xmath108 and @xmath109 show @xmath110 and @xmath111 . the results shown here are very similar to the analogous @xmath129 results in figure [ fig : beta20b2s ] , consistent with the dominance of the mirror instability at high @xmath154.,width=340 ] as in the @xmath129 case , the saturation @xmath221 is @xmath222 . figure [ fig : mubeta80 ] shows the evolution of the ion magnetic moment for the @xmath223 simulations . the evolutions of @xmath116 and @xmath224 are the same for the two magnetizations presented ( @xmath149 and @xmath219 ) , and show the same behavior seen in the @xmath129 case , which is discussed in detail in the previous subsection . and @xmath94 show the evolution of the ion magnetic moment @xmath116 for our @xmath223 runs with @xmath149 and @xmath219 ( runs beta80mag670 and beta80mag1340 , respectively ) . panels @xmath50 and @xmath101 compare the rate of change of @xmath116 with that of the mean magnetic field . in all panels , we show the evolution of the true average magnetic moment @xmath116 @xmath117 ( black line ) and an effective global magnetic moment , @xmath118 @xmath119 ( red line ) , where the subscript @xmath120 denotes an average over all particles . the magnetic moment begins to decrease when the fluctuations reach @xmath206 ( see fig . [ fig : beta80b2s ] ) . all of the key results here are very similar to the @xmath129 results shown in figure [ fig : mubeta20].,width=340 ] , @xmath225 , and @xmath226 ( run betai20magi240mass10 ) . the ion results here are similar to the corresponding @xmath227 results in figures [ fig : beta20b2s ] and [ fig : mubeta20 ] , indicating that the pair plasma results capture most of the key ion physics . panel @xmath93 shows the evolution of @xmath228 and @xmath229 , in solid and dotted lines , respectively . the small contribution of @xmath195 implies the dominance of mirror modes . panel @xmath94 shows the evolution of @xmath111 ( red ) and @xmath110 ( black ) with @xmath230 , implying that the mirror modes have @xmath231 . panel @xmath50 shows the evolution of the ion pressure anisotropy @xmath232 ( black - solid ) . the linear mirror threshold for mirror modes with @xmath233 in a @xmath234 , @xmath235 plasma is also shown ( red line ) along with the analogous threshold for @xmath236 for comparison ( red dotted line ) . panel @xmath101 shows the evolution of @xmath116 ( @xmath237 , black - solid ) and @xmath118 ( @xmath238 , red - solid ) ; see eq . [ eq : mu ] and associated text for details.,width=340 ] in this section we show that our use of @xmath83 in the previous sections does not have a significant effect on the evolution of the ion pressure anisotropy under the influence of the ic and mirror instabilities . in figure [ fig : beta20b2smime4 ] we present the results for a simulation with @xmath235 , @xmath225 , and @xmath226 ( run betai20magi240mass10c ) , which is analogous to the previous @xmath83 simulations that use @xmath239 . a comparison of figure [ fig : beta20b2smime4]@xmath93 with figures [ fig : beta20b2s]@xmath93 , [ fig : beta20b2s]@xmath94 , and [ fig : beta20b2s]@xmath50 shows that for both mass ratios the evolution of @xmath135 is dominated by the mirror instability , with the initial exponential regime being followed by secular growth . figure [ fig : beta20b2smime4]@xmath94 shows the evolution of @xmath111 and @xmath110 . as in the @xmath83 case ( see figures [ fig : beta20b2s]@xmath198 , [ fig : beta20b2s]@xmath240 , and [ fig : beta20b2s]@xmath199 ) , @xmath230 in the secular stage , implying that mirror modes with @xmath231 are dominant . the maximum amplitude of @xmath241 is also in good agreement with the @xmath83 simulations . the evolution of the ion pressure anisotropy @xmath232 is shown in figure [ fig : beta20b2smime4]@xmath50 . as in the @xmath83 runs ( see figures [ fig : beta20b2s]@xmath242 ) , @xmath243 follows the linear threshold of the mirror instability fairly closely . this is seen in figure [ fig : beta20b2smime4]@xmath50 , which shows the linear mirror threshold expected for modes with @xmath233 , @xmath225 , and @xmath235 and 1836 ( red and dotted - red lines , respectively ) . the small difference between our results and the @xmath235 and 1836 thresholds strongly suggests that the use of @xmath83 for most of our study captures reasonably well the key physics of the mirror and ic instabilities in a real electron - ion plasma . finally , figure [ fig : beta20b2smime4]@xmath101 shows the evolution of @xmath116 ( black solid line ) and @xmath118 ( red solid line ) . the results are quantitatively similar to the @xmath227 results in figures [ fig : mubeta20]@xmath93 and [ fig : mubeta20]@xmath94 . this shows that the role played by pitch - angle scattering in the saturation of the mirror instability is well captured by our @xmath83 simulations . we have used pic simulations to study the nonlinear , long - term evolution of ion velocity space instabilities in collisionless plasmas . we have focused on instabilities driven by pressure anisotropy with @xmath244 : the ion cyclotron ( ic ) and mirror instabilities . these instabilities are expected to arise when turbulence and/or mhd instabilities amplify @xmath7 , rendering @xmath244 due to conservation of the ion magnetic moment @xmath4 . alternatively , in the solar wind @xmath245 can increase due to perpendicular heating and/or parallel cooling of the plasma . in contrast to previous studies , we do not consider the initial value problem of the evolution of a given initial ion pressure anisotropy . instead , we self - consistently induce the growth of an ion pressure anisotropy by continuously amplifying the mean magnetic field in our computational domain . this setup allows us to study the long - term , saturated state of the plasma , rather than the ( much shorter ) initial , exponential stage . the former is of most interest for the heliospheric and astrophysical applications of our work . we have focused on the regime @xmath246 and plasma magnetizations @xmath43 in the range @xmath247 ( the magnetization is the ratio of the initial ion cyclotron frequency to the background shear frequency ) . most of our simulations used @xmath83 , but we showed that our results for the ion scale physics are fairly independent of the mass ratio ( see [ sec : compmime ] ) . so long as @xmath45 , our results are also relatively independent of the precise value of the magnetization ( see , in particular , figs . [ fig : beta20b2s ] & [ fig : mubeta20 ] ) . thus we believe that our basic conclusions about the saturation of the mirror and ic instabilities in continuously driven systems can be applied to physical problems where @xmath21 is many orders of magnitude larger than @xmath37 . our primary results are as follows : * for the @xmath246 regime we have studied , the turbulent fluctuations in the saturated state are dominated by mirror modes . this is despite the fact that ic and mirror instabilities are both present at early phases ( and have similar linear instability thresholds ) when @xmath248 a few . a plausible explanation for this dominance of the mirror instability in the saturated state even at @xmath249 a few is that the ion velocity distribution departs from a bi - maxwellian spectrum in the presence of the ic instability . this can in turn increase the threshold for the ic instability @xcite making it less important than the mirror instability in the nonlinear regime . we indeed see more significant departures from a bi - maxwellian in the low @xmath30 regime where the ic instability is present ( figs . [ fig : spectrabeta6 ] & [ fig : spectrabeta20 ] ) . by contrast , the ion distribution function remains relatively bi - maxwellian in the high @xmath30 regime in which the mirror instability dominates over the ic instability at all times ( fig . [ fig : spectrabeta20 ] ) . the dominance of the mirror instability at @xmath248 a few is consistent with the nonlinear study of @xcite , which makes use of hybrid simulations of an expanding box . * small - scale fluctuations driven by the mirror instability initially grow exponentially but saturate in a secular phase in which @xmath250 ( figures [ fig : beta6 ] & [ fig : beta20b2s ] ) . the fluctuations in the saturated state have @xmath231 ( see , e.g. , figure [ fig : fieldsbeta6 ] ) consistent with linear theory expectations for mirror modes . * the ion pressure anisotropy @xmath232 in the saturated state is reasonably well described by the marginal stability condition for the mirror instability ( e.g. , figs [ fig : beta6 ] & [ fig : beta20b2s ] ) . note , however , that the marginal stability state of the mirror mode in our simulations is not the same as in real systems because of the comparable ion and electron masses and the smaller @xmath43 used in our calculations . this is important to bear in mind when comparing our results to heliospheric measurements . * the total thermal energy of the plasma increases in time at the theoretically predicted rate given the background shear in the plasma and the anisotropic stress associated with the pressure anisotropy ( fig . [ fig : anistress ] ) . this is consistent with the idea that the `` viscous '' heating rate in a collisionless plasma is set by how velocity space instabilities regulate the pressure anisotropy @xcite . * when the growth time of the mirror instability is long compared to the ion cyclotron period as is the case in our simulations where the shear slowly amplifies the mean magnetic field the ion magnetic moment is initially reasonably well conserved . this is fundamentally because there is no mechanism to violate magnetic moment conservation during the linear phase of mirror growth . so the modes continue to grow to larger and larger amplitudes , given the sustained free energy source created by the background velocity shear amplifying the magnetic field and continuously generating pressure anisotropy . this regime lasts until @xmath3 and @xmath251 , and is consistent with the secular regime " found by @xcite where @xmath116 is nearly conserved . * saturation of the mirror modes starts to happen once the fluctuations attain @xmath3 . in this stage the variation in magnetic field on the scale of the larmor radius is sufficiently large that the ion magnetic moment is no longer conserved . this is consistent with theories of stochastic ion motion in large amplitude turbulent fluctuations ( e.g. , * ? ? ? * ; * ? ? ? in this phase the ion magnetic moment changes on nearly the same timescale as the mean magnetic field ( fig . [ fig : muderivs ] ) , which is consistent with pitch angle scattering maintaining marginal stability . in this stage the magnetic fluctuations only experience a modest additional growth , reaching a maximum amplitude of @xmath252 . * the need for finite amplitude fluctuations with @xmath3 in order to stop magnetic moment conservation implies the existence of a the temporal delay of @xmath185 between the onset of the ic and mirror instabilities and the onset of efficient pitch angle scattering ( @xmath37 is the rate at which the mean magnetic field is growing ) . our results have a number of consequences for modeling the large scale dynamics of nearly collisionless astrophysical plasmas and for interpreting heliospheric measurements . we briefly mention a few of these applications but defer a more detailed analysis to future work . our results are consistent with a number of in situ heliospheric measurements . for instance , our maximum saturated value of @xmath253 is in reasonable agreement with mirror modes observations in different environments like planetary magnetosheaths ( where large amplitude mirror modes are frequently found . see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the solar wind ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and the heliosheath ( e.g. , * ? ? ? * ; * ? ? ? . corresponds to @xmath254 , where @xmath132 stands for volume average . this means that our results still allow for _ local _ magnetic fluctuations of even larger amplitude . it is also important to stress that if a plasma is rapidly driven to have a temperature anisotropy well in excess of the mirror threshold ( e.g. , at the earth s bow shock ; * ? ? ? * ) then the resulting amplitude of the mirror modes could significantly exceed what we find in our calculations . in this case , a standard initial value calculation of the mirror evolution is likely to better capture the resulting dynamics than our model in which the anisotropy slowly increases in time . ] our results also show that the dominant wavelength is initially close to a few ion larmor radii , with a subsequent migration to longer wavelengths through a coalescence process . this wavelength growth is consistent with observations ( e.g. , * ? ? ? * ; * ? ? ? * ) and with theoretical models that predict the growth of mirror mode wavelengths due to magnetic diffusion @xcite . the fact that our @xmath255 calculations find that the mirror instability dominates the nonlinear state of the fluctuations ( even though the ic and mirror instabilities have comparable linear thresholds and growth rates for a bi - maxwellian distribution ) is also consistent with heliospheric observations . in particular , solar wind measurements at @xmath256 show that @xmath257 is limited by the linear threshold of the mirror instability ( instead of the ic instability . see * ? ? ? * ; * ? ? ? the observations of @xcite also show enhanced magnetic fluctuations and magnetic compressibility that follow the linear mirror threshold fairly well . it is important to emphasize , however , that the observations presented by @xcite and @xcite showing dominance of the mirror instability also include the regime @xmath258 , which is not considered in our study . also , the dominance of the mirror instability could be affected by properties of the solar wind that are usually not included in calculations of instability thresholds . these include the presence of @xmath259 of alpha particles ( see , e.g. , figure 7 of * ? ? ? * ) , the drift of alpha particles with respect to protons @xcite , and the frequent presence of a proton beam @xcite . we interpret the dominance of the mirror modes in our simulations as being due to the suppression of the ic waves , which is caused by the ion velocity distribution function departing from bi - maxwellian @xcite . this interpretation also requires the ic modes to have finite amplitudes ( although smaller than the one of the mirror modes ) so that they can maintain , via pitch - angle scattering , an ion distribution function that is marginally stable to the growth of ic waves . to the best of our knowledge , these ic modes have not been found by in - situ heliospheric measurements during mirror dominated events . this is consistent with the fact that in our simulations the mirror and ic modes only have comparable amplitudes during the initial , exponential regime ( fig . [ fig : beta6]@xmath93 and @xmath94 ) . in the saturated regime ( which is likely to be more representative of the observed state of the plasma in the heliospheric and astrophysical systems ) , we find that the mirror modes have significantly larger amplitudes . moreover , this dominance increases with the magnetization @xmath43 , as can be seen by comparing figures [ fig : beta6]@xmath93 and @xmath94 . can be accounted for by the fact that the scattering rate of cyclotron - frequency fluctuations is expected to be @xmath187 . if we equate this to the shear parameter @xmath37 ( which is , roughly , the rate at which pitch - angle scattering takes place , as can be seen from the changing rate of @xmath116 in fig . [ fig : muderivs ] ) , one obtains that @xmath260 . ] thus , for realistic values of @xmath43 , we do not anticipate easily detectable ic waves by in - situ measurements in cases where the mirror modes dominate . the departure from the bi - maxwellian ion distribution function has not been confirmed by in - situ heliospheric measurements of mirror - dominated events either . we believe that this is due to the high sensitivity of resonant instabilities to gradients in the velocity distribution function . thus , in cases where ic and mirror modes can have similar growth rates , only small deviations from the bi - maxwellian distribution can significantly reduce the importance of the ic modes and give way to the dominance of the mirror modes . we , therefore , do not anticipate easily detectable departures from bi - maxwellian associated to this effect . nevertheless , future careful observations comparing the nature of the fluctuations and the ion distribution functions in the solar wind can help to unveil these effects . a second application of our results is to low - collisionality accretion flows onto black holes . in these systems pressure anisotropies can produce an anisotropic stress that can increase the transport of angular momentum and the heating of the plasma @xcite . our results are consistent with @xcite s hypothesis that the mirror and/or ic instabilities would provide enhanced pitch angle scattering and regulate the evolution of the plasma pressure tensor . axisymmetric kinetic simulations have also demonstrated this directly @xcite though the restriction to axisymmetry limits the quantitative applicability of these initial simulations to real disks . one of the key conclusions of our study in this paper is that even modest plasma magnetization is sufficient to quantitatively capture the nonlinear saturation of the mirror instability . for example , our @xmath261 results in figures [ fig : beta20b2s ] & [ fig : mubeta20 ] show that the evolution of the ion pressure anisotropy , ion magnetic moment , and turbulent fluctuations are quite similar for @xmath114 and @xmath149 . the primary difference is that in the lower magnetization run the plasma has a larger `` overshoot '' of the saturation pressure anisotropy set by the linear mirror instability threshold . this weak dependence on magnetization is encouraging because it is computationally infeasible to reach @xmath149 in three - dimensional kinetic simulations of astrophysical plasmas ( e.g. , accretion disks ) . one of the interesting questions not fully addressed by our work is the saturation of velocity space instabilities in turbulent low - collisionality plasmas where the shear rate of the mean magnetic field itself fluctuates in time and the magnitude of the large - scale turbulent fluctuations satisfies @xmath262 . our calculations in this paper have focused on the case where an imposed velocity shear amplifies a background magnetic field by order unity or more . if , on the other hand , the shear acts for less time such that the background field is amplified by significantly less , the plasma may never become mirror / ic unstable ( depending on @xmath30 and @xmath221 ) or the background shear may reverse sign before the mirror / ic fluctuations have reached sufficiently large amplitudes for pitch angle scattering to set in . this regime will be explored in future work . another key question not addressed by our study is isotropization of the electron pressure anisotropy . in future work we will study how fast electrons interact with the large - scale mirrors generated by the ion pressure anisotropy , and whether there are separate electron - scale fluctuations that produce significant electron pitch - angle scattering ( e.g. , the whistler instability ) . we thank tobi heinemann , ben chandran , phil isenberg and , in particular , matt kunz for fruitful discussions . eq is supported in part by nasa htp grant nnx11aj37 g , nsf grant ast-1333682 , a simons investigator award from the simons foundation , the david and lucile packard foundation , and the thomas alison schneider chair in physics . mr thanks support from the chilean comisin nacional de investigacin cientfica y tecnolgica ( conicyt ; proyecto fondecyt iniciacin n@xmath263 11121145 ) . dv is supported in part by nasa htp grant nnx11aj37 g and nasa grant nnx12ab27 g . 99 anderson , b. j. , fuselier , s. a. , & murr , d. 1991 , geophysical research letter , 18 , 1955 balbus , s. a. , hawley , j. f. 1991 , , 376 , 214 balbus , s. a. 2001 , , 562 , 909 bale , s. d. , kasper , j. c. , howes , g. g. , quataert , e. , salem , c. , & sundkvist , d. 2009 , phys . 103 , 211101 baumgartel , k. , sauer , k. , & dubinin , e. 2003 , geophys . , 30 , 1761 califano , f. , hellinger , p. , kuznetzov , e. a. , passot , t. , sulem , p. l. , & travnicek , p. m. 2008 , j. geophys . res . , 113 , a08219 chen , l. , lin , z. , & white , r. 2001 , phys . of plasma , 8 4713 johnson , j. r. , & cheng , c. z. 2001 , geophys . lett . , 28 , 4421 buneman , o. 1993 , `` computer space plasma physics '' , terra scientific , tokyo , 67 enriquez - rivera , o. , blanco - cano , x. , russell , c. t. , jian , l. k. , & luhmann , j. g. 2010 , aip conf . proc . , 1216 , 276 gary , s. p. 1992 , j. geophys . res . , 97 , 8519 gary , s. p. & lee , m. a. 1994 , j. geophys . res . , 99 , 11297 guo , j. , yang , z. , lu , q. , & wang , s. 2009 , plasma sci . technol . , 11 , 274 hasegawa , a. 1969 , phys . fluids , 12 , 2642 hasegawa , a. & tsurutani , b. t. 2011 , , 107 , 245005 hawley , j. f. , gammie , c. f. , & balbus , s. a. 1995 , , 440 , 742 hellinger , p. , travnicek , p. , kasper , j. c. , & lazarus , a. j. 2006 , geophys . lett . , 33 , 9 hellinger , p. & travnicek , p. m. 2008 , j. geophys . res . , 113 , a10109 hellinger , p. , kuznetzov , e. a. , passot , t. , sulem , p. l. , & travnicek , p. m. 2009 , geophys . res . lett . , 36 , l06103 heuer , m. & marsch , e. 2007 , , 112 , a03102 horbury , t. s. & lucek , e. a. 2009 , , 114 , a05217 isenberg , p. a. 2012 , phys . plasmas 19 , 032116 isenberg , p. a. , maruca , b. a. , & kasper , j. c. 2013 , , 773 , 164 joy , s. p. , kivelson , m. g. , walker , r. j. , khurana , k. k. , russell , c. t. , & paterson , w. r. 2006 , j. geophys . res . , 111 , a12212 kunz , m. , schekochihin , a. a. , & stone , j. m. 2014 , phys . rev 112 , 205003 liu , y. , richardson , j. d. , belcher , j. w. , kasper , j. c. , & skoug , r. m. 2006 , j. geophys . res . , 111 , a09108 lyutikov , m. 2007 , , 668 , l1 marsch , e. , muhlhauser , k. -h . , rosenbauer , h. , schewnn , r. , & neubauer , f. m. 1982 , j. geophys . res . , 87 , 35 marsch , e. , muhlhauser , k. -h . , schewnn , r. , rosenbauer , h. , pilipp , w. & neubauer , f. m. 1982 , j. geophys . res . , 87 , 52 maruca , b. a. , kasper , j. c. & bale , s. d. 2011 , phys . 107 , 201101 matteini , l. , landi , s. , hellinger , p. , pantellini , f. , maksimovic , m. , velli , m. , goldstein , b. , & marsch , e. 2007 , geophys . lett . , 34 , l20105 matteini , l. , hellinger , p. , landi , s. , travnicek , p. m. , & velli , m. 2007 , space sci . 172 , 373 narayan , r. , mahadevan , r. , & quataert , e. 1998 , theory of black holes acretion disks pokhotelov , o. a. , sagdeev , r. z. , balikhin , m. a. , & treuman , r. a. 2004 , j. geophys . , 109 , a09213 quataert , e. 2008 , , 673 , 758 remya , b. , reddy , r. v. , tsurutani , b. t. , lakhina , g. s. , & echer , e. 2013 , jgra , 118 , 785 riquelme , m. a. , quataert , e. , sharma , p. , & spitkovsky , a. 2012 , , nn , nn russell , c. t. , blanco - cano , x. , jian , l. k. , & luhmann , j. g. 2009 , geophys . lett . , 36 , l05106 schekochihin , a. a. , cowley , s. c. , kulsrud , r. m. , hammett , g. w. , & sharma , p. 2005 , , 629 , 139 schekochihin , a. a. , cowley , s. c. , kulsrud , r. m. , rosin , m. s. , & heinemann , t. 2008 , phys . lett . , 100 , 081301 schmid , d. , volwerk , m. , plaschke , f. , voros , z. , zhang , t.l . , baumjohann , w. , & narita , y. 2014 , ann . geophys . , 32 , 651 schwartz , s. j. , burgess , d. , & moses , j. j. , 1996 , ann . , 14 , 1134 sharma , p. , hammett , g. w. , quataert , e. , & stone , j. 2006 , , 637 , 952 shoji , m. , omura , y. , tsurutani , b. t. , verkhoglyadova , o. p. , & lembege , b. 2009 , , 114 , a10203 soucek , j. , lucek , e. , & dandouras , i. 2008 , j. geophys . res . , 113 , a04203 southwood , d. j. , & kivelson , m. g. 1993 , , 98 , 9181 spitkovsky , a. 2005 , aip conf . proc , 801 , 345 , astro - ph/0603211 travnicek , p. , hellinger , p. , taylor , m. , escoubet , c. , dandouras , i. , & lucek , e. 2007 , geophys . lett . , 34 , l15104 tu , c .- y . & marsch , e. 2002 , , 107 , a9 tsurutani , b. t. , smith , e. j. , anderson , r. r. , ogilvie , k. w. , scudder , j. d. , baker , d. n. , & bame , s. j. 1982 , , 87 , 6060 tsurutani , b. t. , lakhina , g. s. , verkhoglyadova , o. p. , echer , e. , guarnieri , f. l. , narita , y. , constantinescu , d. o. 2011 , , 116 , a02103 tsurutani , b. t. , echer , e. , verkhoglyadova , o. p. , lakhina , g. s. , guarnieri , f. l. 2011 , journal of atmospheric and solar - terrestrial physics , 73 , 1398 verscharen , d. , bourouaine , s. , chandran , b. d. g. , & maruca , b. a. 2013 , , 773 , 8 volwerk , m. , zhang , t. l. , delva , m. , voros , z. , baumjohann , w. , & glassmeier , k.h . 2008 , geophys . lett . , 35 , previous 1d studies of the mirror instability suggest that its growth and saturation are dominated by the resonant " scattering of @xmath218 particles ( where @xmath264 is the particle velocity component parallel to the magnetic field ; * ? ? ? * ; * ? ? ? * ) , leading to a flattening of the distribution function near @xmath218 . in contrast , our 2d runs discussed in the main text show that scattering by mirror modes is fairly momentum independent , with no significant flattening of @xmath265 . to compare our results more directly to the previous literature , in this appendix we compare the mirror saturation in 1d and 2d for different values of @xmath154 , paying especial attention to the behavior of @xmath218 particles . we show that , compared with the 1d case , 2d runs significantly change the way pitch - angle scattering occurs , modifying the saturation mechanism . to best compare with the previous literature , we use initial - value simulations in this appendix , where a linearly unstable pressure anisotropy is imposed as an initial condition . figure [ fig:1d2d2.3 ] compares simulations with @xmath266 , @xmath267 , @xmath268 in 1d and 2d . in the 1d simulation , the initial magnetic field @xmath34 forms an angle of @xmath269 with the resolved dimension of the simulation . apart from the different number of dimensions , these simulations also have the same numerical parameters . panels [ fig:1d2d2.3]@xmath93 and [ fig:1d2d2.3]@xmath94 show @xmath116 and @xmath224 as a function of time for the 1d and 2d cases , respectively . in 1d , @xmath116 decreases significantly slower than @xmath224 , which means that there is a significant correlation between @xmath163 and @xmath123 ( as expected for the mirror instability ) . in 2d , on the other hand , the difference between @xmath116 and @xmath224 is very small , implying that pitch - angle scattering occurs more efficiently in 2d . this difference in the scattering efficiency can be explained considering the modes that grow in the 1d and 2d simulations . indeed , in the 1d case , mirror modes with wave vectors pointing in only one direction can grow . in the 2d case , on the other hand , both ic and mirror modes appear ( with the mirror modes forming both positive and negative angles with respect to @xmath34 ) . this can be seen from panels [ fig:1d2d2.3]@xmath50 and [ fig:1d2d2.3]@xmath101 , which show the evolution of the magnetic energy contained in fluctuations parallel ( @xmath167 ) and perpendicular ( @xmath270 ) to @xmath34 for the 1d and 2d cases . whereas the 1d case is dominated by @xmath167 ( characteristic of mirror modes ) , the 2d case is dominated by @xmath270 , showing the dominant presence of the ic modes . panels [ fig:1d2d2.3]@xmath108 and [ fig:1d2d2.3]@xmath109 show snapshots of the parallel velocity distribution @xmath265 for the 1d and 2d runs at the saturated state @xmath271 ( black lines ) , and compare them with a bi - maxwellian distribution ( red line ) . in the 1d case , we can see flattening of @xmath265 for @xmath218 . this flattening is consistent with the results of @xcite , who argue that it is due to the resonant interaction of @xmath218 particles with the mirror modes . the 2d run , however , does not have this flattening , implying that the saturation process in two dimensions is quite different . this is indeed expected from the dominant role of the ic instability . since at higher @xmath272 the ic instability is expected to play a subdominant role , we also explored the case where initially @xmath273 . figure [ fig:1d2d20 ] compares simulations with @xmath266 , @xmath274 , and @xmath275 in 1d and 2d , with both cases having the same numerical parameters . panels [ fig:1d2d20]@xmath93 and [ fig:1d2d20]@xmath94 show the evolution of @xmath116 and @xmath224 for both cases . in the 1d case these two quantities differ significantly , implying an important correlation between @xmath123 and @xmath163 . this correlation almost disappears in the 2d case , which shows the presence of more efficient pitch - angle scattering in the 2d case . since for @xmath273 the ic instability plays a subdominant role , this is likely a property of the mirror instability itself rather than the presence of the ic instability as was the case for the lower @xmath154 results shown in figure [ fig:1d2d2.3 ] . the higher rate of scattering in 2d also affects the maximum amplitude of the fluctuations , which are shown in panels [ fig:1d2d20]@xmath50 and [ fig:1d2d20]@xmath101 for the 1d and 2d cases . although in both cases @xmath276 ( implying that mirror dominates ) , in 2d the amplitude of the modes is smaller , which means that field rearrangement plays a less important role in the approach to marginal stability . finally , figure [ fig:1d2d20 ] also compares @xmath265 at the saturated state ( @xmath277 ) in the 1d and 2d cases . there is very little flattening for @xmath218 in both cases . this confirms our result from the main text that the pitch - angle scattering in the mirror dominated case is fairly independent of particle momentum . it is interesting , however , that we do not see flattening even in 1d . since @xcite reported the appearance of flattening only near the threshold of the mirror instability ( and only for certain box sizes ) , we tested simulations with the same plasma parameters having anisotropies as low as @xmath278 and found similar results .
the space - time geometry that is commonly used today arose from classical physics . an interesting question is what geometry is appropriate for quantum physics . it was suggested that the universal symmetry group elements which act on all hilbert spaces may be appropriate for constructing a physical geometry for quantum theory @xcite . i also proposed the systematic study of all the fundamental interactions _ operationally _ from their effects on quantum interference @xcite . the purpose of this paper is to attempt to bring together these two approaches . the modular variables introduced by aharonov , pendleton and petersen @xcite will play a useful role in this . in section 2 , i shall review noether s theorem and its converse in a generalized form in which the conserved quantities are elements of a group and not the generators of this group as usually stated . this will suggest a quantum geometry by relations defined by the universal group elements , which constitute the symmetry group of physics , that act on all hilbert spaces , as discussed in section 3 . the classical limit of this geometry will be obtained in section 4 as the usual space - time geometry from the quantum state metric in hilbert spaces and the universality of the action of the translational group in every hilbert space . in section 5 , the non locality of fundamental interactions in quantum physics implied by this approach , as shown physically by the aharonov - bohm effect @xcite and its generalizations @xcite , will be studied . the study of the gravitational aharonov - bohm effect around a cosmic string in particular suggests that the use of universal group elements as quantum distances may be appropriate . the usual statement of noether s theorem is that for every continuous symmetry of the equations of motion ( determined by the lagrangian or hamiltonian ) there exists a conserved quantity . although this is easy to prove , the meaning of this theorem is more readily evident in the hamiltonian than in the lagrangian formulation . a symmetry of the equations of motion or time evolution is a transformation @xmath0 such that in any experiment if @xmath0 is applied to the initial state of the physical objects ( fields , particles , etc . ) participating in the experiment then the final state of the transformed experiment must be the same as @xmath0 applied to the final state of the original experiment . for example , spatial translational symmetry implies that if the apparatus is translated to a new spatial location then the same experiment should give the same result . suppose @xmath1 is the time - evolution operator , and @xmath2 , @xmath3 are the initial and final states , i.e. @xmath4 . then the above definition of @xmath0 being a symmetry of the time evolution is @xmath5 for every initial state @xmath2 . this is equivalent to the commutator @xmath6=0 . \label{symmetry}\ ] ] but ( [ symmetry ] ) states also that @xmath0 is conserved during the time - evolution . therefore , the statements that @xmath0 is a symmetry and that @xmath0 is conserved are the _ same _ statement ( [ symmetry ] ) , and there is nothing to prove ! now suppose that there is a continuous symmetry generated by @xmath7 . then ( [ symmetry ] ) is satisfied with @xmath8 for all @xmath9 , and therefore @xmath10=0 . \label{commutator}\ ] ] hence , @xmath7 is conserved , which is noether s theorem . furthermore , if the hamiltonian @xmath11 is independent of time @xmath12 , as it is for an isolated system , then @xmath13 . if ( [ commutator ] ) is valid for all @xmath12 , then @xmath14=0 . \label{hcommutator}\ ] ] the above results may be extended to classical physics by turning the above commutators into poisson brackets in classical phase space . these classical results may be regarded as the classical limit of the quantum results by recognizing that the symplectic structure that gives the poisson brackets are the classical limit @xcite of a symplectic structure in quantum theory @xcite that gives the commutators . but the conservation of @xmath0 that follows from ( [ symmetry ] ) is more general than the usual form of noether s theorem . there are at least two situations in which ( [ symmetry ] ) is valid but there are no corresponding ( [ commutator ] ) or ( [ hcommutator ] ) . first , as is well known , in both classical and quantum physics , @xmath0 may be a discrete symmetry instead of a continuous symmetry . for example , @xmath0 may be parity , which is a symmetry and therefore conserved for all interactions except the weak interaction , as far as we know . another example is that ( [ symmetry ] ) is satisfied for @xmath15 for a discrete set of values @xmath16 only . second , in quantum physics the mean value of @xmath8 has more information than the mean value of all the moments of @xmath7 , namely @xmath17 where @xmath18 is any positive integer @xcite . this is unlike in classical physics where the mean value of a transformation generated by @xmath7 may be obtained using the mean values of all the @xmath17 . both these situations will be considered in section 5 . since in ( [ symmetry ] ) @xmath1 and @xmath0 occur symmetrically , it follows that the converse of the generalized noether s theorem is also true : _ a transformation @xmath0 that is conserved must be a symmetry of the equations of motion . _ the usual view is that @xmath1 is more fundamental than @xmath0 because @xmath1 is determined by the dynamical laws which are regarded as primary , whereas the symmetries such as @xmath0 are obtained secondarily as the symmetries of these laws . but the concise form ( [ symmetry ] ) of the connection between the dynamical laws and symmetries , in which @xmath1 and @xmath0 are on an equivalent footing , suggest that we may equally turn the usual view around and regard the transformations @xmath19 as primary and @xmath1 as derived from them to satisfy ( [ symmetry ] ) so that @xmath19 are the symmetries @xcite . the possibility of regarding symmetries as fundamental relations between quantum states by associating them with a quantum geometry will be explored in the next section . the concept of space originates from our common experience of translating objects and from the possible states they can occupy . if we translate a cup , for example , in various possible ways , classically we may say that the different configurations or states of the cup are `` immersed '' in `` space '' . this space is _ universal _ in the sense that it is regarded as independent of the objects which are `` contained '' in it . but quantum mechanically it is not clear what is meant by the cup being `` immersed '' in space . the cup consists of electrons , protons and neutrons ( or the quarks and gluons which make up the protons and neutrons ) , and the states of these particles belong to the corresponding hilbert spaces which are different from the physical space or the phase space of classical physics . the translation of a cup therefore needs to be represented by the corresponding translation operators that act on these hilbert spaces . the fact that all the particles constituting the cup move together in some approximate sense suggests the introduction of universal translation group elements that are represented by operators that act on each hilbert space . it is this _ universality _ of the translation group that gives us the concept of `` space '' that is independent of the particular system that partakes in it . also , it is well known that we can not operationally determine the metric in space- time , or even the points of space - time , using quantum probes @xcite because of the uncertainty principle . if one tries to obtain the space - time geometry using a clock and radar light signals , which is possible in classical physics @xcite , the uncertainty in the measurement of time in quantum physics is @xmath20 , where @xmath21 is the uncertainty in the energy of the clock . if we try to decrease this uncertainty by increasing @xmath22 , then @xmath22 causes a corresponding uncertainty in the geometry of space- time . the total uncertainty in the measurement of space - time distances is then @xmath23 . the minimum value of this uncertainty as @xmath21 is varied is @xmath24 planck - length @xmath25 . hence , space- time geometry is only approximately valid in quantum theory , with an uncertainty of the order of planck length . however , a geometry for quantum theory may be defined by relations determined by a universal group @xmath26 , which generalizes the above translation group . this is universal in the sense that the same @xmath26 has a representation in each hilbert space . but @xmath26 may have subgroups which may have trivial representations in some hilbert spaces and not in others . an object may be displaced by any @xmath27 , which means that @xmath0 acts on each of the hilbert spaces of the particles or fields constituting that object through the corresponding representation of @xmath26 . each @xmath28 in each of these hilbert spaces is mapped to a corresponding @xmath29 by this action of @xmath0 , and the group element @xmath0 that determines the _ relation _ between @xmath28 and @xmath29 is independent of the hilbert space and is therefore universal . in the example of a cup considered above , @xmath0 is an element of the translation group , @xmath28 and @xmath29 are the states of each particle constituting the cup before and after the translation , and the relation between each such pair is universal in the sense that the entire cup has undergone this translation , or any other object that could take the place of the cup . this quantum geometry can not be subject to the above criticisms of the space - time geometry because the action of @xmath26 on each hilbert space is not subject to any uncertainty . it is reasonable to require that this geometry is preserved in time in the absence of interactions . then each @xmath27 is conserved and the converse of noether s theorem stated in the last paragraph of the previous section implies that @xmath0 is also a symmetry of the time evolution . since the evolution equations are now determined by the standard model , @xmath26 may be the symmetry group of the present day standard model , namely @xmath30 , where @xmath31 is the poincare group . but if the standard model is superseded by new physics that has a different symmetry group then @xmath26 should be this new group and the above statements would all be unaffected . as an illustration of the geometrical relations proposed here , consider the experimentally known quantization of electric charge : all known charges are integral multiple of the fundamental charge @xmath32 . an aspect of this is that the magnitudes of the charges of the electron and the proton are experimentally known to be equal to an amazing precision . to obtain charge quantization , take @xmath0 above to be an arbitrary element of the electromagnetic @xmath33 group , which is a subgroup of @xmath26 . this universal @xmath33 group is a circle parametrized by @xmath34 , say , that varies from @xmath35 to @xmath36 so that @xmath35 and @xmath36 represent the same point on this group , chosen to be the identity . since @xmath33 is abelian , it has only one - dimensional representations . hence the action of @xmath37 on an arbitrary state gives @xmath38 where @xmath7 corresponds to the particular representation of @xmath33 in the hilbert space in which @xmath28 belongs to . but since @xmath39 , which is due to the compactness of the @xmath33 group , @xmath40 for all representations @xcite . hence , @xmath41 or @xmath42 where @xmath18 is an integer . to interpret @xmath7 , consider the physical implementation of the transformation @xmath0 . this may be done by sending each of the particles through the same electromagnetic field with 4-vector potential @xmath43 in a particular gauge so that the effect of the electromagnetic field alone on the particle is given by @xmath44 which is a @xmath33 transformation . indeed , the statement that the electromagnetic field is a @xmath33 gauge field may be taken to mean that it is physically possible to implement a @xmath33 gauge transformation using the electromagnetic field in this way . then @xmath9 has the interpretation of the electric charge . comparing ( [ em ] ) with ( [ u1 ] ) , we may take @xmath34 to be @xmath45 in which case @xmath46 . hence , from ( [ quantization ] ) , @xmath47 where @xmath48 is a universal constant that is determined experimentally to be @xmath49 , where @xmath50 is the charge of the electron . the exact equality of the magnitudes of the charges of the electron and the proton may now be understood as due to them belonging to representations corresponding to @xmath51 and @xmath52 , respectively . the above argument also provides a reason for the introduction of planck s constant which is purely geometrical . the exponent in ( [ em ] ) must be dimensionless because the expansion of the exponential has all powers of the exponent . now , @xmath53 is meaningful in classical physics . but to turn it into a dimensionless quantity , it is necessary to introduce a new scale , which is provided by @xmath54 . from the present point of view , this is needed in order to physically implement the @xmath33 group elements that define relations between states which are part of the quantum geometry . also , from ( [ quant ] ) , @xmath9 is proportional to @xmath50 , and @xmath43 is also proportional to @xmath50 because the charges that generate @xmath43 via maxwell s equations are proportional to @xmath50 . hence , the exponent in ( [ em ] ) is proportional to the fine - structure constant @xmath55 . this argument may be extended to gauge fields in general , and dimensionless coupling constants are obtained for all of them . from now onwards , units in which @xmath56 will be used , and the metric convention is @xmath57 . the relation defined by ( [ em ] ) is not gauge invariant . hence , it can not be used to define an invariant geometrical ` distance ' . consider again the translation of a cup which may be performed by acting on all the quantum states of the particles constituting the cup by a universal group element @xmath58 , where @xmath59 is a generator of translation . the action of this group element on a wave function is also not gauge invariant . but we may combine the two transformations to define the gauge - covariant transformation @xmath60 , where @xmath61 stands for @xmath62 or equivalently @xmath63 , and @xcite @xmath64 where at present the integral is taken along the straight line joining @xmath65 and @xmath66 for simplicity . then , clearly , @xmath67 it is easy to show , using ( [ p ] ) and ( [ reversed ] ) , that the set of operators @xmath68 is a group under multiplication . from ( [ p ] ) , under a gauge transformation , @xmath69 , where @xmath70 , and @xmath71 , @xmath72 transforms to @xmath73 on using @xmath74 it follows that @xmath75 hence , @xmath76 acts gauge covariantly on the hilbert space . @xmath76 may also shown to be gauge covariant from the fact that @xmath77 is gauge invariant because the integrand is gauge invariant @xcite . it also follows from ( [ < f > ] ) that the operator ( [ p ] ) is observable . for example , in the josephson effect , where the current depends on the gauge invariant phase difference across the junction , if @xmath78 is chosen to be the vector across the junction then @xmath72 is observable from the current @xcite . a gauge may be chosen , even if the field strength is non vanishing , so that the component of @xmath79 in the direction of @xmath78 is zero . then in this gauge @xmath80 , which is conserved for an isolated system . since @xmath72 is gauge covariant , it follows that ( [ p ] ) is conserved for an isolated system in every gauge . @xmath72 is a gauge covariant generalization of the modular momentum @xcite , and may be called the modular kinetic momentum . when two systems interact , there would be an exchange of modular kinetic momentum . indeed , this may be regarded as the definition of two systems interacting @xcite . as will be seen in section [ interactions ] , this exchange may happen even when there are no forces between the two systems , which makes the latter interaction more general than the usual interaction via forces . if @xmath81 is the momentum space wave function of @xmath28 , then clearly @xmath82 hence , the change in @xmath83 results in a change in the distribution of the probability density @xmath84 of momentum @xmath85 @xcite . since in the above gauge the gauge invariant @xmath86 is the same as ( [ ft ] ) , it follows that , in an arbitrary gauge , an interaction results in a change in the distribution of the gauge invariant kinetic momentum @xmath87 . similarly , the modular energy @xcite may be generalized gauge - covariantly as follows . define the transformation on the hilbert space @xmath88 by @xmath89 , where @xmath90 and @xmath91 denotes time - ordering . to prove that @xmath88 is gauge - covariant under the above gauge transformation , note first that for a solution @xmath92 of schrdinger s equation , @xmath93 therefore , under the gauge transformation @xmath94 @xmath95 using ( [ se ] ) . hence , @xmath96 where @xmath97 is the gauge transformed hamiltonian . the last equation is equivalent to @xmath98 under this gauge transformation , @xmath99 transforms to @xmath100 on using ( [ h ] ) . hence , @xmath88 acts gauge - covariantly on the hilbert space . more generally , for an arbitrary gauge field the above arguments hold with @xmath101 replaced by @xmath102 where @xmath103 generate the gauge group , and @xmath104 is the corresponding local gauge transformation . the gauge field exponential ( parallel transport ) operator will be generalized to be along an arbitrary piece - wise differentiable curve @xmath105 in @xmath106 . path ordering is necessary because @xmath107 do not commute in general . the energy - momentum operator @xmath108 is defined by @xmath109 . now define the transformation @xmath110 on the hilbert space by @xmath111 , where @xmath112 with p denoting path ordering , and @xmath113 is a curve in space - time that is congruent to @xmath105 while @xmath114 is the curve @xmath113 traversed in the reverse order . in ( [ g ] ) , @xmath113 begins at @xmath61 and ends at @xmath115 , where @xmath116 is a fixed vector ( independent of @xmath62 ) , and @xmath114 therefore begins at @xmath115 and ends in @xmath61 . then @xmath117 under a local gauge transformation , ( [ g ] ) transforms to @xmath118 where @xmath119 . now write @xmath120 as a product of infinitesimal exponentials of the form @xmath121 , and use ( [ p ] ) and ( [ h ] ) in their infinitesimal forms . then ( [ g ] ) implies @xmath122 hence , the operator @xmath110 acts gauge covariantly on the hilbert space . it follows from this proof that @xmath110 would be gauge - covariant also when @xmath114 in ( [ g ] ) is replaced by any piece - wise differentiable curve that would connect @xmath115 to @xmath61 . also , on using ( [ conjugate ] ) , @xmath123 where @xmath124 . the expectation value ( [ expectation ] ) is gauge invariant because the integrand is gauge invariant . it may be observable , in principle , by the josephson effect for a non abelian gauge theory proposed in @xcite . more experimental consequences of @xmath110 will be discussed in section v. the energy - momentum operator @xmath108 may be generalized to relativistic quantum field theory by defining its components as the conserved quantities obtained via noether s theorem from the invariance of the lagrangian under space - time translations : @xmath125 where @xmath126 is the conserved , normal - ordered energy - momentum tensor . then the canonical commutation relations imply that @xmath108 generates space - time translations . it therefore follows from arguments analogous to the above that @xmath110 , given by ( [ g ] ) with the quantum field theoretic @xmath108 , is gauge covariant . since @xmath108 transforms as a covariant vector under lorentz transformations , it also follows that @xmath110 is lorentz invariant . moreover , @xmath110 may be generalized to curves @xmath113 in @xmath127 ( @xmath18 is any positive integer ) , which would correspond to an @xmath128dimensional space - time . it may be reasonable to take @xmath110 as a quantum distance that replaces the classical space - time distance along the curve @xmath113 . but to do so it would be necessary to obtain the classical distance in an appropriate limit from @xmath110 , which will be studied in the next section . to take the classical limit of this geometry , note that classical space - time is constructed with measuring instruments consisting of particles that have approximate position and momentum . it is therefore reasonable to represent them by gaussian wave packets which have minimum uncertainty . for a particle with mean position at the origin and mean momentum zero , the normalized wave function of such a state up to an arbitrary phase factor is @xmath129 where @xmath130 is the uncertainty in position . this may be a state of a molecule in the cup , mentioned in section 3 , in a harmonic oscillator potential in which case it would not spread . as the cup is displaced , the above wave function becomes @xmath131 up to a phase factor . in ( [ p ] ) the second factor may be made the identity by choosing a gauge in which @xmath132 . therefore , @xmath133 may be regarded as a special case of ( [ p ] ) or ( [ g ] ) , and hence as defining a quantum distance between the states @xmath134 and @xmath135 . we may therefore expect a metric to be defined on the translation group to which this operator belongs to and use that to define a metric in space . but this group , being abelian , has no natural metric on it . there are two ways , however , that a metric may be defined on it . one is to use the casimir operator of the poincare group , @xmath136 , to define a metric on it , which locally may be associated with the space - time metric @xcite . the other method , which will be used here , is to utilize the overlap of the two wave functions to obtain a measure of the displacement between them , which would then give an equivalent metric in the translation group . this is possible if the space of wavefunctions on which this group acts has an inner product , which would give a measure of the overlap and therefore how far a state has been translated . such a measure is given by the fubini - study metric in the quantum state space , or the set of rays , of every hilbert space . this is the unique metric , up to multiplication by an overall constant , that is invariant under unitary ( and anti - unitary ) transformations . this may therefore be written in the form @xcite @xmath137 where @xmath138 is the infinitesimal distance between two neighboring states ( rays ) represented by normalized state vectors @xmath28 and @xmath139 . clearly , @xmath138 is zero when the states are the same , and it increases when the overlap between the states decreases . it is also invariant under unitary transformations , and must therefore be the fubini- study metric . the factor @xmath140 in ( [ fs ] ) is just a convention which ensures that this metric in the state space of the hilbert space spanned by @xmath28 and @xmath139 is the metric on a sphere of unit radius . now substitute @xmath141 and @xmath142 as @xmath28 and @xmath139 in ( [ fs ] ) . then , @xmath143 neglecting higher order terms in @xmath144 because it is infinitesimal . hence @xmath145 , which is the same for all hilbert spaces , may be used as a metric on the @xmath146 dimensional translational group that is parametrized by the components of the vector @xmath147 . locally , this metric may be regarded as a metric in the physical space of classical physics . this result may be generalized to an arbitrary state @xmath28 as follows . require that @xmath148 to be close to the identity , where @xmath149 therefore expand @xmath150 where @xmath151 is the momentum component in the direction of @xmath152 . substituting this in ( [ fs ] ) , @xmath153 where @xmath154 is the uncertainty in @xmath155 : @xmath156 . for the gaussian wave packet , @xmath157 and therefore ( [ metric2 ] ) gives ( [ metric ] ) in this case . time is measured by a clock . since the clock must have moving parts , the uncertainty @xmath21 of its hamiltonian @xmath158 must be non zero . neglecting any external interaction of the clock , @xmath158 is a constant . the infinitesimal time evolution of the the clock is given by @xmath159 the fubini - study distance @xmath138 along the evolution curve in the quantum state space corresponding to @xmath160 is obtained analogous to the derivation of ( [ metric2 ] ) to be @xcite @xmath161 a quantum clock directly measures the fubini - study distance @xmath26 and the time @xmath12 is then inferred from @xmath26 using ( [ clock ] ) . the appearance of the same @xmath12 in ( [ clock ] ) in all hilbert spaces is due to the universality of the action of the time translation @xmath162 in every hilbert space . this is analogous to the universality of the spatial displacements which was used earlier to obtain the spatial metric . the spatial translation ( [ spatial ] ) and the active time translation corresponding to ( [ evolution ] ) may be written covariantly as @xmath163 where @xmath164 . this transformation is the infinitesimal version of ( [ g ] ) in a special gauge that makes the second factor in ( [ g ] ) the identity @xcite . in an arbitrary gauge , the same space and time metrics are obtained by replacing the operator in ( [ covariant ] ) with the infinitesimal version of ( [ g ] ) in the above treatment . in relativistic quantum theory , owing to the transformation property of @xmath165 mentioned at the end of section iii , these metrics give a space - time metric that is invariant under lorentz transformations . in the presence of gravity , this lorentzian metric may be obtained locally by doing the above space- time translations in a freely falling einstein elevator , which then globally gives a curved pseudo - riemannian metric . if the quantum geometry is determined by relations between states that are group elements , and if these group elements , which are our observables , are made dynamical the way einstein made space - time distances dynamical in order to obtain gravity , then this would give both gravity and gauge fields @xcite . also , quantum mechanics has an inherent non locality , which may also be understood as due to these group elements being the basic observables . the combination of these two statements imply that gauge fields and gravity should affect quantum states in a non local manner as in the aharonov - bohm effect @xcite , as will be discussed later . first consider the non locality of quantum theory , which may be illustrated by the following example : electrons with initial momentum in the @xmath166 direction go through an infinite diffraction grating in the @xmath167plane of a cartesian coordinate system , with the length of the slits along the @xmath168-direction . then the grating destroys continuous translational symmetry for the electrons in the @xmath169 direction . however , if the distance between successive slits in the @xmath170 direction is @xmath171 then @xmath172 satisfies ( [ symmetry ] ) , where @xmath155 is the momentum operator for electrons in the @xmath169direction which generates translations in the @xmath169direction . ( for simplicity , here and henceforth , the @xmath173 over operators is omitted . ) hence , it follows from the generalized noether s theorem in section 2 that @xmath174 is conserved although @xmath155 is _ not _ conserved . indeed , it is well known that the interference fringes on a screen that is parallel to and far away from the @xmath175- plane is given by @xmath176 , where @xmath177 is the wave length and @xmath18 is an integer . therefore , the possible values of the momentum for an electron in the @xmath169 direction after the interaction are @xmath178 , i.e. @xmath179 . hence , @xmath180 is conserved during the passage of electrons through the grating . the above operator @xmath0 is equivalent to the modular momentum @xmath181 introduced by aharonov et al on the basis of the above example @xcite . but here i shall treat @xmath0 as an element of a universal group that is used to define a quantum geometry as in section 3 . @xmath0 may be obtained from experiments by measuring the hermitian observables @xmath182 , s_i \equiv { 1\over 2i}[\exp(-i { p \ell\over \hbar})- \exp(i { p \ell\over \hbar})].\ ] ] therefore , the unitary operator @xmath0 may also be regarded as an observable . it is important to note that this is a _ non - local _ observable , unlike @xmath155 . this non locality in quantum mechanics may be illustrated also in the simple interference experiment of two coherent wave packets . suppose that the two wave packets are moving in the @xmath166direction and have no overlap at time @xmath12 . for simplicity , assume that they are the same except that their centers are separated by a displacement @xmath171 in the @xmath169direction . let @xmath183 be the phase difference between the wave packets . the wave is then a superposition of these two wave packets : @xmath184 now no local experiments performed on the two wave packets at the two slits could determine the phase factor @xmath185 . for example , the expectation values of the local variables @xmath186 , where @xmath18 is any positive integer , give no information about @xmath187 @xcite . this is easily verified by writing @xmath188 in the coordinate representation . but @xmath189 this means that the momentum distribution at time @xmath12 does depend on the phase factor @xmath190 , i.e. if @xmath155 is measured then the probability distribution for obtaining the individual eigenvalues of @xmath155 is changed by this phase factor ( but the average @xmath191 is unchanged ) . and this may be experimentally verified by letting the wave packets interfere and observing the shift in interference fringes . hence , @xmath192 contains more information than the expectation values @xmath193 of any of the moments of momentum @xmath186 . this is basically due to the linear structure of the hilbert space , which physically corresponds to the principle of superposition , and the fact that @xmath28 is not an analytic function . this fundamental non locality of quantum mechanics translates into a non locality of the effect of all the fundamental interactions on the wave function . this has been shown for the aharonov - bohm effect due to a magnetic field by aharonov et al @xcite . it may be illustrated in the above described interference of two wave packets as follows . suppose the above two wave packets @xmath194 and @xmath195 are those of an electron and they pass on the two sides of a solenoid containing a magnetic flux @xmath196 . the gauge may be chosen so that the vector potential is non zero only along a thin strip bounded by two planes indicated by the dotted lines in figure 1 . then when the line @xmath197 passes the solenoid , the wave packet @xmath194 acquires a phase difference @xmath198 with respect to the wave packet @xmath195 . therefore , the expectation value of the modular momentum @xmath0 , given by ( [ alpha ] ) , which was @xmath199 before the line @xmath197 passed the solenoid is now @xmath200 . it was pointed out by aharonov @xcite that in the last statement @xmath0 may be replaced by the gauge invariant modular kinetic momentum @xmath72 given by ( [ p ] ) . this is because before and after the wave packets pass the solenoid the vector potential @xmath79 is zero along the line @xmath197 and therefore @xmath72 is the same as @xmath0 . since @xmath201 is gauge covariant , @xmath202 is gauge invariant . hence , the above mentioned change in @xmath202 by the factor @xmath203 , as the line @xmath197 crosses the solenoid , is the same in every gauge . vector aharonov - bohm effect in which a wave packet of an electron is split coherently into two wave packets at the beam splitter @xmath204 which are then made to interfere at @xmath205 . when the imaginary line joining the centers of the wave packets @xmath194 and @xmath195 sweeps across the magnetic flux in the solenoid ( shaded region ) the modular momentum or the modular kinetic momentum associated with this line changes , as pointed out by aharonov @xcite . however , we may replace @xmath76 in the above arguments by the more general gauge covariant operator @xmath110 given by ( [ g ] ) with @xmath113 here being an arbitrary ( piece - wise differentiable ) space - like curve that joins @xmath194 and @xmath195 . then @xmath206 where the integral is from @xmath207 to @xmath208 along @xmath113 . hence , @xmath209 which is gauge invariant . then @xmath210 changes by the factor @xmath211 as a portion of @xmath113 crosses the solenoid . it follows that there is nothing special about the straight line @xmath197 crossing the solenoid in the experiment described in figure 1 . the arbitrariness of the choice of @xmath113 joining @xmath194 and @xmath195 in the above argument reflects the topological nature of the aharonov - bohm effect . what is ultimately observed in this effect is the phase factor @xmath212 where the integral is around the solenoid or more generally around the region in which the field strength is non vanishing . since this is an integral of the @xmath213form @xmath43 along a curve , it contains no information about the metric of space - time@xcite . to determine the straight line , or a geodesic in general , the metric is needed , and the above phase factor therefore can not show any preference to a geodesic such as the line @xmath197 . the integrand of ( [ gamma ] ) ( with @xmath50 replaced by 2@xmath50 ) was previously used in ref . @xcite to study the josephson effect due to the enclosed magnetic flux in a superconducting ring that has a josephson junction , which is also an aharonov - bohm effect . consider now the scalar aharonov - bohm effect . a wave packet traveling in the @xmath166 direction is partially transmitted and reflected by a beam splitter @xmath204 ( figure 2a ) . the resulting two wave packets which travel in opposite directions are reflected by two mirrors @xmath214 and @xmath215 situated along the @xmath166axis and they interfere subsequently . meanwhile a pair of oppositely charged capacitor plates @xmath216 is separated and closed so that there is a non zero electric field in the region enclosed by the world- lines of the centers of the wave packets in the @xmath217plane . the same experiment is viewed in the rest frame of the reflected wave packet in figure 2b . figure 2 . a scalar aharonov - bohm effect shown schematically in the @xmath218plane . ( a ) the modular energy and modular kinetic energy associated with the imaginary lines @xmath219 and @xmath220 are different , partly because of the scalar @xmath197 phase shift due to the region of non zero electric field in the capacitor that is open and shut ( shaded region ) . ( b ) the same experiment viewed in the rest frame of the wave packet that is reflected at the beam splitter @xmath204 . when the imaginary line @xmath197 joining the wave packets sweeps across the region of non zero electric field in the capacitor the modular energy and modular kinetic energy associated with this line changes due to the ab phase shift . we may choose a gauge in which the vector potential is non zero only along a strip between the dotted lines parallel to the time - axis in figure 2a or 2b . then the wave packet @xmath194 at time @xmath221 develops a phase shift @xmath222 with respect to @xmath195 at time @xmath12 as the imaginary line @xmath197 crosses the space - time region containing the electric field , where @xmath223 and @xmath216 is the region in which the electric field @xmath224 is non zero . this is a non local effect which may be understood using @xmath225 , where @xmath91 denotes time ordering and @xmath226 is the time interval between the events @xmath227 and@xmath228 . we shall assume that the time dependence of @xmath11 comes only from the vector potential contained in @xmath11 . suppose the wave function of the electron is @xmath229 where @xmath230 and @xmath231 are the wave functions of the above two localized wave packets . then @xmath232 changes by the factor @xmath233 due to the electric field @xmath234 in the region @xmath216 as the imaginary line @xmath197 sweeps across the small region @xmath216 where @xmath234 is non zero ( fig . 2b ) . in the last statement and in ( [ scab ] ) we may replace @xmath1 by the gauge covariant modular kinetic energy operator @xmath88 given by ( [ h ] ) , with @xmath50 replacing @xmath9 . this is because @xmath235 is zero along @xmath236 and @xmath237 , which are outside the strip that contains the non zero @xmath43 . however , is gauge invariant and therefore its change mentioned above may be obtained in any gauge . but since there are no forces acting on the electron , there is no change of its kinetic energy @xmath238 or any of its moments . hence , @xmath239 contains non local aspects which can not be obtained by measurements of the kinetic energy operator or any of its moments . thus the scalar aharonov - bohm effect may be viewed as a quantum effect which is _ non local in time_. the above results are easily generalized to the aharonov - bohm effects due to non abelian gauge fields@xcite by replacing the electromagnetic fluxes by fluxes of yang - mills field strength , and using the expectation values ( [ expectation ] ) of the gauge covariant operator @xmath110 given by ( [ g ] ) . in all cases , the gauge invariant @xmath240 changes as @xmath113 passes a cross - section of the gauge field flux . if the gauge field flux has a singular cross - section , then @xmath113 passes the flux at an event , which of course is independent of the inertial frame in which the effect is being described . when this happens , in this idealized case , @xmath240 changes to @xmath241 , where @xmath216 is a closed curve going around the yang - mills flux . the group element ( [ g ] ) belongs to the group @xmath242 , where for a closed system @xmath243 is the translation group and @xmath244 is the gauge group . but it has an asymmetry in that the part of ( [ g ] ) that belongs to @xmath244 is dynamical , whereas the part that belongs to @xmath243 is fixed . since i proposed that the fundamental interactions should correspond to the universal group element ( [ g ] ) being dynamical , consistency requires that the part of ( [ g ] ) that belongs to @xmath243 should be dynamical as well , i.e. @xmath116 should be made dynamical . but the classical space - time geometry was constructed in section 4 using the latter group elements . it follows therefore that making @xmath116 dynamical would make the space - time metric dynamical and not fixed as it is in minkowski space - time . therefore , the interaction that corresponds to making the @xmath243 group elements dynamical gives , in the classical limit , the well known geometrical description of gravity in classical general relativity . a generalization of it is obtained by replacing @xmath243 with @xmath245 , where @xmath18 is any positive integer . thus the present approach requires the existence of gravity . i now give a simple illustration of the above unified way of treating gravity and gauge fields by considering the gravitational analog of the above vector aharonov - bohm effect . the geometry surrounding a cosmic string in the two dimensional section normal to the axis of the string at a given time is that of a cone whose center is at the axis , which is seen by solving the classical gravitational field equations @xcite . the space - time geometry of a non- rotating cosmic string is obtained by simply adding to this plane the extra dimension in the direction of the axis and the time dimension ; then the curvature outside the string is zero everywhere ( figure 3 ) . it is known that this geometry is similar to the electromagnetic field around a solenoid because the curvature is zero outside the string and yet there is a non trivial holonomy around it . consider now two wave packets separated by a distance @xmath171 and whose centers move along initially parallel lines such that the geodesic line @xmath197 joining their centers meets the conical singularity @xmath26 at its midpoint . but there is another geodesic that connects the same pair of points @xmath194 and @xmath195 of length @xmath246 , shown by the line @xmath247 in figure 3 , where @xmath248 is identified with @xmath194 . ( actually , there are two geodesics connecting @xmath194 and @xmath195 when the angle @xmath249 on the left side exceeds @xmath250 but is less than @xmath251 . the corresponding angles @xmath249 on the right side are @xmath251 and @xmath250 , respectively . the figure 3 . a gravitational analog of the experiment in fig . the conical geometry around a cosmic string that is normal to the plane through @xmath26 is represented by cutting off the wedge @xmath252 from flat space and identifying the planes along which it is cut . the wave packets moving at @xmath194 ( same as @xmath248 ) and @xmath195 are focused by this geometry to interfere at @xmath205 . just before @xmath197 crosses @xmath26 , there are two geodesics connecting @xmath194 and @xmath195 of lengths @xmath171 and @xmath253 in the curvature free region . but just after the crossing there is a unique geodesic @xmath247 of length @xmath253 joining @xmath194 and @xmath195 . singularity may be replaced by a small smooth cap that merges with the rest of the cone smoothly . then there would be a third geodesic joining @xmath194 and @xmath195 though this cap . ) just after @xmath197 crosses the conical singularity , there is only one geodesic joining @xmath194 and @xmath195 , whose length is @xmath253 . this is somewhat analogous to the change in modular kinetic momentum when @xmath197 crosses the solenoid in fig . 1 , because @xmath254 now translates locally along parallel geodesic line segments of length @xmath171 , and this length has gotten shorter after the crossing . in figure 4 , this result is generalized to the case of @xmath26 not being the midpoint of the geodesic @xmath255 , and it is also seen to be ` gauge ' independent in the sense of being invariant under the rotation of the ` wedge ' mentioned above . there is then a phase shift @xmath256 due to the difference in path lengths traveled by the wave packets given by @xmath257 for small @xmath258 , where and @xmath259 and @xmath260 are the path lengths @xmath261 and @xmath262 , @xmath263 is the mass per unit length of the cosmic string , and @xmath264 is the initial momentum of the particle . if the particle carries spin , then there is also a phase shift due to the coupling of spin to the curvature . a generalization of the gedanken experiment shown schematically in fig . 3 to unequal lengths @xmath265 and @xmath266 . the wedge @xmath267 has an arbitrary orientation . but since @xmath268 and @xmath269 are identified , and @xmath270 in order for @xmath271 to be a geodesic , the distances along the geodesics @xmath255 and @xmath271 are respectively @xmath272 and @xmath273 , as shown in fig . ( b ) , independently of the orientation of the wedge . this and other phase shifts for interference of two wave packets around a cosmic string are studied elsewhere and are understood as being due to the poincare holonomy around the string @xcite . in the present approach , these phase shifts may be understood by associating with each curve an element of the poincare group element @xcite whose expectation value changes as the curve crosses the cosmic string . the change in geodesic distances between the wave packets due to the cosmic string , in section 5 , is not surprising because gravity changes distances , according to general relativity , and the cosmic string is a purely general relativistic object without a newtonian analog . what may be more interesting is the similar change of the ` quantum distances ' due to the electromagnetic field in the usual aharonov - bohm effect and its generalization to non abelian gauge fields . both these effects may be treated in a somewhat analogous manner if the modular kinetic energy - momentum ( [ g ] ) , regarded here as universal group elements , may be interpreted as ` distances ' in a quantum geometry as proposed earlier . the fact that space - time distances may be obtained from them approximately , as shown in section 4 , reinforces this view . the treatment of ( [ g ] ) as an observable , which implies the above mentioned non local effects in quantum theory , perhaps removes the mystery of why although the interactions are local as they occur in the hamiltonian or lagrangian , nevertheless there are non- local effects such as the aharonov - bohm and its generalizations to non abelian gauge fields and gravitation @xcite . a criticism that may be made against regarding the universal group elements ( [ g ] ) as quantum distances is that their action on wave functions seems to require that @xmath105 be interpreted as a set of space - time curves , whereas it was argued in section 3 that space - time geometry is not appropriate for quantum theory . however , as discussed in that section , this limitation becomes critical only at the planck scales . but at the planck scale these group elements need not be associated with curves in space - time . they may be defined simply as operators acting on quantum states defining the quantum geometry and representing the fundamental interactions . thus it is possible to have a quantum geometry even at scales in which the space - time geometry breaks down and this would also give a quantum description of all the interactions . the action of @xmath274 on an analytic wave function @xmath28 may be understood as the usual taylor expansion of @xmath28 . if @xmath28 is not analytic , then @xmath275 acts on the momentum space wave function @xmath81 according to @xmath276 in either case , we obtain @xmath277 . this explains why in the non relativistic limit for the charged particle , the aharonov - bohm phase shift that it experiences due to the phase factor @xmath212 remains the same . this is because the latter phase factor is unaffected by the change in the space - time metric that takes place in the non relativistic limit . this reconciles the views of y. aharonov and c. n. yang , who regarded the aharonov - bohm effect as being non local and local , respectively , in proceedings of the international symposium on the foundations of quantum mechanics , tokyo , august 1983 , edited by s. kamefuchi et al . ( physical society of japan , tokyo , 1984 ) 65 - 73 .
the origin of mass is a central mystery of the standard model ( sm ) . in particular , why are the masses of w , z , and the top quark exceptionally large compared with those of other particles in the sm ? the idea of the top quark condensate @xcite explains naturally the large top mass of the order of the electroweak symmetry breaking ( ewsb ) scale . in the explicit formulation of this idea @xcite , often called the `` top mode standard model '' ( tmsm ) , the scalar bound state of @xmath9 plays the role of the higgs boson in the sm . ( for reviews , see , e.g. , refs . @xcite . ) there are several problems in the original version of the tmsm : we need to introduce ad hoc four - fermion interactions of the top quark in order to trigger the ewsb . the top mass @xmath10 , given as a decreasing function of the composite scale @xmath11 , is predicted about 10%30% larger than the experimental value , even if we take @xmath11 to the planck or the gut scale @xcite . such a huge @xmath11 also causes a serious fine - tuning problem . as a possible solution to these problems , following the line of an earlier attempt @xcite of the tmsm in the tev - scale extra dimension scenario @xcite , arkani - hamed , cheng , dobrescu and hall ( acdh ) @xcite proposed an interesting version of such where the sm gauge bosons and the third generation of quarks and leptons live in the @xmath12-dimensional bulk , while the first and second generations are confined in the 3-brane ( 4-dimensional minkowski space - time ) . gauge interactions in higher dimensions than four become strong in a certain high - energy region . bulk gauge interactions are expected to trigger the top condensation without adding ad hoc four - fermion interactions , in contrast to the original version of the tmsm . however , the dynamics of bulk gauge theories was not concretely analyzed in ref . in particular , as it turned out @xcite ( see also refs . @xcite ) , the bulk qcd coupling , which is the most relevant interaction for the top condensation , has an ultraviolet fixed point ( uv - fp ) or upper bound within the same @xmath13-scheme of the truncated kaluza - klein ( kk ) effective theory @xcite as that ref . @xcite was based on . thus , it is quite nontrivial whether the top condensation is actually realized or not . in refs . @xcite , we have studied the dynamical chiral symmetry breaking ( d@xmath14sb ) in bulk gauge theories , based on the ladder schwinger - dyson ( sd ) equation.sb in other scenario for extra dimensions was investigated in ref . switching off the electroweak interaction in the bulk , we then found that the bulk qcd coupling can not become sufficiently large to trigger the top condensation for @xmath2 , while the top condensation can be realized for @xmath1 . for the purpose of model building , we further need to study the effect of the bulk electroweak interactions : since the bulk @xmath15 interaction grows very quickly due to the power - like running behavior and reaches immediately its landau pole @xmath16 , it may affect the most favored channel for condensate , i.e. , the most attractive channel ( mac ) @xcite . we also need to study whether or not the prediction of the top mass agrees with the experiments . in this paper , we demonstrate a possibility that the top condensate is actually the mac even including all of the bulk sm gauge interactions . this is quite nontrivial , because inclusion of the strong bulk @xmath15 interaction may favor the tau condensation rather than the top condensation . in order for only the top quark to acquire the dynamical mass of the order of the ewsb scale , the binding strength should exceed the critical binding strength @xmath17 only for the top quark ( `` topped mac '' or `` tmac '' ) . namely , our scenario works only when @xmath18 where @xmath19 , @xmath20 and @xmath21 denote the binding strengths of the top , bottom , and tau condensates at the scale @xmath11 , respectively . we refer to the scale @xmath11 satisfying eq . ( [ top - cond ] ) as the tmac scale @xmath22 . for the mac analysis , we study binding strengths @xmath23 by using the one - loop renormalization group equations ( rges ) of dimensionless bulk gauge couplings . it is in contrast to the analysis of acdh @xcite where all of bulk gauge couplings are assumed equal ( and strong enough for triggering the ewsb ) . in order to check reliability of our mac analysis , we also study the regularization - scheme dependence of the binding strengths . we calculate gauge couplings in two prescriptions , the @xmath13-scheme of the truncated kk effective theory and the proper - time ( pt ) scheme @xcite . there are some varieties in the estimation of @xmath17 : the naive dimensional analysis ( nda ) @xcite implies @xmath24 , while the ladder sd equation yields much smaller value @xmath25 @xcite . as the estimate of @xmath17 increases in eq . ( [ top - cond ] ) , the region of the tmac scale gets squeezed . even if we adopt the lowest possible value of @xmath17 given by the ladder sd equation @xcite , we find that _ the tmac scale does not exist for the simplest scenario with @xmath2_. on the other hand , _ the tmac scale does exist in @xmath1 _ for the value of the ladder sd equation , @xmath26@xmath27 , where the compactification scale @xmath28 is taken to be 1100 tev . for @xmath3 , the mac analysis significantly depends on the regularization scheme . once we obtain the tmac scale @xmath22 , we can easily predict the top mass @xmath10 and the higgs mass @xmath29 by using the renormalization group equations ( rges ) for the top yukawa and higgs quartic couplings , and the compositeness conditions @xcite at the scale @xmath30 . this is in contrast to the earlier approach @xcite ( see also ref . @xcite ) where the composite scale @xmath11 is treated as an adjustable free parameter and fixed so as to reproduce the experimental value of @xmath10 . without such an adjustable parameter , we predict the top quark mass @xmath31 for @xmath1 and @xmath32 1100 tev . this agrees with the experimental value , @xmath33 gev @xcite . we find that the value of @xmath10 near the compactification scale @xmath28 is governed by the quasi infrared fixed point ( ir - fp ) for the top yukawa coupling @xmath34 @xcite , which is approximately obtained as @xmath35 with the 4-dimensional qcd coupling @xmath36 , the number of color @xmath37 , the quadratic casimir of the fundamental representation @xmath38 , and @xmath39 . the suppression factor @xmath40 in @xmath34 arises from one - loop effects of the bulk top quark which is equivalent to a tower of kk modes ( massive vector - like fermions ) . the mechanism suppressing the top mass prediction is thus similar to that of the top seesaw @xcite . we also predict the higgs boson mass as @xmath41 thanks to the ir - fp property , the prediction for @xmath10 and @xmath29 is stable . the higgs boson with such a characteristic mass can be distinguished clearly from that of supersymmetric or other typical dynamical ewsb models simply through its mass determination in experiments . it will also be discovered immediately after the physics run of lhc . the paper is organized as follows . in sec . ii , we study running effects of bulk gauge couplings in the @xmath13- , and pt - schemes . in sec . iii , we identify the tmac scale . in sec . iv , we predict @xmath10 and @xmath29 for @xmath1 . sec . v is devoted to summary and discussions . in appendix a , we present details of our procedure for the orbifold compactification . we also count the total number of kk modes below the renormalization point . appendix b is for details of bulk gauge couplings in the pt - scheme . let us consider a simple version of the tmsm with extra dimensions where the sm gauge group and the third generation of quarks and leptons are put in @xmath42-dimensional bulk , while the first and second generations live on the 3-brane ( 4-dimensional minkowski space - time ) . the @xmath42-dimensions consist of the usual 4-dimensional minkowski space - time and extra @xmath43 spatial dimensions compactified at a tev - scale @xmath28 . the number of dimensions @xmath42 is taken to be even , @xmath44 , so as to introduce chiral fermions in the bulk . in order to obtain a 4-dimensional chiral theory and to forbid massless gauge scalars , we compactify extra dimensions on the orbifold @xmath45 ( see appendix a ) . we emphasize that there is no elementary field for higgs in our model . the chiral condensation of bulk fermions may generate dynamically a composite higgs field , instead . hence we investigate rges of bulk gauge couplings including loop effects of the composite higgs . we expand bulk fields into kk modes and construct a 4-dimensional effective theory . in this subsection , we study running of gauge couplings in the `` truncated kk '' effective theory @xcite based on the @xmath13-scheme . below the compactification scale @xmath28 , rges of 4-dimensional gauge couplings @xmath46 are given by those of the sm , @xmath47 with @xmath48 and @xmath49 for one ( composite ) higgs doublet . we need to take into account contributions of kk modes in @xmath50 . since the kk modes heavier than the renormalization scale @xmath51 are decoupled in the @xmath13-rges , we only need summing up the loops of the kk modes lighter than @xmath51 . within the truncated kk effective theory , we obtain rges for gauge couplings @xmath46 : @xmath52 where rge coefficients @xmath53 are given by @xmath54 for @xmath55 , @xmath56 for @xmath57 , and @xmath58 for @xmath15 . here @xmath59 is defined as @xmath60 in eqs . ( [ b3kk])([bykk ] ) , @xmath61 denote the total number of kk modes for gauge bosons , gauge scalars , dirac ( 4-component ) fermions , and composite higgs bosons below @xmath51 , respectively . the number of generations in the bulk is @xmath62 , which is unity in our model . in the following analysis , we assume one composite higgs doublet , @xmath63 . @xmath64 for @xmath65 are not necessarily equal , since the orbifold boundary conditions are depending on @xmath66 . for details , see appendix a. we show numerical values for @xmath67 , ( the solid lines ) , @xmath68 , ( the dashed lines ) , and @xmath69 , ( the dotted lines ) in figs . [ fig - g-6d](b ) , [ fig - g-8d](b ) and [ fig - g-10d](b ) . matching the 4-dimensional action to the bulk action , we find the relation between the 4-dimensional gauge coupling @xmath70 and the _ dimensionful _ bulk gauge coupling @xmath71 , @xmath72 . we define the _ dimensionless _ bulk gauge coupling @xmath73 as @xmath74 and thereby obtain @xmath75 combining eq . ( [ hat - g ] ) with eq . ( [ rge_ed ] ) , we find rges for @xmath76 , @xmath77 . \label{rge_ed2}\end{aligned}\ ] ] in figs . [ fig - g-6d](a ) , [ fig - g-8d](a ) and [ fig - g-10d](a ) , we show typical behavior of the dimensionless bulk gauge couplings . we have used the couplings @xmath78 at @xmath79 as inputs of rges : @xcite @xmath80 @xmath81 and @xmath82 , i.e. , @xmath83 we note here that the @xmath15 has the landau pole @xmath16 . the bulk gauge coupling @xmath84 rapidly grows due to the power - like behavior of the running . as a result , the landau pole @xmath16 is close to the compactification scale @xmath28 . a cutoff smaller than the landau pole @xmath16 needs to be introduced in our model . a handy approximation for @xmath85 is widely used , @xmath86 2^{\delta/2 - 1 } n_{\rm kk}(\mu ) & \text{for $ i = f,$ } \end{cases } \label{nkki_app}\ ] ] with @xmath87 we show @xmath88 in figs . [ fig - g-6d](b ) , [ fig - g-8d](b ) and [ fig - g-10d](b ) with the dash - dotted line . the deviation of eq . ( [ nkki_app ] ) from the numerical calculations of @xmath64 is negligibly small for @xmath2 , whereas it is sizable for @xmath89 in @xmath90 . we do not use the approximation of eq . ( [ nkki_app ] ) for the mac analysis ) in the previous report @xcite . while the mac analysis is somewhat affected by the approximation , the predictions of @xmath10 and @xmath29 are not . ] . here we comment on the uv - fp discussed in ref . @xcite , which relies on the approximation of eq . ( [ nkk_app ] ) . ( see also refs . @xcite . ) using eqs . ( [ nkki_app ] ) and ( [ nkk_app ] ) , ref . @xcite gave a rge formula @xmath91 with @xmath92 being the loop factor in @xmath42 dimensions , @xmath93 the rge coefficients @xmath94 are @xmath95 the rge ( [ rge_ed3 ] ) leads to the uv - fp @xmath96 @xcite , @xmath97 for @xmath98 . in the case of @xmath99 and @xmath100 , @xmath101 become negative and hence the bulk qcd has such a uv - fp . on the other hand , in @xmath102 the bulk qcd coupling has a landau pole at @xmath103 . there are several manners to define bulk gauge couplings . in this subsection , we study bulk gauge couplings based on the proper - time ( pt ) regularization scheme with a matching condition slightly different from that in ref . @xcite . in the pt - scheme , one - loop contributions of all kk modes are smoothly ( exponentially ) suppressed , whereas the contributions of kk modes heavier than the renormalization scale @xmath51 are sharply cut off in the truncated kk effective theory . we here note that the landau pole @xmath16 is not so far from the compactification scale @xmath28 , particularly for @xmath89 . under such a situation , numerical differences between @xmath13- , and pt - couplings may be significant . we briefly present our definition of the pt - coupling and give the matching condition to the @xmath13-coupling through the effective charge . see appendix b for details . our 4-dimensional pt - coupling @xmath104 is a bare quantity defined at the cutoff @xmath11 , in sharp contrast to the @xmath13-coupling @xmath105 , which is a renormalized quantity . the effective charge @xmath106 with euclidean momentum @xmath107 is obtained in the pt - scheme as @xmath108 with @xmath109 , @xmath110 and a constant @xmath111 to be discussed later . the coefficient @xmath112 of the logarithmic divergent term from zero modes is given by @xmath113 where @xmath114 is the quadratic casimir of the adjoint representation , @xmath115 the trace of the product of two generator matrices , @xmath116 , and @xmath117 the number of flavor of 4-component fermions . the constant term @xmath118 in eq . ( [ eff_pt_q ] ) arises from zero modes and is given by @xmath119 with @xmath120 the kk mode summation of the vacuum polarization function @xmath121 is calculated in the pt - scheme , @xmath122e^{-tx(1-x)q_e^2}k_g(t ) \nonumber \\ & & \hspace*{-4mm}- \frac{\delta}{2 } \ , \frac{c_a}{(4\pi)^2 } \int_0 ^ 1dx\int_{r\lambda^{-2}}^\infty\frac{dt}{t}\ , ( 2x-1)^2 e^{-tx(1-x)q_e^2 } k_{gs}(t ) \nonumber \\ & & \hspace*{-4mm}- 8t_r \ , \frac{n_f}{(4\pi)^2 } \int_0 ^ 1dx\int_{r\lambda^{-2}}^\infty\frac{dt}{t}\ , x(1-x ) e^{-tx(1-x)q_e^2 } k_f ( t ) \nonumber \\ & & \hspace*{-4mm}- t_r \ , \frac{n_h}{(4\pi)^2 } \int_0 ^ 1dx\int_{r\lambda^{-2}}^\infty\frac{dt}{t}\ , ( 2x-1)^2 e^{-tx(1-x)q_e^2 } k_h ( t ) , \nonumber \\ \label{pi_pt_q}\end{aligned}\ ] ] where @xmath123 s are defined as @xmath124_1 } \left[\,\frac{1}{2}(\vartheta_3 - 1)\,\right ] + { \cal n}_i^{\delta,[n]_2 } \left[\,\frac{1}{2}(\vartheta_3 - 1)\,\right]^2 + \cdots\ ] ] for @xmath65 , with @xmath125 being the jacobi @xmath125 function , @xmath126 . the definition of the factor @xmath127_k}$ ] is given in appendix a. ( for values of @xmath127_k}$ ] in @xmath128 , see table [ tab1][tab3 ] . ) on the other hand , the effective charge @xmath106 is also calculated in the @xmath13-scheme @xcite , @xmath129 where @xmath130 and @xmath131 \nonumber \\ & & + \frac{\delta}{2 } \ , c_a \ , i_s^r(q_e^2;m_{\vec n}^2 ) \nonumber \\ & & + 8 t_r \ , n_f \ , i_f^r(q_e^2;m_{\vec n}^2 ) \nonumber \\ & & + t_r \ n_h \ , i_s^r(q_e^2;m_{\vec n}^2 ) , \end{aligned}\ ] ] with @xmath132 the constant term from zero modes is given by @xmath133 now , we impose the matching condition to relate the pt- and @xmath13-couplings : @xmath134 i.e. , @xmath135 following ref . @xcite , we further impose a condition , @xmath136 in @xmath137 . the parameter @xmath111 of the pt - cutoff is then determined as @xmath138 the matching condition in ref . @xcite roughly corresponds to @xmath139 with @xmath140 . there is an ambiguity associated with the matching scale @xmath141 . on the other hand , eq . ( [ match_cond ] ) includes effects of the finite parts @xmath142 and hence is less ambiguous about the matching scale . the matching condition ( [ match_cond ] ) determines values of pt - couplings at @xmath143 tev , @xmath144 while the corresponding values in the @xmath13-scheme at @xmath145 tev are @xmath146 where inputs are eqs . ( [ qcd - mz ] ) and ( [ su2-mz ] ) . at the compactification scale @xmath28 , the scheme dependence between the pt- , @xmath13-couplings are not significant for @xmath128 . in order to discuss the scheme dependence at the scale beyond @xmath28 , we define the dimensionless bulk gauge coupling in the pt - scheme @xmath147 , @xmath148 in the same way as eq . ( [ hat - g ] ) , where @xmath149 is the 4-dimensional coupling . in the next section , we will discuss the scheme dependence of the couplings , @xmath147 and @xmath150 : the scheme dependence near the landau pole @xmath16 is small for @xmath151 , while it is significant for @xmath3 . in this section , we analyze the energy scale @xmath22 ( tmac scale ) where the top condensate is the mac and only in the @xmath9-channel the binding strength exceeds the critical value , i.e. , @xmath152 where @xmath23 are given in terms of the sm gauge couplings in the bulk such as those shown in figs . [ fig - g-6d](a ) , [ fig - g-8d](a ) , [ fig - g-10d](a ) and the critical binding strength @xmath17 is determined by the ladder sd equation @xcite . this is contrasted to the mac analysis in ref . @xcite where it was assumed that @xmath153 and that the value is large enough to trigger the d@xmath14sb . however , as we have shown in figs . [ fig - g-6d](a ) , [ fig - g-8d](a ) and [ fig - g-10d](a ) , values of the couplings at the scale where @xmath154 are not necessarily large , @xmath155 , for @xmath128 . it is thus highly non - trivial whether or not the tmac scale @xmath22 exists . as usual , we discuss the mac based on the one - gauge - boson - exchange approximation @xcite . the binding strength @xmath156 of a @xmath157 channel is given by @xmath158 where @xmath159 , @xmath160 are the generators of @xmath161 , and @xmath162 is the hypercharge . noting the identity @xmath163 with @xmath164 being the quadratic casimir for the representation @xmath111 of the gauge group , we calculate the binding strengths : @xmath165 for the top , bottom and tau condensates , respectively , where @xmath166 is the quadratic casimir of the fundamental representation of @xmath55 . in the following analysis , we study these three channels . we next turn to the estimation of the critical binding strength @xmath17 . often applied is the naive dimensional analysis ( nda ) @xcite , @xmath24 . on the other hand , the value of @xmath17 is fairly smaller than that of the nda in the approach of the improved ladder sd equation @xcite , @xmath167 there are actually some issues in the estimation of @xmath17 through the ladder sd equation : * non - ladder corrections may push the value of @xmath17 down as much as 120% , as in the analysis of the 4-dimensional walking technocolor @xcite . * another momentum identification of the ( dimensionful ) running coupling constant requires the nonlocal gauge fixing @xcite , which pushes up the value of @xmath17 as @xcite , @xmath168 * the values in eqs . ( [ kd_sd1])([kd_sd2 ] ) were obtained under the assumption that the dimensionless bulk gauge coupling @xmath73 is constant . as shown in figs . [ fig - g-6d](a ) , [ fig - g-8d](a ) and [ fig - g-10d](a ) , all of bulk gauge couplings @xmath169 are monotonously increasing functions . this means that the attractive force is overestimated under the simplification of @xmath170 const . , i.e. , the momentum dependence of @xmath73 makes values of @xmath17 larger than eqs . ( [ kd_sd1])([kd_sd2 ] ) . * effects of the compactification were also neglected in the estimate of eqs . ( [ kd_sd1])([kd_sd2 ] ) . if such effects are included , the value of @xmath17 becomes also larger . taking these issues into account , we may regard the value of @xmath17 in eq . ( [ kd_sd1 ] ) to be a reference value as the best compromise . + now , we are ready to investigate the tmac scale . we show behavior of the binding strengths @xmath23 both in the @xmath13- and pt - couplings in figs . [ fig - tmac-6d][fig - tmac-10d ] to study the scheme dependence . we first discuss the tmac scale based on the @xmath13-coupling . we compare attractive forces of top , bottom , and tau condensates with @xmath17 in figs . [ fig - tmac-6d](a ) , [ fig - tmac-8d](a ) and [ fig - tmac-10d](a ) for @xmath171 tev . as the value of @xmath17 becomes larger , the tmac scale gets squeezed . for @xmath2 , we find that _ the tmac scale is squeezed out _ , if we use _ the reference value _ @xmath172 eq . ( [ kd_sd1 ] ) . the conclusion is unchanged against varying the compactification scale as far as @xmath173 tev . actually , the tau condensation instead of the top condensation becomes the mac . in order for the tmac scale to survive , the value of @xmath172 should be lower than the reference value by about 20% for @xmath32 1100 tev , which is highly unlikely . thus the scenario does not work for @xmath2 . for @xmath1 , on the other hand , the tmac scale @xmath22 satisfying eq . ( [ top - cond - sec3 ] ) does exist , @xmath174 for the reference value of @xmath17 . if we vary the compactification scale as @xmath175100 tev , we find the tmac scale , @xmath176 for @xmath3 , the tmac scale does not exist for @xmath177 tev . if we admit @xmath178 tev , we barely find the tmac scale @xmath179@xmath180 for @xmath181@xmath182 tev . we have studied the tmac scale so far by using the @xmath13-coupling . in order to clarify the scheme dependence , we also investigate the pt - coupling in the mac analysis . we show binding strengths @xmath23 in the pt - scheme in figs . [ fig - tmac-6d](b ) , [ fig - tmac-8d](b ) and [ fig - tmac-10d](b ) for @xmath171 tev . we find that the scheme dependence is negligibly small for @xmath2 . although behavior of the binding strengths @xmath23 for @xmath1 somewhat depends on the regularization scheme , there does exist an overlapped region of the tmac scale between the @xmath13- and pt - schemes . for @xmath3 , however , the scheme dependence is rather large : there is no overlap of the tmac scale for @xmath175@xmath183 tev . in conclusion we have shown that the region of the tmac scale is squeezed out for @xmath2 , while the tmac scale does exist for @xmath1 without much ambiguity . for @xmath3 , we can not obtain a reliable result because of the significant scheme dependence . in this section , we calculate the top mass @xmath10 and the higgs mass @xmath29 by using rges of the top yukawa and higgs quartic couplings with the compositeness conditions @xcite in a way used by acdh @xcite . let us first recapitulate the compositeness conditions in ref . @xcite : assume that the top condensation takes place in the bulk . then , at the compositeness scale @xmath11 , a scalar bound state @xmath184 ( composite higgs ) , @xmath185 is formed in the bulk , where @xmath186 $ ] and @xmath187 are bulk fermions . at this stage , the composite higgs @xmath184 does not propagate in the bulk . its kinetic term is expected to develop at the scale below @xmath11 . it is assumed that the effective theory below @xmath11 is described by the bulk sm . in the truncated kk effective theory of the bulk sm , the compositeness conditions then read @xmath188 where @xmath34 and @xmath189 denote the top yukawa and higgs quartic couplings , respectively . we shall use the above procedure to calculate @xmath10 and @xmath29 . in contrast to the acdh analysis , however , we have shown existence of the tmac scale for @xmath1 without much ambiguity and hence the extra dimension scenario of the tmsm is actually possible . furthermore , while the compositeness scale @xmath11 in ref . @xcite was treated as a free parameter to be adjusted for reproducing the experimental value of @xmath10 , we identify the compositeness scale @xmath11 with the tmac scale @xmath22 , @xmath190 which is no longer an adjustable parameter but constrained as eq . ( [ tmac ] ) . thus we can test our model by comparing the predicted @xmath10 with the experimental value . within the truncated kk effective theory , rges for @xmath34 and @xmath189 read ) are missing in the rges of ref . @xmath191 & ( 4\pi)^2 \mu \frac{d \lambda}{d \mu } = \beta_\lambda^{\rm sm } + \beta_\lambda^{\rm kk } , \label{rge_lam}\end{aligned}\ ] ] where @xmath192 , \\[2 mm ] \beta_y^{\rm kk } & = & \left ( 6 { n_{\rm kk}^f}+\frac{3}{2 } { n_{\rm kk}^h}\right)\,y^3 \nonumber \\ & & \quad -{n_{\rm kk}^g}\left ( 8 g_3 ^ 2 + \frac{9}{4 } g_2 ^ 2 + \frac{17}{12 } g_y^2 \right)\,y \nonumber \\ & & \quad - \delta { n_{\rm kk}^{gs}}\left(\frac{4}{3 } g_3 ^ 2 - \frac{3}{8 } g_2 ^ 2 - \frac{1}{72 } g_y^2 \right)\,y , \label{rge_y - ed } \\ \beta_\lambda^{\rm sm } & = & 12 \left(\lambda y^2-y^4\right ) + 12\lambda^2 \nonumber \\ & & \quad + \frac{3}{4}(3 g_2 ^ 4 + 2g_2 ^ 2 g_y^2 + g_y^4 ) \nonumber \\ & & \quad -3(3g_2 ^ 2+g_y^2)\lambda , \\ \beta_\lambda^{\rm kk } & = & 24 { n_{\rm kk}^f}\left(\lambda y^2-y^4\right ) + 12 { n_{\rm kk}^h}\lambda^2 \nonumber \\ & & \hspace*{-0.5 cm } + { n_{\rm kk}^g}\left[\ , \frac{3}{4}\left(3 g_2 ^ 4 + 2g_2 ^ 2 g_y^2 + g_y^4\right ) - 3(3g_2 ^ 2+g_y^2)\lambda\,\right ] \nonumber \\ & & \hspace*{-0.5 cm } + \frac{\delta}{4 } { n_{\rm kk}^{gs}}(3 g_2 ^ 4 + 2g_2 ^ 2 g_y^2 + g_y^4 ) . \end{aligned}\ ] ] note that @xmath193 and @xmath194 stand for contributions of zero modes and kk modes , respectively . we calculate the rges by using the uv - bcs eq . ( [ comp - cond ] ) and determine @xmath10 and @xmath29 through the conditions , @xmath195 with @xmath196 gev . + + + we show results of @xmath10 and @xmath29 in figs . [ mt - mh-6d ] , [ mt - mh ] , [ mt - mh-10d ] for @xmath128 , @xmath171 tev for various values of the compositeness scale @xmath11 . as we have shown in sec . iii , the tmac scale does exist only for @xmath1 without much ambiguity , @xmath26@xmath27 . identifying @xmath11 with @xmath22 , we depict the region of the tmac scale for @xmath1 by the shaded area in fig . [ mt - mh ] . there is no shaded region in figs . [ mt - mh-6d ] and [ mt - mh-10d ] , because a sensible tmac scale is absent for @xmath197 . for @xmath1 we predict @xmath198 and @xmath199 for the range of the compactification scale @xmath32 1100 tev . the uncertainties in eqs . ( [ 8d - mt ] ) and ( [ 8d - mh ] ) also include error of @xmath200 @xcite . remarkably , the prediction eq . ( [ 8d - mt ] ) for @xmath10 is consistent with the reality , the @xmath13-mass @xmath201 gev which is calculated from the observed value of the pole mass , @xmath202 gev @xcite . it should be emphasized that the compositeness scale @xmath11 in ref . @xcite is an arbitrary parameter and was adjusted to reproduce the experimental value of the top mass . on the contrary , our compositeness scale @xmath11 is fixed by the tmac scale @xmath22 by requiring that the top quark condensation actually takes place , while other condensations do not . hence the top mass as well as the higgs mass is the prediction in our approach . compared with the prediction of ref . @xcite , @xmath203 gev for @xmath1 , eq . ( [ 8d - mh ] ) is determined more sharply . this is because our compositeness scale @xmath11 is given by the tmac scale @xmath22 which is more severely constrained than @xmath11 of ref . @xcite , the value determined from the top mass within 3@xmath204 of the experimental value . we now discuss the reason why the value of eq . ( [ 8d - mt ] ) is significantly smaller than that of the original tmsm in four dimensions which predicted @xmath205 gev . let us consider a simplified rge for @xmath34 neglecting the electroweak gauge interactions : @xmath206 with @xmath207 , @xmath208 . using eq . ( [ nkki_app ] ) as an approximation for @xmath209 and equating the r.h.s . of eq . ( [ y_app ] ) to zero , we find the quasi ir fixed point @xmath210 @xcite @xmath211 the value of eq . ( [ qir ] ) decreases as @xmath212 increases at @xmath213 , since the value of @xmath214 is determined within the sm independently of @xmath59 . in the original tmsm with @xmath215 the prediction of @xmath10 is governed by eq . ( [ qir ] ) with @xmath216 @xcite . as a result , the prediction of @xmath10 with @xmath217 is substantially lower than that of the original tmsm with @xmath218 . the mechanism is still operative even including the electroweak gauge interactions : in fig . [ ir - fp ] , we show the quasi ir fixed point and the behavior of @xmath34 based on the full one - loop rge ( [ rge_y ] ) with various boundary conditions at @xmath11 . we also show the pendleton - ross ( pr ) fixed point @xcite determined from @xmath219 : @xmath220 with @xmath101 being @xmath221 and @xmath222 , @xmath223 . for @xmath224 the pr fixed point is identical to the quasi ir fixed point . as far as @xmath225 @xmath226 , the value of the pr fixed point is smaller than that of the quasi ir fixed point , @xmath227 , while @xmath228 for @xmath3 . the top yukawa coupling at @xmath28 for @xmath151 is actually between @xmath229 and @xmath230 for a sufficiently large top yukawa , @xmath231 , at high energy scale @xmath232 . we note here that the actual prediction of @xmath10 with @xmath151 is even smaller than the value expected from @xmath230 . we also comment that the predicted values of @xmath10 and @xmath29 would be stable thanks to these fixed points , even if the estimate of the tmac scale were somewhat changed from ours for some reason . as was already pointed out in ref . @xcite , the lower value prediction of @xmath10 than that of the original tmsm can also be understood as follows : since kk modes of the top quark ( @xmath233 ) as well as its zero mode ( @xmath234 ) contribute to the vev @xmath235 , @xmath236 the condensate @xmath237 is suppressed compared with the original tmsm and so is the top mass . now we discuss implication of our higgs mass prediction eq . ( [ 8d - mh ] ) . the upper limit of @xmath29 from radiative corrections in the sm is @xmath238 gev at 95% cl @xcite . the prediction eq . ( [ 8d - mh ] ) is still below this upper limit . in order to discriminate the present model from the sm in experiments , we need further information of physics at the compactification scale . however , the higgs boson in such a mass range , @xmath7 gev , is characteristically small within the framework of the dynamical ewsb . yet the value is substantially larger than that of typical supersymmetric models , @xmath239 gev . thus the present scenario is clearly distinguished from most of the models beyond the sm simply through the higgs mass observation . the higgs boson of this mass range decays into weak boson pair almost 100% . it will be immediately discovered in @xmath8 once the lhc starts . we have argued a viable top mode standard model ( tmsm ) with tev - scale extra dimensions where bulk @xmath240 sm gauge interactions ( without ad hoc four - fermion interactions ) trigger condensate of only the top quark , but not of other quarks and leptons . in order for such a situation to be realized , the binding strength @xmath156 should exceed the critical binding strength @xmath17 only for the top quark ( tmac ) , @xmath241 the binding strengths @xmath23 were calculated by using rges for bulk sm gauge couplings . comparing @xmath23 with @xmath17 , we searched for the region of the tmac scale @xmath22 satisfying eq . ( [ top - cond - fin ] ) . we then found that the region of the tmac scale is squeezed out for @xmath2 for the reference value of @xmath17 in ref . @xcite , while it does exist for @xmath1 , @xmath242 ( 3.53.6)@xmath28 . we were not able to draw a reliable conclusion for @xmath3 , since the mac analysis for @xmath3 strongly depends on the regularization scheme . for @xmath1 , we predicted the top mass @xmath10 and the higgs mass @xmath29 : @xmath243 and @xmath244 by using rges for the top yukawa and higgs quartic couplings with the compositeness conditions at the tmac scale @xmath22 . our predictions are governed by the quasi ir - fp and hence are stable against varying the composite scale . the predicted values would not be changed so much , even if the region of the tmac scale got wider than our estimate for some reason . why is the value of eq . ( [ mt - fin ] ) significantly smaller than that of the original tmsm in four dimensions which predicted @xmath205 gev ? the value of the top yukawa coupling at the quasi ir - fp is @xmath245 . thanks to the suppression factor @xmath40 in @xmath34 , the mass of the top quark decreases as the number of dimensions increases . the predicted higgs mass eq . ( [ mh - fin ] ) is substantially larger than that of typical supersymmetric models , while it is distinctively small as a dynamical ewsb scenario . we are thus able to distinguish the present model from typical supersymmetric or dynamical ewsb scenarios simply through the higgs mass observation . we also note that the higgs boson in this mass range will be immediately discovered in @xmath8 once the lhc starts . the upper limit of @xmath29 from radiative corrections in the sm is @xmath238 gev at 95% cl @xcite and the predicted value eq . ( [ mh - fin ] ) is below this upper limit , however . it is thus difficult to discriminate the higgs in the present model from that of the sm only from its mass . we definitely need experimental information other than @xmath29 in order to establish the present model . many issues remain to be explored : \1 ) our results on the tmac scale are sensitive to the value of @xmath17 . although we used the reference value of @xmath17 in the approach of the ladder sd equation neglecting the effect of the compactification , it would be more preferable if we can determine @xmath17 more reliably . for such a purpose , we should take into account effects of the compactification scale @xmath28 which turned out not so small compared with the tmac scale @xmath22 in our analysis . we also need running of bulk gauge couplings beyond one - loop perturbation . \2 ) we incorporated only one composite higgs doublet into rges , assuming other possible bound states such as vector / axial - vector bosons are irrelevant . in order to justify the assumption , we need to solve bound state problems in the bulk gauge theories . once such a composite scalar exists , it should be a tightly bound state formed by strong short distance dynamics with large anomalous dimension . such a system is expected to resemble the gauged nambu - jona - lasinio ( gnjl ) model where the compositeness condition is explicitly formulated . actually , as it happened in the 4-dimensional case @xcite , the pure gauge dynamics strong at short distance in our case can also induce strong four - fermion interactions which may become relevant operators due to large anomalous dimensions , @xmath246 @xcite . we will report elsewhere the phase structure of the bulk gnjl model @xcite . \3 ) there are potential constraints on our model from precision electroweak measurements . the summation of kk modes below the cutoff @xmath11 contributes to @xmath247 as @xmath248 @xcite . in our case with @xmath30 , @xmath249 for @xmath1 , we thus need to take @xmath250 , which may be subtle about the fine tuning . more involved estimate will be done elsewhere . \4 ) masses of other quarks and leptons have not been dealt with in this paper . in the original tmsm , these masses are descended from the top condensate through ad hoc flavor - breaking four - fermion interactions @xcite . the origin of such four - fermion interactions will be highly hoped for in the present scenario . \5 ) our scenario crucially relies on the short distance strong dynamics around the composite scale . we thus need a better - controlled theory in the uv - region . it would be interesting to study a deconstructed / latticized version @xcite of our model . the work is supported in part by the jsps grant - in - aid for the scientific research ( b)(2 ) 14340072 ( k.y . and m.t . ) and by krf pbrg 2002 - 070-c00022 ( m.h . ) . the torus compactification of extra dimensions leaves vector - like zero - modes in the 4-dimensional space - time , even if we start with a chiral gauge theory in the bulk . we thus need to compactify the extra dimensions to an orbifold in order to obtain the conventional sm particles as zero modes . an example of such a compactification was explicitly constructed in ref.@xcite based on the @xmath252 orbifold for @xmath253 ( @xmath254 even ) dimensions . in this appendix , we discuss the massive kk spectrum in the orbifold compactification of ref.@xcite . let us start with the 6-dimensional case ( @xmath253 , @xmath256 ) . the 6-dimensional space - time is decomposed into conventional and extra dimensions , @xmath257 with @xmath258 , @xmath259 and @xmath260 . we assume the compactification radius @xmath261 and the fields at @xmath262 and @xmath263 are identified with the field at @xmath264 . we thus impose the periodic boundary conditions , @xmath265 for a gauge field and @xmath266 for a fermion . the @xmath267 identification of the orbifold @xmath268 leads to the orbifold boundary conditions ( bcs ) , @xmath269 here @xmath270 and @xmath271 are chirality matrices in four- and six - dimensions , respectively to that in ref . ] , @xmath272 it is easy to show that the gauge - vector field @xmath273 is decomposed into its kk - modes , @xmath274 } \cos\left(\dfrac{n_1 y^5}{r}\right ) \nonumber\\ & & + \sum_{n_1>0 } a_{\mu , 0c}^{[n_1 ] } \cos\left(\dfrac{n_1 y^6}{r}\right ) \nonumber\\ & & + \sum_{n_1 , n_2 > 0 } a_{\mu , cc}^{[n_1 , n_2 ] } \cos\left(\dfrac{n_1 y^5}{r}\right ) \cos\left(\dfrac{n_2 y^6}{r}\right ) \nonumber\\ & & + \sum_{n_1 , n_2 > 0 } a_{\mu , ss}^{[n_1 , n_2 ] } \sin\left(\dfrac{n_1 y^5}{r}\right ) \sin\left(\dfrac{n_2 y^6}{r}\right),\end{aligned}\ ] ] with @xmath275 , @xmath276 being positive integers , where we omitted the arguments for the kk - modes at @xmath277 , except for the zero mode @xmath278 . we note that the orbifold condition eq.([eq : z2_gauge_vector ] ) leads to @xmath279 } = a_{\mu , s0}^{[n_1 ] } = a_{\mu , cs}^{[n_1,n_2 ] } = a_{\mu , sc}^{[n_1,n_2 ] } = 0\ ] ] in the kk - spectrum . in order to investigate the kk - spectrum , we define sets of the kk - fields , @xmath280 } & \equiv & \ { a_{\mu , c0}^{[n_1]}(x ) , \ , a_{\mu , 0c}^{[n_1]}(x ) \ } , \\ { \cal a}_\mu^{\delta=2 , [ n_1,n_2 ] } & \equiv & \ { a_{\mu , cc}^{[n_1,n_2]}(x ) , \ , a_{\mu , ss}^{[n_1,n_2]}(x ) \}.\end{aligned}\ ] ] [ eq : gauge_vector_set ] the masses of these kk - states are given by @xmath281 for a state contained in @xmath282}$ ] and @xmath283 for @xmath284}$ ] . using the number of elements of these sets , @xmath285 } = { \cal n}_g^{\delta=2,[n_1,n_2 ] } = 2,\ ] ] the number of kk - modes of gauge bosons @xmath273 with @xmath286 is written as @xmath287 } \cdot n_1(\mu ) + { \cal n}_g^{\delta=2,[n_1,n_2 ] } \cdot n_2(\mu)\end{aligned}\ ] ] for @xmath255 dimensions , where @xmath288 and @xmath289 are given by @xmath290 respectively . it is straightforward to perform similar analysis for the gauge scalar @xmath291 . since the zero - mode of @xmath291 is projected out from the kk - spectrum due to the orbifold bc ( [ eq : z2_gauge_scalar ] ) , the kk decomposition of @xmath291 is given by @xmath292 } \sin\left(\dfrac{n_1 y^5}{r}\right ) \nonumber\\ & & + \sum_{n_1>0 } a_{m , 0s}^{[n_1 ] } \sin\left(\dfrac{n_1 y^6}{r}\right ) \nonumber\\ & & + \sum_{n_1 , n_2 > 0 } a_{m , cs}^{[n_1 , n_2 ] } \cos\left(\dfrac{n_1 y^5}{r}\right ) \sin\left(\dfrac{n_2 y^6}{r}\right ) \nonumber\\ & & + \sum_{n_1 , n_2 > 0 } a_{m , sc}^{[n_1 , n_2 ] } \sin\left(\dfrac{n_1 y^5}{r}\right ) \cos\left(\dfrac{n_2 y^6}{r}\right).\end{aligned}\ ] ] sets of the kk - fields are defined as @xmath293 } & \equiv & \ { a_{m , s0}^{[n_1]}(x ) , \ , a_{m , 0s}^{[n_1]}(x ) \ } , \\ { \cal a}_m^{\delta=2 , [ n_1,n_2 ] } & \equiv & \ { a_{m , cs}^{[n_1,n_2]}(x ) , \ , a_{m , sc}^{[n_1,n_2]}(x ) \}\end{aligned}\ ] ] in a similar manner to eqs.([eq : gauge_vector_set ] ) , where the number of elements are given by @xmath294 } = { \cal n}_{gs}^{\delta=2,[n_1,n_2 ] } = 2.\ ] ] we can now easily count the number of kk - modes of the gauge scalars @xmath291 , @xmath295 } \cdot n_1(\mu ) + { \cal n}_{gs}^{\delta=2,[n_1,n_2 ] } \cdot n_2(\mu).\end{aligned}\ ] ] we next consider the kk spectrum of the bulk chiral fermion @xmath296 , @xmath297 in @xmath255 dimensions , which has obviously 4-components . the orbifold bc ( [ eq : z2_fermion ] ) reads @xmath298 the kk expansion of @xmath299 is thus given by @xmath300}(x ) \cos\left(\dfrac{n_1 y^5}{r}\right ) \nonumber\\ & & + \sum_{n_1>0 } \psi_{+,0c}^{[n_1]}(x ) \cos\left(\dfrac{n_1 y^6}{r}\right ) \nonumber\\ & & + \sum_{n_1,n_2>0 } \psi_{+,cc}^{[n_1,n_2]}(x ) \cos\left(\dfrac{n_1 y^5}{r}\right ) \cos\left(\dfrac{n_2 y^6}{r}\right ) \nonumber\\ & & + \sum_{n_1,n_2>0 } \psi_{+,ss}^{[n_1,n_2]}(x ) \sin\left(\dfrac{n_1 y^5}{r}\right ) \sin\left(\dfrac{n_2 y^6}{r}\right ) \nonumber\\ & & + \sum_{n_1>0 } \psi_{+,s0}^{[n_1]}(x ) \sin\left(\dfrac{n_1 y^5}{r}\right ) \nonumber\\ & & + \sum_{n_1>0 } \psi_{+,0s}^{[n_1]}(x ) \sin\left(\dfrac{n_1 y^6}{r}\right ) \nonumber\\ & & + \sum_{n_1,n_2>0 } \psi_{+,cs}^{[n_1,n_2]}(x ) \cos\left(\dfrac{n_1 y^5}{r}\right ) \sin\left(\dfrac{n_2 y^6}{r}\right ) \nonumber\\ & & + \sum_{n_1,n_2>0 } \psi_{+,sc}^{[n_1,n_2]}(x ) \sin\left(\dfrac{n_1 y^5}{r}\right ) \cos\left(\dfrac{n_2 y^6}{r}\right ) \nonumber\\ & & \end{aligned}\ ] ] with 2-component fermions . in particular , we find that the zero - mode is right - handed under the 4-dimensional chiral rotation , thanks to eq.([eq : z2_fermion ] ) , @xmath301 other kk modes include both of right , and left - handed fermions . we thus define sets of kk - fields , @xmath302 } & \equiv & \ { \psi_{+,c0}^{[n_1]}(x ) , \ , \psi_{+,0c}^{[n_1]}(x ) \ } , \\ \psi_{+,r}^{\delta=2 , [ n_1,n_2 ] } & \equiv & \ { \psi_{+,cc}^{[n_1,n_2]}(x ) , \ , \psi_{+,ss}^{[n_1,n_2]}(x ) \ } , \\ \psi_{+,l}^{\delta=2 , [ n_1 ] } & \equiv & \ { \psi_{+,s0}^{[n_1]}(x ) , \ , \psi_{+,0s}^{[n_1]}(x ) \ } , \\ \psi_{+,l}^{\delta=2 , [ n_1,n_2 ] } & \equiv & \ { \psi_{+,cs}^{[n_1,n_2]}(x ) , \ , \psi_{+,sc}^{[n_1,n_2]}(x ) \}.\end{aligned}\ ] ] the kk fields in @xmath303}$ ] and @xmath304}$ ] ( @xmath305}$ ] and @xmath306}$ ] ) are right - handed ( left - handed ) . these kk - modes acquire their dirac masses among @xmath303}$ ] and @xmath305}$ ] ( @xmath304}$ ] and @xmath306}$ ] ) in four dimensions . the number of these kk dirac fermions is therefore given by @xmath307 } \cdot n_1(\mu ) + { \cal n}_f^{\delta=2,[n_1,n_2 ] } \cdot n_2(\mu ) \label{nkk_f}\end{aligned}\ ] ] with @xmath308 } & \equiv & \ # \psi_{+,r}^{2,[n_1 ] } \nonumber\\ & = & \ # \psi_{+,l}^{2,[n_1 ] } = 2 , \\ { \cal n}_f^{\delta=2,[n_1,n_2 ] } & \equiv & \ # \psi_{+,r}^{2,[n_1,n_2 ] } \nonumber\\ & = & \ # \psi_{+,l}^{2,[n_1,n_2 ] } = 2 . \end{aligned}\ ] ] we note @xmath309 for @xmath255 dimensions . it is straightforward to apply the above procedure to the bulk chiral fermion @xmath310 , @xmath311 we easily find that the zero mode of @xmath310 is left - handed . we thus identify the right[left]-handed fermions in the sm to @xmath296 [ @xmath310 ] . the number of the kk dirac fermions for @xmath313 is equal to eq . ( [ nkk_f ] ) . for the ( composite ) higgs field , we take the same @xmath267 identification as eq . ( [ eq : z2_gauge_vector ] ) . this always leads to @xmath314 for the number of kk modes of the higgs field . hereafter , we abbreviate @xmath59 in the arguments of @xmath315 , @xmath316 , when the number of @xmath59 is obvious . let us next turn to the 8-dimensional case ( @xmath253 , @xmath318 ) , @xmath319 we impose the orbifold bcs : @xmath320 and @xmath321 with @xmath322 noting the relations , @xmath323 we can easily see that the zero - mode of the 8-dimensional bulk chiral fermion @xmath299 is chiral under the 4-dimensional chiral symmetry . the bulk gauge field is decomposed into its kk - modes in a similar manner with the @xmath255 case . we obtain @xmath324_1 } = } \nonumber\\ & & \ { a_{\mu , c000}^{[n]_1 } , \ , a_{\mu , 0c00}^{[n]_1 } , \ , a_{\mu , 00c0}^{[n]_1 } , \ , a_{\mu , 000c}^{[n]_1 } \ } , \\ \lefteqn { { \cal a}_\mu^{\delta=4,[n]_2 } = } \nonumber\\ & & \ { a_{\mu , cc00}^{[n]_2 } , \ , a_{\mu , 00cc}^{[n]_2 } , \ , a_{\mu , ss00}^{[n]_2 } , \ , a_{\mu , 00ss}^{[n]_2 } , \ , \nonumber\\ & & \quad a_{\mu , c0c0}^{[n]_2 } , \ , a_{\mu , c00c}^{[n]_2 } , \ , a_{\mu , 0cc0}^{[n]_2 } , \ , a_{\mu , 0c0c}^{[n]_2 } \ } , \\ \lefteqn { { \cal a}_\mu^{\delta=4,[n]_3 } = } \nonumber\\ & & \ { a_{\mu , ccc0}^{[n]_3 } , \ , a_{\mu , cc0c}^{[n]_3 } , \ , a_{\mu , c0cc}^{[n]_3 } , \ , a_{\mu , 0ccc}^{[n]_3 } , \ , \nonumber\\ & & \quad a_{\mu , ssc0}^{[n]_3 } , \ , a_{\mu , ss0c}^{[n]_3 } , \ , a_{\mu , c0ss}^{[n]_3 } , \ , a_{\mu , 0css}^{[n]_3 } \ } , \\ \lefteqn { { \cal a}_\mu^{\delta=4,[n]_4 } = } \nonumber\\ & & \ { a_{\mu , cccc}^{[n]_4 } , \ , a_{\mu , ccss}^{[n]_4 } , \ , a_{\mu , sscc}^{[n]_4 } , \ , a_{\mu , ssss}^{[n]_4 } \},\end{aligned}\ ] ] for the kk - expansion of @xmath273 . here @xmath325_i$ ] stands for @xmath326_1\equiv [ n_1 ] , \quad [ n]_2\equiv [ n_1 , n_2 ] , \quad [ n]_3\equiv [ n_1 , n_2 , n_3 ] , \cdots.\ ] ] we thus find @xmath327_1 } & = & 4 , \quad { \cal n}_g^{\delta=4,[n]_2 } = 8 , \\ { \cal n}_g^{\delta=4,[n]_3 } & = & 8 , \quad { \cal n}_g^{\delta=4,[n]_4 } = 4.\end{aligned}\ ] ] for @xmath291 ( @xmath260 ) we obtain @xmath328_1 } = } \nonumber\\ & & \ { a_{m , s000}^{[n]_1 } , \ , a_{m , 0s00}^{[n]_1 } \ } , \\ \lefteqn { { \cal a}_m^{\delta=4,[n]_2 } = } \nonumber\\ & & \ { a_{m , sc00}^{[n]_2 } , \ , a_{m , cs00}^{[n]_2 } , \ , a_{m , s0c0}^{[n]_2 } , \ , a_{m , s00c}^{[n]_2 } , \ , \nonumber\\ & & \quad a_{m , 0sc0}^{[n]_2 } , \ , a_{\mu , 0s0c}^{[n]_2 } \ } , \\ \lefteqn { { \cal a}_m^{\delta=4,[n]_3 } = } \nonumber\\ & & \ { a_{m , scc0}^{[n]_3 } , \ , a_{m , sc0c}^{[n]_3 } , \ , a_{m , csc0}^{[n]_3 } , \ , a_{m , cs0c}^{[n]_3 } , \ , \nonumber\\ & & \quad a_{m , s0cc}^{[n]_3 } , \ , a_{m , 0scc}^{[n]_3 } , \ , a_{m , s0ss}^{[n]_3 } , \ , a_{m , 0sss}^{[n]_3 } \ } , \\ \lefteqn { { \cal a}_m^{\delta=4,[n]_4 } = } \nonumber\\ & & \ { a_{m , sccc}^{[n]_4 } , \ , a_{m , cscc}^{[n]_4 } , \ , a_{m , scss}^{[n]_4 } , \ , a_{m , csss}^{[n]_4 } \}.\end{aligned}\ ] ] it is straightforward to obtain similar expression for @xmath329 ( @xmath330 ) . we find @xmath331_1 } & = & 2 , \quad { \cal n}_{gs}^{\delta=4,[n]_2 } = 6 , \\ { \cal n}_{gs}^{\delta=4,[n]_3 } & = & 8 , \quad { \cal n}_{gs}^{\delta=4,[n]_4 } = 4.\end{aligned}\ ] ] the numbers of kk - modes of gauge fields and gauge scalars are given by @xmath332_k } \cdot n_k(\mu ) , \\ { n_{\rm kk}^{gs}}(\mu ; \delta=4 ) & = & \sum_{k=1}^4 { \cal n}_{gs}^{4,[n]_k } \cdot n_k(\mu ) , \end{aligned}\ ] ] where @xmath333 s are defined in eq.([eq : n1n2 ] ) for @xmath334 and @xmath335 for @xmath336 . for a bulk chiral fermion @xmath299 in @xmath253 ( @xmath318 ) dimensions , we find @xmath337_1 } & \equiv & \ # \psi_{+,r}^{4,[n]_1 } \nonumber\\ & = & \ # \psi_{+,l}^{4,[n]_1 } = 4 , \\ { \cal n}_f^{\delta=4,[n]_2 } & \equiv & \ # \psi_{+,r}^{4,[n]_2 } \nonumber\\ & = & \ # \psi_{+,l}^{4,[n]_2 } = 12 , \\ { \cal n}_f^{\delta=4,[n]_3 } & \equiv & \ # \psi_{+,r}^{4,[n]_3 } \nonumber\\ & = & \ # \psi_{+,l}^{4,[n]_3 } = 16 , \\ { \cal n}_f^{\delta=4,[n]_4 } & \equiv & \ # \psi_{+,r}^{4,[n]_4 } \nonumber\\ & = & \ # \psi_{+,l}^{4,[n]_4 } = 8.\end{aligned}\ ] ] the number of 4-component dirac kk - modes is then given by @xmath338_k } \cdot n_k(\mu).\ ] ] here , we note @xmath339 and @xmath340 in our orbifold compactification , because of @xmath341_2 } \ne { \cal n}_{gs}^{\delta=4,[n]_2 } , \quad { \cal n}_g^{\delta=4,[n]_2 } \ne \frac{1}{2}{\cal n}_f^{\delta=4,[n]_2}.\ ] ] the differences between them impact on the running of @xmath342 around @xmath343 . the bulk qcd coupling in eq . ( [ rge_ed2 ] ) grows slowly , comparing with that in the approximation ( [ rge_ed3 ] ) . as a result , the region of the tmac scale is slightly suppressed , if the identical value for @xmath344 at the compactification scale @xmath28 is used in both rges . ( see fig . [ fig - tmac-8d](a ) and fig . 2 in ref . ) we also comment on the relation @xmath345 in our orbifold compactification . it is now trivial task to extend the analysis to @xmath253 ( @xmath347 ) dimensions . the number of the kk - modes for @xmath347 is given by @xmath348_k } \cdot n_k(\mu)\ ] ] for @xmath65 . we summarize the number of elements among sets of kk - fields in tables [ tab1 ] , [ tab2 ] and [ tab3 ] for @xmath349 , respectively . .@xmath127_k}$ ] for @xmath256 ( @xmath2 ) . the column shows @xmath127_k}$ ] for @xmath350 ( gauge bosons ) , @xmath351 ( gauge scalars ) , @xmath352 ( dirac fermions ) , and @xmath353 ( higgs ) . [ tab1 ] [ cols="^,^,^",options="header " , ] we may take yet another choice for evaluating the vacuum polarization function , instead of the truncated kk effective theory @xcite . in this appendix , we consider the proper - time ( pt ) scheme as well as the @xmath13-scheme in the truncated kk effective theory . the effective charge is defined by @xmath354 where @xmath355 is the bare coupling and @xmath356 denotes the vacuum polarization function , @xmath357 the effective charge is closely related to physical quantities such as scattering amplitudes . through the effective coupling , we determine the relation of gauge couplings calculated in the two regularization schemes . we consider bulk gauge theories with bulk fermions and bulk scalars . in four dimensions , it is straightforward to calculate the contributions of fermions and scalars : @xmath358 for @xmath117 pieces of dirac fermions of the fundamental representation , and @xmath359 for @xmath360 pieces of complex scalars of the fundamental representation , respectively , where @xmath116 and @xmath361 is the mass of fermions and scalars . for bulk fields , we perform to sum over their kk modes . on the other hand , it is slightly complicated to compute the loop correction of massive gauge bosons within the 4-dimensional theory . instead higgsing gauge theory in four dimensions , we directly calculate the loop correction in extra dimensions . in order to keep the gauge invariance , we use the background field method with the feynman gauge . taking effects of compactification into account , we replace the loop integral in @xmath42 dimensions to @xmath362 with the @xmath42-dimensional ( dimensionful ) coupling @xmath71 and the 4-dimensional one @xmath70 . we here decomposed the loop momentum @xmath363 to the 4-dimensional one @xmath364 and the discrete one @xmath365 in extra @xmath59 dimensions , @xmath366 where masses of kk modes @xmath367 are given by @xmath368 of course , the external momentum @xmath369 does not have the extra momentum , @xmath370 . we thus obtain contributions of gauge bosons and gauge scalars as @xmath371 , \label{pibare - g}\end{aligned}\ ] ] and @xmath372 respectively . we separate the vacuum polarization function to the contributions of zero modes @xmath373 , which corresponds to the sm corrections , and kk modes @xmath374 : @xmath375 with @xmath376 and @xmath377 we briefly review the results in refs . @xcite . in the truncated kk effective theory , we calculate the vacuum polarization function by using the dimensional regularization ( taking the loop integral to @xmath378 instead of @xmath379 ) : @xmath380 , \\ ( 4\pi)^2 \pi_{gs}^{\overline{\rm ms}}(-q^2;m^2 ) & = & - \frac{\delta}{2 } \ , i_s(-q^2;m^2 ) , \\ ( 4\pi)^2 \pi_f^{\overline{\rm ms}}(-q^2;m^2 ) & = & - 8 t_r \ , n_f \ , i_f(-q^2;m^2 ) , \\ ( 4\pi)^2 \pi_h^{\overline{\rm ms}}(-q^2;m^2 ) & = & - t_r \ , n_h \ , i_s(-q^2;m^2 ) , \end{aligned}\ ] ] where we defined @xmath381^{-\epsilon } , \\ i_s(-q^2;m^2 ) & \equiv & \frac{\gamma(\epsilon)}{(4\pi)^{-\epsilon } } \int_0 ^ 1 dx \ , ( 2x-1)^2 \left[\,m^2-x(1-x)q^2\,\right]^{-\epsilon},\\ i_f(-q^2;m^2 ) & \equiv & \frac{\gamma(\epsilon)}{(4\pi)^{-\epsilon } } \int_0 ^ 1 dx \ , x(1-x ) \left[\,m^2-x(1-x)q^2\,\right]^{-\epsilon } .\end{aligned}\ ] ] we renormalize the bare coupling in the @xmath13-scheme : @xmath382 with the rge coefficient of zero mode @xmath383 , @xmath384 and that of kk modes @xmath385 , @xmath386 where @xmath387 denote the total number of kk modes for gauge bosons , gauge scalars , fermions , and scalars ( composite higgs fields ) below @xmath51 , respectively . we then find the effective coupling in the truncated effective theory based on the @xmath13-scheme , @xmath388 with @xmath389 where we defined @xmath390 + \frac{\delta}{2 } \ i_s^r(-q^2;m^2 ) \nonumber \\ & & \quad + 8 t_r \ , n_f \ , i_f^r(-q^2;m^2 ) + t_r \ , n_h \ , i_s^r(-q^2;m^2 ) , \\ i_g^r ( -q^2;m^2 ) & \equiv & \int_0 ^ 1 dx \ln \left(\,1-x(1-x)\frac{q^2}{m^2}\,\right ) , \\ i_s^r ( -q^2;m^2 ) & \equiv & \int_0 ^ 1 dx \ , ( 2x-1)^2 \ln \left(\,1-x(1-x)\frac{q^2}{m^2}\,\right ) , \\ i_f^r ( -q^2;m^2 ) & \equiv & \int_0 ^ 1 dx \ , x(1-x ) \ln \left(\,1-x(1-x)\frac{q^2}{m^2}\,\right),\end{aligned}\ ] ] and obtained the constant term @xmath391 we identify the mass of the heaviest kk state to the cutoff @xmath11 . in the limit of @xmath392 , we easily find @xmath393 with @xmath394 where @xmath395 is replaced by @xmath396 for @xmath256 . since we can not take so large value to the cutoff @xmath11 due to the landau pole for @xmath15 , the effects of the finite parts ( [ ms - q ] ) summing over kk modes are negligible in a certain low energy scale , comparing with the finite term arising from zero modes , @xmath397 . numerically , we find the finite sums for qcd at @xmath398 , @xmath399 for @xmath128 , respectively , where we took @xmath175 tev , @xmath400 gev , @xmath401 for @xmath128 , respectively . of course , the effects of eq . ( [ ms - q ] ) get smaller for larger values of @xmath28 . at the zero momentum , the effects of finite terms via kk modes come to be completely negligible thanks to eq . ( [ ms - q ] ) . we thus identify @xmath402 to the sm value . we also find the @xmath403-function , @xmath404 we calculate the vacuum polarization function in the pt - scheme . using the trick with the pt - parameter @xmath405 , @xmath406 and performing the momentum integral , @xmath407 we can rewrite vacuum polarization functions in eqs . ( [ pibare - f ] ) , ( [ pibare - s ] ) , ( [ pibare - g ] ) , and ( [ pibare - gs ] ) as follows : @xmath408 , \\ ( 4\pi)^2 \pi_{gs}^{\rm pt}(q_e^2;m^2 ) & = & - \frac{\delta}{2 } \ , c_a \ , j_s(-q^2;m^2 ) , \\ ( 4\pi)^2 \pi_f^{\rm pt}(q_e^2;m^2 ) & = & - 8 t_r \ , n_f \ , j_f(q_e^2;m^2 ) , \\ ( 4\pi)^2 \pi_h^{\rm pt}(q_e^2;m^2 ) & = & - t_r \ , n_h \ , j_s(q_e^2;m^2 ) , \end{aligned}\ ] ] with @xmath409 , \\ j_s(q_e^2;m^2)&= \int_0 ^ 1 dx \int_0^\infty \frac{dt}{t } ( 2x-1)^2 \exp[-t(x(1-x)q_e^2+m^2 ) ] , \\ j_f(q_e^2;m^2)&= \int_0 ^ 1 dx \int_0^\infty \frac{dt}{t } x(1-x ) \exp[-t(x(1-x)q_e^2+m^2)],\end{aligned}\ ] ] where the suffix @xmath410 denotes the euclidean momentum . since the pt - integral diverges near @xmath411 , we regularize the integral with the cutoff @xmath11 : @xmath412 we first calculate @xmath373 . performing the feynman parameter integration , we obtain @xmath413,\end{aligned}\ ] ] where we used the confluent hypergeometric function @xmath414 and the relation between the confluent hypergeometric function and the error function , @xmath415 here , we note the kummer s transformation , @xmath416 and the recurrence formula , @xmath417 = -\frac{1}{c}{}_1f_1(1,c+1;-t ) .\ ] ] since the behaviors of @xmath414 around @xmath418 and @xmath419 are given by @xmath420 and @xmath421 respectively , we separate the pt - integration to the singular part and finite ones : @xmath422 \nonumber \\ & & - \int_0^{\frac{r q_e^2}{4\lambda^2 } } \frac{dt}{t } \left[\,{}_1f_1(1,c ;- t)-1\,\right ] + \int_1^\infty \frac{dt}{t}{}_1f_1(1,c ;- t ) , \\ & = & -\ln \frac{r q_e^2}{4\lambda^2 } + \frac{1}{c } \int_0^{\frac{r q_e^2}{4\lambda^2 } } dt\ , { } _ 1f_1(1,1+c ;- t ) \nonumber \\ & & -\frac{1}{c } \int_0 ^ 1 dt\ , { } _ 1f_1(1,1+c ;- t ) + \int_1^\infty \frac{dt}{t}{}_1f_1(1,c ;- t ) .\end{aligned}\ ] ] noting the integration of the confluent hypergeometric function , @xmath423 we find @xmath424 with @xmath425 we can read @xmath426 from the above results : @xmath427 with @xmath428 here , we note @xmath429 thanks to @xmath430 . now , we study the contributions of the kk modes . we can perform the summation over all kk states by using the jacobi @xmath125 function , @xmath431 which has the remarkable property , @xmath432 we then obtain the summation of the vacuum polarization via kk modes : @xmath433\exp[-tx(1-x)q_e^2]k_g(t ) \nonumber \\ & & - \frac{\delta}{2 } \cdot \frac{c_a}{(4\pi)^2 } \int_0 ^ 1dx\int_{r\lambda^{-2}}^\infty\frac{dt}{t}\ , ( 2x-1)^2 \exp[-tx(1-x)q_e^2]k_{gs}(t ) \nonumber \\ & & - 8t_r \cdot \frac{n_f}{(4\pi)^2 } \int_0 ^ 1dx\int_{r\lambda^{-2}}^\infty\frac{dt}{t}\ , x(1-x ) \exp[-tx(1-x)q_e^2]k_f ( t ) \nonumber \\ & & - t_r \cdot \frac{n_h}{(4\pi)^2 } \int_0 ^ 1dx\int_{r\lambda^{-2}}^\infty\frac{dt}{t}\ , ( 2x-1)^2 \exp[-tx(1-x)q_e^2]k_h ( t ) , \end{aligned}\ ] ] where we defined @xmath124_1 } \left[\,\frac{1}{2}(\vartheta_3 - 1)\,\right ] + { \cal n}_i^{\delta,[n]_2 } \left[\,\frac{1}{2}(\vartheta_3 - 1)\,\right]^2 + { \cal n}_i^{\delta,[n]_3 } \left[\,\frac{1}{2}(\vartheta_3 - 1)\,\right]^3 + \cdots\ ] ] with @xmath65 and @xmath126 . the values of @xmath127_k}$ ] are given in table [ tab1][tab3 ] . combining with eq . ( [ pt0 ] ) , we obtain the effective coupling as @xmath434 with @xmath435 where we neglected the corrections of the order of @xmath436 . in the limit of @xmath437 , we find @xmath438 where @xmath395 is replaced by @xmath396 for @xmath256 . we regard the bare coupling @xmath355 as a quantity defined at the cutoff @xmath11 : @xmath439 since the effective coupling @xmath106 should not depend on the cutoff @xmath11 at the zero momentum @xmath440 , we can deduce the `` @xmath403-function '' for @xmath149 : @xmath441 with the `` rge coefficient '' @xmath442 , @xmath443 matching the rge for the @xmath13-coupling with @xmath444 in the limit of @xmath445 , we easily find @xcite @xmath446 we identify the effective coupling at the zero momentum calculated in the pt - scheme to the sm one with @xmath447 : @xmath448 although the coefficients of @xmath449 in the @xmath13-scheme and the pt - scheme are accidentally identical in @xmath451 , they have generally the regularization dependence and their difference is the factor @xmath452 in eq . ( [ pt - q2 ] ) . y. nambu , enrico fermi institute report no . 89 - 08 , 1989 ; in _ proceedings of the 1989 workshop on dynamical symmetry breaking _ , edited by t. muta and k. yamawaki ( nagoya university , nagoya , japan , 1990 ) . k. yamawaki , in _ proceedings of the 14th symposium on theoretical physics `` dynamical symmetry breaking and effective field theory '' _ , cheju island , korea , july 21 - 26 , 1995 , ed . j. e. kim ( minumsa pub . co. , korea , 1996 ) p.43 - 86 , [ arxiv : hep - ph/9603293 ] . c. t. hill , and e. h. simmons , phys . * 381 * , 235 ( 2003 ) , [ arxiv : hep - ph/0203079 ] . n. arkani - hamed , h. c. cheng , b. a. dobrescu and l. j. hall , phys . rev . * d62 * , 096006 ( 2000 ) , [ arxiv : hep - ph/0006238 ] . m. hashimoto , m. tanabashi and k. yamawaki , phys . rev . * d64 * , 056003 ( 2001 ) , [ arxiv : hep - ph/0010260 ] . k. agashe , jhep * 0105 * , 017 ( 2001 ) , [ arxiv : hep - ph/0012182 ] . v. gusynin , m. hashimoto , m. tanabashi and k. yamawaki , phys . rev . * d65 * , 116008 ( 2002 ) , [ arxiv : hep - ph/0201106 ] . h. abe , k. fukazawa , and t. inagaki , prog . * 107 * , 1047 ( 2002 ) , [ arxiv : hep - ph/0107125 ] ; h. abe , and t. inagaki , phys . rev . * d66 * , 085001 ( 2002 ) , [ arxiv : hep - ph/0206282 ] . s. raby , s. dimopoulos and l. susskind , nucl . b169 * , 373 ( 1980 ) . m. hashimoto , m. tanabashi , and k. yamawaki , in _ proceedings of the 2002 international workshop on strong coupling gauge theories and effective field theories ( scgt 02 ) _ , nagoya , japan , dec 10 - 13 , 2002 , ed . m. harada , y. kikukawa , and k. yamawaki ( world scientific pub . co. , singapore 2003 ) , p.291297 , [ arxiv : hep - ph/0304109 ] . a. manohar and h. georgi , nucl . b234 * , 189 ( 1984 ) . b. a. dobrescu and c. t. hill , phys . lett . * 81 * , 2634 ( 1998 ) , [ arxiv : hep - ph/9712319 ] ; r. s. chivukula , b. a. dobrescu , h. georgi and c. t. hill , phys . * d59 * , 075003 ( 1999 ) , [ arxiv : hep - ph/9809470 ] . see also h - j . he , c. t. hill and t. m. p. tait , phys . rev . * d65 * , 055006 ( 2002 ) , [ arxiv : hep - ph/0108041 ] . h. georgi , e. h. simmons , and a. g. cohen , phys . b236 * , 183 ( 1990 ) ; e. h. simmons , phys . rev . * d42 * , 2933 ( 1990 ) ; t. kugo , and m. g. mitchard , phys . lett . * b282 * , 162 ( 1992 ) . b. pendleton and g. g. ross , phys . b98 * , 291 ( 1981 ) . w. a. bardeen , c. n. leung and s. t. love , phys . lett . * 56 * , 1230 ( 1986 ) ; c. n. leung , s. t. love and w. a. bardeen , nucl . b * 273 * , 649 ( 1986 ) . kondo , s. shuto and k. yamawaki , mod . . lett . * a6 * , 3385 ( 1991 ) . p. nath and m. yamaguchi , phys * d60 * , 116004 ( 1999 ) , [ arxiv : hep - ph/9902323 ] . n. arkani - hamed , a. g. cohen and h. georgi , phys . lett . * 86 * , 4757 ( 2001 ) , [ arxiv : hep - th/0104005 ] ; c. t. hill , s. pokorski and j. wang , phys . rev . * d64 * , 105005 ( 2001 ) , [ arxiv : hep - th/0104035 ] .
as mobile broadband wireless networks evolve , increasingly ambitious targets have been set for spectral efficiency . exploiting the capabilities of multiple antennas at the mobiles and base transceiver station ( bts ) is crucial to meet those goals . an effective approach is to use spatial precoding at the transmitters and filtering at receivers to adapt to the variations of the multiple - input multiple - output ( mimo ) channel and interference conditions by optimizing a metric such as the sum rate , signal - to - interference - and - noise ratio ( sinr ) , or mean squared error ( mse ) . two challenges need to be addressed : first , the precoders and filters should be _ jointly _ optimized ; and second , the optimization should be carried out without _ a priori _ knowledge of the channel and interference conditions . while joint optimization of precoders and receivers in cellular networks has attracted considerable attention in recent years ( e.g. , @xcite ) , much of the work has assumed linear precoders for the downlink , and has considered channel estimation methods for estimating those precoders with unknown channels . here we consider joint optimization and adaptation of _ non_linear precoders and receivers . specifically , we assume that the users are successively decoded and cancelled at the bts for the uplink , and that successive interference pre - cancellation with tomlinson - harashima precoding is applied by the bts for the downlink . the mobiles use linear filtering and precoding . a time - division duplex ( tdd ) network is assumed in order to exploit the reciprocity between uplink and downlink channels . the precoder and receiver filters are optimized according to a sum - mse criterion assuming mimo fading channels . this criterion is motivated by previous results showing that the minimum mean squared error ( mmse ) receiver with successive interference cancellation achieves the sum capacity of the multiaccess channel @xcite , and that the precoding structure considered ( generalized decision - feedback equalizer with interference pre - compensation ) achieves the sum capacity of the gaussian broadcast channel @xcite . furthermore , the mse criterion leads directly to adaptive approaches for filter estimation in the presence of unknown channels and interference . for a single cell we first show that alternating optimization of precoders and receiver filters ( for either the uplink or downlink ) converges to a local optimum . although this optimization must be applied separately for the uplink and downlink to obtain the optimal precoders and receiver filters for each direction , we observe that the filters optimized for one direction achieve near - optimal performance for the other direction as well when the uplink and downlink power constraints and noise power are similar . with this in mind we consider a simplified scheme in which the uplink ( downlink ) receive filters are also used as downlink ( uplink ) precoders . with this constraint we again show that alternating optimization of the filters at the bts and mobiles converges to a fixed point . in the absence of channel state information ( csi ) the preceding results lead to an iterative bi - directional training scheme in which uplink and downlink pilots are alternately transmitted to estimate the bts and mobile filters , respectively . the estimated bts forward / backward filters then nonlinearly precode both the downlink data and pilots for interference pre - cancellation , and the estimated mobile receiver filters precode both the uplink data and pilots . with sufficient training , the estimated filters approach the corresponding mmse filters in which case training in each direction corresponds to an mmse update of the corresponding filters . these uplink / downlink updates are then iterated until convergence , or a complexity constraint is reached . this scheme extends the bi - directional training scheme presented in @xcite , where the uplink / downlink pilots are used to estimate linear filters / precoders without interference cancellation . a key feature of these schemes is that the pilots are transmitted _ synchronously _ in each direction ( across all cells ) , and are used to estimate the filters _ directly _ via a least squares criterion , rather than directly estimating the channel . as discussed in @xcite , this scheme is _ distributed _ in the sense that each node ( bts or mobile ) autonomously computes its own filters from the received signal and pilots . in contrast , prior schemes for precoding / filtering based on channel estimation ( e.g. , @xcite ) are more suitable for centralized optimization in which all csi for direct- and cross - channels is passed to a single remote location , which computes all filters and then passes those back to the corresponding nodes . numerical results are presented that illustrate the performance of the bi - directional training scheme as a function of the amount of training and number of forward - backward iterations . it is shown that the achievable rate with nonlinear filters approaches the uplink sum capacity as the number of iterations increases . for a fixed number of training symbols , the nonlinear filters significantly outperform linear filters when the number of iterations is limited even though the nonlinear filters require more parameters to be estimated . an example for a multi - cell scenario shows that more iterations are required for the precoders and receive filters to converge than for a single cell . nonlinear filters again significantly outperform linear filters with few iterations in the high - snr regime . related work on joint precoder and receive filter design has been presented in @xcite . in @xcite , iterative algorithms are proposed for simultaneous diagonalization of the downlink multiuser channels to enable interference - free spatial - division multiplexing . in @xcite , algorithms are presented to jointly optimize linear filters and precoders for the uplink using a sum - mse criterion . mse duality results for broadcast and multiaccess channels are presented in @xcite . duality ensures that any uplink minimum mse scheme has a downlink counterpart . in @xcite , the downlink sum - mse minimization problem is solved by creating a `` virtual '' uplink channel , which has the same transmit power and noise level as for the actual downlink channel . however , those methods can not be implemented in a distributed manner in the setting of this paper , where the uplink and downlink channels may have different power constraints and noise levels . in @xcite , iterative mmse beamforming algorithms are presented to minimize the sum power subject to a set of sinr constraints , where the precoder and receive filter updates are done sequentially across users . in contrast , all users update simultaneously in this paper . consider a single cell consisting of a bts with @xmath0 antennas and @xmath1 users , where user @xmath2 has @xmath3 antennas . the uplink channel between user @xmath2 and the bts is denoted by @xmath4 of size @xmath5 , each entry of which is a unit - variance circularly symmetric complex gaussian ( cscg ) random variable . each user is restricted to one data stream , although the algorithms to be presented are easily extended to the scenario with multiple data streams per user . the received signal vector at the bts is @xmath6 where @xmath7 is the data symbol from user @xmath2 and @xmath8\!=\!1 $ ] , @xmath9 is the @xmath10 precoder with @xmath11 , where @xmath12 is the power constraint for user @xmath2 , and @xmath13 is the cscg noise vector each entry of which has zero mean and variance @xmath14 . similarly , assuming tdd with channel reciprocity , the downlink received signal at user @xmath2 can be written as @xmath15 where @xmath16 is the data symbol for user @xmath17 , @xmath18 is the corresponding transmit beam , and @xmath19 denotes the cscg noise vector with @xmath2th entry having variance @xmath20 . the filters satisfy the bts power constraint @xmath21 . we first discuss joint optimization of transmit precoders and receive filters at a central controller with perfect csi . adaptive versions with training are subsequently discussed in section [ s : training ] . first consider the case where the received signal in ( [ upsignal ] ) is processed by linear filters @xmath22 $ ] , where @xmath23 denotes the receive filter for user @xmath2 . the estimated signal for user @xmath2 is given by @xmath24 and the corresponding mse for user @xmath2 is @xmath25\\ & \!=\!1\!+\!\sigma^2\|\boldsymbol{g}_k\|^2\!-\!2\re\big\{\boldsymbol{g}_k^\dagger\boldsymbol{h}_k\boldsymbol{v}_k\big\}\!+\!\sum_{i=1}^{k}\big|\boldsymbol{g}_k^\dagger\boldsymbol{h}_i\boldsymbol{v}_i\big|^2,\label{upmse}\end{aligned}\ ] ] where @xmath26 takes the real part of a complex number . the optimization problem is stated as : [ upproblem ] @xmath27 [ prop : fullpow ] the sum mmse is minimized when all users transmit with full power . the proof is given in the appendix . hence the decrease in sum mse for a particular user @xmath2 due to an increase in power @xmath12 dominates the increase in sum mse due to the increase in interference at neighboring receivers . we will assume that each user transmits at the maximum power so that the constraints ( [ powercon ] ) are binding . the lagrangian for problem ( [ upproblem ] ) is @xmath28 where @xmath29 is the lagrange multiplier associated with the power constraint for user @xmath2 . hence the optimal solution to problem ( [ upproblem ] ) must satisfy : @xmath30 for @xmath31 , where @xmath29 in ( [ eq : uplnfilter2 ] ) is chosen to satisfy ( [ powercon ] ) . the nonlinear successive cancellation filter at the bts is the decision - feedback detector ( dfd ) shown in fig . [ uplink ] . the received signal is input to the linear feedforward filter @xmath32 , and the feedback filter @xmath33 implements the successive interference cancellation . assuming the decoding order is from user @xmath34 to user @xmath1 , the feedback matrix @xmath35 is lower triangular with diagonal elements equal to one . the estimated symbol for user @xmath2 is then @xmath36 where @xmath37 denotes the ( @xmath2,@xmath17)-th element of @xmath35 and @xmath38 denotes the slicer function in the forward loop . for user @xmath2 , if decoding of the previous symbols is correct , i.e. , @xmath39 , and @xmath40 , then the interference from users 1 to user @xmath41 can be removed . the achievable mse for user @xmath2 is then @xmath42 the problem becomes [ dfeproblem ] @xmath43 in this scenario it may not be optimal ( in the sum - mse sense ) for all users to transmit with full power . we will nevertheless assume that the power constraints are binding since this simplifies the problem , and because transmitting at full power is known to achieve the uplink sum capacity . writing the lagrangian as before , the corresponding first - order conditions are @xmath44 for @xmath31 , where the lagrange multipliers @xmath29 s are chosen to enforce the power constraints ( [ con : fullpow ] ) . the summation indices differ from ( [ eq : uplnfilter1 ] ) and ( [ eq : uplnfilter2 ] ) due to successive cancellation . the expressions for the downlink filters can be obtained in a similar fashion as for the uplink . the uplink and downlink problems are different because the snrs for the uplink and downlink channels are different in general , and because the power constraints are on individual users in the uplink , while the constraint is on the sum over all users in the downlink . for the downlink , the estimated symbol for user @xmath2 is @xmath45 where @xmath46 is the receiver filter . the corresponding mse is @xmath47 the downlink sum - mse minimization problem is [ dproblem ] @xmath48 the corresponding optimal filters satisfy : @xmath49 for @xmath31 , where the lagrange multiplier @xmath50 is chosen to satisfy . with nonlinear filtering at the bts @xcite , i.e. , dirty paper coding , assuming the encoding order is from user @xmath1 to user @xmath34 , the signal received by user @xmath2 can be written as @xmath51 the corresponding mse for user @xmath2 at the output of the linear receiver @xmath46 is then @xmath52 the downlink sum - mse minimization problem is [ dproblem_ipc ] @xmath53 the corresponding optimal filters satisfy : @xmath54 for @xmath31 , where the lagrange multiplier @xmath50 is chosen to satisfy . focusing on the uplink the precoders and filters in ( [ eq : uplnfilter1])([eq : uplnfilter2 ] ) are coupled , evading an explicit solution in terms of the channels only . moreover , problem is non - convex when @xmath55 , thus there is no known efficient numerical algorithm for finding the global optimum . , or if @xmath55 and there is no constraint on the number of data streams per user , then this problem can be transformed into a convex program @xcite . ] as pointed out in @xcite , the precoders and filters can be computed iteratively . with fixed transmit precoders @xmath56 , the lagrangian ( [ lagrangian ] ) is strictly convex in receive filters @xmath57 , and ( [ eq : uplnfilter1 ] ) gives the optimal solution . conversely , with fixed @xmath57 , we compute @xmath56 and @xmath58 together from with the requirement that @xmath59 for every @xmath2 . the search for each @xmath29 can be efficient because the optimization of @xmath60 is decoupled across users . because @xmath29 is chosen to guarantee @xmath61 , the lagrangian in ( [ lagrangian ] ) is always equal to the sum - mse . thus the lagrangian as well as the sum - mse monotonically decrease with each update , and must converge since they are lower bounded by zero . in case multiple @xmath29 s enforce @xmath59 , we choose the one that yields the minimum lagrangian ( [ lagrangian ] ) . if multiple @xmath29 s achieve the same minimum , then we can randomly choose one . since the preceding procedure dictates a one - to - one mapping between @xmath56 and @xmath57 , it is not difficult to show that the set of precoders and filters @xmath62 must also converge to a fixed point . the fixed point must be a local optimum for problem ( [ upproblem ] ) , although it may not be globally optimal . similar to the linear case , the iterative approach previously described can also be applied to compute ( [ eq : upnlnfilter1])([eq : upnlnfilter2 ] ) and the convergence of the filters follows from the monotonic decrease of the lagrangian and the one - to - one mapping between the precoders and the receiver filters . for the downlink the set of linear filters can be computed by iterating ( [ eq : downlnfilter1 ] ) and ( [ eq : downlnfilter2 ] ) , and the set of nonlinear filters can be computed by iterating ( [ eq : downnlnfilter2 ] ) and ( [ eq : downnlnfilter1 ] ) . according to similar arguments as for the uplink , all the filters will again converge to a fixed point . the iterative optimization procedure in section [ sec : iterativeseparate ] was applied separately to compute the set of uplink filters , given by a solution to ( [ upproblem ] ) , and the set of downlink filters , given by a solution to ( [ dproblem ] ) . here we present a suboptimal iterative algorithm that computes all uplink / downlink precoders and filters simultaneously . numerical results show that the performance of this algorithm is very close to that of separate uplink and downlink optimizations with practically relevant snrs . let @xmath63 and @xmath64 denote the uplink precoders and receive filters as in section [ sec : linearup ] , and let @xmath65 and @xmath66 denote the downlink precoders and receive filters . taking linear filters as an example , we apply the following four steps in each iteration : 1 . let each uplink @xmath67 be the downlink receiver @xmath68 , and set @xmath8\!=\!\rho$ ] . 2 . compute @xmath69 . 3 . let each downlink @xmath18 be the uplink receiver @xmath70 , and set @xmath71\!=\!\beta$ ] . 4 . compute @xmath72 . that is , we use the same update formulas as for the uplink and downlink optimizations discussed in section [ sec : linearup ] , but constrain the precoders and receive filters at each terminal to be the same . the uplink and downlink powers @xmath73 and @xmath74 are assumed to be given _ a priori_. this assumption is needed to prove the following convergence result , but will be relaxed in the next section . for any fixed positive @xmath73 and @xmath74 , the precoders and filters @xmath75 computed according to the preceding iterative algorithm converge to a fixed point as the number of iterations goes to infinity . we provide a positive potential function of the precoders and filters and show that this function decreases every iteration . the uplink mse , normalized by @xmath73 , for user @xmath2 is @xmath76 and the downlink mse , normalized by @xmath74 , for user @xmath2 is @xmath77 we define the functions : @xmath78 and @xmath79 since @xmath80 and @xmath81 , it can be verified that @xmath82 for any given set of filters @xmath83 . the convergence of @xmath84 ( or @xmath85 ) can then be verified by observing that @xmath84 ( or @xmath85 ) monotonically decreases with an update for any @xmath23 at the bts or any @xmath46 for a particular user , and it is lower bounded by zero . since in each step the mapping from @xmath56 to @xmath57 or from @xmath86 to @xmath87 is unique , it is easy to show that the filters must converge to a fixed point . with similar arguments this proposition can be extended to nonlinear filters and to interference networks with multiple users and multiple cells . in the case of interference networks , the potential function can be chosen as the the sum of the achievable mse s plus the scaled norms of the precoders at all transmitters . ideally , the parameters @xmath73 and @xmath74 are such that the transmit power constraints at the bts and mobiles are satisfied at convergence . the values are of course unknown _ a priori_. a straightforward power scaling method is to normalize the precoders directly after each iteration so that the transmit power constraint is met . this is analogous to the max - sinr algorithm proposed in @xcite for interference networks . although we can not prove the convergence of the corresponding optimization algorithm , convergence has always been observed for numerical examples . in fact the performance of this algorithm is very close to that of separate optimization of the uplink problem and its downlink counterpart . in this section , we assume the channels are unknown to the bts and mobiles _ a priori_. we propose a bi - directional training scheme for adapting the precoders at the mobiles and the filters at the bts with the goal of minimizing the uplink sum - mse . this bi - directional training scheme can be implemented in a distributed manner . specifically , estimates of ( [ eq : uplnfilter1 ] ) and ( [ eq : upnlnfilter1 ] ) can be obtained by training in the uplink , and estimates of ( [ eq : uplnfilter2 ] ) and ( [ eq : upnlnfilter2 ] ) can be obtained by training in the downlink . a similar scheme can be applied to the downlink ; since the uplink and downlink problems are similar , we focus on the uplink in this section . we first observe the equivalence between iteratively computing ( [ eq : uplnfilter1])-([eq : uplnfilter2 ] ) and bi - directional _ optimization _ , in which the following steps are iterated , as illustrated in fig . [ bidirectional ] : 1 . _ forward optimization : _ fix the uplink precoders @xmath9 and select the receive filters @xmath57 to minimize the output mse . backward optimization : _ reverse the direction of transmission , fix the downlink precoders at the bts as @xmath57 and select the receive filter @xmath9 to minimize the output mse subject to a norm constraint . in the first step , the optimal linear receive filters are given by ( [ eq : uplnfilter1 ] ) . in the second step , let @xmath71\!=\!\gamma$ ] , where @xmath88 is chosen such that the total downlink transmit power @xmath89 does not exceed the constraint . by channel reciprocity , the signal received by user @xmath2 is given by ( [ eq : downrecsig ] ) with @xmath18 replaced by @xmath70 . the mse , normalized by @xmath88 , with receive filter @xmath9 is then @xmath90 in the second step we wish to [ dlmseproblem ] @xmath91 by constraining the norm of @xmath9 , we account for the uplink power constraint . namely , the solution to problem ( [ dlmseproblem ] ) is given by @xmath92 where the lagrange multiplier @xmath93 is chosen to satisfy ( [ con : fullpowtraindown ] ) . if there are multiple @xmath93 s that satisfy ( [ con : fullpowtraindown ] ) , then the one which minimizes ( [ downmse ] ) should be chosen . since this solution is the same as in ( [ eq : uplnfilter2 ] ) , the previous iterative procedure for minimizing the lagrangian in ( [ lagrangian ] ) is equivalent to bi - directional optimization . note that with the constraint ( [ con : fullpowtraindown ] ) , the term @xmath94 in ( [ dlmse ] ) is a constant so that the solution to problem ( [ dlmseproblem ] ) does not depend on the noise level at user @xmath2 . this is expected since problem ( [ upproblem ] ) is to minimize the uplink sum - mse and the solution should not depend on the noise level at the user side . with unknown channels the mse optimization criterion can be replaced by a least squares cost function given the transmitted pilot symbols . in the uplink , we assume that the users _ synchronously _ transmit sequences of @xmath95 training symbols given by the matrix @xmath96 where @xmath97 $ ] and @xmath98 is the @xmath99 row vector containing the training symbols for user @xmath2 . the received signal at the bts is then given by ( [ upsignal ] ) with @xmath100 . the corresponding sequence of estimated symbols is @xmath101 where @xmath102 $ ] . the receive filter is then chosen to minimize the cost function @xmath103 , and the solution is given by @xmath104 to estimate the uplink precoders , a similar training scheme can be applied on the downlink where the training symbols are passed through the precoders @xmath57 and superposed before transmission at the bts . the received signal at user @xmath2 is then given by ( [ eq : downrecsig ] ) with @xmath105 and @xmath18 replaced by @xmath70 . at each user , the cost function is given by @xmath106 where an additional norm constraint is added , and @xmath107 $ ] . the solution is given by @xmath108 where @xmath29 is chosen to satisfy the norm constraint @xmath59 . if multiple @xmath29 s satisfy the constraint , then the one which minimizes the cost function should be chosen . we next present a bi - directional training scheme for implementing the updates ( [ eq : upnlnfilter1])([eq : upnlnfilter2 ] ) with nonlinear filters in the absence of channel information . we assume a dfd in the uplink , and the analogous interference pre - compensation scheme for the downlink as illustrated in fig . [ backward_training ] . downlink interference is pre - compensated in the _ reverse _ order as for the uplink , so that the effective downlink channel is triangular . ( that is , for the downlink user @xmath2 receives no interference from users @xmath109 , whereas for the uplink it receives no interference from users @xmath110 @xcite . ) as before , we first relate alternating optimization of ( [ eq : upnlnfilter1])-([eq : upnlnfilter2 ] ) to bi - directional optimization . namely , fixing the uplink precoders @xmath56 , the optimal feedforward filters @xmath57 are given by ( [ eq : upnlnfilter1 ] ) . to optimize the uplink precoders , consider downlink transmission with perfect interference pre - compensation using precoders @xmath57 at the bts , and with encoding order from user @xmath1 to user @xmath34 . the received symbol at user @xmath2 can be written as ( [ eq : downrecsignln ] ) with @xmath111 . with receive filter @xmath9 , the mse ( normalized by @xmath88 ) at user @xmath2 is given by @xmath112 in analogy with the linear case we wish to @xmath113 the solution to this problem is given by ( [ eq : upnlnfilter2 ] ) . due to the constraint ( [ fullpow ] ) , the term @xmath114 in ( [ ipcmse ] ) is a constant so that the solution to this problem again does not depend on the noise level @xmath115 at user @xmath2 . in the absence of csi , the previous iterative optimization can be implemented by iterating the following steps : 1 . _ forward training : _ the @xmath1 users synchronously transmit uplink training symbols using the @xmath9 s as uplink precoders , and the bts optimizes the feedforward filters @xmath57 with perfect interference cancellation . 2 . _ backward training : _ the bts transmits training symbols using @xmath57 as downlink precoders with perfect interference pre - compensation in the _ reverse _ user order as for the uplink . user @xmath2 optimizes its receive filter @xmath9 subject to a norm constraint . iterate steps 1 and 2 . downlink interference pre - compensation can be achieved with dirty paper coding . however , this has high computational complexity . here we will instead consider two alternative approaches for pre - compensating interference at the bts during the training phase : sequential training and tomlinson - harashima precoding ( thp ) @xcite . thp for mimo channels is described in @xcite and can be used to pre - compensate downlink interference in a symbol - by - symbol manner . specifically , the dfd used in the uplink can be reused as the precoders for thp . as before , we replace the mse optimization criterion with a least squares criterion in adaptive mode . in the forward training step the users synchronously transmit sequences of @xmath95 training symbols given by the matrix @xmath96 where @xmath116 $ ] and @xmath98 is the @xmath99 row vector containing the training symbols . the received signal at the bts is then given by ( [ upsignal ] ) with @xmath100 . to obtain the feedforward and feedback receive filters for user @xmath2 , the known training symbols for users @xmath117 can be included in the least squares objective as proposed in @xcite . the corresponding sequence of estimated symbols can be written as @xmath118 where @xmath102 $ ] , @xmath119 is a row vector of dimension @xmath120 that contains the nonzero elements of the @xmath2-th row of @xmath121 , and @xmath122^t$ ] . the receive filters are then chosen to minimize the cost function @xmath123 , and the corresponding solution is given by @xmath124 we emphasize that in this way the filters are estimated _ directly _ , as opposed to estimating the channels first followed by computation of the filters . we next describe two downlink training methods that can be used to estimate the uplink precoders , accounting for the interference cancellation at the bts . _ sequential training : _ training of the receive filter at each user can be implemented by sequentially scheduling transmission of training symbols to the users in order @xmath31 . specifically , we layer the training sequences for user 1 , then add user 2 , then add user 3 , and so on , until user @xmath1 . to optimize the receiver filter at user @xmath2 , training symbols for users @xmath125 are therefore simultaneously transmitted with linear precoding filters @xmath126 so that the received signal vector at user @xmath2 is given by ( [ eq : downrecsignln ] ) with @xmath111 and @xmath127 ( using the notation defined in section [ sec : lineartrain ] ) . as for the linear case , user @xmath2 can estimate its receive filter @xmath9 by minimizing the least squares cost function @xmath106 and the corresponding solution is the same as in ( [ lsuser ] ) . this sequential training method perfectly pre - compensates for interference . however , a drawback is that when the filter for user @xmath2 is being estimated , the symbols corresponding to users @xmath117 are needed only to generate the appropriate interference ; they are not used to update the corresponding filters . hence this method will generally require more training than the simultaneous training method described next . _ training with thp : _ this method is illustrated in fig . [ downlink ] , which shows the uplink dfd filters being used as the precoders for thp . this can be viewed as moving the feedback part of the dfd in fig . [ uplink ] to the transmitter with the slicer function replaced by a modulo operation , which depends on the signal constellation used for modulation ( see also @xcite ) . because the estimates of the symbols in the uplink are biased , the dfd feedback filter must be modified to remove this bias when used in the thp - precoder . specifically , the feedback filter in the downlink precoder is the hermitian transpose of the matrix @xmath128 where @xmath129 and @xmath130 is the bias factor for estimating @xmath7 in ( [ dfesignal ] ) . we now describe the interference pre - compensation for the downlink using the feedforward receive filters @xmath57 as precoders and @xmath9 ( the precoder for user @xmath2 in the uplink ) as the receiver filter at user @xmath2 . denoting the input to the thp encoder as @xmath131 , the output is then @xmath132 where @xmath133 denotes the modulo operation and @xmath134 denotes the corresponding offset due to the modulo operation , which depends on the current symbol @xmath131 , the previous encoded symbols @xmath135 , and the feedback matrix . in general , each real and imaginary part is an integer multiple of the maximum distance between two constellation points ( see @xcite ) . the received symbol at user @xmath2 at the output of the receive filter is given by @xmath136 since @xmath137 and @xmath138 , the last summation in ( [ drecsignal ] ) is zero , i.e. , the interference to user @xmath2 generated by the users @xmath139 to @xmath1 can be completely removed , which means that user @xmath2 experiences interference only from users @xmath34 through @xmath41 . the effective desired symbol is then @xmath140 . we emphasize that thp in the downlink is implemented in the _ reverse _ user order as cancellation in the uplink . with the modulo operation , the power of the encoded symbols may increase relative to the original power ( prior to thp ) to a small extent , and the increment decreases as the constellation size increases @xcite . in ( [ drecsignal ] ) , since the powers of the effective training symbols and the residual interference are different from the corresponding powers in ( [ eq : downrecsignln ] ) ( with @xmath111 ) , training with thp in the downlink is an approximate approach to estimating ( [ eq : upnlnfilter2 ] ) . the training symbols for the downlink for different users are precoded according to ( [ modoper ] ) , where the bias factor for @xmath141 can be estimated as @xmath142 and @xmath143 is given by ( [ estalpha ] ) once the uplink receive filters have been obtained . the input to the feedforward filters @xmath23 s ( used as precoders ) at the bts is then given by ( [ modoper ] ) with @xmath16 replaced by the training symbol @xmath144 and @xmath145 replaced by @xmath146 . taking the desired symbol in ( [ drecsignal ] ) as @xmath147 , the receiver filter is then obtained by minimizing the least squares cost function @xmath148 , where @xmath149 $ ] denotes the received signal vectors and each entry @xmath150 is given by ( [ eq : downrecsig ] ) with @xmath151 and @xmath111 . the corresponding solution is given by @xmath152 where @xmath29 is chosen to satisfy the constraint @xmath59 . this expression for @xmath9 depends on the offset @xmath134 , which is introduced at the bts , and therefore must be estimated at the receivers . here we detect @xmath134 first before estimating @xmath9 . for downlink training the detection can be based on the output of the filter @xmath9 ( uplink precoder ) , which from ( [ drecsignal ] ) is given by @xmath153 the estimate of @xmath134 can be obtained by slicing @xmath154 . since the points in the discrete set of possible @xmath134 s have a minimum distance that is greater than the maximum distance between constellation points in each real or imaginary dimension , we expect that the performance loss due to errors in detecting @xmath134 is small . this is verified by the numerical examples in the next section . 5(a ) shows uplink sum rates versus snr achieved by different schemes in a single cell with @xmath155 users . there are @xmath156 antennas at the bts and two antennas at each user @xmath157 . the sum rate curve is obtained by choosing the filters @xmath9 s to maximize the achievable sum rate @xmath158 . the other four curves show the performance of linear and nonlinear filters with 100 forward - backward iterations , and with only two iterations . in each iteration an mmse update is performed , corresponding to an infinite amount of training . equal transmit power is assumed for all the users and the curves are averaged over a thousand random channel realizations . for the nonlinear precoder , thp with quadrature phase - shift keying ( qpsk ) is used for downlink training . the results show that with 100 forward - backward iterations , the sum rate with nonlinear filters is quite close to the uplink sum capacity , and is somewhat higher than with linear filters . with two iterations the gap between linear and nonlinear filters increases , especially at high snrs . hence nonlinear filters are more attractive when the number of forward - backward iterations is highly constrained . 5(b ) compares the sum rates achieved by limited and unlimited bi - directional training . two forward - backward iterations are used and the training sequence for each user is a randomly generated qpsk sequence . with 20 training symbols per iteration , the performance is close to that with mse updates ( infinite amount of training ) . this figure indicates that nonlinear filters still outperform linear filters with finite training . however , the gap in sum rate for nonlinear filters with infinite and finite training is slightly larger compared to the analogous gap with linear filters . 5(c ) shows the effects of varying the number of training symbols on nonlinear filters , again with two iterations . [ f : up_sum_rate_3nr ] shows the achievable sum rate in an overloaded system with @xmath159 receive antennas at the bts and @xmath155 users . each iteration is an mmse update ( infinite training ) . with linear filters the sum rate saturates as the received snr increases since there are not enough degrees of freedom to null the interference . however , with nonlinear filters the maximum degrees of freedom ( asymptotic slope of the sum rate ) of three can be achieved as the received snr increases . the sum rate approaches the capacity upper bound as the number of forward - backward iterations increases . ( more forward - backward iterations are needed at high snrs to approach the sum capacity . ) with three receive antennas and linear filters , full degrees of freedom can also be achieved if the bts schedules transmissions from only three out of the four users . the corresponding achievable sum rate is also included for comparison where the three users are scheduled randomly . it is observed that nonlinear filters ( with four users scheduled ) still outperform linear filters over a wide range of snrs in this scenario . however , at high snrs ( above 22 db ) the linear filters outperform the nonlinear filters , since the nonlinear filters require more iterations to reach the sum capacity . antennas at the bts and @xmath160 antennas at each user . the curves with nonlinear filtering correspond to @xmath155 users.,scaledwidth=50.0% ] fig . [ f : sum_rate_unified ] compares the uplink and downlink sum rates achieved by separately minimizing the mse for each with the corresponding rates achieved with simultaneous optimization , as described in section [ sec : uniapproach ] . for simultaneous optimization the transmit power at each node is normalized to satisfy the power constraint in each forward - backward iteration . while there are slight differences in the curves , simultaneous optimization performs essentially the same as separately optimizing the uplink or downlink precoders and receive filters . therefore the simultaneous approach is preferable from the perspective of reducing the training overhead . antennas at the bts , @xmath160 antennas at each user , and two forward - backward iterations.,scaledwidth=50.0% ] it is straightforward to extend bi - directional training for linear and nonlinear filters discussed in section [ s : training ] to the scenario with multiple cells . joint optimization of linear precoders and receivers in a cellular system has been considered in many references , including @xcite . achieving the maximum degrees of freedom in the multiple - cell scenario requires interference alignment across cells , as discussed in @xcite . that is , the other - cell interference lies in a subspace with lower dimension than the number of other - cell users . a related bi - directional training method was presented in @xcite . here we apply a nonlinear dfd at each bts for the uplink to cancel the interference from intra - cell users , while mitigating inter - cell interference via the precoders and feedforward filters . similarly , for the downlink we can apply thp as described in the preceding section to precompensate for the intra - cell interference while avoiding inter - cell interference . the training procedures are the same as described in section [ sec : nonlinear ] , where the users across cells synchronously transmit on the uplink and the bts s synchronously transmit on the downlink . the algorithms and results in sections [ sec : linearup][s : training ] can be easily extended to the multiple - cell scenario . [ f : sum_rate ] shows uplink sum rate versus snr in a two - cell scenario with three antennas at each bts and two dual - antenna users in each cell . all of the direct- and cross - channels are assumed to be of equal strength . the necessary conditions for interference alignment presented in @xcite are satisfied with equality with these parameters . in this scenario , the inter - cell interference in each cell can be aligned in a one - dimensional subspace @xcite , so that two degrees of freedom can be achieved in each cell . 8(a ) compares sum rates for linear and nonlinear filters with different numbers of iterations . in the high - snr regime there are substantial differences among the results with two , 100 , and 1000 forward - backward iterations , which indicates that it takes many more iterations for the precoders and filters to converge than for the single - cell scenario . that is , fig . 5(a ) shows that for the single - cell scenario , the achievable sum rate with two iterations is comparable to that with 100 iterations . in fig . 8(b ) , the performance of finite bi - directional training is presented with two forward - backward iterations . for these results downlink training with nonlinear filters uses thp with qpsk . with 20 training symbols per iteration , the performance is close to that with mmse updates ( infinite training ) . nonlinear filters give a noticeable gain over linear filters with either infinite or finite training over a wide range of snrs . [ f : sum_rate_3users ] shows the achievable uplink sum rate with two cells , three users in each cell , and three antennas at each user . here the necessary conditions for interference alignment in @xcite are not satisfied , so that interference alignment is not achievable . with linear filters the sum rate saturates as the snr increases irrespective of the number of forward - backward iterations . in contrast , with nonlinear filters the sum rate increases with snr given enough forward - backward iterations ( so that the filters can converge ) , and the maximum of two degrees of freedom per cell can be achieved . ( as in the last example , other - cell interference occupies a subspace of one dimension leaving two dimensions for the desired signals @xcite . ) users in each cell , @xmath159 antennas at each bts , and three antennas at each user.,scaledwidth=50.0% ] a distributed iterative bi - directional training approach has been proposed for jointly optimizing the filters in base transceiver stations and user equipments in cellular systems . the approach allows for a nonlinear decision feedback canceller in the uplink and a feedback precoder for downlink interference pre - cancellation . in the absence of channel state information , the filters can be directly estimated by synchronously transmitting uplink pilots from all users and downlink pilots from all bts s . downlink training may be nonlinearly precoded as well to enable simultaneous training of all user filters ( uplink precoders ) . this adaptive scheme avoids the need to stagger pilots across cells , and to exchange channel estimates across users and btss . with nonlinear filters , it is observed that the uplink sum rate achieved by bi - directional training in a single cell approaches the sum capacity with increasing number of iterations . although there are more parameters to be estimated compared to linear filters , with nonlinear filters finite training can significantly outperform linear filters with a limited number of forward - backward iterations . furthermore , the nonlinear filters have been observed to achieve the available degrees of freedom even when interference alignment is infeasible with linear precoders and filters . remaining open issues are how to trade off the number of iterations and the amount of training per iteration , and how to determine the optimum number of data - streams for each user if multiple data - streams are allowed . the proof of convergence for the simultaneous approach to uplink / downlink filter optimization in which the filters are normalized at each step also remains open . finally , an important issue in practice is how to combine such joint beamforming schemes with the scheduling of users across time - frequency resources . it is enough to show that given any set of precoders @xmath56 , to achieve the minimum uplink sum mmse each user should transmit with full power . denote @xmath161 $ ] , @xmath162 $ ] , and @xmath163 $ ] . the covariance matrix for the mmse estimation error @xmath164 is given by @xmath165 taking the trace of both sides gives the sum mse @xmath166 where the @xmath167 s are the eigenvalues of @xmath168 . assume that the transmit power of user @xmath2 is scaled up by @xmath169 , which implies the norm of the @xmath2-th column of @xmath170 is scaled by @xmath171 . define the new channel matrix as @xmath172 $ ] . denote @xmath173 as the @xmath2-th largest eigenvalue of @xmath174 . since @xmath175 and the matrix @xmath176 is positive semidefinite , by ( * ? ? ? * corollary 4.3.3 ) , we must have @xmath177 for each @xmath17 . hence increasing the power for any user must decrease the sum mse . s. shi , m. schubert , and h. boche , `` downlink mmse transceiver optimization for multiuser mimo systems : duality and sum - mse minimization , '' _ ieee trans . signal process . _ , vol . 55 , pp . 54365446 , nov . 2007 . d. j. love , r. w. heath jr . , v. k. n. lau , d. gesbert , b. d. rao , and m. andrews , `` an overview of limited feedback in wireless communication systems , '' _ ieee j. select . areas commun . _ , vol . 26 , pp . 13411365 , oct . 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. select . areas commun . _ , vol . 28 , pp . 13801408 , dec . 2010 . m. f. hanif , le - nam tran , a. tolli , m. juntti , s. glisic , `` efficient solutions for weighted sum rate maximization in multicellular networks with channel uncertainties '' , _ ieee trans.signal process . 56595674 , nov . 2013 . r. osawa , h. murata , k. yamamoto , and s. yoshida , `` performance of two - way channel estimation technique for multi - user distributed antenna systems with spatial precoding , '' in _ proc . ieee vehicular technology conference fall _ , sep . 2009 . h. weingarten , y. steinberg , and s. shamai ( shitz ) , `` the capacity region of the gaussian multiple - input multiple - output broadcast channel , '' _ ieee trans . inform . theory _ , vol . 52 , pp . 39363964 , aug . k. s. gomadam , v. r. cadambe , and s. a. jafar , `` a distributed numerical approach to interference alignment and applications to wireless interference networks , '' _ ieee trans . theory _ , vol . 57 , pp . 33093322 , jun . 2011 .
the success of symplectic integration algorithms ( ) as tools for the numerical solution of hamiltonian systems illustrates the importance of numerical algorithms that inherit the fundamental physical constraints of the systems to which they are applied ( `` geometric integrators '' ) . time - reversal symmetry plays a central role in physics ( e.g. , davies 1974 ) . to define reversibility formally and without reference to coordinates , let @xmath0 denote the state of the system governed by the differential equations = f(x ) . [ eq : basicde ] let @xmath1 be an involution , i.e. an operator such that @xmath2 . the system ( [ eq : basicde ] ) is @xmath1-reversible if @xmath1 reverses the direction of time , that is , if = -f(tx ) . in standard phase space @xmath3 and @xmath4 . the trajectory of the system governed by the differential equations ( [ eq : basicde ] ) is defined by a map @xmath5 such that a system located at @xmath0 at time 0 is found at @xmath6 at time @xmath7 ; by definition @xmath8 . then , @xmath1-reversibility implies that g_ttg_t = t tg_t = g_t^-1t = g_-tt . reversible systems may or may not be hamiltonian . however , the general features of motion in reversible and hamiltonian systems show many similarities , including the existence of families of kam tori ( e.g. , moser 1973 ; arnold 1984 ; roberts & quispel 1992 ) . these considerations motivate us to examine reversible integration algorithms ( ) . a one - step numerical integration algorithm with timestep @xmath9 is defined by an operator @xmath10 that is intended to be a close approximation to @xmath11 , and the algorithm is an if when it is applied to a reversible system _ note that in contrast to @xmath5 , @xmath12 is not necessarily equal to @xmath13 . let us first briefly review ; for more detail see channell & scovel ( 1990 ) , yoshida ( 1993 ) , marsden et al . ( 1996 ) , and sanz - serna & calvo ( 1994 ) . the standard example of a hamiltonian system is motion in a conservative potential ( this is also reversible ) . the hamiltonian is h(,)=^2+u ( ) [ eq : sepham ] and the equations of motion are = , = -u , or = -u . [ eq : second ] the most popular for such systems is the leapfrog or verlet algorithm ( e.g. , hockney & eastwood 1981 ) . this is a map from the phase - space coordinates @xmath14 at time @xmath15 to coordinates @xmath16 at time @xmath17 , defined by =_n+h_n , _ n+1=_n - hu( ) , _ n+1=+h_n+1 . leapfrog is an explicit second - order method ( i.e. the one - step error is o@xmath18 ) , but higher - order explicit can be constructed by concatenating leapfrog steps of different sizes , including backwards steps . leapfrog can be generalized to any separable hamiltonian of the form @xmath19 where each @xmath20 is integrable . for general hamiltonians , can be constructed by two methods . the first is based on expanding the scalar generating functions for the symplectic transformation @xmath21 in powers of @xmath9 ( kang 1986 ; channell & scovel 1990 ) . the simplest example is the first - order _ n+1=_n+hh(_n,_n+1)_n+1 , _ n+1=_n - hh(_n,_n+1)_n , which is implicit in general but explicit for hamiltonians of the form ( [ eq : sepham ] ) . the second approach is to construct symplectic implicit runge - kutta algorithms . for the differential equation ( [ eq : basicde ] ) , the @xmath22-stage runge - kutta method is defined by ( e.g. , ralston & rabinowitz 1978 ) k_i = f(x_n+h_j=1^sa_ijk_j),x_n+1=x_n+h_i=1^s b_jk_j . [ eq : rkimp ] runge - kutta methods are symplectic if they satisfy a simple algebraic test ( sanz - serna & calvo 1994 ) b_ia_ij+b_ja_ji - b_ib_j=0 , 1i , j s. [ eq : criterion ] on setting @xmath23 in ( [ eq : criterion ] ) , it is evident that all symplectic runge - kutta methods are necessarily implicit . the best known examples are the gauss - legendre runge - kutta methods ( e.g. , sanz - serna & calvo 1994 ) . the simplest can be written x_n+1=x_n+hf , [ eq : midptt ] which is the second - order implicit midpoint method . a symplectic fourth - order two - stage runge - kutta method is given by a_11=a_22= , a_12=- , a_21=+ , b_1=b_2=. [ eq : rk ] more generally , the gauss - legendre runge - kutta method is the unique @xmath22-stage method with order @xmath24 , and this method is always symplectic . these attractive features are offset by the computational cost of solving the implicit equations ( [ eq : rkimp ] ) . a limitation of is that they are difficult to generalize to variable timesteps . when a standard variable timestep prescription is applied to an , its performance is no better than that of non - symplectic integrators . the reason is simply that the mapping @xmath25 is usually not symplectic even when @xmath26 is . this is a serious problem , since most applications benefit from a variable timestep . one way to introduce variable timestep is by extending the phase space ( mikkola 1997 ) . suppose we wish to use a timestep @xmath27 . we define an extended phase space @xmath28 by @xmath29 and @xmath30 , where @xmath7 is time and @xmath31 is energy . we then define a new hamiltonian ( , ) = g ( , ) . the equations of motion for the hamiltonian @xmath32 in the extended phase space with fictitious time @xmath33 are the same as the equations of motion for the hamiltonian @xmath34 in the original phase space , supplemented by the condition @xmath35 . we may now integrate the hamiltonian @xmath32 using an with fixed timestep @xmath36 , which corresponds to @xmath37 . a limitation of this approach is that the hamiltonian @xmath32 is generally not separable , so that leapfrog and its generalizations can not be applied . several well - known one - step second - order algorithms for the differential equation ( [ eq : basicde ] ) are : for example , the implicit midpoint method ( eq . [ eq : midptt ] ) , the trapezoidal method , x_n+1=x_n+h , [ eq : trapp ] and the explicit midpoint method , x_n+2=x_n+hf(x_n+1 ) . [ eq : expmidpt ] some are also when they are applied to reversible hamiltonians . leapfrog and its generalizations are reversible . a reversible version of leapfrog with variable timestep @xmath27 is given by ( huang & leimkuhler 1997 ; calvo et al . 1998 ) @xmath38 quinn et al . ( 1997 ) describe a closely related algorithm based on a discrete set of timesteps that depend only on @xmath39 or @xmath40 and are separated by factors of two . a different approach to constructing is to modify non - reversible integration algorithms . for example , the following two operators are reversible even if @xmath10 is not : _ ^ -1 t , ( 1+t_ht)^-1(1+_h ) ; if we write @xmath41 then both of these operators are equal to @xmath42 to first order in @xmath9 . hut et al . ( 1997 ) describe numerical experiments with the second of these operators , calling it a `` time - symmetrization meta - algorithm '' since it can be applied to any one - step numerical integration algorithm . reversible algorithms remain reversible with a variable timestep so long as the timestep is determined symmetrically by the location of the system at the start and end of the step , e.g. @xmath43 $ ] ( hut et al . 1995 ) . special second - order differential equations of the form = f(x ) [ eq : fff ] are reversible , and so can profitably be integrated using . a useful source of high - order is linear multistep methods ( e.g. , henrici 1962 ; gear 1971 ; lambert 1973 ) , which have the form _ j=0^k ( a_k - jx_n - j - h^2b_k - jf_j)=0 , where @xmath44 , @xmath45 . we may assume without loss of generality that @xmath46 . the method is explicit if @xmath47 and otherwise implicit . linear multisteps include the classic strmer ( explicit ) and cowell ( implicit ) methods , which are characterized by @xmath46 , @xmath48 , @xmath49 , and @xmath50 for @xmath51 . it is easy to show that the requirement for reversibility is that @xmath52 and @xmath53 where @xmath54 ; thus the strmer and cowell methods are generally not reversible . multistep are discussed by lambert & watson ( 1976 ) , quinlan & tremaine ( 1990 ) , and fukushima ( 1998 , 1999 ) . in general , their performance over long time intervals on reversible dynamical systems is much better than strmer - cowell methods , although for occasional unfortunate choices of timestep their performance is ruined by timestep resonances ( quinlan 1999 ) . systems of first - order differential equations such as ( [ eq : basicde ] ) can be integrated by linear multistep methods of the form _ j=0^k(_k - jx_n - j - h_k - jf_n - j)=0 , [ eq : basic ] where @xmath44 , @xmath55 . we can assume without loss of generality that @xmath56 ; explicit methods have @xmath57 . these include the classic adams - bashforth ( explicit ) and adams - moulton ( implicit ) methods , which have @xmath58 and @xmath59 for @xmath51 ( e.g. , henrici 1962 ; gear 1971 ) . multistep methods for first - order differential equations are more general than multistep methods for special second - order equations , since any second - order equation can be written as a set of first - order equations . moreover , implementing variable timesteps is easy in first - order equations if we wish to use a timestep @xmath60 we introduce a fictitious time @xmath33 by the relation @xmath61 , and equation ( [ eq : basicde ] ) can be rewritten as = g(x)f(x ) , which can be integrated using unit timestep in @xmath33 . the aim of this paper is to investigate linear multistep for first - order differential equations . an important earlier investigation is that of cano & sanz - serna ( 1998 ) , who found that such methods typically possess grave numerical instabilities . we review general linear multistep methods in [ sec : intro ] , multistep in [ sec : reverse ] , together with their instabilities in [ sec : zero ] . we show in [ sec : zmm ] that it is possible to construct some linear multistep that are not subject to the cano & sanz - serna instabilities . a general discussion of stable multistep with up to 6 steps is given in [ sec : onetwo ] [ sec : fivesix ] . finally , [ sec : kep ] describes numerical examples based on integrating kepler orbits and [ sec : disc ] discusses our results . following henrici ( 1962 ) and lambert ( 1973 ) , we associate with ( [ eq : basic ] ) the linear operator l[x(t),h]=_j=0^k\{_k - jx[t+(k - j)h]- h_k - jx[t+(k - j)h]}. [ eq : linearoperator ] we can expand @xmath62 in a taylor series to obtain l[x(t),h]=_q=0^c_qh^qx^(q)(t ) , [ eq : tay ] where c_0=_l=0^k_l , c_q=1q!_l=0^k_ll^q - 1(q-1)!_l=0^k_ll^q-1,q=1,2, the order @xmath63 of the multistep is the unique integer for which @xmath64 , @xmath65 . the constant @xmath66 is known as the error constant and is a measure of the local truncation error . now define the characteristic polynomials ( ) = _ j=0^k_j^j,()= _ j=0^k_j^j . [ eq : poly ] if @xmath67 , then l[x(t),h]=e^t[(e^h)-h(e^h ) ] , which together with equation ( [ eq : tay ] ) implies ( 1+z)-(1+z)(1+z)=c_p+1z^p+1+(z^p+2 ) , [ eq : series ] where @xmath63 is the order and @xmath68 . order @xmath69 requires ( 1)=_j=0^k_j=0 , [ eq : rhoone ] and order @xmath70 requires (1)=(1 ) . [ eq : rel ] a multistep method is zero - stable if and only if the roots of @xmath71 lie inside the unit circle in the complex plane , or are on the unit circle and simple ( proofs of this are given by henrici 1962 , gear 1971 and lambert 1973 ) . zero - stability ensures that the parasitic solutions generated by the additional roots of the difference equation ( which is of order @xmath72 , while the original differential equation ( [ eq : basicde ] ) has order one ) do not grow , at least in the limit of zero timestep . we will discuss other forms of stability shortly in [ sec : zero ] . we assume that the polynomials @xmath71 and @xmath73 have no common roots other than 1 , for the following reason . suppose that @xmath74 is a common root , so that @xmath75 is a common factor . then @xmath76 , @xmath77 where @xmath78 and @xmath79 are polynomials of degree @xmath80 . then equation ( [ eq : series ] ) may be written ( 1+z)-(1+z)(1+z)=c_p+11-_0z^p+1 + ( z^p+2 ) . [ eq : seriesa ] thus @xmath81 define a @xmath82-step method with the same order and the same error constant ( @xmath83 $ ] ) as the original @xmath72-step method , and there is no obvious reason why the simpler method should not be used instead . there are @xmath84 unknown coefficients in equation ( [ eq : basic ] ) ( since @xmath56 ) , and therefore in principle these coefficients can be chosen so that the order is @xmath85 ( or @xmath86 if the method is explicit , with @xmath57 ) . however , the maximum order of a zero - stable multistep method is @xmath87 if @xmath72 is odd and @xmath88 if @xmath72 is even ( henrici 1962 ) . suppose that a trajectory @xmath89 , @xmath90 satisfies the difference equation ( [ eq : basic ] ) . the same segment of the time - reversed trajectory is given by @xmath91 , @xmath92 , where @xmath1 is the time - reversal operator and @xmath93 . we shall assume that @xmath94 and that @xmath95 . then @xmath4 and @xmath96 . in an the time - reversed trajectory should also satisfy the difference equation ( [ eq : basic ] ) , that is _ j=0^k(_k - jtx_n - k+j - h_k - jg_n - k+j)=0 . since @xmath97 , we may operate with @xmath1 to obtain _ j=0^k(_k - jx_n - k+j+h_k - jf_n - k+j)=0 , or , equivalently , _ j=0^k(_jx_n - j+h_jf_n - j)=0 . [ eq : reverse ] if ( [ eq : reverse ] ) is to be satisfied whenever ( [ eq : basic ] ) is satisfied , then we must have _ k - j = c_j,_k - j =- c_j , [ eq : cond ] where @xmath98 is a constant . applying this relation twice gives @xmath99 so @xmath54 . if @xmath100 , we shall say that the multistep method has even parity ; if @xmath101 , the parity is odd . the characteristic polynomials ( [ eq : poly ] ) now satisfy ( ) = c^k(^-1 ) , ( ) = -c^k(^-1 ) [ eq : symm ] . now let @xmath102 be the roots of @xmath71 . equation ( [ eq : symm ] ) implies that if @xmath103 is a root then so is @xmath104 . for zero - stability , the roots can not lie outside the unit circle therefore they must lie on the unit circle , and moreover they must be simple . equation ( [ eq : rhoone ] ) guarantees that @xmath105 is a root for any method with order @xmath69 . this root is simple if @xmath106 , and equation ( [ eq : rel ] ) then requires that @xmath107 for any method with order @xmath108 . however , even - parity methods have @xmath109 . therefore , even - parity methods are not zero - stable and we can restrict our attention to odd - parity methods , @xmath101 . if the reversibility criterion ( [ eq : cond ] ) is satisfied , then l[x(t),h]=cl[x(t+kh),-h ] . using ( [ eq : tay ] ) and the taylor series expansion for @xmath110 , we may derive the constraint c_p = c_n=0^p(-1)^p - nk^nn!c_p - n . we can re - write this equation as c_p[1-c(-1)^p]=_n=1^p(-1)^p - nk^nn!c_p - n . if @xmath111 , then this equation implies that @xmath112 if @xmath113 and thus the order can not be @xmath114 . therefore , odd - parity methods have even order . the stability of multistep methods for oscillatory problems can be parametrized by the interval of periodicity introduced by lambert & watson ( 1976 ) . suppose the right - hand size of equation ( [ eq : basicde ] ) is @xmath115 , with @xmath116 real . then , the multistep method ( [ eq : basic ] ) becomes a linear difference equation , with solution @xmath117 . here , @xmath118 and @xmath119 are complex constants , the latter satisfying ( ) -ih()=0 g ( ) -i(e^i)(e^i)=h . [ eq : period ] the interval of periodicity is the largest value of @xmath120 such that all of the roots of the first of equations ( [ eq : period ] ) lie on the unit circle . outside the interval of periodicity , the solution grows exponentially and hence is unstable . for reversible methods with odd parity , we may write g()=_j=0^k_j_j=0^k_j . if the multistep method is zero - stable , then @xmath71 has @xmath72 distinct roots on the unit circle . thus , @xmath121 has @xmath72 distinct roots on the interval @xmath122 $ ] . for sufficiently small values of @xmath120 , the equation @xmath123 will still have @xmath72 roots . as @xmath120 is increased , eventually a pair of these roots will disappear . this occurs at the smallest local maximum of @xmath121 and marks the end of the interval of periodicity . so , a plot of @xmath121 can be used to determine the interval of periodicity ( fukushima 1998 ) . the growth of errors in multistep methods is discussed by henrici ( 1962 ) , gear ( 1971 ) and especially by cano & sanz - serna ( 1998 ) . for our purposes , a limited heuristic version of these analyses is sufficient . let @xmath124 be a solution of the multistep method ( [ eq : basic ] ) . now perturb the solution to @xmath125 , @xmath126 . linearizing equation ( [ eq : basic ] ) , we have _ j=0^k(_k - j - h_k - jf_n - j)e_n - j=0 , [ eq : basica ] where @xmath127 . now look for solutions of the form @xmath128 where @xmath129 is smooth and @xmath119 is a complex constant . in the limit @xmath130 , @xmath131 , where @xmath132 is the accurate solution of the differential equation ( [ eq : basicde ] ) and therefore is smooth . then ( [ eq : basica ] ) becomes _ j=0^k^jy(t+jh)=0 , [ eq : basicb ] where @xmath133 . now expand in a taylor series in @xmath9 . to order @xmath134 we have @xmath135 , which requires that @xmath119 is one of the roots @xmath102 of the polynomial @xmath136 . to order @xmath9 we have = ( _ j=0^k_j^j _ j=0^kj_j^j)y(t ) fx(x(t))= ( x(t ) ) . thus , as @xmath137 , the linearized difference equation ( [ eq : basica ] ) has solutions of the form e_n=_j=1^k a_j_j^ny_j(t_0+nh ) , where @xmath138 satisfies the differential equation = _ jf(x(t))xy_j(t ) , [ eq : lvar ] and the growth parameters are _ j=(_j)_j(_j),j=1, ,k . [ eq : growth ] this should be contrasted with the variational equation of the orbit itself . if the trajectory @xmath139 is perturbed to @xmath140 , @xmath141 , then = f(x(t))xe(t ) . [ eq : var ] cano & sanz - serna ( 1998 ) point out that the variational equations ( [ eq : lvar ] ) and ( [ eq : var ] ) generally have quite different solutions , and hence the former can be unstable even if the latter is stable . however , this instability can be evaded for special values of the growth parameters @xmath142 : for @xmath143 , equations ( [ eq : growth ] ) and ( [ eq : var ] ) are the same , for @xmath144 , equation ( [ eq : growth ] ) is simply the time - reversed version of equation ( [ eq : var ] ) , and for @xmath145 , the solution of equation ( [ eq : growth ] ) is simply @xmath146const . we use the term `` zero - growth '' methods to denote multistep methods such that @xmath147 . cano ( 1996 ) has shown that for the reversible planar kepler problem , these are the only possible choices for the growth parameters @xmath142 to ensure stability , although for other potentials this may not be the case . multistep methods for integrating the special second - order equation @xmath148 ( lambert & watson 1976 ; quinlan & tremaine 1990 ) are not generally subject to this sort of instability , for the following reason . the expansion of the analog to equation ( [ eq : basicb ] ) in this case yields @xmath135 to order @xmath134 and @xmath149 to order @xmath9 . satisfying these two equations simultaneously requires that @xmath119 is a double root of @xmath71 ; therefore there is no instability if all of the roots of @xmath71 are simple ( cano & sanz - serna 1998 ) . in this section , all the zero - growth multistep algorithms with up to six steps are found . let us begin with some general results for stable , odd - parity , multistep , specified by the characteristic polynomials @xmath71 and @xmath73 . let @xmath102 be the roots of @xmath71 ; for zero - stable reversible methods , these are distinct , lie on the unit circle and are either real ( @xmath150 or @xmath151 ) or appear in complex - conjugate pairs . equation ( [ eq : rhoone ] ) implies that @xmath152 is always a root , which we denote @xmath153 . therefore , @xmath151 is also a root if and only if the total number of roots @xmath72 is even ; when present , we denote this root as @xmath154 . we normalize the multistep method so that @xmath56 and ( ) = _ j=1^k(-_j ) . we now demand that the growth parameter associated with the root @xmath155 is @xmath156 ( note that @xmath157 by eq . [ eq : rel ] ) . from equation ( [ eq : growth ] ) , this requires that ( _ l)=_l_l(_l)=_l_l_j = l(_l-_j ) . [ eq : sigjj ] the unique polynomial of order @xmath72 that passes through the points ( [ eq : sigjj ] ) and satisfies the symmetry condition ( [ eq : symm ] ) is ( ) = _ l=1^k _ l(+_l)_j = l(-_j ) , [ eq : sigdef ] with @xmath158 when @xmath159 . an alternative form is ( ) = ( ) _ l=1^k _ l+_l-_l . [ eq : sigdefa ] note that if @xmath160 then @xmath155 is a root of both @xmath71 and @xmath73 . in this case , there is a zero - growth multistep with the same order and error constant but fewer steps ( see [ sec : intro ] ) . therefore , we restrict our attention to methods with @xmath161 . the method is explicit if @xmath162 , which in turn requires _ j=1^k _ j=0 , [ eq : www ] a condition that can only be satisfied if @xmath72 is even . a @xmath72-step method is completely specified by the @xmath163-order polynomials @xmath71 , @xmath73 . in a stable multistep with @xmath72 even , the roots of @xmath71 are @xmath150 , @xmath151 , @xmath164 where @xmath165 , @xmath166 . for @xmath72 odd , the root at @xmath151 is not present and the index runs to @xmath167 . in a zero - growth method , the polynomial @xmath73 is determined by equation ( [ eq : sigdef ] ) once @xmath71 and the signs @xmath142 are specified . we now discuss all stable , zero - growth , multistep of @xmath168 steps . we label the interesting methods by sz@xmath72p where @xmath72 is the number of steps and `` p '' is an optional suffix that distinguishes different methods with the same @xmath72 . since @xmath105 is always a root , for @xmath169 we must have @xmath170 . since @xmath157 by equation ( [ eq : rel ] ) , the zero - growth requirement is automatically satisfied . equation ( [ eq : sigdef ] ) requires that @xmath171 , which yields the trapezoidal method ( eq . [ eq : trapp ] ) x_n+1=x_n+h(f_n+1+f_n ) ; [ eq : trap ] the first few terms of the taylor series ( [ eq : tay ] ) are c_0=0 , c_1=0 , c_2=0 , c_3=- ; thus the error constant is @xmath172 and the method is second - order . the function @xmath121 ( eq . [ eq : period ] ) is @xmath173 , so the interval of periodicity is infinite . for @xmath174 the roots of @xmath71 can only be @xmath175 ( @xmath105 is always a root ; @xmath154 must be real since @xmath176 is also a root and there are only two roots ; @xmath154 must be on the unit circle and distinct from @xmath153 ; thus @xmath177 ) . thus , the first characteristic polynomial is ( ) = ^2 - 1 . there are two choices for the second characteristic polynomial @xmath73 , depending on @xmath178 : ( ) = 2,()=^2 + 1 . [ eq : sigtwo ] the first of these gives the explicit midpoint method ( eq . [ eq : expmidpt ] ) , x_n+1=x_n-1 + 2hf_n . [ eq : mid ] the first few terms of the taylor series ( [ eq : tay ] ) are c_0=0 , c_1=0 , c_2=0 , c_3= ; thus the error constant is @xmath179 and the method is second - order . the function @xmath121 is @xmath180 so the interval of periodicity is @xmath181 . the second equation for @xmath73 in ( [ eq : sigtwo ] ) yields x_n+1=x_n-1+h(f_n+1+f_n-1 ) , [ eq : traptwo ] which is the same as the trapezoidal method ( [ eq : trap ] ) , with a timestep of @xmath182 . all of these two - step methods are special cases of a one - parameter family of multistep , x_n+1=x_n-1+h[_0f_n+1 + 2(1-_0)f_n+_0f_n-1 ] . [ eq : btwo ] the explicit midpoint method ( eq . [ eq : mid ] ) has @xmath183 and the trapezoidal method with timestep @xmath182 ( eq . [ eq : traptwo ] ) has @xmath184 . when @xmath185 , the method is simply the composition of two trapezoidal steps ( eq . [ eq : trap ] ) . milne s method ( @xmath186 ) is the two - step method of maximum possible order ( @xmath187 ) , but its disadvantage is that its growth parameter @xmath188 and so it is susceptible to numerical instability . the interval of periodicity @xmath189 for @xmath190 and infinity for @xmath191 . to compare these methods we have integrated a kepler orbit with unit semimajor axis and period @xmath192 for 100 time units using various eccentricities and timesteps ( fig . [ fig : multi2 ] ) . even for timesteps as low as @xmath193 , the relative energy error @xmath194 , except near the zero - growth methods at @xmath195 . in particular , the second - order zero - growth methods even the explicit method at @xmath183 exhibit much better behavior than the fourth - order milne s method . for @xmath196 , the distinct roots of @xmath119 can only be @xmath105 and @xmath197 where @xmath198 and @xmath199 . thus , ( ) = ^3-(2u+1)^2+(2u+1)-1 . there are two choices for @xmath73 . the first choice is to take @xmath200 , so that we have ( ) = ^3-(+u)^2-(+u)+ . the first few terms of the taylor series ( [ eq : tay ] ) are c_0=0 , c_1=0 , c_2=0 , c_3=-+u . @xmath201 has no solution for @xmath198 so the method is second - order ; the interval of periodicity is infinite . the second choice is to take @xmath202 so that we have ( ) = -^3+(-u)^2+(-u)- . the first few terms of the taylor series ( [ eq : tay ] ) are c_0=0 , c_1=0 , c_2=0 , c_3=+u . once again , @xmath201 has no solution for @xmath198 so the method is second - order . the interval of periodicity ranges from @xmath203 as @xmath204 to 0 as @xmath205 . for @xmath206 , the distinct roots of @xmath119 can only be @xmath105 , @xmath177 and @xmath207 where @xmath198 and @xmath199 . thus , ( ) = ^4 - 2u^3 + 2u-1 . there are four choices for @xmath73 : @xmath208 , @xmath209 . none of these yield methods of order @xmath210 for @xmath198 . the three- and four - step methods have no obvious advantage over the trapezoidal or explicit midpoint methods , and we will not explore them further . for @xmath211 , the distinct roots of @xmath119 are @xmath105 , @xmath212 and @xmath213 , where @xmath214 and @xmath215 . there are three choices for @xmath73 : @xmath216 , or @xmath217 ( note that @xmath218 is not a distinct choice ) , with @xmath219 and @xmath220 . the choices @xmath221 yield only second - order methods . the choice @xmath222 yields a fourth - order method if u_2=1 + 11u_113-u_1 , u_1(-1,1 ) , u_2(-,1 ) . [ eq : l5pm ] the integration formula is @xmath223 \qquad\qquad\hbox{method sz5}. \label{eq : s5pm}\end{aligned}\ ] ] the error constant is c = c_5(1)=17u_1 + 1031440(u_1 - 1 ) . for @xmath224 , the distinct roots of @xmath119 are @xmath105 , @xmath177 , @xmath225 and @xmath226 , where @xmath214 and @xmath215 . there are six choices for @xmath73 : @xmath227 or @xmath150 and @xmath228 , @xmath217 , with @xmath229 and @xmath230 . the choices @xmath231 or @xmath217 yield only second - order methods . the choice @xmath232 yields a fourth - order method if u_2=1 + 2u_14-u_1 , u_1(-1,1 ) , u_2(-,1 ) . [ eq : l6ppm ] the error constant is c = c_5(1)=14+u_145(u_1 - 1 ) . the integration formula is @xmath233 \qquad\qquad \hbox{method sz6i } , \label{eq : s6ppm}\end{aligned}\ ] ] the label `` i '' stands for `` implicit '' . the choice @xmath234 yields an explicit fourth - order method if u_2=7u_1 - 1u_1 + 5 , u_1(-,1 ) , u_2(-1,1 ) . [ eq : l6mpm ] the error constant is c = c_5(1)=19 + 11u_1180(1-u_1 ) . the integration formula is @xmath235\qquad\qquad \hbox{method sz6e}. \label{eq : s6mpm}\end{aligned}\ ] ] the intervals of periodicity for the three fourth - order methods are shown in figure [ fig : period ] . these are largest for @xmath236 near @xmath151 and decrease rapidly for @xmath237 . thus , only methods with negative @xmath236 are of practical use . at best , the interval of periodicity is significantly smaller than that of comparable multistep for special second - order equations ( lambert & watson 1976 ; quinlan & tremaine 1990 ; fukushima 1998 ) . this is the price of the greater generality of the present methods . this section supplements our theoretical analysis of multistep with numerical experiments using the keplerian hamiltonian @xmath238 . the equations of motion are = v_x , = v_y , = -xr^3 , = -yr^3 , where @xmath239 and the gravitational constant @xmath240 and the attracting mass @xmath241 have been set to unity . our test integrations follow an orbit with eccentricity @xmath242 , unit semi - major axis , and orbital period @xmath192 . the orbit is started at apocenter on the @xmath0-axis , so the initial conditions are x=1+e , y=0 , v_x=0 , v_y=[(1-e)/(1+e)]^1/2 . the interval of periodicity shown in figure [ fig : period ] gives the maximum stable timestep for the harmonic oscillator potential only . it is of greater relevance to astronomers to establish the timesteps that can be safely used to integrate circular or mildly eccentric orbits in the kepler potential . figure [ fig : periodkepler ] shows the largest timestep for which the relative energy error is less than @xmath243 after 100 orbital periods . the results displayed are for circular orbits , but the curves for mildly eccentric orbits ( @xmath244 ) are almost indistinguishable . a comparison between figures [ fig : period ] and [ fig : periodkepler ] reveals that the timestep needed for accurate integration of keplerian near - circular orbits is typically at least a factor of five smaller than would be naively inferred from the interval of periodicity . figure [ fig : multi56 ] shows a further test of the multistep . the standard kepler problem is integrated for 100 time units with timestep @xmath245 and eccentricities 0.2 and 0.5 . a striking feature is that the error is almost independent of the parameter @xmath236 over a limited range , while outside this range the method is unstable . the stable range shrinks as the eccentricity increases . the dependence of energy error on timestep is explored in figure [ fig : hvar56 ] . in this figure , we have specialized to a single value of the parameter @xmath236 : @xmath246 for sz5 and sz6i and @xmath247 for sz6e . all three methods display the expected dependence @xmath248 for @xmath249 . for larger values of @xmath9 , the methods are unstable . we have also plotted the behavior of the classic fourth - order adams - bashforth ( explicit ) and adams - moulton ( implicit ) multistep methods . the classic methods exhibit similar errors for @xmath249 but remain stable at much larger timesteps . the advantage of the reversible multistep methods only becomes clear over longer time intervals . figure [ fig : long56 ] shows that the maximum energy error in the reversible methods is constant at large times , while the energy error in the classical methods grows linearly . variable timesteps are useful for eccentric kepler orbits . to incorporate these , we introduce a fictitious time @xmath33 by the relation @xmath250 , so the equations of motion become = g(x , y)v_x , = g(x , y)v_y , = -g(x , y)xr^3 , = -g(x , y)yr^3 , = g(x , y ) . [ eq : kepvar ] these are integrated using unit timestep in @xmath33 . we typically use @xmath251 since this is the characteristic free - fall time from radius @xmath252 . we parameterize the timestep by the number of force evaluations per unit time ( or per radian , since the orbital period is @xmath192 ) . this is distinct from the number of steps per radian because implicit methods require several iterations per step , and because some explicit methods though not multistep methods require several force evaluations per step . figure [ fig : trap ] shows the fractional energy errors resulting from integrating kepler orbits using the trapezoidal method ( eq . [ eq : trap ] , solid lines ) and explicit adaptive leapfrog ( eq . [ eq : calvo ] , dashed lines ) . each step of the trapezoidal method was iterated to convergence , with the first approximation taken from the first - order euler method typically 3 iterations were required at the smallest timesteps , rising to @xmath253 at the largest . the energy errors are the maximum over 1000 time units or @xmath254 orbits . the results for shorter integrations over 100 time units are almost indistinguishable , showing that there is no secular energy drift with either integration method . adaptive leapfrog generally gives energy errors that are smaller by about an order of magnitude , mostly because it is explicit so there are fewer force evaluations per timestep . we have conducted similar experiments with the explicit midpoint method ( eq . [ eq : mid ] , not shown in figure ) but this is much less successful at following high - eccentricity orbits , presumably because of its smaller interval of periodicity . we next investigate the fourth - order multistep . these have relatively small intervals of periodicity and can not reliably integrate high - eccentricity orbits . however , they are successful at following orbits with moderate eccentricities . figure [ fig : fourth ] shows the fractional energy errors resulting from integrating a kepler orbit with eccentricity @xmath255 using four methods : the explicit six - step method with @xmath256 ( method sz6e , eq . [ eq : s6mpm ] ) ; the implicit six - step method with @xmath257 ( method sz6i , eq . [ eq : s6ppm ] ) ; the implicit five - step method with @xmath257 ( method sz5 , eq . [ eq : s5pm ] ) . in all panels , the heavy solid line is the error resulting from integrating the eccentric orbit with variable timestep , the light solid line is the error from integrating the same orbit with fixed timestep , and the dashed line is the error from integrating a circular orbit with fixed timestep . for large timesteps , the implicit methods do not converge and in this case no energy error is plotted . for all four methods a variable timestep yields errors that are 12 orders of magnitude smaller than a fixed timestep ( at the same number of function evaluations per unit time ) . the explicit multistep method performs much better than the implicit methods , in part because the implicit methods require 38 force evaluations per step to iterate to convergence . however , even if the convergence of the implicit multistep methods could be achieved in 23 iterations they would not perform significantly better than the explicit method on this test . a surprising result is that there is a range of the horizontal axis in which the multistep methods with variable timestep integrate eccentric orbits with smaller energy error than circular orbits . toomre ( 1990 , private communication ) and quinlan ( 1999 ) showed that multistep for the special second - order differential equation ( [ eq : fff ] ) suffer instabilities at unlucky timesteps at which there is a resonance between the oscillation frequencies of the solution and the method . this phenomenon must also occur in the multistep for first - order differential equations that are investigated here . suppose the angular speed of the circular orbit given by the numerical method is @xmath116 . if the number of steps in the multistep is at least 5 ( or 6 for second - order differential equations ) , then there are two or more complex - conjugate pairs of roots of the characteristic polynomial ( [ eq : poly ] ) on the unit circle . these we write as @xmath258 and @xmath259 . quinlan ( 1999 ) shows that the troublesome timesteps @xmath9 satisfy _ i - _ j = 2h . [ eq : gerry ] in practice these resonant instabilities are never a concern for our zero - growth multistep . this is because the timesteps ( [ eq : gerry ] ) lie outside the interval of periodicity for nearly circular keplerian orbits . let us demonstrate this by considering one example in more detail namely , the 6-step implicit method sz6i defined by ( [ eq : l6ppm ] ) and ( [ eq : s6ppm ] ) . this is the most susceptible to resonant instabilities of the zero - growth methods we have examined , both because it has the largest interval of periodicity and because , as @xmath260 , two of the spurious roots coalesce as @xmath261 . we recall that there are five spurious roots on the unit circle , namely @xmath177 , and the two complex - conjugate pairs @xmath225 and @xmath262 , where @xmath215 . using ( [ eq : l6ppm ] ) , the angular separations between the spurious roots on the unit circle can be readily computed as a function of @xmath236 . figure [ fig : final ] shows the locations of the potentially troublesome timesteps @xmath9 , together with the maximum possible timestep ( already presented in figure [ fig : periodkepler ] ) . the resonant instabilities always occur at timesteps that are outside the interval of periodicity for keplerian orbits . similar results hold for the methods sz5 and sz6e . the generic instabilities of multistep for first - order differential equations identified by cano & sanz - serna ( 1998 ) are a grave problem and render many of these methods unusable . the main contribution of this paper is the elaboration of a class of multistep the zero - growth methods that successfully evade the instabilities . this select group includes two familiar second - order multisteps , namely the trapezoidal method ( [ eq : trap ] ) and the explicit midpoint method ( [ eq : mid ] ) , as well as entirely new multistep . in particular , we have identified three one - parameter families of zero - growth , fourth - order namely , the five - step implicit method sz5 defined by equations ( [ eq : l5pm ] ) and ( [ eq : s5pm ] ) , the six - step implicit method sz6i defined by ( [ eq : l6ppm ] ) and ( [ eq : s6ppm ] ) , and the six - step explicit method sz6e defined by ( [ eq : l6mpm ] ) and ( [ eq : s6mpm ] ) . of these , the explicit method sz6e eliminates the need to iterate to convergence at each timestep without a major sacrifice in the interval of periodicity . the stability of the zero - growth methods has been understood at a theoretical level and confirmed at a practical level by integrations of circular and eccentric keplerian orbits . there are several advantages of multistep for first - order differential equations over the multistep for special second - order differential equations examined by lambert & watson ( 1976 ) , quinlan & tremaine ( 1990 ) , and fukushima ( 1998 , 1999 ) . they permit the introduction of variable timesteps without spoiling reversibility or long - term energy conservation . they are also more general , since any ordinary differential equations can always be reduced to a system of first - order equations ; thus , for example , the methods described here can be used to follow orbits in rotating frames of reference or the motions of rotating rigid bodies . a third advantage is that they are less susceptible to timestep resonances . the price paid for these attractive features is that ( i ) the interval of periodicity of the higher order methods such as sz5 , sz6i and sz6e is rather small , forcing the use of smaller timesteps than other methods ; ( ii ) more steps are needed for a method of given order ; there are no fourth - order methods with fewer than five steps and no fifth or sixth - order methods with 8 or fewer steps ( the limit of our explorations ) . nonetheless , the six - step explicit method ( sz6e ) remains a competitive option for problems in which it is critical to have variable timesteps and to avoid irreversible numerical errors : examples include long - term integrations of asteroid or comet orbits , or orbits near the centers of galaxies . 1 . are there zero - growth with more than @xmath224 steps that have higher order or a larger interval of periodicity ? we have explored all @xmath263 but have found no of order greater than 4 . gautschi ( 1961 ) developed a class of methods closely related to the linear multisteps . just as a linear multistep is defined by the requirement that the operator ( [ eq : linearoperator ] ) annihilate all algebraic polynomials of order @xmath264 , so the gautschi multisteps annihilate certain trigonometric polynomials up to a given degree . these methods are particularly appealing when the solution is known to be periodic and a reasonable estimate for the period of the orbit can be guessed in advance requirements that are often satisfied in solar system integrations . can we find reversible gautschi multisteps ? integration methods are generally designed to maximize the order @xmath63 , defined so that the one - step error in following a system with characteristic frequency @xmath116 is proportional to @xmath265 . perhaps it is more sensible to `` economize '' the error , by minimizing the maximum value of the error over a range of frequencies @xmath266 . in this case the error would be nearly proportional to a chebyshev polynomial @xmath267 with argument @xmath268 and degree @xmath269 , instead of @xmath270 . we are indebted to gerry quinlan for communicating his results on resonant instabilities in advance of publication . this research was supported in part by nasa grant nag5 - 7310 . nwe thanks both the canadian institute for theoretical astrophysics and the department of astrophysical sciences , princeton university for their hospitality during working visits . his research is supported by the royal society . hut , p. , funato , y. , kokubo , e. , makino , j. , & mcmillan , s. 1997 , in computational astrophysics : 12th kingston meeting on theoretical astrophysics , eds . d. a. clarke & m. j. west ( san francisco : astronomical society of the pacific ) , 26
there is current interest ( see , for instance , refs . ) in studying inverse transitions , melting or freezing , motivated by the quite unconventional situation present in such transitions in which the ordered phase is more entropic than the disordered one @xcite . this apparent counter - intuitive transition has become even more interesting since there are now various physical systems displaying this kind of transition , as for example , magnetic films @xcite and , particularly , high-@xmath0 superconductors @xcite . therefore , the knowledge about which conditions are necessary for the existence of inverse transitions is a challenging issue @xcite . among the proposed classical magnetic models which can show inverse transitions , the blume capel ( bc ) @xcite and the gathak - sherrington ( gs ) @xcite models can be useful to clarify what would be such conditions . for instance , the phase diagram of the bc model @xcite displays inverse melting with a phase transition from the ferromagnetic ( @xmath1 ) phase to the paramagnetic ( @xmath2 ) one when the temperature is decreased . however , in order to obtain inverse melting in the bc model , it is strictly necessary to impose in the problem the entropic advantage of the interacting state through parameter @xmath3 , where @xmath4 and @xmath5 are the degeneracy of @xmath6 and @xmath7 spins states , respectively @xcite . in contrast with the bc model , the gs model @xcite - which is similar to the the bc model except that it has the spins couplings given as random gaussian variables - shows naturally inverse freezing . in other words , there is a reentrant first order boundary phase separating the spin glass ( @xmath8 ) and @xmath2 phases . however , there is no need of any entropic advantage of interacting states @xcite . the previous discussion suggests that the randomness is responsible for producing naturally inverse freezing in the gs model as compared with the inverse melting case in the bc model . in fact , to be more precise , it seems that the presence of non - trivial form of randomness , which means frustration , @xcite would be the essential ingredient which allows inverse freezing in the gs model with no need of entropic advantage . that sets some questions : what is the actual role of non - trivial randomness to produce naturally inverse freezing ? is this natural inverse freezing robust when the level of frustration is diminished ? can models with trivial randomness present naturally inverse transitions ? the purpose of the present work is to investigate what is the actual role of frustration as a basic condition to produce naturally inverse freezing . that could be achieved by comparing two distinct situations , the trivial randomness and the highly frustrated regime and also studying how natural inverse freezing would be affected when the level of frustration is varied . therefore , the problem is not only to find a random magnetic model which presents naturally inverse freezing , but also to get one which allows the level of frustration to vary . quite recently , the fermionic ising spin - glass ( fisg ) model @xcite has been proposed as a model able to display naturally inverse freezing in a phase diagram temperature _ versus _ the chemical potential @xmath9 . this model has the spin operators @xmath10 given as bilinear combinations of creation and destruction fermionic operators which have four eigenstates , two of them non - magnetic @xcite . the spin - spin coupling @xmath11 , likewise the gs model , is a random variable which follows a gaussian distribution . indeed , the existence of inverse freezing in this model could be expected since there is a close relationship between the gs and the fisg models @xcite . for instance , the partition function of these two models can be related by a mapping between the anisotropy constant @xmath12 of the gs model and @xmath9 @xcite . the fisg model is also useful to examine how quantum effects , included by the presence of a transverse field @xmath13 @xcite can affect inverse freezing . in ref . , it has been shown that , when @xmath13 is increased , the reentrance in the @xmath14 first order boundary phase in the phase diagram temperature @xmath15 _ versus _ the chemical potential @xmath9 gradually disappears . this scenario also suggests that @xmath13 in the fisg model plays the opposite role of the rather artificial parameter @xmath16 in the bc model discussed in ref . . nevertheless , the fisg model , in its original form with gaussian random couplings , is a strongly frustrated model . therefore , it is not possible to investigate in this model how the change of the level of frustration could affect the inverse freezing . an alternative route to accomplish the previously mentioned investigation would be to use the fisg model in a version in which one could adjust the level of frustration . in that sense , the classical hopfield spin glass @xcite can be quite useful . in this model , the interactions @xmath11 between classical ising spins are given below : @xmath17 where @xmath18 ( @xmath19 or @xmath20 , @xmath21 is the number of sites ) are independent random distributed variables . this type of interaction has been intensively used to study complex systems @xcite . there are clearly two extreme cases for the classical hopfield ising spin glass model . when @xmath22 , it becomes the mattis model @xcite , which is a well known example of trivial randomness @xcite . the second one is when @xmath23 . in that limit , the thermodynamics corresponds to the strongly frustrated regime given by a random gaussian distributed @xmath11 @xcite . in particular , the replica symmetric mean field solution of the classical hopfield spin glass model @xcite can provide a useful method for the purposes of the present work . it is described in terms of two order parameters , the usual spin glass @xmath24 and @xmath25 which indicates the presence of mattis states that have a thermodynamics similar to the usual ferromagnetism @xcite . most important , there is a parameter called here degree of frustration , @xmath26 which allows controlling the level of frustration in the theory . it is important to remark that there is a particular value @xmath27 which for @xmath28 the effects of frustration are dominant @xcite . as a consequence , it is possible to interpolate the thermodynamics from the trivial randomness ( @xmath29 ) to the strongly frustrated regime ( @xmath28 ) . therefore , to answer the questions arisen previously , we use the hopfield fermionic ising spin glass ( hfisg ) model . in this case , the spin operators @xmath30 are defined as the fisg model , but the random spin coupling @xmath11 is given as eq . ( [ hebb ] ) . the partition function is obtained in the functional integral formalism using grassmann fields where the disorder is treated with the replica method @xcite . the problem can be reduced to a one - site problem by using a similar procedure to solve the classical hopfield spin glass @xcite within the static approximation ( sa ) @xcite and one - step replica symmetry breaking ( 1s - rsb ) @xcite . it should be remarked that in the hfisg model , as the fisg one @xcite , the replica diagonal component of the sg order parameter @xmath31 appears as an additional order parameter to be solved with the others . particularly , this order parameter represents the spin self - interaction that has an imaginary time dependence for quantum sg models @xcite . however , the spin operators @xmath32 commute with the hamiltonian operators for the present hfisg model . consequently , the @xmath31 has no dynamic , which means that the sa is exact in the present work@xcite . one important point in the present work is how to locate first order boundaries phases found for different degrees of frustration . the criterion adopted here follows closely that one suggested in ref . for the classical @xmath33 model . hence , we selected , from the set of spin glass solutions in the transition , that one which meets continuously with the spin glass solution for small value of the chemical potential @xmath9 . in fact , that corresponds to the largest spin glass order parameter which gives the lowest grand canonical potential ( see also the discussion in ref . ) . then , by equating the grand canonical potential of the spin glass and paramagnetic solutions , the first order boundary phase is located . the use of the 1s - rsb scheme also deserves some remarks since our main interest is to obtain boundaries of phase transitions . it has been found in previous works @xcite that the use of the replica breaking symmetry ( rsb ) schemes in gs or fisg models essentially preserves the reentrance in the @xmath14 first order boundary phase associated to the inverse freezing . there are only small differences which appear mainly at very low temperatures . even so , for the hfisg model , we decided to use 1s - rsb scheme to check for intermedianted values of @xmath34 whether or not these differences will remain unimportant . moreover , we also analyzed the stability of the rs solutions for a sake of completeness of the work . this paper is structured as follow : in section 2 , we derived the thermodynamics and the set of coupled equations for the saddle point order parameters . in section 3 , phase diagrams temperature _ versus _ the chemical potential are presented for several values of @xmath34 . the entropy behavior as function of temperature and the grand canonical potential as function of chemical potential are also discussed . finally , section 4 is reserved to conclusions . the hamiltonian considered here is a hopfield fisg ( hfisg ) model @xmath35 where @xmath36 is given in eq . ( [ hebb ] ) and the random @xmath37 follows the distribution @xmath38 in this model , @xmath39 $ ] is the spin operator , with @xmath40 as the number operator , @xmath41 are fermions creation ( destruction ) operators and @xmath42 or @xmath43 indicate the spin projections . the partition function in the grand canonical ensemble is given in the lagrangian path integral formalism where the spin operators are represented as bilinear combinations of anticommuting grassmann fields ( @xmath44 ) @xcite @xmath45 where @xmath46 \phi_{j\sigma}(\tau ) \\ & - h(\phi^{*}(\tau),\phi(\tau ) ) \ } \end{split}\ ] ] where @xmath9 is the chemical potential and @xmath47 . @xmath48 is fourier transformed in time which results in @xmath49 with @xmath50\phi_{j\sigma}(\omega ) \label{eqao}\ ] ] @xmath51 and @xmath52 where @xmath9 is chemical potential , @xmath53 , @xmath54 are the matsubara s frequencies and @xmath55 if @xmath56 . in this work , the problem is analyzed within sa which considers only the term when @xmath57 in eq . ( [ eqs ] ) @xcite . the grand canonical potential is obtained by using the replica method : @xmath58 where @xmath59 and @xmath60 means the configurational averaged over @xmath61 . thus : @xmath62 \label{zn1}\ ] ] where after using @xmath11 given in eq . ( [ hebb ] ) , the action @xmath63 can be written as @xmath64 with @xmath65 denoting the replica index and @xmath66 . the average over @xmath67 given in eq . ( [ zn1 ] ) is discussed in detail in the appendix a. in the present work , the one - step replica symmetry breaking ( 1s - rsb ) ansatz is adopted , in which the replica matrix @xmath68 and the matrix @xmath69 are parametrized as : @xmath70 and order parameters @xmath71 are invariant with respect to permutations of replicas : @xmath72 , where @xmath73 . therefore , the parametrization ( [ 1srsb ] ) is used in eqs . ( [ med1])-([e25 ] ) . as a consequence , the 1s - rsb grand canonical potential is found as @xmath74}{1-\beta j(\bar{q}-q_1 ) } \\ & -\frac{1}{2 } \frac{\beta j a q_0}{1-\beta j[\bar{q}-q_1 + x(q_1-q_0 ) ] } + \frac{\beta j m^2}{2 } + \frac{a}{2 } \\ & \times\ln [ 1-\beta j(\bar{q}-q_1 ) ] -\lim_{n\rightarrow 0}\frac{1 } { n}\ln\langle\langle \theta(\{r\},m,\xi ) \rangle\rangle_{\xi } , \end{split}\ ] ] where @xmath75 \\&\times\exp \left [ ( r_1-r_0)\sum_{l=1}^{n / x}\left(\sum_{\alpha=(l-1)x+1}^{lx}s^\alpha\right)^2 \right ] \\&\times\exp\left [ r_0(\sum_{\alpha=1}^n s^{\alpha})^2 + \beta j \sum_{\alpha=1}^{n } ( \xi m s^{\alpha } ) \right ] \end{split}\ ] ] with @xmath26 . the quadratic forms into the function @xmath76 can be linearized by hubbard - stratonovich transformations where new auxiliary fields are introduced in the problem . therefore , one has : @xmath77^{n / x } , \end{split } \label{e28}\ ] ] with @xmath78 ( @xmath79 ) and @xmath80 the local spin glass component of the random field @xmath81 is defined by @xmath82 . \end{split } \label{e30}\ ] ] the functional integral over the grassmann variables , as well as the sum over the matsubaras frequencies in eq . ( [ e28 ] ) , can be performed following closely the procedure given in references @xcite . finally , the grand canonical potential within 1s - rsb approximation can be written as @xmath83 } + \frac{a}{2}\ln [ 1-\beta j(\bar{q } % -q_1 ) ] \\&-q_1)]+\frac{a}{2x}\ln \frac{1-\beta j[\bar{q}-q_1 + x(q_1-q_0)]}{1-\beta j(\bar{q}-q_1 ) } \\ & -\beta\mu-\frac{1}{x}\int dz\langle\langle\ln \int dv\left(2k(z , v|\xi)\right)^x\rangle\rangle_{\xi } \label{eq1 } \end{split}\ ] ] where @xmath84}\cosh\bar{h}(z , v,\xi ) \label{kzv}\ ] ] and @xmath85 . \label{eq20}\ ] ] the set of equations for the order parameters @xmath86 , @xmath87 , @xmath88 , @xmath89 and the block size parameter @xmath90 can be found from eq . ( [ eq1 ] ) using the saddle point conditions . in particular , the elements of matrix r are given by @xmath91\}^2 } \label{r01}\ ] ] @xmath92\{1-\beta j[\bar{q}-q_1+x(q_1-q_0)]\}}\ ] ] @xmath93}. \label{rbarra}\ ] ] the average over @xmath61 in the grand canonical potential can be done using the parity properties of the functions dependent on @xmath94 and @xmath95 . the entropy can also be obtained directly from the grand canonical potential . the numerical solutions for the coupled set for the saddle point order parameters @xmath88 , @xmath87 , @xmath89 and @xmath86 are displayed in phase diagrams @xmath103 _ versus _ @xmath104 given below for several values of the parameter @xmath34 , where @xmath15 is the temperature and @xmath105 is defined in eq . ( [ hebb ] ) . for the numerical results , @xmath106 is used . the rs scheme for the order parameters is also calculated when @xmath107 . the stability of such solution is investigated by calculating the almeida - thouless eigenvalue @xmath108 given in eqs . ( [ linhaat1])-([linhaat2 ] ) . such analysis shows that the @xmath2 solutions and mattis states ( @xmath109 ) present @xmath110 while the @xmath8 ones have @xmath111 . this result could indicate that it is necessary to use rsb schemes to locate more adequately the @xmath14 first order boundary phase . however , the results shown in figs . ( [ fig1])-([fig2 ] ) have found that , in such first order boundary phase , the reentrance and , therefore , the inverse freezing are not essentially affected by the use of rs or 1s - rsb schemes similar to the fisg model @xcite . in fig . ( [ fig1 ] ) , it is presented phase diagrams which illustrate two quite distinct situations concerning the degree of frustration . the first one ( @xmath29 ) , which is shown in fig . ( [ fig1]-a ) , corresponds to trivial randomness @xcite , which means that there are no effects of frustration . in this phase diagram , for low @xmath103 and small @xmath104 , one gets @xmath109 as solution . for higher @xmath103 and/or larger @xmath104 , it is found the @xmath2 solution . for small @xmath104 , the boundary phase transition between @xmath109 and @xmath2 phase ( called here @xmath112 ) is second order . however , when @xmath104 increases , it appears a tricritical point located at @xmath113 and @xmath114 ( see appendix b ) . most important , there is no reentrance in the subsequent first order part of the boundary phase transition @xmath115 . in fig . ( [ fig1]-b ) , the degree of frustration is increased ( @xmath116 ) and , for that case , effects of frustration become dominant . thus , the @xmath109 solution is replaced by a @xmath8 one in which , @xmath117 , @xmath118 , @xmath119 and @xmath120 , while the @xmath2 phase has @xmath117 , @xmath118 , @xmath121 and @xmath122 . the freezing temperature has also a second order part @xmath123 . as for the trivial randomness case , a tricritical point appears , with value @xmath124 and @xmath125 ( see appendix b ) . nevertheless , the first order part of the freezing temperature @xmath126 now shows a reentrance which indicates the existence of inverse freezing , as it can be seen in the insert in fig . ( [ fig1]-b ) , which displays the entropy as function of @xmath103 . there , it is shown that entropy of the @xmath8 phase is larger than the @xmath2 one for that particular degree of frustration . it should be remarked that this phase diagram is quite similar to that one found for the fisg model @xcite . the location of the boundaries phases in fig . ( [ fig1 ] ) has been checked in the limits @xmath127 and @xmath128 as it can be seen in appendices b and c. there is a more complex scenario as compared with that one described in fig . ( [ fig1]-b ) when @xmath129 . in fig . ( [ fig2]-a ) ( @xmath130 ) , the solution for the order parameters shows a phase diagram which illustrates this new scenario . for a small @xmath104 , when @xmath103 is decreased , there is a second order phase transition between the @xmath2 and @xmath8 phases . however , for even lower temperatures , there is another phase transition , which is now a first order one , between the @xmath8 phase and @xmath109 region , which is given now by @xmath131 , @xmath118 , @xmath119 and @xmath120 . for this particular value of @xmath34 , @xmath8 and @xmath109 solutions occupy approximately equal sizes in the phase diagram . the freezing temperature has a similar behavior to that one found in fig . ( [ fig1]-b ) . it has a second order part @xmath123 for small @xmath9 , then it appears a tricritical point at @xmath132 and @xmath133 ( see appendix b ) . below this point , @xmath126 presents a reentrance which allows , for an adequate constant @xmath9 , crossing from the @xmath8 phase to the @xmath2 one when the temperature is decreasing . nonetheless , this first order boundary phase has a complex nature . it appears a triple point at ( @xmath134,@xmath135 ) where @xmath109 , @xmath2 and @xmath8 phases coexist . below this point , the first order boundary phase @xmath115 diplays no reentrance , as in the case @xmath29 . for smaller values of @xmath34 , as shown in figs . ( [ fig2]-b)-([fig2]-d ) , the region where @xmath109 solutions are found becomes increasingly larger than the @xmath8 one which is consistent with earlier results found in the classical hopfield spin glass model which displays , in a phase diagram @xmath15 _ versus _ @xmath34 , a dominance of the mattis states when @xmath136 0@xcite . even so , in this new situation , the location @xmath123 is not affected so much . however , the tricritical point is displaced for smaller and lower values of @xmath104 and @xmath103 , respectively . in comparison , the triple point is displaced for smaller values of @xmath104 and higher values of @xmath103 . as a consequence , @xmath126 and @xmath115 appear in a decreasing and increasing range of temperature , respectively . nevertheless , most important , the reentrance in @xmath137 is gradually suppressed when @xmath129 . however , the insert in fig ( [ fig2]-d ) shows that even when the level of frustration is very weak , and @xmath123 appears in a very short range of temperature , a reentrance in such first order boundary phase is still preserved . _ versus _ @xmath104 for @xmath138 and two values of @xmath116 . the full , dotted and dashed - dotted lines represent the @xmath139 of sg , pm and fe solutions , respectively . , width=327 ] @xmath103 for several values of @xmath34 and @xmath104 . the labels @xmath115 and @xmath126 indicate the first order transition between the @xmath140 and @xmath14 phases , respectively . the full vertical line indicates the second order transition between @xmath14 phases . , width=325 ] the procedure to locate the first order boundary lines in the previous phase diagrams is illustrated in fig . ( 3 ) , where the grand canonical potential _ versus _ @xmath104 is plotted for @xmath138 . in fig . ( 3 ) , the set of multiple @xmath8 solutions is represented by branches of full lines , where the chosen @xmath8 solution is that one which meets continuously with the only one @xmath8 solution available for small values of @xmath104 . from that @xmath8 solution , the first order boundary phase is obtained by equating the grand canonical potential of sg , pm and fe solutions . the corresponding behavior of the entropy as a function of @xmath103 for the values of @xmath34 used in fig . ( [ fig2 ] ) is shown in fig . ( [ fig4 ] ) . this figure illustrates the gradual suppression of inverse freezing when @xmath129 . for instance , in fig . ( [ fig4]-a ) , the value of @xmath141 is chosen to cross the reentrant first order boundary phase in @xmath126 in fig . ( [ fig2]-a ) . the figure shows that the entropy of @xmath8 phase is larger than the entropy of the @xmath2 one in the first order transition which , as in the inset of fig . ( [ fig1]-b ) , indicates the existence of inverse freezing . the same procedure has been adopted in figs . ( [ fig2]-b)-([fig2]-d ) , values of @xmath9 are adjusted to be close to @xmath135 . the result found in fig . ( [ fig2]-b ) is similar to that one found in fig.([fig2]-a ) . however , fig . ( [ fig2]-c ) displays the entropy behavior when @xmath126 and @xmath115 are now crossed . the first crossing is in the reentrant line transition @xmath126 giving an inverse freezing as in figs . ( [ fig2]-a)-([fig2]-b ) . the second one in @xmath115 gives a usual phase transition in which the @xmath2 phase is more entropic than the @xmath109 one , as already found in fig . ( [ fig1]-a ) . in fig . ( [ fig2]-d ) the reentrance in @xmath126 is almost suppressed . thus , for the chosen value of @xmath9 , there is only one crossing in @xmath115 which gives the entropy behavior of a usual phase transition as that one found in fig . ( [ fig1]-a ) . in the present work , it has been studied the hfisg model in which the random spin - spin coupling @xmath11 , instead of the usual gaussian distribution , is given by eq . ( [ hebb ] ) . for this particular choice of @xmath11 , the problem can be treated in a mean field framework which allows adjusting the level of frustration . therefore , it is possible to study not only the existence of a natural inverse freezing in the limits of trivial randomness and strong frustration but also how changing the level of frustration could affect such transition . in this approach , within 1s - rsb scheme , besides the replica non - diagonal @xmath8 orders parameters @xmath87 , @xmath88 and the block size parameter @xmath90 , there is also the replica diagonal order parameter @xmath89 . the set of order parameters is completed with @xmath86 which corresponds to the presence of mattis states ( @xmath109 ) @xcite . the coupled equations for @xmath87 , @xmath88 , @xmath90 , @xmath89 and @xmath86 are solved for several values of degree of frustration @xmath34 in a phase diagram @xmath15 _ versus _ chemical potential @xmath9 . in particular , the rs solution is also obtained when @xmath107 . the comparison among the several scenarios displayed in figs . ( [ fig1])-([fig2 ] ) elucidates the role of frustration as the essential ingredient responsible for producing naturally inverse freezing in the hfisg model . in other words , an inverse freezing without any need of an artificial entropic advantage . for instance , the comparison between results given in figs . ( [ fig1]-a ) and ( [ fig1]-b ) shows that only in the case @xmath116 ( the strong frustrated regime ) a reentrant first order boundary phase appears . the trivial randomness is not able to generate naturally a reentrance and , therefore , an inverse transition . furthermore , in fig . ( [ fig2 ] ) , when the level of frustration is diminished , the range in temperature of the @xmath14 first order boundary phase is decreased but , most importantly , the reentrance in this first order boundary phase is gradually suppressed , what implies in the gradual suppression of inverse freezing until the complete disappearing in the trivial randomness case ( @xmath29 ) . nevertheless , whatever the level of frustration , there is always a reentrance in @xmath14 first order boundary phase which gives an inverse freezing . in contrast , the range of the @xmath140 first order boundary phase is simply increased , nothing else happens . these previous features lead to the conclusion that frustration in any level is the necessary condition to create naturally the entropic advantage of the @xmath8 phase as compared with the @xmath2 one which generates an inverse freezing . albeit , these results are restrict to a particular model , we suggest that the role of frustration as an essential ingredient to rise naturally inverse freezing could be more general . one last remark must be done . the present approach could be used directly in gs model with the coupling @xmath11 as given in eq . ( [ hebb ] ) replacing the original gaussian distributed one @xcite . however , the great advantage of the hfisg model is that it would also allow studying the present problem in the presence of a transverse magnetic field @xmath13 @xcite . in ref . , it has been shown that @xmath13 tends to destroy the inverse freezing . therefore , the presence of an additional @xmath13 would lead to another question : what would happen with the inverse freezing in the hfisg model when the level of frustration and the strength of quantum effects are simultaneously changed ? in that case , it would possible to investigate the robustness of the inverse freezing by adjusting simultaneously the degree of frustration and @xmath13 . this question is currently under investigation . sgm acknowledges the hospitality of the departamento de ciencias de la tierra y la materia condensada - universidad de cantabria where this work was concluded and the support of fundacion carlina / spain . this work was partially supported by the brazilian agencies capes ( coordenao de aperfeioamento de pessoal de nivel superior ) and cnpq ( conselho nacional de pesquisas cientficas ) . in this appendix , the averaging procedure of the partition function of the hfisg model is introduced following closely ref . . the first term in the action @xmath63 ( see eq . ( [ asg ] ) ) can be linearized by a hubbard - stratonovich transformation by introducing @xmath142 auxiliary fields @xmath143 which are splitted in two subsets with @xmath144 and @xmath145 terms . therefore , @xmath146 \\ & \times \int_{-\infty}^{\infty } d m_{\nu}^{\alpha } \exp\left\lbrace \tau \sum_{\nu=1}^{l}\sum_{\alpha}\left [ -\frac{1}{2}(m_{\nu}^{\alpha})^2 + \eta_{i,\alpha}^{\nu } \right ] \right\rbrace \\ & \times \int_{-\infty}^{\infty } d m_{\varrho}^{\alpha } \exp\left\lbrace \tau \sum_{\varrho = l+1}^{p}\sum_{\alpha}\left [ -\frac{1}{2}(m_{\varrho}^{\alpha})^2 + \eta_{i,\alpha}^{\varrho } \right ] \right\rbrace , \label{mu } \end{split}\ ] ] where @xmath147 , @xmath148 and @xmath149 . it is assumed that the relevant contributions come from @xmath150 which are order unity , while @xmath143 is of order @xmath151 . therefore , the average over the @xmath152 independent random variables @xmath153 can be done using @xmath154 given in eq . ( [ gaussian ] ) which results in : @xmath155 \rangle\rangle_{\xi}= \nonumber \\ \exp\left [ \sum_{i}\sum_{\varrho = s}^{p } \ln\left ( \cosh ( \beta j \sum_{\alpha}s_{i}^{\alpha}m_{\varrho}^{\alpha})\right ) \right ] . \label{mu1}\end{aligned}\ ] ] the argument of the exponential in the right hand side of eq . ( [ mu1 ] ) can be expanded up to second order in @xmath143 . the result is a quadratic term of the spins variables @xmath156 in the last exponential of eq . ( [ mu ] ) . this term can be linearized by introducing the spin glass order parameter @xmath157 using the integral representation of the delta function as @xmath158\\ & = \delta(q_{\alpha\beta}-\frac{1}{n}\sum_{i}s_{i}^{\alpha}s_{i}^{\beta } ) . \label{mu2 } \end{split}\ ] ] therefore , the exponential involving @xmath143 in eq . ( [ mu ] ) can be written as : @xmath159\right\}\\&= \int_{-\infty}^{\infty } \prod_{\alpha\beta } \frac{dq_{\alpha\beta } d\tilde{r}_{\alpha\beta}}{2\pi } \exp \left [ \frac{\beta}{2 } \sum_{\varrho = l+1}^{p } m_{\varrho}^{\alpha}\lambda_{\alpha\beta } m_{\varrho}^{\beta } d_{\alpha\beta } \right ] , \label{e20 } \end{split}\ ] ] with @xmath160 where the matrix element @xmath161 introducing eqs . ( [ e20])-([e21 ] ) into eq . ( [ mu ] ) , the @xmath143 fields can be integrated to give : @xmath162 \\&\times \langle\langle \int^{+\infty}_{-\infty } dm_{\nu}^{\alpha } \exp\left\{\tau \sum_{\nu=1}^{l}\sum_{\alpha } \left [ -\frac{1}{2}(m_{\nu}^{\alpha})^{2}+ \eta_{i,\alpha}^{\nu } \right]\right\ } \rangle\rangle_{\xi } \\ & \times \int^{\infty}_{-\infty}\prod_{\alpha\beta } \frac{dq_{\alpha\beta}{\tilde{r}}_{\alpha\beta}}{2\pi } \exp\ { d_{\alpha\beta}- \frac{1}{2}(p - l)tr\ln\underline{\underline{\lambda } } ) \}. \label{e22 } \end{split}\ ] ] assuming @xmath163 in eq . ( [ e22 ] ) , the averaged partition function is given as @xmath164 \\ & \times \exp \left [ -\frac{p-1}{2 } tr\ln\underline{\underline{\lambda } } \right ] \langle\langle \theta(\tilde{r}_{\alpha\beta},\tilde{r}_{\alpha\alpha},m^{\alpha}_{1})\rangle\rangle_{\xi } \end{split } \label{med1}\ ] ] where the matrix @xmath165 is defined in eq . ( [ e21 ] ) and @xmath166 \label{e25 } \end{split}\ ] ] with @xmath167 and @xmath168 . the trace of the matrix @xmath165 given in eq . ( [ e21 ] ) is obtained in terms of its eigenvalues . the grand canonical potential is found introducing eq . ( [ med1 ] ) in eq . ( [ replica ] ) which is evaluated at the saddle point . thus , @xmath169 and @xmath170 this appendix presents a procedure to obtain the tricritical point @xmath171 for both cases : @xmath29 and @xmath172 . for @xmath29 , the landau expansion of the grand canonical potential , eq . ( [ eq1 ] ) , in powers of @xmath86 gives @xmath173 with @xmath174/2,\ ] ] @xmath175 sech [ ( \beta \mu)/2]^4/96.\ ] ] the tricritical point is obtained when @xmath176 . the position of @xmath177 in the fig . ( [ fig1]-a ) for @xmath127 can be also checked from the equation @xmath178 . for @xmath179 , the @xmath180 phase transition is investigated . in this case , it is assumed that there is no essential difference between rs and 1s - rsb schemes concerning the location of the tricritical point.@xcite therefore , we start with the thermodynamic potential within the rs solution ( @xmath181 , @xmath182 and @xmath183 ) written explicitly as a function of the order parameters @xmath16 and @xmath184 , and with @xmath117 : @xmath185 - \int dz \\ \times \ln [ \cosh \beta\mu + \mbox{e}^{[\frac{\beta^2 j^2 a}{2}(\bar{r}-r)-\frac{\beta j a}{2 } ] } \cosh\beta j \sqrt{ar}z ] \label{aa1 } \end{split}\ ] ] where @xmath16 and @xmath184 are given by eqs . ( [ r01 ] ) and ( [ rbarra ] ) within the rs solution , respectively . equation ( [ aa1 ] ) is expressed as an expansion in powers of @xmath16 , which is related to the @xmath8 order parameter . therefore , @xmath186 where @xmath187 is obtained by a saddle point solution of @xmath139 . in this case , @xmath187 can also be written in the form of a series @xmath188 with @xmath189 as a result , @xmath190^{-1 } , \\ \bar{r}_2=-\frac{2+\beta^6 j^6 a^2 \bar{r}_0 ^ 3 x_0 ^ 2(1-x_0)}{\bar{r}_0[2-\beta^4 j^4 a\bar{r}_0 ^ 2 x_0(1-x_0)]},\end{aligned}\ ] ] and @xmath191 now , eq . ( [ rbr ] ) is introduced into the coefficients of expression ( [ omegae ] ) , which are expanded in powers of @xmath16 again . the resulting expression is then expressed in powers of the @xmath8 order parameter @xmath24 by expanding @xmath16 : @xmath192 with @xmath193 @xmath194 -6\beta j ( 1-\beta j \bar{q}_0)f_3 \end{split}\ ] ] where @xmath195 $ ] . the @xmath180 second order phase transition occurs when @xmath196 and @xmath197 with the tricritical point located when @xmath196 and @xmath198 . in particular , one can use the condition @xmath196 with eqs . ( b7 ) and ( b9 ) to obtain the critical temperature @xmath199 by solving @xmath200}\ ] ] where @xmath201 . for @xmath127 , it is recovered the result @xmath202 for @xmath116 here a procedure is presented to obtain the first order boundary phases of @xmath140 phases @xmath203 and @xmath14 phases @xmath204 at @xmath128 within the rs and 1s - rsb solutions . for @xmath29 , the grand canonical potential of fm solution at @xmath128 is @xmath205 . by comparing @xmath206 with the grand potential potential of pm phase , @xmath207 , the first order boundary is located at @xmath208 . for the cases shown in fig . ( [ fig2 ] ) , it is recovered the grand canonical potential for @xmath29 . for @xmath28 , the effects of frustration are dominants . in this case for rs solution , @xmath209 for @xmath210 . therefore , close to the transition , which means @xmath211}$ ] , the @xmath212 at @xmath213 is @xmath214 \label{omegax}\ ] ] where @xmath215 ( @xmath216 is the error function ) , @xmath217 and @xmath218 . from the condition @xmath219 , @xmath220 and @xmath16 can be solved which allows finding @xmath221 . particularly , @xmath222=0.689 for @xmath116 . where @xmath227 \\ & + e^{\gamma \vartheta_{-}}\left [ 1 + \mbox{erf } \left ( \frac { \eta_{- } } { \sqrt{2 a ( r_{1}-r_{0 } ) } } \right ) \right ] \\ & + \mbox{erf}\left ( \frac { \bar{u}+ \sqrt{a r_{0}}z } { \sqrt{2 a ( r_{1}-r_{0 } ) } } \right ) + \mbox{erf}\left ( \frac { \bar{u } - \sqrt{a r_{0}}z } { \sqrt{2 a ( r_{1}-r_{0 } ) } } \right ) \end{split } \label{1rscompl}\ ] ] with @xmath228 , @xmath229 and @xmath230 . the previous results for rs solution are recovered from eqs . ( [ 1srsbsch])-([1rscompl ] ) when @xmath231 . again , equations for @xmath232 , @xmath233 , @xmath234 , @xmath235 , @xmath236 and @xmath237 can be obtained from @xmath212 and solved . in this case , @xmath222=0.682 for @xmath116 . 00 n. schupper and n. m. shnerb , phys . e * 72 * , 046107 ( 2005 ) ; n. schupper and n. m. shnerb , phys . lett . * 93 * , 037202 ( 2004 ) . a. crisanti and l. leuzzi , phys . lett . * 95 * , 087201 ( 2005 ) . m. sellitto , phys . b * 73 * , 180202(r ) ( 2006 ) . s. prestipino , phys . e * 75 * , 011107 ( 2007 ) . l. leuzzi , philos . mag . * 87 * , 543 ( 2007 ) . r. angelini , g. ruocco , s. de panfilis , phys . e * 78 * , 020502(r ) ( 2008 ) . s. g. magalhaes , c. v. morais , f. m. zimmer , phys . b * 77 * , 134422 ( 2008 ) . a. l. greer , nature * 404 * , 134 ( 2000 ) . o. portmann , a. vaterlaus , and d. pescia , nature ( london ) * 422 * , 701 ( 2003 ) . n. avraham et al . , nature ( london ) * 411 * , 451 ( 2001 ) . m. r. feeney , p. g. debenedetti , f. h stillinger , j. chem . * 119 * , 4582 ( 2003 ) . m. blume , phys . 141 * , 517 ( 1966 ) ; h. w. capel , physica ( amsterdan ) * 32 * , 966 ( 1966 ) . s. k. ghatak and d. sherrington , j. phys . c * 10 * , 3149 ( 1977 ) . e. fradkin , b. a. huberman , and s. h. shenker , phys . b * 18 * , 4789 ( 1978 ) . r. oppermann and a. muller - groeling , nucl . phys . b * 401 * , 507 ( 1993 ) . a. theumann , a. a. schmidt and s. g. magalhes , physica a * 311 * , 498 ( 2002 ) . w. wiethege , d. sherrington , j. phys . c : solid state phys . * 19 * , 6893 ( 1986 ) . i. p. castillo , d. sherrington , phys . b * 72 * , 104427 ( 2005 ) . h. feldmann and r. oppermann , j. phys . a * 33 * , 1325 ( 2000 ) . f. m. zimmer , s. g. magalhaes , phys . b * 74 * , 012202 ( 2006 ) . d. j. amit , _ modelling brain function . the world of attractor neural networks _ ( cambridge university press , cambridge , england , 1989 ) . d. j. amit , h. gutfreund , h. sompolinsky , phys rev . a * 32 * , 1007 ( 1985 ) . d. j. mattis , phys . lett . * 56a * , 421(1977 ) . j. p. provost and g. vallee , phys . * 50 * , 598 ( 1983 ) . d. sherrington and s. kirkpatrick , phys . rev lett * 35 * , 1792 ( 1975 ) ; s. kirkpatrick and d. sherrington , phys . b * 17 * , 4384 ( 1978 ) . a. j. bray and m. a. moore , j. phys . c * 13 * , l655 ( 1980 ) . g. parisi , j. phys . * 13 * , 1101 ( 1980 ) . alba theumann and m. vieira gusmo , phys . let . * 105a * , 311 ( 1984 ) . j. miller and d. a. huse , phys . rev . 70 , 3147 ( 1993 ) . f. a da costa , c. s. o. yokoi and s. r. a. salinas , j. phys . a * 27 * , 3365 ( 1994 ) . p. mottishaw , europhys . lett . * 1 * , 409 ( 1986 ) . j. w. negele , h. orland , _ quantum many - particle systems _ ( addison - wesley publishing company , eua , 1988 ) . f. m. zimmer , s. g. magalhes , physica a * 359 * , 380 ( 2006 ) . d. j. amit , h. gutfreund and h. sompolinsky , ann . phys . * 173 * , 30 ( 1987 ) .
the continuous electron beam accelerator facility ( cebaf ) at jefferson lab ( jlab ) delivers a highly polarized electron beam in the 1 11 gev energy range for currents up to @xmath2 200 @xmath1a . a key part of the jlab physics program consists of high - precision parity violating electron scattering experiments @xcite that require a fast and precise electron beam polarization measurement . conventional polarimetry techniques such as mller @xcite and mott @xcite are destructive for the beam properties due to solid targets they use and can not be operated simultaneously with the physics experiments . they can only be operated at low intensity ( @xmath3 10 @xmath1a ) and at low energy ( @xmath3 5 mev ) respectively ; therefore , the physics experiments have to assume that beam polarization remains constant when the intensity or the energy of the beam is several orders of magnitude higher . alternatively compton polarimetry has been used with great success at jlab @xcite and elsewhere @xcite to measure the electron beam polarization in a continuous and non - destructive manner . however , unlike at high - current ( @xmath4 20 ma ) storage rings @xcite and at high - energy ( @xmath4 50 gev ) colliders @xcite , the relatively low - current and low - energy nature of the cebaf electron beam has required the use of high - finesse optical cavities @xcite to enhance the laser power beyond commercially available lasers to provide sufficient electron - photon collision luminosity . compton polarimetry determines the electron beam polarization based on a well - known spin - dependent scattering cross - section @xcite between polarized electrons and photons . the longitudinal electron beam polarization @xmath5 is extracted from the experimental compton scattering asymmetry @xmath6 , according to , @xmath7 where @xmath8 is the theoretical compton scattering asymmetry that can be derived by combining the calculated spin - dependent compton cross section with the appropriate experimental response function @xcite , and @xmath9 is the circular photon beam polarization . the scattering asymmetry @xmath6 can be measured based on the detection of either the scattered photons or the scattered electrons and depends on the energies of the colliding electron and photon beams . for an electron beam of 1.0 gev , the maximum scattering asymmetry for infrared ( 1064 nm ) photons is 1.8% , whereas for green ( 532 nm ) photons , it is 3.5% . for an electron beam of 6 gev , these values are about 10% and 18% , respectively . therefore , compton polarimetry at low beam energies ( @xmath3 2 gev ) is difficult to perform . + external optical cavities have frequently been employed to increase the laser power from a light source through a process of coherent addition . recently optical cavities have been widely employed in many areas such as x-/@xmath10-ray generation via inverse - compton scattering @xcite , high harmonic generation @xcite and electron beam polarimetry @xcite . according to the best of our knowledge , there are only three facilities in the world have ever built a compton polarimeter with an optical cavity ; mainz microtron ( mami ) @xcite , hadron electron ring accelerator ( hera ) @xcite and jlab @xcite . in hall a of jlab , a compton back - scattering polarimeter had been operational since 1999 @xcite . its photon source had a typical circulating power of @xmath2 1.5 kw inside a monolithic fabry - perot cavity where the circularly polarized photons scattered from the longitudinally polarized electrons . it provided 1.0% statistical ( and 1.2% systematic ) precision with @xmath2 25 minutes data taking for 4.6 gev , 40 @xmath1a electrons @xcite . this polarimeter used a continuous - wave ( cw ) infrared ( ir , 1064 nm ) laser as its photon source and had serious limitations in providing a good signal - to - noise ratio for beam energies below 2 gev @xcite . under this circumstance , it is highly desirable to have a shorter - wavelength and higher - power light source since this will result in larger back - scattered electron and photon energies and a larger measured scattering asymmetry , and hence increased ability to control systematic uncertainties . + the lead radius experiment ( prex ) @xcite at jlab aims to provide the first model - independent evidence of the existence of a significant neutron skin in @xmath11pb . prex proposed to obtain a statistical and systematic precisions of 3.0% and 2.0% respectively on the parity - violating electroweak asymmetry of polarized electrons from @xmath11pb to determine the neutron radius of @xmath11pb at 1.0% accuracy . the uncertainty goal for beam polarimetry at 1.06 gev was 1.0% . in order to meet the high - precision requirement , the existing compton polarimeter underwent an upgrade @xcite in 2010 . the upgrade includes a new photon detector with an upgraded integrating data acquisition ( daq ) system , a new electron detector , and a new compton photon source that includes a green laser ( 532 nm ) with a new high - finesse fabry - perot cavity @xcite . + this paper will describe the upgraded experimental setup in hall a at jlab , with an emphasis on the frequency - doubled green laser system and optical cavity . a unique technique to extract the laser beam polarization inside a resonant fabry - perot cavity will be given and finally , the first electron beam polarization measurement results from the prex experiment @xcite at a beam energy and current of 1.06 gev and 50 @xmath1a will also be presented . with an enhancement factor and intra - cavity power of 3,800 and 3.7 kw , we report the highest power compton polarimetry facility operating at 532 nm as compared to existing facilities such as mami @xcite and hall c at jlab @xcite . , d@xmath12 , d@xmath0 and d@xmath13 ) that deflect the primary electron beam and send it through the fabry - perot cavity . the scattered compton electrons and photons will be detected by electron and photon detectors , respectively.,width=294 ] the upgraded compton polarimeter in hall a consists of a magnetic chicane , a photon source that is composed of a laser system and fabry - perot cavity , a photon detector , and an electron detector as shown in fig . [ fig : chicane ] . the electron beam enters from the left . when the four identical dipoles of the chicane , referred to as d@xmath14 , d@xmath12 , d@xmath0 and d@xmath13 , are energized , the beam is deflected vertically and crosses the photon beam enhanced in the fabry - perot cavity that is located at the center of the chicane . the fabry - perot cavity is enclosed in a vacuum chamber that is connected to electron beam pipes from both ends and sits on an optical table where the laser and other optical elements are located . the compton interaction point ( cip ) is at the center of the beam crossing in the cavity . the crossing angle between the electron and photon beams is 24 mrad . the electrons undergo compton scattering with circularly polarized photons resonating in the fabry - perot cavity fed by a cw green laser ( 532 nm ) . the backscattered photons are detected in the photon detector while the scattered electrons can be detected in the electron detector located a few mm above the primary beam in front of d@xmath13 . approximately one electron in every one billion undergoes compton scattering . unscattered electrons , separated from the scattered particles by d@xmath0 in the chicane , continue on into the hall a target for the primary experiment . independent electron and photon analyses give complementary determinations of the electron beam polarization . a scale 3d view of the compton polarimeter in the hall a tunnel at jlab is shown in fig . [ fig : chicane3d ] . + the photon detector is located under dipole d@xmath13 and consist of a single crystal of ce - doped gd@xmath12sio@xmath15 ( gso ) and a single pmt ( photomultiplier tube ) . it is mounted on a motorized table with remote - controllable motion along both axes ( horizontal and vertical ) transverse to the electron beam direction . the scattered photons from the cip go through a thin vacuum window and lead disk in the vacuum pipe before they hit the detector window . the photon detector daq uses a customized 12-bit flash analog - to - digital converter ( fadc ) that integrates the scintillation photon signals from the pmt and samples the data at rates of 200 mw @xcite . + the electron detector consists of four silicon microstrip planes spaced horizontally 1 cm from each other with a 200 micron vertical rising offset between planes . each plane consists of 192 strips of silicon with a width of 240 microns . the planes are inclined at an angle of 58 mrad from the vertical position . the detector is mounted on a translator with a remote - controllable stepper motor and can travel vertically up to 120 mm from the main beam . the optical elements of the polarimeter are mounted on a vibration - damped optical table between dipoles d@xmath12 and d@xmath0 ( fig . [ fig : chicane3d ] ) . a schematic of the optical system of the polarimeter is illustrated in fig . [ fig : opticalsetup ] . based on their main functionality , we can categorize the optical elements into four groups necessary for achieving the required power and circular polarization in the cavity . the first group is the laser source that provides a green beam at the wavelength of 532 nm . the second group consists of the elements that transport , align and focus the incident beam for coupling to the optical cavity . the third group includes elements for controlling and measuring the polarization . the last group consists of elements which allow the use of the reflected beam from the cavity in the electronic feedback system to achieve frequency locking of the laser to the cavity . as shown in fig . [ fig : opticalsetup ] , the seed laser is a narrow - linewidth ( @xmath3 5.0 khz ) diode pumped nd : yag nonplanar ring oscillator ( npro ) laser ( lightwave 126 ; jdsu ) that delivers a linearly polarized cw ir ( 1064 nm ) beam of up to 250 mw . the output of this laser has been fiber - coupled to a ytterbium - doped fiber amplifier ( ydfa , yar-10k-1064-lp - sf ; ipg photonics ) through a single - mode polarization - maintaining ( pm ) fiber . the ydfa takes @xmath2 10 mw of seed laser and provides a linearly polarized ( extinction ratio : 20 db ) output with a maximum power of 10 w at 1064 nm in cw mode . the frequency doubling crystal is bulk periodically poled mgo : linbo@xmath0 ( ppln , hc photonics ) doped with 5% mgo in order to minimize photo - refractive damage . the crystal is 0.5 mm thick , 3 mm wide and 50 mm long , and the quasi - phase matching ( qpm ) period is 6.92 @xmath1 m with a 50% duty cycle . the input and output surfaces are antireflection ( ar ) coated for 1064 nm and 532 nm , respectively . the crystal is placed in an externally controlled , temperature - stabilized oven developed in - house . the oven is mounted on a stage composed of a four - axis tilt aligner ( 9071-v ; newport ) in order to establish a precise alignment for phase matching . a temperature controller ( tecsource 5305 ; arroyo instruments ) with a nominal resolution of 0.01 @xmath16c provides the required phase - matching temperature for second harmonic generation ( shg ) . the crystal optical axis is matched with the direction of the output polarization ( vertical ) of the ydfa . a pair of 0.5 inch lenses ( l@xmath17 and l@xmath18 in fig . [ fig : opticalsetup ] ) with focal lengths of 13.8 mm and 15 mm ( with a total effective focal length of @xmath2 120 mm ) are then used to focus the beam waist diameter ( @xmath2 80 @xmath1 m ) into the center of the crystal . this makes the rayleigh length of the pump laser focus to be around 5 mm and ensures the beam size at the crystal entry and exit faces to be around 400 @xmath1 m . this critical focusing also ensures the incoming beam is not clipped by the tiny crystal thickness while keeping the shg efficiency close to the maximum . the generated green beam is separated from the residual ir beam after the crystal via a pair of 1.0 inch dichroic mirrors ( @xmath19 99% at 532 nm , @xmath20 95% at 1064 nm ) noted as dc@xmath14 and dc@xmath12 ( altos photonics ) . the lenses , oven and dichroic mirrors are all mounted on a separate linear translation stage seated on a rail , and the whole system is contained in an enclosure box . + m and ir pump power of 5 w.,width=283 ] with the experimental setup as described above , we have measured several properties , such as crystal temperature tuning curve , beam quality and power stability , of the second harmonic beam . the measured temperature tuning curve at 5 w ir pump power is shown in fig . [ fig : tempscanfit ] . the red circles show the experimental data while the blue line represents the theoretical fit for the first - order qpm interaction predicted by the sellmeier equation for ppln crystal @xcite . the experimental results give the full width at half - maximum ( fwhm ) phase - matching temperature bandwidth @xmath21 = 0.6 @xmath16c at the phase - matching temperature 64.0 @xmath16c . + ) is used to extract the normalized shg conversion efficiency.,width=291 ] fig . [ fig : ppln_conveff ] shows the average green power ( solid circles ) output from ppln and corresponding phase - matching temperature ( open squares ) at different ir pump powers from the fiber amplifier . it is important to note that the value of the phase - matching temperature changes slightly when the ir pump power is changed . the phase - matching temperature ranges between 64.0 @xmath16c and 64.3 @xmath16c for ir pump power between 200 mw and 5 w. at each power level the crystal is maintained at the peak phase - matching temperature to ensure maximum green power . for a loosely focused gaussian beam , the normalized conversion efficiency @xmath22 for first - order qpm is defined as @xcite , @xmath23 @xmath24 where @xmath25 and @xmath26 are the fundamental and shg powers , @xmath27 is the frequency of the fundamental beam , @xmath28 is the nonlinear coefficient of ppln , @xmath29 is the permittivity of free space , and @xmath30 and @xmath31 are the refractive indexes at these wavelengths . in fig . [ fig : ppln_conveff ] , as the theoretical fit ( green line ) defined by eq . ( [ eq : coneff ] ) shows , the green power varies quadratically with the ir pump power with a normalized conversion efficiency @xmath22 of 1.37%w@xmath32cm@xmath32 . it is necessary to point out that all powers are direct measurements after dc@xmath12 by an optical power meter ( pm140 ; thorlabs ) without correcting for the residual reflection at the crystal face or losses in the dichroic mirrors . the experimental results show that there is no sign of saturation in the shg power at 5 w ir pump power . this is mostly due to a larger beam waist size at the crystal center so that there is no pump depletion . therefore , the achieved conversion efficiency is somewhat lower than the maximum conversion efficiency of 2.62%w@xmath32cm@xmath32 provided by the vendor . all of these considerations suggest that thermal lens effects are negligible . the maximum achieved green power at 5 w ir pump power is 1.74 w with a total conversion efficiency of 34.8% . a primary goal of this work is to develop a stable and good - quality green beam at the watt level , so we did not pursue higher conversion efficiencies . + ) and vertical ( @xmath34 ) directions and continuous lines show the theoretical fits to extract the @xmath35 @xcite values in both directions.,width=294 ] the quality of the green beam is of major importance in power enhancement with an optical cavity , as any deviation from a perfect gaussian profile will lead to a rejection of power by the high - finesse cavity and will not contribute to the power buildup inside the fabry - perot cavity . in order to check the beam quality , we focused the green beam after the ppln into a 200 @xmath1 m waist diameter with an additional lens and measured the beam size at different axial positions . [ fig : ppln_m2 ] shows the green beam profile at 1.74 w monitored by a laser beam profiler ( lba - fw - scor20 ; spiricon ) along with the theoretical fit . closed and open circles are the measured beam diameters in the horizontal ( @xmath36 ) and vertical ( @xmath34 ) directions . the result of the theoretical fits ( solid lines ) to the experimental data determines an @xmath35 @xcite value of @xmath2 1.1 in both dimensions confirming the tem@xmath37 spatial mode in the shg process . however , there is a small astigmatism mainly arising from the difference in refractive indexes along the horizontal and vertical axes of the ppln crystal . as shown in fig . [ fig : powerstability ] , the peak - to - peak stability of green power at 1.74 w is 0.8% for the entire period of 12 hours , while the fundamental ir power stability from the ydfa is measured to be better than 0.6% over 4 hours . + the frequency - doubled green beam based on single - pass shg in a ppln crystal pumped by a ydfa has been used with the high - finesse fabry - perot cavity as a photon source for the compton polarimeter during the three - month run period of the prex experiment @xcite in 2010 . in the following we describe how we establish a frequency locking of this beam to a resonant frequency of the high - finesse fabry - perot cavity . the green beam is propagated through a series of optical elements ( fig . [ fig : opticalsetup ] ) to properly match the fundamental cavity mode ( tem@xmath37 ) at the injection point . after exiting the ppln crystal and reflecting from two dichroic mirrors ( dc@xmath38 and dc@xmath39 ) , the green beam is focused by a lens l@xmath40 ( @xmath41 = 75 mm ) at 6.0 cm from the frequency doubling setup with waist diameters 370 @xmath1 m and 450 @xmath1 m in the horizontal and vertical directions . the focusing of the incident beam at the cip is accomplished by three lenses denoted l@xmath14 , l@xmath12 and l@xmath0 , respectively . a diverging lens l@xmath38 ( @xmath42 = -1.0 m ) at 405 mm from the frequency - doubling setup nearly collimates the beam after it passes through an optical isolator ( isolation : 40 db , io-3 - 532-hp ; thorlabs ) and a half - wave plate ( hwp ) . in order to achieve the correct beam waist size at the cavity center , we have two more lenses , l@xmath39 and l@xmath43 ( @xmath44 = -50 mm , @xmath45 = 200 mm ) , installed at 820 mm and 1.0 m from the frequency - doubling setup . here in our setup , l@xmath38 and l@xmath39 are on fixed mounts and l@xmath43 is mounted on a remote controlled translation stage ( m-605.1dd ; physik instrumente ) . this allows fine tuning and precise alignment of the laser beam to the cavity by using another pair of remotely controlled mirrors ( m@xmath38 and m@xmath39 ) that allow four degrees of motion ( 2 translations , 2 rotations ) for the laser beam with respect to the optical axis of the cavity . both the proper focusing and precise alignment of the impinging laser beam to the cavity are needed in order to achieve good mode matching . the mirrors m@xmath46 , m@xmath47 , m@xmath48 and m@xmath49 are fixed at 45@xmath16 with respect to the incident beam . a camera ( ccd@xmath14 ) facing the mirror m@xmath47 monitors the positions of the incident and reflected beams from the cavity . another camera ( ccd@xmath12 ) at the cavity exit is used for monitoring the profile of the transmitted beam from the cavity . + as shown in fig . [ fig : opticalsetup ] , our cavity consists of two ( m@xmath50 , m@xmath51 ) identical highly reflective ( r = 99.975% ) concave mirrors ( advanced thin films ) separated by 850 mm . the mirrors have a typical transmittance of 240 ppm ( parts per million ) and total nominal ( scattering and absorption ) loss of @xmath3 10 ppm . this leads to a maximum power enhancement of @xmath52 = 4,000 . the coating is made of alternating dielectric quarter - wave layers of sio@xmath39 and ta@xmath39o@xmath53 on a super polished ( surface rms roughness = 0.5 @xmath54 ) fused silica substrate ( thickness 4 mm , diameter 7.75 mm ) by an ion beam sputtering technique . the mirrors have a typical radius of curvature of 0.5 m that requires a gaussian beam waist diameter of 348 @xmath1 m at the center of the cavity @xcite . this geometry was chosen to maximize the electron - photon collision luminosity when the laser light is frequency locked to the cavity and crosses with the electron beam . + a careful study of beam propagation from the frequency doubling setup to the cavity through each optical element has been performed using optocad @xcite . [ fig : beamtransport ] shows the calculated beam size versus the distance along the beam path from the frequency doubling setup . experimental verification of the laser waist size at the center of the cavity is not possible with the beam profiler . therefore , we created an auxiliary optical path ( 2.375 m ) that has the same path length from the doubling setup to the cavity with all the optical elements in place . instead of m@xmath50 , we used an uncoated mirror substrate made of the same fused silica material with the same radius of curvature . we experimentally checked the beam waist size and location after this mirror substrate , and compared it to our calculation . a small correction with l@xmath43 ( few mm ) is necessary in order to get an average beam size of 350 @xmath1 m at a distance of 425 mm ( corresponding to the cavity center ) from this mirror substrate . at the exit of the cavity , there is a set of optical elements composed of a quarter - wave plate ( qwp@xmath39 ) , a holographic beam sampler ( hbs ) and a wollaston prism ( fig . [ fig : opticalsetup ] and fig . [ fig : opticstable ] ) that are used for measuring and monitoring the laser polarization ( discussed in section [ sec : laserpolarization ] ) . the strongest requirements on the design of a fabry - perot cavity in a compton polarimeter are that it must be rigid and must have a small crossing angle with the electron beam . the small crossing angle increases the electron - photon interaction luminosity . the geometry of the cavity is determined by the total distance between the two mirrors ( 850 mm ) , the radius of curvature of the mirrors ( 500 mm ) and the crossing angle between the laser and electron beams . with this design the crossing angle is 24 mrad , the gap between the electron beam and mirror edge is 6.12 mm and the laser beam size on the cavity mirror is 898 @xmath1 m . + a mechanical drawing of the cavity is shown in fig . [ fig : opticstable ] . the cavity and all the other optical elements are mounted on an optical table with a footprint of 1.5 m @xmath55 1.2 m. this table is placed on a laminar flow damping system consisting of four pneumatic posts with auto - leveling valves to isolate the vibrations from the ground . the optical table is located inside a small room equipped with a laminar flow fan filter unit . in order to ensure thermal and mechanical stability , an invar ( 64feni ) frame consisting of three cylindrical rods attached to two vertical plates , each with an octagonal cutout , forms the backbone of the cavity frame . the cavity vacuum chamber is a cylindrical vessel made of stainless steel with a diameter of 4.5 inches ( fig . [ fig : opticstable ] ) connected to two octagonal gimbal mounts through two soft bellows . the bellows allow the gimbal mounts to tilt freely in both the horizontal and vertical planes when they are adjusted by four piezo actuators under atmosphere . however , in our experiment , the bellows were too stiff under vacuum for the piezo actuators to move the gimbals remotely . therefore the procedure involves locking the cavity under atmosphere , then locking the gimbals and drawing vacuum . the cavity mirror is mounted on a mirror holder attached to these gimbal mounts which are also made of invar . each gimbal mount is supported by four stainless steel cylindrical bearings ( c - flex bearing co. ) that form two axes ( horizontal and vertical ) for each gimbal mount and allow them to tilt freely around each axis . the mirror holder is machined such that the rotation axis of each mirror lies on the same axis as the gimbals . the bearings are 0.25 inch in diameter and 0.4 inch in length , and each can support a load up to 100 lbs . two remote controlled picomotors ( 8302 ; newfocus ) attached to each gimbal mount are used for aligning each cavity mirrors by tilting the gimbal mounts in both the vertical and transverse planes with respect to the laser beam propagation direction . a pair of counteracting spring plungers ( maximum load 13 lbs ) are attached to the gimbal mounts to keep the alignment in position . the picomotors are interfaced to the epics @xcite slow control system ( discussed in section [ sec : feedback ] ) that allows remote alignment of the cavity mirrors . + a pair of vacuum window substrates ( w@xmath56 and w@xmath57 in fig . [ fig : opticalsetup ] ) that are made of fused silica ( 3 mm thick and 0.7 inch in diameter ) allow the laser beam to enter and exit the cavity via a pair of 0.5 inch turning mirrors ( m@xmath48 and m@xmath49 in fig . [ fig : opticalsetup ] ) oriented at 45@xmath16 with respect to the incident laser beam . the vacuum windows are ar coated for 532 nm and welded to stainless steel flanges ( fig . [ fig : opticstable ] ) by the glass - metal soldering technique . two soft bellows on both sides of each gimbal mount are connected to both sides of the cavity vacuum chamber through a pair of flanges and allow the gimbals to tilt freely . each of the 45@xmath16 turning mirrors is mounted to an aluminum holder that is attached to a stainless steel flange mounted on a post . each aluminum holder has a slit 4 cm long and 1 cm wide that allows the electron beam to pass through . when the cavity is installed in the electron beam line ( fig . [ fig : cavityinstalledhall ] ) , the stainless steel flanges are connected to the beam pipe by another set of soft bellows that isolates any vibrations from the rest of the electron beam pipe . two beam position monitors located on both sides of the cavity are used to monitor changes in beam position of the electron beam during beam tuning and data taking . two ion pumps can provide a vacuum of @xmath2 10@xmath58 torr inside the cavity after several hours of bake - out at @xmath2 100 @xmath59c . maintaining the resonance and therefore the enhancement in the cavity requires feedback control of the laser frequency . the optical frequency stabilization must be better than the free spectral range ( @xmath60=176.5 mhz ) divided by the finesse ( @xmath61=12,000 ) , which is @xmath215 khz ( cavity bandwidth ) . this is equivalent to a stabilization of the 850 mm cavity length to better than the laser wavelength divided by the finesse , which is @xmath2 0.04 nm . the electronic feedback system that controls the laser frequency uses the pound - drever - hall technique @xcite . the error signal is created by phase detection between the photodiode ( pdr in fig . [ fig : cavcontrl ] ) output of the modulated reflection signal and the modulation signal from the function generator ( fg220 , yokogawa ) . the error signal includes information on cavity resonance , and is then used to build the feedback signals supplied on both the fast and slow control ports of the seed laser . the electronic feedback system has been designed and built by saclay @xcite . the simplified flow chart is shown in fig . [ fig : cavcontrl ] . + the optical elements that are critical to beam alignment and polarization control and measurement , the lasers ( seed laser , fiber amplifier and ppln doubling system ) and the locking electronics are all interfaced to a remotely controlled system through a local workstation . electronics specific to the feedback control loop , the function generator ( ramp generator , modulation signal generator ) , the oscilloscope and the workstation are located in the control room . the lasers , stepper and servo motors with their control units , photodiodes , ccd cameras and the preamplifiers are located on the optical table in the hall a tunnel . the monitors related to the cavity resonance mode and beam alignment are connected to cameras ccd@xmath39 and ccd@xmath38 through coaxial bnc cables and are located in the control room . [ fig : cavityfunctionalview ] shows a functional view of the laser and cavity system with control units . + an automatic switching system from open - loop " mode to closed - loop " mode around the cavity resonance region permits the system to transition automatically between the resonant and locked states . it consists of an electronic circuit and an epics program that manage the laser temperature scan via the slow control port of the seed laser . a 10 v peak - to - peak triangular ramp , together with a sinusoidal modulation signal at @xmath62 = 928 khz with 50 mv amplitude , is supplied on the laser piezoelectric transducer ( pzt ) via the fast control port of the seed laser . + two fast photodiodes ( s1223 ; hamamatsu ) are used for monitoring the reflected ( pdr ) and transmitted ( pdt ) signals . they are held at a constant voltage level of 5 v. the currents from the photodiodes are transformed into voltage signals via a trans - impedance amplifier that allows signal transmission across 100 meters of coaxial cables . a band - pass filter is applied on the reflected signal from the cavity after it is mixed with the modulation signal at frequency @xmath62 and amplified . the mixing and filtering extracts an error signal corresponding to the change in reflected light at the laser modulation frequency . the value of @xmath62 was determined by minimizing the laser residual amplitude modulation ( ram ) @xcite . the error signal is injected into a series of three separate integrators ( proportional - integral - derivative or pid ) common to the slow and fast control loops . the two control modes play complementary roles : the slow mode is for compensating the slow drift in laser frequency while the fast mode allows the efficient reduction of the laser frequency jitter . the output signals of these two modules are applied directly to the two laser control ports that control the seed laser frequency . + the vme ( versa module eurocard ) crate , used for controlling the electronics , is located in experimental hall a. the electronics are controlled from a workstation in the control room with the help of interface cards . the interface cards permit us to transmit numerical signals over a network to the optical table area which is about 100 meters from where the vme crate is located . the epics slow control program is used for regulating and activating the entire system from the workstation in the control room . it allows real - time control of analog and digital inputs and outputs to the various crates and modules . the vme crate houses the following cards : an icv150 card that measures the voltages of the control signals ( pdr , pdt , fast ramp , slow ramp ) ; an icv196 card that provides the digital interface between the electronic - card sequencer and the control screens by sending and receiving transistor - transistor logic ( ttl ) signals ; and an mv712 card that allows rs-232 commands to control devices such as the laser , the ppln temperature controller and the fiber amplifier . a servo motor controller is used for controlling the steering mirrors m@xmath38 , m@xmath39 and the focusing lens l@xmath43 . beam position variations are monitored with two quadrant cell photodiodes , noted as 4q@xmath14 and 4q@xmath12 , that detect a small amount of transmitted light ( @xmath63 ) behind m@xmath47 and m@xmath56 . the analog signals read by their electronics are fed to the icv150 card and monitored by the control system . each quarter wave plate ( qwp@xmath38 , qwp@xmath39 ) is controlled by a separate stepper motor controller ( ks491 - 2p , suruga seiki ) . with the help of the slow control interface , we can control the entire locking assembly from the workstation : the locking threshold level , the servo loop gain , the state of the feedback loop and the laser frequency scan program . once the optical cavity is installed in the electron beam line , it is important to check various parameters ( such as cavity finesse @xmath61 , bandwidth @xmath64 and the intra - cavity power @xmath65 ) that are essential to the performance of an operational cavity . two methods have been employed to characterize the cavity : a graphical method and a decay - time measurement method . + the graphical method analyzes the shape of the cavity resonance waveforms ( transmission , reflection and error signals ) while the seed laser frequency is being scanned by a triangular ramp voltage applied to the laser pzt . in this method , we first determine the fwhm of the reflection / transmission peak that corresponds to the cavity bandwidth . then , one converts from time units to frequency units by setting the gap , 2@xmath21 = @xmath66 + @xmath67 , between the two sidebands ( @xmath68 , @xmath69 ) equal to twice the modulation frequency @xmath62 . [ fig : bandwidth ] shows a theoretical fit @xcite to the transmission ( fig . [ fig : bandwidth](a ) ) and reflection ( fig . [ fig : bandwidth](b ) ) signals that are used to extract the cavity bandwidth and therefore the cavity finesse . the measurement shows that the cavity exhibits a typical bandwidth of @xmath2 13 khz with a corresponding finesse of @xmath2 13,300 . since the typical width of the resonance is @xmath2 0.04 nm , the measurement is susceptible to external disturbances and fluctuates for each measurement . in addition , due to a hysteresis exhibited in the laser pzt , the modulation sidebands would not be symmetric around the main peak as shown in fig . [ fig : bandwidth ] . therefore , the finesse measured in this method has a large error at the order of 10@xmath70 20@xmath70 . + once the laser beam and cavity are aligned with respect to each other , the laser frequency can be locked to the fundamental cavity resonance mode via its feedback system . [ fig : scopesignal ] shows a digital oscilloscope trace that includes waveforms corresponding to the locked ( feedback on ) and unlocked ( feedback off ) states of the cavity . in the cavity decay - time measurement method , while the cavity is locked , we switch off the laser diode and immediately record the transmitted power via a fast photodiode ( s1223 ; hamamatsu ) and a digital oscilloscope ( tds3054b ; tektronix ) . for a high - finesse cavity , the cavity decay time @xmath71 is directly related to the cavity parameters by @xcite , @xmath72 where @xmath73 is the cavity length and @xmath74 is the speed of light . here , it is necessary to take into account the decay time of the laser diode ( @xmath75 ) itself in order to determine @xmath76 correctly . s.,width=283 ] fig . [ fig : decaytime ] shows a typical cavity decay curve measured ( black dots ) for the optical cavity and an exponential fit @xcite ( red line ) to extract the decay time . here , after correcting the fitted value of the decay time to the measured laser decay time ( 6 @xmath1s ) @xcite and taking the sampling error of the oscilloscope as 0.5 @xmath1s , we have @xmath71 = 11.40 @xmath77 0.5 @xmath1s with the corresponding @xmath61 = 12,637 @xmath77 554 and @xmath78 = 14.00 @xmath77 0.32 khz from this measurement . + [ cols="<,<,<,<",options="header " , ] after taking into account all the systematic errors described , the total systematic error in the determination of the intra - cavity laser polarization was estimated to be 0.70% . the averaged statistical errors during three - month period of the prex experiment were 0.1% and 0.13% for the left- and right - circularly polarized states respectively . based on the transfer function measurement , the average left- and right - laser polarizations ( @xmath9 = @xmath79 ) with uncertainties during the prex experiment are summarized as @xmath80 , @xmath81 . + so far we have discussed how to make a reliable photon target with a pure circular polarization inside a high - finesse optical cavity for the electron beam so that there is efficient compton scattering . the green cavity has been combined with the photon arm @xcite of the upgraded compton polarimeter to continuously measure the electron beam polarization . the electron beam polarization measurement can be achieved by one of the three ways at the down stream of the optical cavity : detecting the backscattered compton photons , detecting the scattered compton electrons , and detecting both the compton photons and compton electrons simultaneously . at the time we commissioned our laser system and cavity , the electron detector had not been fully commissioned and there is no way to detect the scattered compton electrons . therefore , the results presented here are solely based on the data from the upgraded photon detector and daq that are described in ref . @xcite . in the following section , without going into the details , we will briefly present the results of the electron beam polarization measurement conducted for the first time during the prex experiment with electron beam parameters of 1.06 gev and 50 @xmath1a @xcite . the upgraded compton polarimeter measures the electron beam polarization based on the detection of scattered compton photons during electron helicity reversal at a rate of 120 hz @xcite . the helicity of the electron beam at jlab can be flipped at a rate of between 1 ms to 33 ms at the injector with a pseudo - random pattern . [ fig : triggerrate ] shows the scattered photon rates versus accelerator helicity timing signal ( called the mps signal ) measured with the photon detector for an electron beam energy and current of 1.06 gev and 50 @xmath1a . during the data taking , the laser polarization is flipped periodically between right- and left - circular state and the cavity is locked for @xmath2 90 seconds in each polarization state . between each polarization state , the cavity is out of resonance for @xmath2 40 seconds to measure the background rate , which is subtracted from the signal rate measured when the cavity is locked . lower background rates can be achieved by several steps . low vacuum pressure in the cavity ( 10@xmath58 torr in our case ) can reduce the scattering between electron beam and residual gas . properly controlling the electron beam parameters upstream and carefully centering and focusing the beam at the interaction point is a key to achieve low background rates . during background rate optimization , the electron beam is scanned vertically until it crosses the photon beam at the cip by using all four dipoles connected in series to the power supplies in the compton chicane ( fig . [ fig : chicane ] ) . + the compton photon energy spectrum is measured using the triggered mode of the daq . in this readout mode , a discriminator with a very low threshold is used to trigger a readout of the fadc samples over a 500 ns window , allowing study of the sampled pulse shape and size . the sum of samples in the pulse represents the total energy deposited in the gso crystal for each detected photon . this energy spectrum is plotted in ( fig . [ fig : cptspectr ] ) , with the horizontal axis in raw adc units . to determine the experimental compton asymmetry @xmath6 , the measured adc response is calibrated to the well known compton scattering photon energy spectrum and the result is compared to a full monte carlo simulation of the photon detector by geant4 @xcite . after confirming good agreement with data , this same simulation is then used to calculate the theoretical asymmetry @xmath8 in eq . ( [ eq : cmptasym ] ) . the integrating photon detector daq ( as described in ref . @xcite ) sums the fadc samples over the entire period of stable beam helicity between flips , using a gate supplied by an accelerator timing board . these integrated signals ( @xmath82 ) are used to calculate the asymmetry @xmath6 as : @xmath83 for each laser helicity period with separate sums of integrals for all electron helicity ( @xmath77 ) windows . a separate sum and difference of integrated signals for the adjacent cavity - unlocked periods , @xmath84 , used to determine background for the cavity - locked period . after taking into account the background and helicity - dependent electron beam parameters at the accelerator source , eq . ( [ eq : asymmegan ] ) becomes , @xmath85 where @xmath86 denotes the mean integral per helicity window over each cavity ( locked / unlocked ) period . here , @xmath87 is the measured integrated signal with the laser locked , and includes both the compton - scattered signal @xmath88 and the background signal @xmath89 . @xmath6 can be extracted separately for each laser polarization using one of the three averaging techniques discussed in ref . [ fig : aexp ] shows a histogram of a background - subtracted compton asymmetry taken for every electron helicity pair in a single one hour run . the electron beam polarization is continuously monitored by the compton polarimeter during the prex experiment . as we described in section [ sec : intro ] , the electron beam polarization @xmath5 is extracted from the measured experimental asymmetry @xmath6 , the laser polarization @xmath9 and the theoretical asymmetry @xmath8 , according to eq . ( [ eq : cmptasym ] ) . [ fig : lasercycle ] shows the electron beam polarization measurement based on an average of two experimental asymmetry numbers obtained for the left and right laser cycles during a typical run . the average experimental asymmetry corresponding to the left and right laser cycles for this run are -1.59% @xmath77 0.02% ( stat ) and 1.56% @xmath77 0.02% ( stat ) . after taking the value of @xmath8 as 1.825% and using the asymmetry numbers above , the extracted average electron beam polarization for this run is 87.40% @xmath77 0.70% ( stat ) . + finally , after taking into account all the uncertainties in other parameters ( such as laser polarization , theoretical asymmetry , photon detector gain shift etc . ) , the average beam polarization measured by the compton polarimeter during the prex experiment is @xmath90 . it was compared with an independent mller polarimeter measurement made at different times during the run which gave an average @xmath91 @xcite . the observed difference in electron beam polarization is consistent within the systematic errors between the two measurements . the green fabry - perot cavity provided hall a of jlab with a unique laser source to carry out precision compton polarimetry . after combining with the upgraded compton daq and photon detector @xcite , it extends the operating energy range of the previous compton polarimeter @xcite from 3 11 gev to 1 11 gev . high - precision electron beam polarimetry was tested for the first time at jlab with beam energy and current of 1.06 gev and 50 @xmath1a , and achieved its 1.0% precision goal for the prex experiment . the system we developed allows us to obtain a reliable intra - cavity power and stable photon beam polarization for the long - term operation of the upgraded compton polarimeter . the cavity mechanical and locking stability was excellent during the three - month period of the prex experiment , despite the concern of a very high - radiation and acoustically noisy environment in hall a at jlab . the frequency - doubled green laser power was stable indicating that the ppln crystal could withstand high power for an extended period of time . + the frequency locking of the green laser generated by seeding the nd : yag npro laser to the ydfa makes the intra - cavity power scalable . this allows a possibility of boosting the intra - cavity power further ( 10 kw or more at 532 nm ) by increasing the injection power to the cavity provided there is no thermal problem in the cavity mirrors . a higher intra - cavity power may also be achievable by replacing the current cavity mirrors with higher finesse mirrors which is possible with the current feedback electronics . regarding the intra - cavity laser polarization , it is one of the dominant errors in our polarimeter with an uncertainty of 0.7% . it is dominated by our ability to through direct measurement , effects that might alter the polarization over the numerous re - circulations within the cavity . other factors include analyzing power and photon detector gain shift can also affect the systematic errors in the final electron beam polarization @xcite . recent studies at hera @xcite and jlab hall c @xcite show that it is possible to get @xmath3 0.3% level precision for the intra - cavity laser polarization with proposed techniques . so far this result is only based on scattered photon analysis . this measurement could be combined with an independent electron analysis as demonstrated in hall c at jlab with a reported precision of 0.6% @xcite . all of this encourages us that , with a careful and dedicated study , it should be feasible to get @xmath3 0.5% precision at electron beam energies below 2 gev . + future high - precision parity - violating electron scattering experiments at jlab , such as the moller @xcite , pvdis @xcite , prex - ii @xcite , and crex @xcite experiments , also require electron beam polarimetry with sub-1% precision and would rely on the green compton polarimeter we built . this requires an improvement of the current precision by at least a factor of two . finally , a compton polarimeter with a mode - locked pulsed - laser - based optical cavity @xcite would give a much higher signal - to - noise ratio in scattered compton events and would make the system even more robust and efficient , especially in the more noisy electron beam environment expected @xcite after the upgrade of the cebaf energy to 12 gev . the authors wish to thank the hall a technical staff for their assistance in installing the upgraded system and the jlab accelerator division for their efforts to achieve and maintain the high electron beam quality necessary for running the compton polarimeter . this work is supported by doe grants de - ac05 - 06or23177 , under which jefferson science associates , llc , operate jefferson lab , and de - fg02 - 84er40146 . 00 s. abrahamyan _ et al . _ ( prex collaboration ) , phys . rev . lett . , 108 ( 2012 ) 112502 . d. androic _ et al . _ ( qweak collaboration ) , phys . lett . , 111 ( 2013 ) 141803 . d. wang _ et al . _ ( pvdis collaboration ) , nature , 506 ( 2014 ) 7468 . b. wagner _ _ , nuclear instruments and methods a , 294 ( 1990 ) 541 . n. sherman , phys . 103 , ( 1956 ) 1601 . m. baylac _ et al . _ , b , 539 ( 2002 ) 8 . a. narayan _ et al . _ , x , 6 ( 2016 ) , 011013 . i. passchier _ _ , nuclear instruments and methods a , 414 ( 1998 ) 446 . _ , nuclear instruments and methods a , 329 ( 1993 ) 79 . _ , nuclear instruments and methods a , 165 ( 1979 ) 177 . sld collaboration , m. woods , the workshop on high energy polarimeters at nikhef , vol . slac - pub-7319 , slac , 1996 . r. zimmermann and w. hillert , proceedings of the international particle accelerator conference , ( 2012 ) 132 . _ , nuclear instruments and methods a , 412 ( 1998 ) 1 . n. falletto _ _ , nuclear instruments and methods a , 459 ( 2001 ) 412 . lipps , h. a. tolhoek , physica , 20 ( 1954 ) 85 . prescott , slac preprint tn-73 - 1,1973 . z. huang and r. d. ruth , phys . lett . , 80 ( 1998 ) 976 . graves , w. brown , f.x . kaertner , d.e . moncton , nuclear instruments and methods a , 608 ( 2009 ) s103 . i. pupeza _ et al . _ , nature photonics , 7 ( 2013 ) 608 . v. brisson _ _ , j. inst . , 5 ( 2010 ) _ , prog . in part . and nucl . , 55 , ( 2005 ) 332 . s. escoffier _ _ , nuclear instruments and methods a , 551 ( 2005 ) 563 . g. bardin _ et al . _ , conceptual design report of a compton polarimeter for cebaf hall a , ( 1996 ) . m. friend _ _ , nuclear instruments and methods a , 676 ( 2012 ) 96 . _ , nuclear instruments and methods a , 728 ( 2013 ) 92 . a. rakhman , _ phd thesis _ , syracuse university , ( 2011 ) . fejer , g.a . magel , d.h . jundt , and r.l . byer , ieee j. quantum . , 28 ( 1992 ) 2631 . jundt , opt . , 22 ( 1997 ) 1553 . hum , m.m . fejer , c. r. physique , 8 ( 2007 ) 180 . siegman , lasers " , university science books , mill valley , ca , ( 1986 ) . optocad , tracing gaussian tem@xmath37 beams through an optical set - up , http://home.rzg.mpg.de/~ros/optocad.html[http://home.rzg.mpg.de/ ros / optocad.html ] experimental physics and industrial control system , http://www.aps.anl.gov/epics/ w.p . et al . _ , b , 31 ( 1983 ) 97 . m.j . lawrence , b. willke , m.e . husman , e.k . gustafson , and r.l . byer , j. opt . b , 16 ( 1999 ) 523 . j. poirson , f. bretenaker , m. vallet , and a. le floch , j. opt . b , 14 , ( 1997 ) 20811 . a.m. de rival _ et al . instrum . , 67 ( 1996 ) 2680 . g. rempe , r.j . thompson , h.j . kimble , and r. lalezari , opt . lett . , 17 ( 1992 ) 363 . hood , h.j . kimble and j. ye , phys . a , 64 ( 2001 ) 033804 . smith and d.i . shernoff , appl . , 24 ( 1985 ) 1722 . jones , j. opt . am . a , 3 ( 1941 ) 488 . h. hurwitz and r.c . jones , j. opt . a , 31 ( 1941 ) 493 . b.e.a . saleh and m.c . teich , fundamentals of photonics " , wiley - interscience , second edition , ( 2007 ) root program and library , https://root.cern.ch p. micossi , f.d . valle , e. milotti , e. zavattini , c. rizzo , and g. ruoso , appl . b , 57 ( 1993 ) 95 . d. jacob , m. vallet , f. bretenaker , a.l . floch , and m. oger , opt . 20 ( 1995 ) 671 . s. moriwaki , h. sakaida , t. yuzawa , n. mio , appl . b , 65 ( 1997 ) 347 . j.e . logan , n.a . robertson , and j. hough , opt . comm . , 107 ( 1994 ) 342 . w. winkler , a. rdiger , r. schilling , k.a . strain , k. danzmann , opt . , 112 ( 1994 ) 245 . et al . _ , phys . rev . beams , 10 ( 2007 ) 023501 . s. agostinelli _ _ , nuclear instruments and methods a , 506 ( 2003 ) 250 . k.s . et al . _ , an ultra - precise measurement of the weak mixing angle using moller scattering , proposal for jefferson lab pac 34 , ( 2008 ) . _ , precision measurement of parity - violation in deep inelastic scattering over a broad kinematic range , proposal for jefferson lab pac 35 , ( 2009 ) . et al . _ , precision parity - violating measurement of the neutron skin of lead , proposal for jefferson lab pac 38 , ( 2011 ) . j. mammei _ et al . _ , parity - violating measurement of the weak charge distribution of @xmath92ca to 0.02 fm accuracy , proposal for jefferson lab pac 40 , ( 2013 ) . _ , j. inst . , 7 ( 2012 ) p01017 . _ , j. inst . , 7 ( 2012 ) p01021 . j. benesch , g.b . franklin , b.p . quinn and k.d . paschke , phys . rev . beams , 18 ( 2015 ) 112401 .
an important trend in the condensed matter physics of the last two decades , has been the use of advanced field - theoretical methods to discuss various subtle and fundamental properties of interacting many - particle systems at low temperatures . there are several reasons for this trend . the first reason is of course , that a traditional topic in statistical and condensed matter physics is the study of phase transitions and critical phenomena , for which the universal properties are independent of the microscopic details of the system and can therefore be determined by a field theory describing only the large - scale properties of the system of interest . since the latter are usually solely determined by symmetry considerations , this has led to the important concept of spontaneous symmetry breaking , which has turned out to be not only highly successful in condensed matter physics , but also in high - energy physics and in particular in the standard model of elementary particles @xcite . a second reason is that soon after the development of the renormalization group methods for critical phenomena @xcite , it was realized that the same methods can in fact be used to describe the large - scale properties of many - particle systems at any temperature and not only near the critical one . moreover , application of the renormalization group ideas does not only lead to an understanding of the static behaviour but also of the dynamical properties @xcite . as a result quantum field theories can be used to conveniently determine the dynamics of many - particle systems close to equilibrium , i.e. , for example the linear hydrodynamical equations of motion . in addition , it can even be used in highly nonequilibrium situations where in general also nonlinearities play an important role . this feature , that also the dynamics of the system can be captured by a quantum field theory , has for instance led in recent years to the study of so - called quantum phase transitions @xcite . finally , the importance of field - theoretical methods in condensed matter physics is associated with the observation that also the effects of imperfections , i.e. , disorder , can be treated in this way @xcite . apart from the technological importance of disorder , for example for superconducting magnets , disorder leads also to fundamentally new physics such as the phenomenon of localization @xcite . in quantum hall systems , a combination of disorder and interaction effects give rise to the realization of various peculiar quantum fluids with fractionally charged excitations @xcite . the application of field theories has proven to be highly successful in this case and has led to a theory of the quantum hall effect in terms of edge states that form a chiral luttinger liquid @xcite . in mesoscopic physics , the study of disorder in small electronic structures has resulted in the so - called random matrix theory @xcite , which has also been of much use in the study of the quantization of classically chaotic systems . in this course we aim to show that quantum field theory is also very convenient for obtaining a detailed understanding of the equilibrium and nonequilibrium properties of trapped atomic gases . after the first observations of bose - einstein condensation in 1995 @xcite , degenerate atomic gases have received again a great deal of attention and are presently the main subject of study of a large number of experimental and theoretical groups around the world . the reason for all this excitement is , first of all , that bose - einstein condensation has never before been observed experimentally in a clear - cut manner , even though this phenomenon was already predicted by einstein in 1925 @xcite . second , it is of fundamental interest because it is the only phase transition that occurs also in the absense of interactions and is , therefore , the textbook example for the use of statistical - physics methods . finally , the goal of achieving bose - einstein condensation turned out to be much more difficult than anticipated at first . just before the breakthrough in 1995 , it had even acquired the nature of a ` quest for the holy grail ' , since the pioneering experiments were already performed in 1980 @xcite . from a theoretical point of view , a quantitative understanding of the experiments with cold atomic gases requires that we take into account the following two effects . first , the gas is magnetically trapped in an , usually axially symmetric , harmonic oscillator potential . this is necessary because , in order to obtain the required record low temperatures of @xmath0 nk , the gas can not be allowed to have any contact with material walls . second , the atoms of the gas interact with each other , which in general dramatically affects the behaviour of the condensed gas , even at the relevant low densities of @xmath1 @xmath2 . an accurate description of these degenerate gases thus requires the solution of a highly inhomogeneous many - body problem . the theoretical challenge posed by these new quantum systems lies therein that the density of the gas is sufficiently small that it should be possible to accurately solve this many - body problem from first principles and to compare the outcome of the theory directly with experiment , i.e. , without any adjustable parameters . in our opinion , quantum field theory is the most simple way in which we are able to meet this challenge . we start our development of the quantum field theory of trapped atomic gases by considering first the equilibrium properties of these gases . we consider both bose and fermi gases , and the ultimate aim of this section is to arrive , for both cases , at an accurate description of the normal and superfluid phases of the gas . although mixtures of bose and fermi gases are also of current interest , we do not consider them explicitly here , because they can be treated by a straightforward generalization of the theory . in section [ nft ] we then turn to the nonequilibrium properties , which are perhaps the most interesting and certainly the least understood at present . the reason for organizing the course in this way , is that the development of the equilibrium theory gives us an opportunity to introduce all the necessary tools that are required for a treatment of the more complicated nonequilibrium case . in particular , we present in detail the way in which we can make use of functional methods . to make also a connection with the more familiar operator language , however , we first briefly summarize the outcome of the second quantization formalism . the atoms of interest to us have internal degrees of freedom due to the electron and nuclear spins . in principle this is very important , because it gives the atom a magnetic moment , which is used to trap the atoms in a magnetic field minumum . during this course , however , we restrict ourselves to atomic gases that are a mixture of at most two hyperfine states . without loss of generality , we can then suppose to have @xmath3 identical atoms with mass @xmath4 and effective spin @xmath5 in an external potential @xmath6 . as a result , the time - dependent schrdinger equation we have to solve is @xmath7 where the hamiltonian is @xmath8 @xmath9_- = i\hbar$ ] , and all other commutators of the positions and momenta vanish . the first term in the right - hand side is the sum of the one - particle hamiltonians , which includes an effective zeeman interaction that accounts for a possible difference in the hyperfine energies . the second term represents the interactions . for simplicity , we have assumed that the interaction @xmath10 is independent of the hyperfine states of the atoms @xmath11 and @xmath12 . this is in general not justified for realistic atomic gases , but is valid for the specific applications that we have in mind . moreover , in section [ hft ] we also discuss the general case . finally , we have also neglected possible three - body forces . this is a result of the fact that we are interested in dilute quantum gases , for which it is highly improbable for three atoms to simultaneously interact with each other . without interactions the eigenstates are , of course , given by the states @xmath13 , where the specific quantum state for each atom is exactly known . here @xmath14 and the nonnegative integers @xmath15 , @xmath16 , and @xmath17 denote the three quantum numbers that are required to specify the one - particle eigenstates in the external potential . the wave functions and energies of these eigenstates are @xmath18 and @xmath19 , respectively , and are found from the time - independent schrdinger equation @xmath20 in addition , the internal state @xmath21 is a shorthand notation for @xmath22 . the many - body wave function @xmath23 , however , has to be symmetric or antisymmetric under permutations for bosonic or fermionic atoms , respectively . therefore it is more convenient to use a properly ( anti)symmetrized version of the above basis , i.e. , the states @xmath24 with the occupation numbers @xmath25 for bosons and @xmath26 for fermions . the hilbert space of all these states , without the constraint @xmath27 , is known as the fock space . clearly , the many - body wave function @xmath23 can be expanded in this basis as @xmath28 where @xmath29 is the amplitude for the gas to be in state @xmath24 at time @xmath30 . in this basis the schrdinger equation becomes @xmath31 which shows that we need the matrix elements of the hamiltonian between different states in the fock space . to calculate these most easily we introduce so - called annihilation operators @xmath32 by @xmath33 from which it follows that the creation operators @xmath34 obey @xmath35 as a result , we see that the operator @xmath36 counts the number of atoms in the state @xmath37 , i.e. , @xmath38 we have also that @xmath39_{\mp } = [ \hat{\psi}^{\dagger}_{{\bf n},\alpha } , \hat{\psi}^{\dagger}_{{\bf n}',\alpha'}]_{\mp}=0 $ ] and most importantly that @xmath40_{\mp } = \delta_{{\bf n},{\bf n } ' } \delta_{\alpha,\alpha'}~.\ ] ] in equations ( [ cr ] ) and ( [ com ] ) the upper sign refers to bosons and the lower to fermions . this will be true throughout the course . from these results we can now easily show , first of all , that the basis in the fock space is given by @xmath41 with @xmath42 the vacuum state containing no atoms . second , the hamiltonian is @xmath43 with @xmath44 the one - particle energies . furthermore , @xmath45 is the amplitude for a collision which scatters two atoms out of the states @xmath46 and @xmath47 into the states @xmath37 and @xmath48 , as schematically shown in figure 1(a ) . introducing the field operators @xmath49 and @xmath50 , that annihilate and create atoms in the spin state @xmath21 at position @xmath51 respectively , we can rewrite this result into @xmath52 note that , due to the ( anti)commutation relations of the creation and annihilation operators and the orthogonality of the wave functions @xmath53 , the field operators obey @xmath54_{\mp}= [ \hat{\psi}^{\dagger}_{\alpha}({\bf x } ) , \hat{\psi}^{\dagger}_{\alpha'}({\bf x}')]_{\mp}= 0 $ ] and @xmath55_{\mp}= \delta({\bf x}-{\bf x}')\delta_{\alpha,\alpha'}$ ] . moreover , it is also important for the following to note that the number operator is @xmath56 and , similarly , that the effective total spin operator is @xmath57 the density of atoms in the state @xmath21 is thus simply @xmath58 . this completes our brief discussion of the second quantization formalism . in principle , we could now proceed to develop an operator formulation of the quantum field theory of interest to us . since the experimentally most important observables can be expressed as appropriate products of the field operators , as we have just seen , this would essentially amount to the study of the ( imaginary ) time evolution of the heisenberg operator @xmath59 at a fixed chemical potential @xmath60 @xcite . put differently , the desired quantum field theory would be defined by the heisenberg equation of motion @xmath61_-~,\ ] ] and we would need to solve this equation in a sufficiently accurate approximation . as mentioned previously , however , we here want to develop feynman s ` path - integral ' formulation of the problem , which will turn out to be much more convenient for our purposes . to do so in a manner that is the same for both bosonic and fermionic atomic gases , we first need to introduce some mathematical background . we have seen that in the case of fermions , we need to make use of anticommuting creation and annihilation operators . this automatically builds in the pauli principle in the theory , because it implies that @xmath62 and thus that the occupation numbers @xmath63 are restricted to be either @xmath64 or @xmath65 . for reasons that will become clear in a moment , it is in that case also convenient to introduce anticommuting complex numbers or grassmann variables . the simplest example is to have two such grassmann variables , say @xmath66 and @xmath67 . the set @xmath68 , and linear combinations thereof with complex coefficients , form then a so - called grassmann algebra . by definition we have @xmath69_+ = [ \phi,\phi^*]_+ = [ \phi^*,\phi^*]_+ = 0 $ ] and thus in particular @xmath70 . therefore , the above set is indeed complete . the complex conjugation in this algebra is defined by @xmath71 , @xmath72 , and @xmath73 . moreover , we can also define an analytic function on this algebra by @xmath74 as a result , it is natural to define also a differentiation by @xmath75 to be more precise this is in fact a left differentiation and the minus sign occurs , because we need to permute @xmath67 and @xmath66 before we can differentiate with respect to @xmath66 . so , similarly , we have @xmath76 and @xmath77 next we also need integrations over these grassmann variables . note that since @xmath78 we have only two possible integrals , namely @xmath79 and @xmath80 . we define these by @xmath81 and @xmath82 this means that integration is equivalent to differentation . the main reason for the above definition is that we want the integration to obey the usual rules of partial integration . in particular , this implies that @xmath83 for any function @xmath84 . it is clear that this condition requires that @xmath85 . the result of @xmath86 is then solely a question of normalization . it turns out that we are primarily interested in integrals of the form @xmath87 as we will see in section [ fi ] . clearly , we can now also consider the grassmann algebra based on the variables @xmath88 and @xmath89 with @xmath90 . what we will need in the following are gaussian integrals over these variables . it is not difficult to show with the above definitions that @xmath91}~.\ ] ] if the variables @xmath88 and @xmath89 were just ordinary complex numbers we would in contrast have the result @xmath92}~.\ ] ] these last two results will be used many times in the following . one immediate use of these grassmann variables is that we can now consider eigenstates of the annihilation operator @xmath32 , also when we are dealing with fermions . these eigenstates are called coherent states . consider the state @xmath93 where @xmath94 is a grassmann variable that also anticommutes with the creation and annihilation operators in our fock space . clearly we have that @xmath95 so @xmath96 is indeed an eigenstate of @xmath32 with eigenvalue @xmath94 . in general we can now make the states @xmath97 that obey @xmath98 . introducing the grassmann - valued field @xmath99 , the latter two relations can be rewritten as @xmath100 and @xmath101 . it is important to note that these coherent states are not orthonormal . in contrast , we find that @xmath102 nevertheless they obey a closure relation , as can be seen explicitly from @xmath103 we write this from now on simply as @xmath104d[\phi]~e^{-(\phi|\phi ) } |\phi\rangle\langle\phi| = \hat{1}~.\ ] ] the interesting observation at this point is that essentially the same formulaes also hold for bosons @xcite . we have only a minus sign difference in @xmath105 but then it can still be easily shown that @xmath106 . a convenient way to do so is , for instance , by noting that the commutation relation @xmath107_- = \delta({\bf x}-{\bf x } ' ) \delta_{\alpha,\alpha'}$ ] implies that @xmath108 acts as @xmath109 on these states . furthermore , the same expressions hold for the overlap @xmath110 and the closure relation , if we define the integration measure by @xmath111d[\phi ] \equiv \int \prod_{{\bf n},\alpha } d\phi^*_{{\bf n},\alpha } d\phi_{{\bf n},\alpha}/(2\pi i)$ ] in this case . summarizing , we have thus found for bosons and fermions that @xmath112 @xmath113 and @xmath104d[\phi]~e^{-(\phi|\phi ) } |\phi\rangle\langle\phi| = \hat{1}~.\ ] ] the last ingredient we need is that in terms of these coherent states the trace of an operator @xmath114 over the fock space can be expressed as @xmath115 = \int d[\phi^*]d[\phi]~e^{-(\phi|\phi ) } \langle\pm\phi|\hat{o}|\phi\rangle~.\ ] ] the minus sign in the fermionic case is easily understood from the fact that we then have @xmath116 , due to the anticommuting nature of the grassmann variables . after this mathematical interlude we can now return to physics and to our goal of arriving at a unified treatment of bosonic and fermionic atomic gases . we are interested in determining the equilibrium properties of a trapped atomic gas at some temperature @xmath117 . from statistical physics we know that all these properties can be obtained from the grand - canonical partition function @xmath118~,\ ] ] where @xmath119 and @xmath60 the chemical potential . we thus need to evaluate this quantity . as mentioned previously , we want to do so by making use of quantum field theory and , in addition , of feynman s path - integral approach to quantum mechanics . we thus want to write the partition function as a functional integral over time - dependent fields @xmath120 , just like the partition function of a single particle in an external potential can be written as a functional integral over all time - dependent paths @xmath121 . this can indeed be achieved with the coherent states that we have introduced in the previous section . we start with using our formula ( [ trace ] ) for the trace of an operator , @xmath122d[\phi]~e^{-(\phi|\phi ) } \langle\pm\phi| e^{-\beta(\hat{h}-\mu \hat{n } ) } |\phi\rangle~,\ ] ] and observe that we are thus faced with the task of calculating the matrix elements @xmath123 with @xmath124 and @xmath125 . this is difficult in general but can be simplified in the following way . we first realize that the operator @xmath126 is identical to the quantum mechanical evolution operator @xmath127 evaluated at @xmath128 . put differently , we thus want to calculate the matrix elements of the imaginary - time evolution operator @xmath129 for @xmath130 . to do so , we next split the time interval @xmath131 $ ] into @xmath132 pieces , with @xmath133 and @xmath134 . so @xmath135 . the procedure is summarized in figure 2 . $ ] needed for feynman s path - integral formulation of the partition function . ] at each intermediate time @xmath136 we then apply a closure relation of the coherent states . this gives @xmath137d[\phi_m]~ e^{-(\phi_m|\phi_m ) } \right ) \prod_{m=1}^m \langle\phi_m| e^{-\delta\tau(\hat{h}-\mu \hat{n})/\hbar } |\phi_{m-1}\rangle~. \nonumber\end{aligned}\ ] ] now we can use that in the limit @xmath138 we only need to know the latter matrix elements up to order @xmath139 , because terms of order @xmath140 lead to corrections of order @xmath141 , which vanish in that limit . hence @xmath142/\hbar ) ~ , \nonumber\end{aligned}\ ] ] with the grand - canonical hamiltonian functional resulting from equations ( [ hamilton ] ) and ( [ number ] ) equal to @xmath143 = \sum_{\alpha } \int d{\bf x}~ \phi^*_{\alpha}({\bf x } ) \left\ { - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) + \epsilon_{\alpha } - \mu \right\ } \phi_{\alpha}({\bf x } ) \hspace*{0.4 in } \\ & & \hspace*{0.85 cm } + \frac{1}{2 } \sum_{\alpha,\alpha ' } \int d{\bf x } \int d{\bf x}'~ \phi^*_{\alpha}({\bf x } ) \phi^*_{\alpha'}({\bf x } ' ) v({\bf x}-{\bf x } ' ) \phi_{\alpha'}({\bf x } ' ) \phi_{\alpha}({\bf x})~ , \nonumber\end{aligned}\ ] ] since @xmath144 and @xmath145 . thus , neglecting terms of order @xmath140 , we can again exponentiate the right - hand side of equation ( [ me ] ) , which leads in first instance to @xmath146/\hbar}~,\ ] ] and therefore for the desired matrix element of the imaginary - time evolution operator to @xmath147d[\phi_m]~ e^{-(\phi_m|\phi_m ) } \right ) \\ & & \hspace*{0.83 in } \times \exp \left\ { \sum_{m=1}^{m } ( ( \phi_m|\phi_{m-1})- \delta\tau h[\phi^*_m,\phi_{m-1}]/\hbar ) \right\}~. \nonumber \end{aligned}\ ] ] this can then be manipulated into the suggestive form @xmath148d[\phi_m]~ \right ) \\ & & \times \exp \left\ { - \frac{1}{\hbar } \sum_{m=1}^{m } \delta\tau \left ( \hbar \frac{(\phi_m|\phi_{m})-(\phi_m|\phi_{m-1 } ) } { \delta\tau } + h[\phi^*_m,\phi_{m-1 } ] \right ) \right\}~. \nonumber \end{aligned}\ ] ] taking now the continuum limit @xmath138 and putting @xmath149 , we find that @xmath150d[\phi]~ e^{-s[\phi^*,\phi]/\hbar}~ , \nonumber\end{aligned}\ ] ] where the so - called euclidean action is @xmath151 \\ & & = \int_0^{\hbar\beta } d\tau~ \left\ { \sum_{\alpha } \int d{\bf x}~ \phi^*_{\alpha}({\bf x},\tau ) \hbar \frac{\partial}{\partial \tau } \phi_{\alpha}({\bf x},\tau ) + h[\phi^*(\tau),\phi(\tau ) ] \right\}~. \nonumber\end{aligned}\ ] ] this is essentially the desired functional integral over the complex fields @xmath120 with the boundary conditions @xmath152 and @xmath153 . it is precisely the field theory analogue of the feynman path integral . to obtain the partition function we only need to put @xmath154 equal to @xmath155 and perform a last integration over @xmath156 and @xmath157 . it then finally becomes @xmath158d[\phi]~ e^{-s[\phi^*,\phi]/\hbar}\ ] ] with the boundary conditions @xmath159 , i.e. , the fields are periodic in @xmath131 $ ] for bosons and antiperiodic for fermions . note that in equation ( [ path1 ] ) we have used the same notation for the integration measure as in equation ( [ path2 ] ) , although there is in principle one more integration in the expression for the partition function . the difference is in the continuum limit accounted for in the boundary conditions , which are in practice usually left implicite . having arrived at an exact identity between the partition function and a functional integral , we are now going to familiarize ourselves with this identity , and with how to perform functional integrals in general , by considering the ideal quantum gases . since the partition functions @xmath160 of the ideal quantum gases are known exactly , they are ideal test cases for our field - theoretical methods . moreover , a thorough knowledge of the ideal quantum gases is also an important first step in understanding experiments with trapped atomic gases , because the effects of the interatomic interaction can essentially be included perturbatively . how this perturbation theory is performed is discussed in detail in section [ ifd ] , but before we can do that we need to understand the noninteracting gases first . in that case , we have @xmath161 = \sum_{\alpha } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \\ & & \hspace*{0.2 in } \times~ \phi^*_{\alpha}({\bf x},\tau ) \left\ { \hbar \frac{\partial}{\partial \tau } \right . - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) + \epsilon_{\alpha } - \mu \right\ } \phi_{\alpha}({\bf x},\tau ) \nonumber\end{aligned}\ ] ] and the partition function is a gaussian integral , which explains why we were interested in gaussian integrals in section [ gv ] . it will be illustrative to evaluate this partition function in three different ways . in the first way , we perform the evaluation of the trace involved in the definition of the partition function at the end of the calculation , and start by considering the matrix element @xmath162 as the functional integral @xmath163d[\phi_{{\bf n},\alpha } ] \right)~ \exp \left\ { \sum_{{\bf n},\alpha } \phi^*_{{\bf n},\alpha}(\hbar\beta ) \phi_{{\bf n},\alpha}(\hbar\beta ) \right\ } \nonumber \\ & & \times \exp \left\{- \frac{1}{\hbar } \int_0^{\hbar\beta } d\tau \sum_{{\bf n},\alpha } \phi^*_{{\bf n},\alpha}(\tau ) \left ( \hbar \frac{\partial}{\partial \tau } + \epsilon_{{\bf n},\alpha } - \mu \right ) \phi_{{\bf n},\alpha}(\tau ) \right\}~ , \nonumber\end{aligned}\ ] ] which is the product for each @xmath164 and @xmath165 of the path integral @xmath166d[\phi]~ \exp \left\ { \phi^*(\hbar\beta ) \phi(\hbar\beta ) - \frac{1}{\hbar } \int_0^{\hbar\beta } d\tau~ \phi^*(\tau ) \left ( \hbar \frac{\partial}{\partial \tau } + \epsilon - \mu \right ) \phi(\tau ) \right\ } \nonumber\end{aligned}\ ] ] with the boundary conditions @xmath167 and @xmath168 . it clearly corresponds to the matrix element @xmath169 for one particular value of @xmath164 and @xmath165 . we calculate this matrix element by performing a shift in the integration variables , i.e. , @xmath170 and @xmath171 , where @xmath172 obeys the ` classical ' equations of motion @xmath173}{\delta \phi^*(\tau ) } \right|_{\phi=\phi_{\rm cl } } = \left ( \hbar \frac{\partial}{\partial \tau } + \epsilon - \mu \right ) \phi_{\rm cl}(\tau ) = 0\ ] ] and similarly for @xmath174 . the solutions with the correct boundary solutions are @xmath175 and @xmath176 , leading to the path integral @xmath177d[\xi]~ \exp \left\ { - \frac{1}{\hbar } \int_0^{\hbar\beta } d\tau~ \xi^*(\tau ) \left ( \hbar \frac{\partial}{\partial \tau } + \epsilon - \mu \right ) \xi(\tau ) \right\ } \nonumber\end{aligned}\ ] ] with the boundary conditions @xmath178 . this means that the last path integral is just equal to @xmath179 and that the desired result is just the prefactor . substituting this for each value of @xmath164 and @xmath165 , we apparently have @xmath180 and the partition function becomes the product @xmath181 this is the correct result , because from the usual thermodynamic identity @xmath182 we find @xmath183 as desired . the second way is easier and more common in practice . we immediately start with @xmath184d[\phi]~ e^{-s_0[\phi^*,\phi]/\hbar}\ ] ] and incorporate the boundary conditions by expanding the fields as @xmath185 where @xmath186 for bosons and @xmath187 for fermions . these are known as the even and odd matsubara frequencies , respectively . using this expansion we have @xmath188 if we also take account of the jacobian involved in the change of integration variables . note that the difference between the jacobians in the bosonic and fermionic case , is a consequence of the fact that for grassmann variables we have that @xmath189 instead of the result @xmath190 that we expect on the basis of ordinary complex integration . note also that these are all again gaussian integrals , so we find in first instance @xmath191 to evaluate the sum over matsubara frequencies we need to add a convergence factor @xmath192 and finally take the limit @xmath193 . the precise reason for this particular procedure can not be fully understood at this point but is explained in section [ ifd ] . however , doing so we indeed find that @xmath194 to see that this is correct , we differentiate the latter equation with respect to @xmath195 . this gives us @xmath196 which can be proved by contour integration in the following way . the function @xmath197 has poles at the even and odd matsubara frequencies with residu @xmath198 . hence , by cauchy s theorem the left - hand side is equal to @xmath199 with @xmath200 a contour that fully encloses the imaginary axis in the direction shown in figure 3 . adding to the contour @xmath200 the contour @xmath201 , which gives no contribution to the integral , we can again apply cauchy s theorem to obtain the desired result . why does the integration over the contour @xmath201 vanish ? the reason for this is that the integrant behaves as @xmath202 for @xmath203 and as @xmath204 for @xmath205 . the integrant thus always vanishes much faster than @xmath206 on the contour @xmath201 for any @xmath207 . the third and last method is simplest and , therefore , most common in practice . we first rewrite the partition function as @xmath208d[\phi]~ \exp \left\ { \sum_{\alpha } \int_0^{\hbar\beta } d\tau \int d{\bf x } \right . \\ & & \times \left . \sum_{\alpha ' } \int_0^{\hbar\beta } d\tau ' \int d{\bf x}'~ \phi^*_{\alpha}({\bf x},\tau ) g^{-1}_{\alpha,\alpha'}({\bf x},\tau;{\bf x}',\tau ' ) \phi_{\alpha'}({\bf x}',\tau ' ) \right\ } \nonumber\end{aligned}\ ] ] and see @xmath209 as a ` matrix ' both in spin space as in coordinate space . we then know that this gaussian integral is just @xmath210^{\mp 1 } = \exp \left\ { \mp { \rm tr}[\ln ( -g^{-1 } ) ] \right\}~.\ ] ] clearly , we have from the action @xmath211 $ ] that @xmath212 or equivalently that @xmath213 which means that @xmath214 is a green s function . we come back to its physical meaning shortly . for now we only need to see that the last equation is solved by @xmath215 thus again @xmath216 in principle , we have slightly cheated in the last step of this derivation , because equation ( [ green ] ) shows that @xmath217 the problem with this last way of calculating the functional integral is that it does not correctly account for the jacobian involved in the change of variables from imaginary time @xmath218 to the matsubara frequencies @xmath219 . however , this problem can be avoided by calculating never a single determinant , but always the ratio of two determinants . in this manner the effect of the jacobian is exactly cancelled . we will see some examples of how this works in sections [ hft ] and [ sc ] . we now return to the physical meaning of the above green s function . to see its meaning , we first consider the time - ordered expectation value @xmath220 \rangle \\ & & \equiv \theta(\tau-\tau ' ) \langle \hat{\psi}_{\alpha}({\bf x},\tau ) \hat{\psi}^{\dagger}_{\alpha'}({\bf x}',\tau ' ) \rangle \pm \theta(\tau'-\tau ) \langle \hat{\psi}^{\dagger}_{\alpha'}({\bf x}',\tau ' ) \hat{\psi}_{\alpha}({\bf x},\tau ) \rangle~. \nonumber\end{aligned}\ ] ] here the expectation value is taken in the grand - canonical ensemble and @xmath221 is the imaginary time heisenberg operator , which is defined by @xmath222 and therefore obeys the heisenberg equation of motion @xmath223_-$ ] . for the noninteracting case it reads @xmath224 as a result @xmath225 \rangle = \hbar \delta(\tau-\tau ' ) \langle [ \hat{\psi}_{\alpha}({\bf x},\tau ) , \hat{\psi}^{\dagger}_{\alpha'}({\bf x}',\tau')]_{\mp } \rangle \hspace*{0.4 in } \\ & & \hspace*{1.8 cm } + \left ( \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } - v^{\rm ex}({\bf x } ) - \epsilon_{\alpha } + \mu \right ) \langle t[\hat{\psi}_{\alpha}({\bf x},\tau ) \hat{\psi}^{\dagger}_{\alpha'}({\bf x}',\tau ' ) ] \rangle~. \nonumber\end{aligned}\ ] ] substituting the equal - time relations that we derived in section [ sq ] , i.e. , @xmath226_{\mp } = \delta({\bf x}-{\bf x } ' ) \delta_{\alpha,\alpha'}$ ] , thus suggests that @xmath227 \rangle~.\ ] ] we can actually also prove this important relation , that bridges the gap between the functional formulation of quantum field theory used here and the more familiar operator formalism . first of all , it is clear from the slicing procedure used in our derivation of the functional integral in section [ fi ] , that it automatically leads to time - ordered expectation values . we should , therefore , only be able to prove that @xmath228d[\phi]~ \phi_{\alpha}({\bf x},\tau ) \phi^*_{\alpha'}({\bf x}',\tau ' ) e^{-s_0[\phi^*,\phi]/\hbar}~. \nonumber\end{aligned}\ ] ] this is most easily achieved in the following way . we introduce a partition function in the presence of external currents @xmath229 and @xmath230 , where in the fermionic case these currents are also grassmann variables . the partition function is taken to be @xmath231 = \int d[\phi^*]d[\phi]~ \exp \left\ { - \frac{1}{\hbar } s_0[\phi^*,\phi ] \raisebox{0.25 in } { } \right . \\ & & \hspace*{0.15 in } + \left . \sum_{\alpha } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ ( \phi^*_{\alpha}({\bf x},\tau ) j_{\alpha}({\bf x},\tau ) + j^*_{\alpha}({\bf x},\tau ) \phi_{\alpha}({\bf x},\tau ) ) \right\}~ , \nonumber\end{aligned}\ ] ] since then we must simply prove that @xmath232 using a short - hand notation we have @xmath233 = \int d[\phi^*]d[\phi]~ \exp \left\ { ( \phi|g^{-1}|\phi ) + ( \phi|j ) + ( j|\phi ) \right\}~.\ ] ] the terms in the exponent can be rewritten as @xmath234 , which is usually called completing the square . performing a shift in the integration variables , we then easily see that @xmath235 = z_0[0,0 ] e^{-(j|g|j ) } = z_0[0,0 ] \exp \left\ { - \sum_{\alpha } \int_0^{\hbar\beta } d\tau \int d{\bf x } \right . \hspace*{0.4 in } \\ & & \hspace*{0.08 in } \times \left . \sum_{\alpha ' } \int_0^{\hbar\beta } d\tau ' \int d{\bf x}'~ j^*_{\alpha}({\bf x},\tau ) g_{\alpha,\alpha'}({\bf x},\tau;{\bf x}',\tau ' ) j_{\alpha'}({\bf x}',\tau ' ) \right\}~ , \nonumber\end{aligned}\ ] ] which after differentiation indeed leads to the desired result . we can now in fact calculate the expectation value of the time - ordered product of any number of operators . with the above expression for @xmath236 $ ] one can easily prove that this results in the sum of all possible products of time - ordered expectation values of two operators . for instance @xmath237 this is the famous wick s theorem , which plays a crucial role in the next section where we start to discuss the profound effects that interactions can have on the results obtained thusfar . the green s function , or one - particle propagator , @xmath214 is one of the most important quantities that we want to determine theoretically , because it gives us the possibility to calculate in principle the expectation value of any one - particle observable . it also gives us the elementary excitations of our system of interest . for example , the average density of the spin state @xmath21 is in an ideal gas given by @xmath238 where we introduced the notation @xmath239 for the limit @xmath193 of @xmath240 . note that this procedure is necessary due to the time - ordening involved in the definition of the green s function and the fact that the field operators do not commute at equal times . it also gives a natural explanation for our previous use of the convergence factor @xmath192 . moreover , from the argument of the bose or fermi distribution function , we see that the elementary excitations have an energy of @xmath241 . if we replace in the fourier transform of the green s function , i.e. , in @xmath242 @xmath243 by @xmath244 we also see that @xmath245 has a pole at @xmath246 . this is a general result : poles in @xmath245 correspond to the energies of the elementary excitations . these energies in general can also have a negative imaginary components , which then correspond to the lifetime of the excitation . the question that arises , therefore , is how to determine also the green s function for an interacting system . this can be done in perturbation theory , as follows . we want to calculate the expectation value @xmath247d[\phi]~ \phi_{\alpha}({\bf x},\tau ) \phi^*_{\alpha'}({\bf x}',\tau ' ) e^{-s[\phi^*,\phi]/\hbar } \nonumber\end{aligned}\ ] ] with the action @xmath248=s_0[\phi^*,\phi]+s_{\rm int}[\phi^*,\phi]$ ] . we now expand both the numerator and the denominator in powers of @xmath249 . up to first order we find for the partition function @xmath250d[\phi]~e^{-s_0[\phi^*,\phi]/\hbar } \left ( 1 - \frac{s_{\rm int}[\phi^*,\phi]}{\hbar } \right ) \\ & & \equiv z_0 \left ( 1 - \frac{1}{\hbar } \langle s_{\rm int}[\phi^*,\phi ] \rangle_0 \right)~. \nonumber\end{aligned}\ ] ] using wick s theorem , we thus have @xmath251 \rangle_0 = \frac{1}{2 } \sum_{\alpha,\alpha ' } \int_0^{\hbar\beta } d\tau \int d{\bf x } \int d{\bf x}'~ g_{0;\alpha,\alpha}({\bf x},\tau;{\bf x},\tau^+ ) \\ & & \hspace*{1.5 in } \times~ \frac{- v({\bf x}-{\bf x}')}{\hbar } g_{0;\alpha',\alpha'}({\bf x}',\tau;{\bf x}',\tau^+ ) \nonumber \\ & & \hspace*{0.52 in } \pm \frac{1}{2 } \sum_{\alpha } \int_0^{\hbar\beta } d\tau \int d{\bf x } \int d{\bf x}'~ g_{0;\alpha,\alpha}({\bf x}',\tau;{\bf x},\tau^+ ) \nonumber \\ & & \hspace*{1.5 in } \times~ \frac{- v({\bf x}-{\bf x}')}{\hbar } g_{0;\alpha,\alpha}({\bf x},\tau;{\bf x}',\tau^+)~. \nonumber\end{aligned}\ ] ] ) up to first order in the interaction . ] to understand the general structure of the perturbation expansion , it is very convenient to represent equations ( [ z1 ] ) and ( [ z2 ] ) in terms of feynman diagrams . the final result is shown in figure 4 , where a wiggly line corresponds to the factor @xmath252 and a thin arrowed line pointing from @xmath253 to @xmath254 represents the noninteracting green s function @xmath255 . in figure 4 we have , for clarity sake , also explicitly indicated the various coordinates and spin degrees of freedom that we have to either integrate or sum over , respectively . it is , however , much more common in practice to suppress in feynman diagrams those degrees of freedom that have to be integrated or summed over , and denote only the degrees of freedom on which the quantity of interest depends explicitly . for example , for the numerator of equation ( [ expec ] ) we obtain @xmath256d[\phi]~ \phi_{\alpha}({\bf x},\tau ) \phi^*_{\alpha'}({\bf x}',\tau ' ) e^{-s_0[\phi^*,\phi]/\hbar } \left ( 1 - \frac{s_{\rm int}[\phi^*,\phi]}{\hbar } \right ) \hspace*{0.4 in } \\ & & = z_0 \left ( - \langle \phi_{\alpha}({\bf x},\tau ) \phi^*_{\alpha'}({\bf x}',\tau ' ) \rangle_0 + \frac{1}{\hbar } \langle \phi_{\alpha}({\bf x},\tau ) \phi^*_{\alpha'}({\bf x}',\tau ' ) s_{\rm int}[\phi^*,\phi ] \rangle_0 \right ) , \nonumber\end{aligned}\ ] ] whose diagrammatic equivalent is then given in figure 5 . up to first order the interacting green s function , which is represented by a thick arrowed line , obeys the equation shown in figure 6 . note that the so - called disconnected diagrams have exactly cancelled . this is a general feature that happens for any expectation value , i.e. , the disconnected diagrams that occur in the numerator are always exactly cancelled by the denominator @xcite . one can also show that the general structure of the perturbation expansion of the green s function is as in figure 7 , in which only one - particle irreducible diagrams contribute to the selfenergy @xmath257 . these one - particle irreducible diagrams can be distinguished from the fact that they do not become disconnected if we cut a single thin arrowed line . this is dyson s equation @xcite . it is particularly insightful for a homogeneous gas or for a trapped gas in the weak - coupling limit , since then the one - particle states @xmath53 are not affected by the interactions and the exact one - particle propagator can be written as @xmath258 note that for realistic trapped atomic gases this weak - coupling limit is essentially always realized in the normal phase of the gas . it is , therefore , also of some experimental interest to consider that limit in more detail first . one then finds that the dyson equation in first instance becomes @xmath259 because @xmath260 and because we have assumed the interactions to be spin independent , we can easily convince ourselves that in perturbation theory it is always true that the selfenergy obeys @xmath261 and as a result also that up to all orders in the interaction @xmath262 . as we will see in section [ hft ] , however , nonperturbative effect can change this result due to the phenomena of spontaneous broken symmetry . ignoring the latter for the moment , we find for each spin state the uncoupled equation @xmath263 or @xmath264 hence @xmath265 which shows that the poles in the green s function are indeed shifted by the interactions . in our lowest order calculation we have found for the selfenergy the diagrammatic result shown in figure 8 . this is the most simple approximation we can think of . in the weak - coupling limit it is mathematically equal to @xmath266 the first term is known as the direct or hartree , and the second term as the exchange or fock contribution to the selfenergy . from equation ( [ self ] ) we conclude that the matrix elements of the interaction enter only in the combination @xmath267 . clearly , this is a refection of the pauli principle , which forces the effective interaction between two fermionic atoms in the same state @xmath37 to vanish . to make the theory selfconsistent we should use in the expression for the selfenergy not the noninteracting propegators , but precisely those that follow from the dyson equation this is the hartree - fock approximation , which is diagrammatically given in figure 9 . it leads in the weak - coupling limit to the new dispersion relation @xmath268 for the ` dressed ' particles or quasiparticles of the gas , where the selfenergy is essentially also found from equation ( [ self ] ) . we only have to replace in the right - hand side @xmath269 by @xmath270 . in this manner we have thus obtained an approximation to the interacting green s function that is nonperturbative in the interaction and effectively sums an infinite number of feynman diagrams . of course , the hartree - fock approximation can also be used in the strong - coupling limit . in that case it diagrammatically still corresponds to the solution of the dyson equation of figure 7 , with a selfenergy as shown in figure 9 . however , we are then no longer allowed to assume that the exact green s function is diagonal in the eigenstates @xmath53 of the external trapping potential . to see more explicitly what this means physically , it is convenient to rederive the hartree - fock theory without making use of perturbation theory . because the hartree - fock theory is very useful in a large number of circumstances , and because we want to illustrate a very useful technique which is nowadays often used in the literature , we are now going to reproduce the above results without making use of a diagramatic expansion . we consider for simplicity a fermionic mixture with an equal number of atoms in two hyperfine states and start by splitting our spin - independent interaction @xmath271 into two spin - dependent parts such that one contributes only to the hartree diagram and the other only to the fock diagram . denoting a spin - dependent interaction by @xmath272 we thus want that @xmath273 with @xmath274 . using operators in spin space , a possible solution to these equations is @xmath275 and @xmath276 where @xmath277 is the exchange operator of the spins of atoms @xmath65 and @xmath278 , and @xmath279 are the usual pauli matrices . clearly , we now have @xmath280 = \frac{1}{2 } \sum_{\alpha,\alpha';\beta,\beta ' } \int_0^{\hbar\beta } d\tau \int d{\bf x } \int d{\bf x}'~ \\ & & \times \left\{\phi^*_{\alpha'}({\bf x},\tau ) \phi_{\alpha}({\bf x},\tau ) v^h_{\alpha',\beta';\alpha,\beta}({\bf x}-{\bf x } ' ) \phi^*_{\beta'}({\bf x}',\tau ) \phi_{\beta}({\bf x}',\tau ) \raisebox{0.15 in } { } \right . \nonumber \\ & & \hspace*{0.5 in } - \left . \raisebox{0.15 in } { } \phi^*_{\alpha'}({\bf x},\tau ) \phi_{\beta}({\bf x}',\tau ) v^f_{\alpha',\beta';\alpha,\beta}({\bf x}-{\bf x } ' ) \phi^*_{\beta'}({\bf x}',\tau ) \phi_{\alpha}({\bf x},\tau ) \right\}~ , \nonumber\end{aligned}\ ] ] which we write as @xmath281 \equiv \frac{1}{2 } ( \phi^*\phi|v^h|\phi^*\phi ) - \frac{1}{2 } ( \phi^*\phi||v^f||\phi^*\phi)~.\ ] ] we next apply a so - called hubbard - stratonovich transformation @xcite to both the hartree and the fock parts of the interaction . first the hartree part . we note that @xmath282}$ ] can be written as a functional integral over the four real fields contained in @xmath283 as @xmath284~ \exp \left\ { \frac{1}{2\hbar } ( \kappa|{v^h}^{-1}|\kappa ) - \frac{1}{2\hbar } ( \kappa|\phi^*\phi ) - \frac{1}{2\hbar } ( \phi^*\phi|\kappa ) \right\}~. \nonumber\end{aligned}\ ] ] if we put @xmath285 = 0 $ ] for a moment , we can substitute this equality in the partition function to obtain @xmath286d[\phi^*]d[\phi ] e^{-s[\kappa,\phi^*,\phi]/\hbar}$ ] with the action @xmath287 = - \frac{1}{2 } ( \kappa|{v^h}^{-1}|\kappa ) + \sum_{\alpha,\alpha ' } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \phi^*_{\alpha}({\bf x},\tau ) \left\ { \left ( \hbar \frac{\partial}{\partial \tau } \right . \right . \hspace*{0.4 in } \\ & & \hspace*{1.0 cm } \left . \left . - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) + \epsilon_{\alpha } - \mu \right ) \delta_{\alpha,\alpha ' } + \kappa_{\alpha,\alpha'}({\bf x},\tau ) \right\ } \phi_{\alpha'}({\bf x},\tau)~. \nonumber\end{aligned}\ ] ] we see that in this manner the action for the fermions has become quadratic with a selfenergy @xmath288 . hence , we can now integrate out the fermion fields to obtain @xmath286 e^{-s^{\rm eff}[\kappa]/\hbar}$ ] and @xmath289 = - \frac{1}{2 } ( \kappa|{v^h}^{-1}|\kappa ) -\hbar { \rm tr}[\ln ( -g^{-1})]~,\ ] ] with @xmath290 also a functional of @xmath291 . up to now we have not made any approximations and have only performed an exact rewriting of the partition function . however , the action @xmath292 $ ] contains all powers of the fields @xmath291 and is thus quite complicated . it may therefore appear that we have not made any progress . this is not the case , because the partition function will be largest for configurations that minimize the action @xmath292 $ ] . to make use of this observation , we expand the action around its minimum , i.e. , we put @xmath293 and require that @xmath294/\delta\kappa_{\alpha,\alpha'}({\bf x},\tau ) |_{\kappa=\langle\kappa\rangle } = 0 $ ] . if we neglect the fluctuations , we find that @xmath295/\hbar}$ ] . this will turn out to be the hartree approximation , as we will show now . first of all we have @xmath296 substituting this in equation ( [ haction ] ) gives us for the terms linear in the fluctuations @xmath297 - ( \delta\kappa|{v^h}^{-1}|\langle\kappa\rangle)~. \nonumber\end{aligned}\ ] ] if @xmath298 is indeed a minimum of the action @xmath292 $ ] , these linear terms have to vanish , which implies that @xmath299 as promised , this is precisely the most general mathematical expression for the hartree contribution to the selfenergy in figure 9 . taking again @xmath300 , we also have that the hartree approximation to the one - particle propagator obeys @xmath301 , with @xmath302 the average total atomic density , and we obtain the usual hartree selfenergy @xmath303 now we again want to include the fock part of the interaction and treat this also by a hubbard - stratonovich transformation . this requires introducing four real fields that depend on two spatial coordinates . they are denoted by @xmath304 . schematically we then use @xmath305~ \exp \left\ { - \frac{1}{2\hbar } ( \lambda||{v^f}^{-1}||\lambda ) + \frac{1}{2\hbar } ( \lambda||\phi^*\phi ) + \frac{1}{2\hbar } ( \phi^*\phi||\lambda ) \right\}~ , \nonumber\end{aligned}\ ] ] which leads to the total selfenergy @xmath306 \delta(\tau-\tau')\ ] ] and after integration over the fermion fields to the effective action @xmath307 = - \frac{1}{2 } ( \kappa|{v^h}^{-1}|\kappa ) + \frac{1}{2 } ( \lambda||{v^f}^{-1}||\lambda ) - \hbar { \rm tr}[\ln ( -g^{-1})]~.\ ] ] requiring now that also @xmath308/ \delta\lambda_{\alpha,\alpha'}({\bf x},{\bf x}',\tau ) |_{\lambda=\langle\lambda\rangle } = 0 $ ] , then leads to the expected fock selfenergy @xmath309 in the symmetric case it simply becomes @xmath310 with @xmath311 and @xmath312 exactly the hartree - fock approximation to the off - diagonal part of the one - particle density matrix . note that to actually perform the hartree - fock calculation in the latter case , we need to be able to determine the green s function @xmath313 . the easiest way to do so , is by realizing that it is the green s function of the operator in the fermionic piece of the action @xmath314 $ ] obtained after the hubbard - stratonovich transformations . if we diagonalize this operator by solving the eigenvalue problem @xmath315 the desired one - particle propagator aquires the ideal gas form @xmath316 with new one - particle energies @xmath317 and eigenstates @xmath318 that incorporate the average effect of the interactions of an atom with all the other atoms in the gas . clearly , in the cases that the eigenstates are not affected by these so - called mean - field effects , we recover the weak - coupling results of the previous section . although we have thus precisely reproduced our diagrammatic result , there are two important advantages in using the hubbard - stratonovich transformation . first , it is in principle exact , and allows us to also calculate corrections to the hartree - fock approximation . for example , if we expand @xmath319 $ ] up to quadratic order in @xmath320 and @xmath321 , and neglect all higher orders , we find the so - called generalized random phase approximation . the latter approach actually gives us also the opportunity to study the density fluctuations and therefore the collective excitations of the gas . we return to this important topic in section [ modes ] . second , it allows for a beautiful way to describe phase transitions . as mentioned previously , in perturbation theory we always find for a homogeneous gas that @xmath322 due to the same feature of @xmath323 . from a fundamental point of view this is a result of the translational invariance of the gas and of the rotational symmetry in spin space . however , we can imagine that in principle we can also find selfconsistent solutions that do not have this property . we then have a spontaneous breaking of symmetry and therefore a phase transition in our system . for example , if below a certain temperature @xmath324 , we are dealing with a transition to a charge density wave or ( wigner ) crystal . if on the other hand @xmath325 the gas is in a ferromagnetic phase . for a spin - density wave we even have that @xmath326 . in all these cases the hubbard - stratonovich approach used above leads in a natural way to the appropriate landau theory of the phase transition . since the landau theory is also very useful for the understanding of superfluidity in atomic gases , which is clearly the primary goal of the present course , we use the next section to give a short introduction into this subject . in this summary of the landau theory of phase transitions , we restrict ourselves in first instance to the homogeneous case because this makes the discussion more transparent . however , at the end of the section we also briefly mention how the inhomogeneity enters the theory . we have seen , for example in the hartree - fock theory discussed above , that the green s function @xmath327 can have a nonzero value of the average spin density @xmath328 , which is usually called the magnetization @xmath329 in this context . this signals a phase transition to a ferromagnetic phase and the magnetization is called the order parameter of this transition . in the previous section we have also seen how we can , in principle exactly , obtain an expression for the partition function as a functional integral over the field @xmath330 , i.e. , @xmath331~e^{-s^{\rm eff}[{\bf s}]/\hbar}~.\ ] ] we just have to integrate out the fields @xmath332 and @xmath333 . for space and time independent values of the magnetization @xmath5 the action @xmath334 $ ] is @xmath335 , where @xmath336 is the landau free - energy density and @xmath337 is the total volume of system . because of the symmetry of @xmath248 $ ] under rotations of the spin , this free - energy density must also be rotationally invariant and can therefore only depend on the magnitude of @xmath5 , which we denote by @xmath338 . if a phase transition occurs the behaviour of @xmath339 can essentially fall only into two categories . at temperatures very high compared to the critical temperature @xmath340 , the system is fully disorded and the free energy @xmath339 must have a single minimum at @xmath341 to make sure that the order parameter @xmath342 is zero . bringing the temperature closer to @xmath340 , however , the free energy can develop a second local minimum . as long as the free energy in this second local minimum is higher than the minimum at @xmath341 , the equilibrium value of @xmath342 will still be zero and no phase transition has occured . lowering the temperature further , the value of the free energy in the second minimum decrease until , precisely at the critical temperature @xmath340 , it is equal to the free energy at @xmath341 . for temperatures below this critical one the second minimum has actually become the global minimum of the free energy , which implies that @xmath343 and we are in the ordered phase . in this scenario the order parameter has always a discontinuity at the critical temperature . as a result , this corresponds to a discontinuous , or first - order , phase transition . the behaviour of the free energy and the order parameter is illustarted in figure 10 . this should be compared with the behaviour of the free energy and the order parameter for a continuous , i.e. , second or higher order , phase transition , which is quite different and depicted in figure 11 . now the landau free energy @xmath339 has always a single minimum , which for temperatures above @xmath340 is at @xmath341 but at temperatures below the critical temperature shifts to a nonzero value of @xmath338 . in particular , the order parameter thus shows no discontinuity at @xmath340 . in the case of a second - order phase transition , we conclude that near the critical temperature @xmath342 is very small . as a result we can then expand the landau free - energy density in powers of @xmath338 . because the free energy must also be symmetric under @xmath344 , we have @xmath345 with @xmath346 and @xmath347 . thus if @xmath348 we have @xmath349 and @xmath350 . but for @xmath351 , @xmath352 becomes negative and we have @xmath353 note that the free - energy density in this minimum is @xmath354 and has a discontinuity in its second derivative with respect to the temperature . historically , this is the reason why the corresponding phase transition was named to be of second order . in fact , the landau theory of second order phase transitions is slightly more involved , because it also considers slow spatial fluctuations in @xmath5 . since a uniform rotation of @xmath5 costs no energy , we must have that the free energy is @xmath355 = \int d{\bf x}~ \left ( \gamma ( \mbox{\boldmath $ \nabla$ } { \bf s}({\bf x}))^2 + \alpha(t ) |{\bf s}({\bf x})|^2 + \frac{\beta}{2 } |{\bf s}({\bf x})|^4 \right)\ ] ] and @xmath356~e^{-\beta f_l[{\bf s}]}$ ] . landau theory now esentially corresponds to minimizing @xmath357 $ ] and taking only quadratic fluctuations into account . taking also higher order fluctuations into account turns out to be very difficult . it requires renormalization group methods , which we are not going to treat here and that can be found in the literature @xcite . the effects of these fluctuation corrections to landau theory are generally known as critical phenomena . interestingly , they do not occur in first - order phase transitions . more importantly for our purposes is , however , that for atomic gases the fluctuations are only important in a small temperature interval around the critical temperature . for many applications it is , therefore , possible to neglect them . what usually can not be negected is the effect of the inhomogeneity of the gas . in the context of landau theory that implies that the coefficients @xmath352 , @xmath358 and @xmath359 in equation ( [ lfe ] ) become also dependend on the spatial position in the trapping potential . we will see several examples of this feature in the following . we finally want to consider two important second - order phase transitions , that are purely due to quantum effects and can be conveniently treated with the methods that we have developed sofar . moreover , they occur very often in nature , for example in metals , in liquid helium and recently of course also in atomic gases of rubidium @xcite , lithium @xcite , sodium @xcite and hydrogen @xcite . let us first consider the last case , which is associated with a gas of spin - less bosons . for the low temperatures of interest the action is @xmath151 \\ & & = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \phi^*({\bf x},\tau ) \left\ { \hbar \frac{\partial}{\partial\tau } - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu \right\ } \phi({\bf x},\tau ) \nonumber \\ & & + \frac{1}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x } ~ v_{{\bf 0 } } \phi^*({\bf x},\tau ) \phi^*({\bf x},\tau ) \phi({\bf x},\tau ) \phi({\bf x},\tau)~ , \nonumber\end{aligned}\ ] ] where @xmath271 is taken to be equal to @xmath360 . the justification for this simplificaties is , roughly speaking , that the thermal de broglie wavelength @xmath361 of the atoms is much larger than the range of the interatomic interaction . in this system the phase transition of interest is bose - einstein condensation . the associated order parameter is @xmath362 , since for time independent @xmath363 the above action has presicely the form of a landau theory with a ` free - energy ' of @xmath364 \\ & & = \int d{\bf x}~ \left ( \frac{\hbar^2}{2 m } |\mbox{\boldmath $ \nabla$}\phi({\bf x})|^2 + ( v^{\rm ex}({\bf x})-\mu ) |\phi({\bf x})|^2 + \frac{v_{{\bf 0}}}{2 } |\phi({\bf x})|^4 \right)~. \nonumber\end{aligned}\ ] ] we conclude that in lowest order the critical temperature is determined by @xmath365 , because then the configurations @xmath366 precisely make a vanishing contribution to the quadratic part of the ` free energy ' . this condition makes sense , because it is exactly the condition that we have a bose - einstein condensation in the ideal case : in the ideal bose gas the number of particles in the one - particle ground state is @xmath367 , which indeed diverges for @xmath368 . to determine the corrections to this result we now explicitly substitute @xmath369 into our functional integral . it is for lateron important to realize that to consistently define the fluctuations @xmath370 in this manner , we also have to require that @xmath371 the physical reason behind this condition is that @xmath370 should contain all the configurations that are orthogonal to @xmath372 . in principle , therefore , it should also contain configurations that in effect only multiply @xmath372 by a global phase . such fluctuations lead to the phenomena of phase ` diffusion ' and are discussed in section [ pd ] . in full detail we find after the above substitution that @xmath373 = \hbar\beta f_l[\phi_{\bf 0}^*,\phi_{\bf 0 } ] + s_0[{\phi'}^*,\phi ' ] + s_{\rm int}[{\phi'}^*,\phi']~,\ ] ] where the linear and quadratic terms are given by @xmath374 \\ & & = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ { \phi'}^*({\bf x},\tau ) \left\ { - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu + v_{{\bf 0 } } |\phi_{\bf 0}({\bf x})|^2 \right\ } \phi_{\bf 0}({\bf x } ) \nonumber \\ & & + \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \phi'({\bf x},\tau ) \left\ { - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu + v_{{\bf 0 } } |\phi_{\bf 0}({\bf x})|^2 \right\ } \phi_{\bf 0}^*({\bf x } ) \nonumber \\ & & + \int_0^{\hbar\beta } d\tau \int d{\bf x}~ { \phi'}^*({\bf x},\tau ) \nonumber \\ & & \hspace*{0.85 in } \times \left\ { \hbar \frac{\partial}{\partial\tau } - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu + 2v_{{\bf 0}}|\phi_{\bf 0}({\bf x})|^2 \right\ } \phi'({\bf x},\tau ) \nonumber \\ & & + \frac{1}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } ( \phi_{\bf 0}({\bf x}))^2 { \phi'}^*({\bf x},\tau ) { \phi'}^*({\bf x},\tau ) \nonumber \\ & & + \frac{1}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } ( \phi_{\bf 0}^*({\bf x}))^2 \phi'({\bf x},\tau ) \phi'({\bf x},\tau)~ , \nonumber\end{aligned}\ ] ] and the cubic and quartic terms by @xmath375 = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } \phi_{\bf 0}({\bf x } ) { \phi'}^*({\bf x},\tau ) { \phi'}^*({\bf x},\tau ) \phi'({\bf x},\tau ) \hspace{0.4 in } \\ & & \hspace*{0.1 in } + \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } \phi_{\bf 0}^*({\bf x } ) { \phi'}^*({\bf x},\tau ) \phi'({\bf x},\tau ) \phi'({\bf x},\tau ) \nonumber \\ & & \hspace*{0.1 in } + \frac{1}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } { \phi'}^*({\bf x},\tau ) { \phi'}^*({\bf x},\tau ) \phi'({\bf x},\tau ) \phi'({\bf x},\tau)~ , \nonumber \end{aligned}\ ] ] respectively . in the bogoliubov approximation we neglect the last three interaction terms @xcite . furthermore , to make sure that @xmath376 or @xmath377 , we need to require that the terms linear in @xmath378 and @xmath379 drop out of the action @xmath380 $ ] . clearly , this implies that @xmath381 which is the same result as obtained from minimizing the landau ` free - energy ' @xmath382 $ ] . in the context of trapped atomic gases , this equation is known as the gross - pitaevskii equation @xcite . it determines the macroscopic wave function of the condensate . the reason for calling @xmath372 the macroscopic wave function follows from the fact that the total density of the gas now obeys @xmath383 the total number of condensate atoms thus equals @xmath384 . as equation ( [ dens ] ) shows , it is in general always smaller that the total number of atoms in the gas due to the effect of the fluctuations . note that in our present formulation the average @xmath385 physically describes not only the depletion of the condensate due to the usual thermal fluctuations known from the ideal bose gas , but also due to the interactions , i.e. , purely due to quantum fluctuations . assuming that we have solved for the gross - pitaevskii equation , the fluctuation corrections are in the bogoliubov theory determined by a quadratic action of the form @xmath386 \\ & & = - \frac{\hbar}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left [ { \phi'}^*({\bf x},\tau ) , \phi'({\bf x},\tau ) \right ] \cdot { \bf g}^{-1 } \cdot \left [ \begin{array}{c } \phi'({\bf x},\tau ) \\ { \phi'}^*({\bf x},\tau ) \end{array } \right]~ , \nonumber\end{aligned}\ ] ] where the associated green s function @xmath387 has now a matrix structure because not only the normal average @xmath388 but also the anomalous average @xmath389 is now unequal to zero . we thus have that @xmath390 \cdot \left [ { \phi'}^*({\bf x}',\tau ' ) , \phi'({\bf x}',\tau ' ) \right ] \right\rangle~.\ ] ] from equation ( [ s0 ] ) we in fact find that in the bogoliubov approximation @xmath391 \delta({\bf x}-{\bf x } ' ) \delta(\tau-\tau')~. \nonumber\end{aligned}\ ] ] with the noninteracting green s function @xmath392 defined by @xmath393\ ] ] and , of course , @xmath394 this is clearly only the lowest order result for the green s function , because a perturbative treatment of @xmath395 $ ] leads , in the same way as in section [ ifd ] , to higher order corrections . in general , the dyson equation is , however , always of the form @xmath396^{-1 } = \left [ \begin{array}{cc } g_0^{-1 } & 0 \\ 0 & g_0^{-1 } \end{array } \right ] - \left [ \begin{array}{cc } \sigma_{11 } & \sigma_{12 } \\ \sigma_{21 } & \sigma_{22 } \end{array } \right]~.\ ] ] the off - diagonal elements are again called anomalous , since they vanish in the normal phase of the gas . diagrammatically the dyson equations for @xmath397 and @xmath398 are shown in figure 12 . the selfenergy in the bogoliubov approximation is @xmath399 \delta({\bf x}-{\bf x } ' ) \delta(\tau-\tau')~.~~\end{aligned}\ ] ] because we have completely negelected the cubic and quartic terms in the action to obtain this selfenergy , the bogoliubov approach is only valid if the fluctuations are sufficiently small . physically , this impies that the deplection of the condensate must be small . it can , therefore , not be applied to liquid helium , but is expected to be valid for a weakly - interacting atomic gas at such low temperature that it essentially only consists of a condensate . under these conditions the bogoliubov theory not only predicts the condensate density profile @xmath400 but also the collective modes of the condensate . both these predictions have been accurately verified in recent experiments @xcite . theoretically , the eigenfrequencies of the collective modes are again determined by the poles in the one - particle green s function . to understand more clearly how these can be determined , we first consider a homogeneous bose gas . in a box with volume @xmath401 the one - particle states are most easily characterized by the wavevector @xmath402 and equal to @xmath403 . the one - particle energies are thus @xmath404 . moreover , the gross - pitaevskii equation reduces to @xmath405 in that case , because the landau ` free - energy ' is minimized for a macroscopic wave function that is independent of the position in the box . making use of this fact , equation ( [ gb ] ) can immediately be solved by a fourier transformation . the result is @xmath406~. \nonumber\end{aligned}\ ] ] clearly there are poles in @xmath407 if the determinant of the right - hand side is zero or if @xmath408 this is the famous bogoliubov dispersion of the collective excitations . note that to finish the calculation we still have to obtain the condensate density @xmath409 . this is determined by the total density of the gas , which obeys @xmath410 with @xmath411 the density of the ` above ' condensate particles . by inverting the right - hand side of equation ( [ gbf ] ) , we find that the noncondensate density is equal to @xmath412 for a given density and temperature , the last two equations thus fully determine the condensate density . note that equation ( [ ncd ] ) explicitly shows that the condensate is indeed depleted both by thermal as well as quantum effects . the generalization to the inhomogeneous case is straightforward . first we again have to solve the gross - pitaevskii equation at a fixed chemical potential . given the condensate wave function , we can then calculate the collective mode frequencies by finding the poles of @xmath387 , or equivalently but more conveniently , the zero s of @xmath413 . clearly , the latter are located at @xmath414 , where @xmath415 is found from the eigenvalue problem @xmath416 \cdot \left [ \begin{array}{c } u_{\bf n}({\bf x } ) \\ v_{\bf n}({\bf x } ) \end{array } \right ] \hspace*{0.4 in } \\ = \hbar\omega_{\bf n } \left [ \begin{array}{cc } 1 & 0 \\ 0 & - 1 \end{array } \right ] \cdot \left [ \begin{array}{c } u_{\bf n}({\bf x } ) \\ v_{\bf n}({\bf x } ) \end{array } \right]~ \nonumber \end{aligned}\ ] ] and we introduced the operator @xmath417 . this is de bogoliubov - de gennes equation that has recently been applied with great succes to the collective modes of a bose condensed rubidium and sodium gas @xcite . note that a special solution with @xmath418 is given by @xmath419 = [ \phi_{\bf 0}({\bf x}),-\phi^*_{\bf 0}({\bf x})]$ ] . as a result we can , by making use of the fact that the left - hand side of the bogoliubov - de gennes equation involves a hermitian operator , easily proof that all the solutions obey @xmath420 as required by the condition in equation ( [ ort ] ) . moreover , we can similary show that the solutions with @xmath421 can always be normalized as @xcite @xmath422 physically , the zero frequency solution @xmath423 describes the dynamics of the global phase of the condensate @xcite . because of the so - called @xmath424 symmetry of the action , i.e. , its invariance under the simultaneous phase changes @xmath425 and @xmath426 , this solution is essentially of no importance to the thermodynamic properties of a macroscopic gas sample and is therefore usually neglected . nevertheless , it has from a fundamental point of view some interesting consequences , as we will see in section [ pd ] . knowing all the eigenstates of @xmath413 we can then easily perform the inversion and finally again determine the density profile of the noncondensed atoms . keeping the physical significance of the zero frequency mode in mind , we ultimately find , @xmath427 which may be compared with equation ( [ ncd ] ) . at temperatures near absolute zero , we have as a good approximation that @xmath428 and the bogoliubov theory applies . however at nonzero temperatures we thermally excite particles and @xmath429 becomes nonzero . if we treat the effect of the noncondensate part of the gas in the hartree - fock approximation , we find that @xmath395 $ ] on average adds @xmath430 = 2 \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } n'({\bf x } ) { \phi'}^*({\bf x},\tau ) \phi_{\bf 0}({\bf x } ) \\ & & + 2 \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } n'({\bf x } ) \phi'({\bf x},\tau ) \phi_{\bf 0}^*({\bf x } ) \nonumber \\ & & + 2 \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{{\bf 0 } } n'({\bf x } ) { \phi'}^*({\bf x},\tau ) \phi'({\bf x},\tau ) ~ , \nonumber \end{aligned}\ ] ] to the action @xmath380 $ ] . in figure 13 we indicate how this can be understood diagramatically . performing the same analysis as before , we conclude that the gross - pitaevskii equation is modified to @xmath431 and the normal selfenergies are changed into @xmath432 . the bogoliubov - de gennes equation for the elementary excitations is , therefore , now given by @xmath433 \cdot \left [ \begin{array}{c } u_{\bf n}({\bf x } ) \\ v_{\bf n}({\bf x } ) \end{array } \right ] \hspace*{0.4 in } \\ = \hbar\omega_{\bf n } \left [ \begin{array}{cc } 1 & 0 \\ 0 & - 1 \end{array } \right ] \cdot \left [ \begin{array}{c } u_{\bf n}({\bf x } ) \\ v_{\bf n}({\bf x } ) \end{array } \right]~. \nonumber \end{aligned}\ ] ] these last two equations in combination with equation ( [ ncda ] ) are known as the popov theory in the recent literature @xcite . it is much studied at present in the context of bose - einstein condensation in atomic gases , and has been applied with succes to the equilibrium density profile of the gas below the critical temperature @xcite . it has also been used to determine the collective mode frequencies of the gas at nonzero temperatures , however , with much less success @xcite . the reason for the failure of the popov theory in this case is that the bogoliubov - de gennes equation in equation ( [ bdgt ] ) describes physically only the motion of the condensate in the presence of a static noncondensed cloud and not the dynamics of the noncondensed cloud itself . how that can be achieved is discussed lateron in section [ modes ] , when we have obtained a better understanding of the nonequilibrium theory . now we briefly want to make a connection to the interaction parameter @xmath434 used in the bogoliubov and popov theories and the specific two - body scattering properties of the atomic gas of interest in a particular experiment . in the previous section we mentioned that the bogoliubov theory , and to a certain extent also the popov theory , agree very well with experiment . it is clear , however , that to apply these theories to an actual experiment we need to know the interaction parameter @xmath434 . in the case of atomic gases it is indeed possible to perform an _ ab initio _ calculation of this quantity , something which for instance can not be done for liquid helium . the reason why @xmath434 can be determined for atomic gases , is that under the experimental conditions of interest the densities are always so low that we only need to consider all two - body processes taking place in the gas and we can neglect three - body and higher - body processes . this implies physically that we only have to calculate and add the quantum mechanical amplitudes for two atoms to scatter of each other an arbitrary number of times . diagrammatically the procedure can essentially be summarized by the t - matrix equation in figure 14 , because by iteration of this equation we easily see that we are indeed summing all two - body interacion processes . denoting the total momentum of the two incoming particles by @xmath435 and the sum of the two matsubara frequencies by @xmath436 , the t - matrix equation in figure 14 can , with our knowledge of how to perform the matsubara sum in the right - hand side , easily be shown to be mathematically equivalent to @xmath437 it is immediately solved by @xmath438 note that this is a somewhat formal result , because the momentum sum in the right - hand side has an ultra - violet divergence , which from a field - theoretical point of view requires a renormalization procedure . the physical reason for the divergence is that we have used the approximate potential @xmath439 instead of the actual interactomic potential @xmath271 . the argument for using the @xmath440-function approximation was that the thermal de broglie wavelength of the atoms is for the ultra - low temperatures of interest always much larger than the typical range of the interaction . we now see that this argument is not fully correct , because if we calculate corrections in perturbation theory , we have to deal with momentum sums which are not always restricted to momenta of order @xmath441 and are therefore sensitive to the precise details of the interaction potential . to cure this disease we note that if we put in equation ( [ tmatrix ] ) the bose occupation numbers equal to zero , we precisely get the t - matrix or lippmann - schwinger equation @xcite for two atoms interacting with the potential @xmath439 . furthermore , if the atoms interact with the potential @xmath271 the solution of the corresponding lippmann - schwinger equation is known from elementary scattering theory to be equal to @xmath442 for small incoming momenta and energies , where @xmath443 is the @xmath338-wave scattering length . we thus conclude that we must interpret @xmath444 as being equal to @xmath445 and , therefore , that the desired t - matrix in principle obeys @xmath446 to obtain an accurate theory for a trapped atomic bose gas that includes the effect of all two - body processes , we should now use instead of @xmath434 the above t - matrix in the bogoliubov or popov theories . of course , in the normal phase the same is true for the hartree - fock theory . if the temperature is not too close to the critical temperature for bose - einstein condensation , it turns out that @xmath447 and we arrive at the conclusion that for an application to realistic atomic gases we must replace everywhere in section [ sf ] the interaction parameter @xmath434 by @xmath442 . for temperatures close to the critical temperature , this is however no longer true and the bose distribution functions have a nonnegligible effect of the t - matrix @xcite . the theory that , by using ` dressed ' one - particle propagators in figure 14 , selfconsistently includes these effects of the medium on the scattering properties of the atoms is known as the many - body t - matrix theory . it results in an effective interaction that , to avoid confusion , is denoted by @xmath448 in section [ cf ] . as we will see then , it plays a crucial role if we want to arrive at a consistent discription of condensate formation in an atomic bose gas . finally , we want to briefly discuss the bardeen - cooper - schrieffer or bcs theory of superconductivity @xcite , which has received considerable attention recently in connection with ongoing experiments with the fermionic isotope of lithium . the reasons for this will become clear shortly . we are in that case dealing with effective spin @xmath449 fermions and the action is taken to be @xmath151 \\ & & = \sum_{\alpha=\uparrow,\downarrow } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \phi^*_{\alpha}({\bf x},\tau ) \left\ { \hbar \frac{\partial}{\partial\tau } - \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } - \mu \right\ } \phi_{\alpha}({\bf x},\tau ) \nonumber \\ & & + \int_0^{\hbar\beta } d\tau \int d{\bf x}~ v_{\bf 0 } \phi^*_{\uparrow}({\bf x},\tau ) \phi^*_{\downarrow}({\bf x},\tau ) \phi_{\downarrow}({\bf x},\tau ) \phi_{\uparrow}({\bf x},\tau)~. \nonumber\end{aligned}\ ] ] note that the use of a single chemical potential implies that we only consider the optimal case of an equal density in each hyperfine state . this situation is optimal in the sense that , for unequal densities in each hyperfine state , it is no longer possible to pair up all the atoms in the gas . as a result the critical temperature of the gas drops dramatically , and in general becomes experimentally inaccessible . the effect is to a large extent analogous to putting superfluid @xmath450he in an homogeneous magnetic field . furthermore , we in first instance consider the homogeneous case , because we want to illustrate in this section the local - density approximation to include the effect of the external trapping potential . physically , this approximation treats the gas as consisting of a large number of independent gases that are in diffusive equilibrium with each other . such an approach only works if the correlation length of the gas is much smaller than the typical length scale associated with changes of the external potential . fortunately , this is almost always the case for realistic trapped atomic gases and for that reason the local - density approximation is often used in practice . the bcs theory is the theory of bose - einstein condensation of so - called cooper pairs . this means that the order parameter is @xmath451 , in analogy with the order parameter @xmath362 for the bose case just discussed . furthermore , it requires that the interaction parameter @xmath434 is negative , since otherwise the formation of pairs would not be energetically favorable . from now on we assume , therefore , that this is the case . the condensate of cooper pairs can also be nicely treated with a hubbard - stratonovich transformation . we now introduce a complex field @xmath452 and use @xmath453d[\delta]~ \exp \left\ { \frac{1}{\hbar } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left ( \frac{|\delta({\bf x},\tau)|^2}{v_{\bf 0 } } \right . \nonumber \\ & & + \left . \left . \raisebox{0.2 in } { } \delta^*({\bf x},\tau ) \phi_{\downarrow}({\bf x},\tau ) \phi_{\uparrow}({\bf x},\tau ) + \phi^*_{\uparrow}({\bf x},\tau ) \phi^*_{\downarrow}({\bf x},\tau ) \delta({\bf x},\tau ) \right ) \raisebox{0.25 in } { } \right\}~. \nonumber\end{aligned}\ ] ] this leads to a partition function with the action @xmath454 = - \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \frac{|\delta({\bf x},\tau)|^2}{v_{\bf 0 } } \\ & & - \frac{\hbar}{2 } \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left [ \phi^*_{\downarrow}({\bf x},\tau ) , \phi_{\uparrow}({\bf x},\tau ) \right ] \cdot { \bf g}^{-1 } \cdot \left [ \begin{array}{c } \phi_{\downarrow}({\bf x},\tau ) \\ \phi^*_{\uparrow}({\bf x},\tau ) \end{array } \right]~ , \nonumber\end{aligned}\ ] ] where @xmath455 \delta({\bf x}-{\bf x } ' ) \delta(\tau-\tau')~ , \nonumber\end{aligned}\ ] ] and the noninteracting green s function @xmath392 is defined by @xmath456\ ] ] and @xmath457 we thus see that the fermionic part has exactly the same matrix structure as in the case of a condensed bose gas , only the selfenergy is now @xmath458 = \frac{1}{\hbar } \left [ \begin{array}{cc } 0 & \delta({\bf x},\tau ) \\ \delta^*({\bf x},\tau ) & 0 \end{array } \right ] \delta({\bf x}-{\bf x } ' ) \delta(\tau-\tau')~.~~\ ] ] if we again integrate out the fermion fields , we get the effective action @xmath459 = - \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \frac{|\delta({\bf x},\tau)|^2}{v_{\bf 0 } } - \hbar { \rm tr}[\ln ( -{\bf g}^{-1})]~,\ ] ] which we can expand in powers of @xmath460 by using @xmath461 and therefore @xmath462 = - \hbar { \rm tr}[\ln ( -{\bf g}_0^{-1 } ) ] + \hbar \sum_{m=1}^{\infty } \frac{1}{m } { \rm tr}[({\bf g}_0 \mbox{\boldmath $ \sigma$})^m]~.\ ] ] explicite calculation @xcite shows that for space and time independent @xmath460 we obtain a ` free - energy ' density of the form of the landau - theory of second - order phase transitions , i.e. , @xmath463 with as expected @xmath464 here @xmath465 is the density of states of a single spin state at the fermi energy @xmath466 and the critical temperature is given by @xmath467 if we use the same renormalization procedure as in section [ ap ] to eliminate the interaction parameter @xmath434 in favor of the negative @xmath338-wave scattering length @xmath443 . below @xmath340 we thus have a nonzero average @xmath468 . using this in the green s function for the fermions and neglecting fluctations , we find in momentum space that @xmath469\ ] ] and therefore poles in @xmath470 if @xmath471 this disperion relation has clearly a gap of magnitude @xmath472 at the fermi surface . as a consequence @xmath472 is known as the bcs gap parameter . note that the critical temperature below which the gap becomes nonzero depends exponentally on the parameter @xmath473 . for typical atomic gases this is a quantity which is much larger than one , and the bcs transition is experimentally inaccesible . the only exception at the moment appears to be @xmath474li , with its anomalously large and negative triplet scattering length of @xmath475 @xcite . this explains the present interest in spin - polarized atomic lithium . in view of this exciting possibility , we have recently studied the equilibrium properties of atomic lithium is a harmonic oscillator potential @xmath476 , with a trapping frequency of @xmath477 hz @xcite . to incorporate the effect of the external potential we have , as mentioned above , applied the local - density approximation . the result is shown in figure 15 . note that the use of the local - density approximation implies that we perform the above outlined homogeneous calculation for each point in space with a chemical potential that is equal to @xmath478 the third term in the right - hand side represents the mean - field effect of the hartree contribution to the selfenergy of the fermions . since our hubbard - stratonovich procedure is in principle exact , it is not immediately clear why such a term must be included in the theory . it can , however , be shown that it arises from the fluctuations of the bcs gap parameter . with this remark , we end our development of the equilibrium field theory of trapped atomic gases . we now turn our attention to the nonequilibrium field theory . and energy gap @xmath479 for a @xmath474li atomic gas consisting of @xmath480 atoms in each spin state a ) at @xmath481 nk , b ) at @xmath482 nk , slightly below @xmath340 , and c ) at @xmath483 nk . the left scale of each plot refers to the density and the right scale to the energy gap . the dotted line in c ) shows the density distribution for an ideal fermi gas with the same number of particles and at the same temperature . ] as we have already seen several times , the equlibrium field theory gives us also information on dynamical properties of the gas by means of the substitution @xmath484 or equivalently @xmath485 . however , the popov theory of bose - einstein condensation has shown us that we have to be very careful with this procedure , because we may not always end up with the correct physics needed for a discription of a particular experiment . it is for these cases that a truly nonequilibrium field theory is required and the development of such a theory is the topic of section [ qkt ] . before embarking on that , however , we first consider two important dynamical problems where the equilibrium theory does give us all the answers . in a bose gas with effectively attractive interactions , i.e. , with a negative scattering length @xmath443 , a condensate will always have the tendency to collapse to a high density state due to the gain in energy that can be obtained in this way @xcite . the most important question in this context is , therefore , if a condensate can exist sufficiently long to be experimentally observed . neglecting the variation of the noncondensate density on the size of the condensate @xcite , we know from the popov theory that the dynamics of the collapse is , apart from an unimportant shift in the chemical potential , determined by the gross - pitaevskii equation @xcite @xmath486 more precisely , this determines only the semiclassical dynamics . if we also want to study the quantum fluctuations , which is necessary if we are also interested in how the condensate tunnels through the macroscopic energy barrier , it is most convenient to calculate the grand canonical partition function of the condensate @xcite . quantizing the gross - pitaevskii equation we obtain for this partition function the functional integral @xmath487d[\phi]~ \exp \left\ { - \frac{1}{\hbar } s[\phi^*,\phi ] \right\}~,\ ] ] over the complex field @xmath363 and with the euclidian action @xmath488 = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \phi^*({\bf x},\tau ) \left ( \hbar \frac{\partial}{\partial \tau } + \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } \right . \\ & & + \left . v^{\rm ex}({\bf x } ) - \mu + \frac{2\pi a\hbar^2}{m } |\phi({\bf x},\tau)|^2 \right ) \phi({\bf x},\tau)~. \nonumber\end{aligned}\ ] ] as always for bose systems , the integration is only over fields that are periodic on the imaginary time axis . although it has recently been shown by freire and arovas that the tunneling process can also be studied in terms of the complex field @xmath363 @xcite , we believe that it leads to somewhat more physical insight if we use instead the fields @xmath489 and @xmath490 that correspond to the density and phase fluctuations of the condensate , respectively . they are introduced by performing the canonical variable transformation @xcite @xmath491 in the functional integral for the partition function . as a result we find @xmath492d[\theta]~ \exp \left\ { - \frac{1}{\hbar } s[\rho,\theta;\mu ] \right\}~,\ ] ] with @xmath493 = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left ( i\hbar \rho({\bf x},\tau ) \frac{\partial \theta({\bf x},\tau)}{\partial \tau } \right . \\ & & + \frac{\hbar^2 \rho({\bf x},\tau)}{2 m } ( \mbox{\boldmath $ \nabla$ } \theta({\bf x},\tau))^2 + \frac{\hbar^2}{8m\rho({\bf x},\tau ) } ( \mbox{\boldmath $ \nabla$ } \rho({\bf x},\tau))^2 \nonumber \\ & & + v^{\rm ex}({\bf x})\rho({\bf x},\tau ) - \mu\rho({\bf x},\tau ) + \left . \frac{2\pi a\hbar^2}{m } \rho^2({\bf x},\tau ) \right)~. \nonumber\end{aligned}\ ] ] next , we notice that this action is only quadratic in the phase fluctuations . the field @xmath490 can therefore be integrated over exactly , because it only involves the evaluation of a gaussian integral . compared to ordinary gaussian integrals there is , however , one slight complication which is associated with the fact that @xmath490 are phase variables . this implies that the periodicity of the original field @xmath363 only constraints the phase field @xmath490 to be periodic up to a multiple of @xmath494 . to evaluate the grand canonical partition function in equation ( [ gr ] ) we must therefore first integrate over all fields @xmath490 that obey the boundary condition @xmath495 and subsequently sum over all possible integers @xmath12 . because these different boundary conditions only affect the zero - momentum part of @xmath490 we first have to evaluate the sum @xmath496~ \exp \left\ { -i \int_0^{\hbar\beta } d\tau~ n_{\bf 0}(\tau ) \frac{\partial \theta_{\bf 0}(\tau)}{\partial \tau } \right\}~ , \nonumber\end{aligned}\ ] ] with @xmath497 the number of condensate particles . after performing a partial integration on the integral in the exponent , we can carry out the path integration over @xmath498 to obtain @xmath499~. \nonumber\end{aligned}\ ] ] as expected , the integration over the global phase of the condensate leads to the constraint of a constant number of condensate particles , i.e. , @xmath500 . moreover , we have @xmath501 , which restricts the number of condensate particles to an integer . putting all these results together , we see that the integration over the zero - momentum part of @xmath489 is only a sum over the number of condensate particles and we have that @xmath502 here we introduced the canonical partition function of the condensate , which is apparently equal to the functional integral @xmath503d[\theta]~ \exp \left\ { - \frac{1}{\hbar } s[\rho,\theta;0 ] \right\}\ ] ] over all the nonzero momentum components of the density and phase fields . the integration over the nonzero momentum components of the phase field @xmath490 is easily performed , because it now involves an ordinary gaussian integral . introducing the green s function for the phase fluctuations @xmath504 by @xmath505 we immediately obtain the desired effective action for the density field @xmath506 = \int_0^{\hbar\beta } d\tau \int d{\bf x } \int d{\bf x}'~ \left(-\frac{\hbar}{2 } \frac{\partial \rho({\bf x},\tau ) } { \partial \tau } g({\bf x},{\bf x}';\rho ) \frac{\partial \rho({\bf x}',\tau ) } { \partial \tau } \right ) \\ & & + \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left ( \frac{\hbar^2}{8m\rho({\bf x},\tau ) } ( \mbox{\boldmath $ \nabla$ } \rho({\bf x},\tau))^2 + v^{\rm ex}({\bf x } ) \rho({\bf x},\tau ) \right . \nonumber \\ & & \hspace*{2.4 in } + \left . \frac{2\pi a\hbar^2}{m } \rho^2({\bf x},\tau ) \right)~. \nonumber\end{aligned}\ ] ] being an action for the density fluctuations of the condensate , @xmath507 $ ] also describes all the collisionless modes of the condensate . this is important for our purposes , because the mode which becomes unstable first , determines precisely how the condensate collapses . moreover , it determines the probability with which the collapse is going to take place , both for quantum and thermal fluctuations , since the energy barrier is smallest in that direction of the configuration space . it should be noted that as long as we can neglect the interaction between the condensate and the thermal cloud , the action @xmath507 $ ] describes also the collective modes of a gas with positive scattering length . for various other theoretical approaches that have been applied under these conditions see , for example , refs . the actual measurements of the collective mode frequencies have been performed by jin _ _ @xcite and mewes _ et al . _ @xcite and are at sufficiently low temperatures indeed in good agreement with the theoretical predictions @xcite . we expect the same to be true for a gas with effectively attractive interactions and , therefore , the action @xmath507 $ ] to be a good starting point for the following discussion . to obtain the collisionless modes explicitly we consider first the case of an ideal bose gas by putting @xmath508 . for the ideal bose gas we expect the gaussian profile @xmath509 to describe an exact mode of the condensate . the reason is that in the noninteracting case we can make a density fluctuation by taking one particle from the condensate and putting that in one of the excited states of the external potential . the corresponding density fluctuation obeys @xmath510 for the experimentally relevant case of an isotropic harmonic oscillator @xcite it is more convenient to use instead of the cartesian quantum numbers @xmath164 , the two angular momentum quantum numbers @xmath511 and @xmath4 and the quantum number @xmath512 that counts the number of nodes in the radial wave function @xmath513 . the density fluctuation then becomes @xmath514 with @xmath515 the excitation energy and @xmath516 the size of the condensate wave function . comparing this now with the expansion of the gaussian profile in equation ( [ gprof ] ) around the groundstate density profile , which is obtained by substituting @xmath517 , we find that @xmath518 has precisely the same form as a density fluctuation in which one particle is taken from the condensate and put into the harmonic oscillator state with quantum numbers @xmath519 . the frequency of this so - called ` breathing ' mode described by the gaussian density profile must therefore be equal to @xmath520 . to proof that this is indeed correct , we need to evaluate the effective action @xmath507 $ ] , and hence the green s function @xmath504 , for a gaussian density profile . substituting such a profile in equation ( [ geq ] ) immediately leads to @xmath521 , with @xmath522 the latter equation can be solved , if we can solve the eigenvalue problem @xmath523 this turns out to be an easy task , because substituting @xmath524 gives essentially the radial schrdinger equation for an isotropic harmonic oscillator with frequency @xmath525 , i.e. , @xmath526 the desired eigenfunctions are therefore @xmath527 , with @xmath528 the properly normalized harmonic oscillator states with the energies @xmath529 , and the corresponding eigenvalues are @xmath530 . introducing finally the ` dual ' eigenfunctions @xmath531 , the green s function @xmath532 is given by @xmath533 note that prime on the summation sign indicates that the sum is over all quantum numbers except @xmath534 . the latter is excluded because the associated eigenfunction @xmath535 is just a constant and thus does not contribute to @xmath504 , which is defined as the green s function for all phase fluctuations with nonvanishing momenta . putting all these results together , we see that the dynamics of the collective variable @xmath536 is determined by the action @xmath537 = \int_0^{\hbar\beta } d\tau~ \left\ { \frac{3mn_{\bf 0}}{4 } \left ( \frac{dq}{d\tau } \right)^2 + n_{\bf 0 } \left ( \frac{3\hbar^2}{4mq^2 } + \frac{3}{4}m\omega^2q^2 \right ) \right\ } \\ & & \equiv \int_0^{\hbar\beta } d\tau~ \left\ { \frac{1}{2 } m^ * \left ( \frac{dq}{d\tau } \right)^2 + v(q ) \right\}~ , \nonumber\end{aligned}\ ] ] that is equivalent to the action of a particle with effective mass @xmath538 in a potential @xmath539 . as expected from our previous remarks , this potential has a minimum for @xmath540 and can be expanded near its minimum as @xmath541 it thus comfirms that the gaussian profile describes a breathing mode with frequency @xmath520 around an equilibrium density profile that is given by @xmath542 . our next task is to investigate how interactions affect this result . considering again only gaussian density profiles , the action @xmath543 $ ] is again that of a particle with effective mass @xmath538 but now in the potential @xcite @xmath544 the physically most important feature of this potential is that it is unbounded from below , since @xmath545 if @xmath546 . hence , the condensate indeed always has the tendency to collapse to the high - density state @xmath547 . however , if the number of condensate particles is sufficiently small , or more precisely if @xcite @xmath548 the condensate has to overcome a macroscopic energy barrier before it can collapse . under these conditions the condensate is therefore really metastable and can in principle be observed experimentally . the most important question in this respect is of course : how metastable is the condensate ? within the gaussian approximation this question is easily answered , because then the dynamics of the condensate is equivalent to the dynamics of a particle in an unstable potential , as we have just seen . we therefore only need to evaluate the wkb - expression for the tunnneling rate @xcite and compare this to the rate of decay due to thermal fluctuations by calculating also the height of the energy barrier . the outcome of this comparison for the conditions of the experiment with atomic @xmath549li is presented in reference @xcite and shows that , for the relatively high temperatures @xmath550 that have been obtained thusfar @xcite , the decay by means of thermal fluctuations over the energy barrier is the dominant decay mechanism of the condensate . more important for our purposes , however , is that sufficiently close to the maximum number of condensate particles @xmath551 the collective decay of the condensate discussed above is always much more probable than the decay due to two and three - body collisions that lead to a spin - flip or the formation of @xmath549li molecules , respectively . as a result the collapse of the condensate should be observable within the finite lifetime of the gas . in fact , on the basis of this separation of time scales we expect the condensate to go through a number of growth and collapse cycles @xcite . physically this picture arises as follows . starting from a gas with a number of atoms @xmath552 , the condensate will initially grow as a response to evaporative cooling . however , if the number of condensate atoms starts to come close to @xmath551 , the condensate fluctuates over the energy barrier and collapses in a very short time of @xmath553 @xcite . during the collapse the condensate density increases rapidly and two and three - body inelastic processes quickly remove almost all the atoms from the condensate . after this has occurred the condensate grows again from the noncondensed part of the gas and a new growth and collapse cycle begins . it is only after many of these cycles that enough atoms are removed for the gas to relax to an equilibrium with a number of condensate particles that is less than @xmath554 . this is shows quantitatively in figure 16 for the experimental conditions of interest . in response to evaporative cooling . the long - time decay of the condensate is due to two and three - body inelastic collisions . ] a final issue which needs to be addressed at this point is the actual dynamics of the collapse and , in particular , how we must include the effect of the inelastic growth and decay processes on this dynamics . unfortunately , the inclusion of these effects is rather complicated and we need the results of section [ cf ] to be able to correctly address this problem . it is , however , of some interest because sackett _ et al . _ have recently observed that after a single collapse there remains a remnant of the condensate with about 10% of the initial number of atoms @xcite . at present , it is an important open problem to theoretically understand the magnitude of this remnant . as we have just seen explicitly , a particularly interesting consequence of the finite size of the gas is that quantum fluctuations play a much more important role . although this is especially true for the case of attractive interactions that we considered in section [ nsl ] , it is also true for a bose gas with repulsive interactions . a striking example in this respect is the phenomenon of phase ` diffusion ' , which was recently discussed by lewenstein and you @xcite . we rederive their results for a trapped bose gas in a moment , but first consider also the same phenomenon for a neutral and homogeneous superconductor . in this manner it is possible to bring out the physics involved more clearly . using the approach of section [ sc ] , it can be shown that at zero temperature the dynamics of the superconducting order parameter , i.e. , the bcs gap parameter @xmath555 that is proportional to the wave function of the condensate of cooper pairs , is in a good approximation determined by a time - dependent ginzburg - landau theory @xcite with the action @xmath556 = \frac{n(0)}{4 } \int dt \int d{\bf x}~ \left\ { \frac{\hbar^2}{|\delta_0|^2 } \left| \frac{\partial \delta}{\partial t } \right|^2 \right . \\ & & - \left . \frac{\hbar^2 v_f^2}{3|\delta_0|^2 } \left| \mbox{\boldmath $ \nabla$ } \delta \right|^2 + 2 |\delta|^2 \left ( 1 - \frac{|\delta|^2}{2|\delta_0|^2 } \right ) \right\}~ , \nonumber\end{aligned}\ ] ] where @xmath465 is the density of states for one spin projection at the fermi energy @xmath557 and @xmath558 is the equilibrium value of the order parameter @xcite . writing the complex order parameter in terms of an amplitude and a phase , we immediately observe that the amplitude fluctuations are gapped @xcite and can , therefore , be safely neglected at large length scales . the long - wavelength dynamics of the superconductor is thus dominated by the phase fluctuations , according to the action @xmath559 = \frac{n(0)\hbar^2}{4 } \int dt \int d{\bf x}~ \left\ { \left ( \frac{\partial \theta}{\partial t } \right)^2 - \frac{v_f^2}{3 } \left ( \mbox{\boldmath $ \nabla$ } \theta \right)^2 \right\}~.\ ] ] this also implies that the global phase @xmath560 of the superconductor has a dynamics that is governed by @xmath561 = \frac{n(0)n\hbar^2}{4n } \int dt~ \left ( \frac{d \theta_{\bf 0}}{dt } \right)^2~,\ ] ] using the fact that the total volume @xmath337 of the system is given by @xmath562 . up to now our discussion has again been semiclassical . to consider also the quantum fluctuations , we have to quantize this theory by applying the usual rules of quantum mechanics . doing so , we find that the wave function of the overall phase obeys a schrdinger equation @xmath563 with a ` diffusion ' constant that can easily be shown to be equal to the quantity @xmath564 @xcite and is , most importantly for our purposes , proportional to @xmath565 . in the thermodynamic limit @xmath566 a state with a well defined stationary phase is clearly a solution and we are then dealing with a system having a spontaneously broken @xmath424 symmetry . however , for a finite ( and fixed ) number of particles the global phase can not be well defined at all times and always has to ` diffuse ' in accordance with the above schrdinger equation . note also that in the groundstate the phase is fully undetermined and @xmath567 . maybe surprisingly , the same calculation is somewhat more complicated for a bose gas because the amplitude fluctuations of the order parameter can not be neglected even at the largest length scales . however , taking these amplitude fluctuations into account properly , we nevertheless arrive at an action that is equivalent to equation ( [ phase ] ) and hence again leads to the phenomenon of phase ` diffusion ' . we start again from the action @xmath568 $ ] for the condensate . the difference with the previous subsection is , however , that now we are not so much interested in the dynamics of the density but in the phase dynamics instead . therefore we now want to integrate over the density field @xmath489 . this can not be done exactly and we therefore here consider only the strong - coupling limit , which was also treated by lewenstein and you @xcite . in that limit we are allowed to neglect the gradient of the average density profile @xcite and the action @xmath568 $ ] is for the longest wavelengths well approximated by @xmath569 = \int_0^{\hbar\beta } d\tau \int d{\bf x}~ \left ( i\hbar \rho({\bf x},\tau ) \frac{\partial \theta({\bf x},\tau)}{\partial \tau } \right . \\ & & + \left . v^{\rm ex}({\bf x})\rho({\bf x},\tau ) - \mu\rho({\bf x},\tau ) + \frac{2\pi a\hbar^2}{m } \rho^2({\bf x},\tau ) \right)~. \nonumber\end{aligned}\ ] ] in equilibrium the average density profile of the condensate thus obeys @xmath570 performing now the shift @xmath571 , we find for the zero - momentum part of the action @xcite @xmath572 = \hbar\beta e_{\bf 0}(\mu ) + \int_0^{\hbar\beta } d\tau \left ( i\hbar \delta n_{\bf 0 } \frac{d \theta_{\bf 0}}{d\tau } + \frac{2\pi a\hbar^2}{mv_{\bf 0}(\mu ) } ( \delta n_{\bf 0})^2 \right)~,\ ] ] where @xmath573 and @xmath574 correspond , respectively , to the energy and the volume of the condensate in the so - called thomas - fermi approximation @xcite . moreover , @xmath575 represents the fluctuations in the total number of condensate particles in that same approximation , implying that the density fluctuations @xmath576 are only nonzero in that region of space where the condensate density does not vanish . performing now the integration over the number fluctuations @xmath577 and the usual wick rotation to real times @xmath485 , we immediately see that the effective action for the global phase of the condensate has precisely the same form as in equation ( [ phase ] ) , i.e. , @xmath578 = \frac{mv_{\bf 0}(\mu)}{8\pi a } \int dt~ \left ( \frac{d\theta_{\bf 0}}{dt } \right)^2~.\ ] ] the appropriate ` diffusion ' constant is therefore equal to @xmath579 , which can easily be shown to be equal to @xmath580 if we make use of the fact that in the thomas - fermi approximation the chemical potential obeys @xmath581 and the radius of the condensate is given by @xmath582 @xcite . hence , the ` diffusion ' constant is proportional to @xmath583 . note that if the condensate where contained in a box the ` diffusion ' constant would be proportional to @xmath584 instead . it is important to note also that , in contrast to the case of a fermionic superfluid , we have to integrate over the amplitude fluctuations of the order parameter to arrive at a quadratic action for the phase fluctuations . this leads to the important conclusion that for a bosonic superfluid it is impossible to be in a state with only phase fluctuations and no density fluctuations , even at the largest length scales . for simplicity we restrict ourselves for the rest of this course to the case of a doubly - polarized atomic bose gas , which for most practical purposes can be seen as a gas of spin - less bosons @xcite . once we have understood the quantum kinetic theory for this particular case , however , we can easily generalize to multi - component bose gases that are receiving considerable attention at present @xcite . furthermore , the same methods can also be used for fermionic gases and have recently already been applied to the problem of cooper - pair formation in a two - component fermion gas @xcite . in textbooks an ideal bose gas is generally discussed in terms of the average occupation numbers of the one - particle states @xmath53 @xcite . given the density matrix @xmath585 of the gas at an initial time @xmath586 , these occupation numbers obey @xmath587~,\ ] ] with @xmath588 and @xmath589 the usual ( heisenberg picture ) creation and annihilation operators of second quantization , respectively . because the hamiltonian of the gas @xmath590 commutes with the number operators @xmath591 , the nonequilibrium dynamics of the system is trivial and the average occupation numbers are at all times equal to their value at the initial time @xmath586 . if we are also interested in fluctuations , it is convenient to introduce the eigenstates of the number operators , i.e. , @xmath592 and to consider the full probability distribution @xmath593~,\ ] ] which is again independent of time for an ideal bose gas . the average occupation numbers are then determined by @xmath594 and the fluctuations can be obtained from similar expressions . as indicated in section [ pd ] , we are also interested in the phase dynamics of the gas . in analogy with the occupation number representation in equation ( [ pn ] ) , we can now obtain a completely different description of the bose gas by making use of coherent states and considering the probability distribution @xmath595 $ ] . although we expect that this probability distribution is again independent of time , let us nevertheless proceed to derive its equation of motion in a way that can be generalized lateron when we consider an interacting bose gas . first , we use that @xmath596 = \int d[\phi^*_0]d[\phi_0]~ p[\phi_0^*,\phi_0;t_0 ] \frac{|\langle\phi;t|\phi_0;t_0\rangle|^2 } { \langle\phi;t|\phi;t\rangle \langle\phi_0;t_0|\phi_0;t_0\rangle}~.\ ] ] this is a particularly useful result , because the time dependence is now completely determined by the matrix element @xmath597 for which the functional integral representation is well known from our discussion in section [ fi ] . it is given by a ` path ' integral over all complex field evolutions @xmath598 from @xmath586 to @xmath30 . more precisely we have @xmath599d[\psi]~ \exp \left\ { \frac{i}{\hbar } s_+[\psi^*,\psi ] \right\}~,\ ] ] with the forward action @xmath600 $ ] given by @xmath601 = \sum_{\bf n } \left\ { -i\hbar \psi^*_{\bf n}(t ) \psi_{\bf n}(t ) \raisebox{0.2 in } { } \right . \\ & & \hspace*{0.6 in } + \left . \int_{t_0}^{t } dt_+~ \psi^*_{\bf n}(t_+ ) \left ( i\hbar \frac{\partial}{\partial t_+ } - \epsilon_{\bf n } \right ) \psi_{\bf n}(t_+ ) \right\}~. \nonumber\end{aligned}\ ] ] in the same manner the matrix element @xmath602 can be written as a ` path ' integral over all possible field configurations @xmath603 evolving backward in time from @xmath30 to @xmath586 , i.e. , @xmath604d[\psi]~ \exp \left\ { \frac{i}{\hbar } s_-[\psi^*,\psi ] \right\}~,\ ] ] with a backward action @xmath605 = \sum_{\bf n } \left\ { -i\hbar \psi^*_{\bf n}(t_0 ) \psi_{\bf n}(t_0 ) \raisebox{0.2 in } { } \right . \\ & & \hspace*{0.6 in } + \left . \int_{t}^{t_0 } dt_-~ \psi^*_{\bf n}(t_- ) \left ( i\hbar \frac{\partial}{\partial t_- } - \epsilon_{\bf n } \right ) \psi_{\bf n}(t_- ) \right\ } \nonumber \\ & & \hspace*{0.1 in } = \sum_{\bf n } \left\ { -i\hbar \psi^*_{\bf n}(t ) \psi_{\bf n}(t ) \raisebox{0.2 in } { } \right . \nonumber \\ & & \hspace*{0.6 in } + \left . \int_{t}^{t_0 } dt_-~ \psi_{\bf n}(t_- ) \left ( -i\hbar \frac{\partial}{\partial t_- } - \epsilon_{\bf n } \right ) \psi^*_{\bf n}(t_- ) \right\}~. \nonumber\end{aligned}\ ] ] putting all these results together , we see that the probability distribution @xmath595 $ ] can in fact be represented by a functional integral over all fields @xmath606 that evolve backwards from @xmath30 to @xmath586 and then forward in time from @xmath586 to @xmath30 . absorbing the factor @xmath607 $ ] into the measure of the functional integral , we thus arrive at the desired result that @xmath608 = \int_{\psi({\bf x},t)=\phi({\bf x})}^{\psi^*({\bf x},t)=\phi^*({\bf x } ) } d[\psi^*]d[\psi]~ \exp \left\ { \frac{i}{\hbar } s[\psi^*,\psi ] \right\}~,\ ] ] where the total ( backward - forward ) action in first instance obeys @xmath609 = s_-[\psi^*,\psi ] + s_+[\psi^*,\psi ] = -i\hbar \sum_{\bf n } \left ( \psi^*_{\bf n}(t ) \psi_{\bf n}(t ) - |\phi_{\bf n}|^2 \right ) \hspace{0.4 in } \\ & & + \sum_{\bf n } \int_{{\cal c}^t } dt'~ \left\ { \frac{1}{2 } \left ( \psi^*_{\bf n}(t ' ) i\hbar \frac{\partial}{\partial t ' } \psi_{\bf n}(t ' ) - \psi_{\bf n}(t ' ) i\hbar \frac{\partial}{\partial t ' } \psi^*_{\bf n}(t ' ) \right ) \right . \nonumber \\ & & \hspace*{2.55 in } - \left . \raisebox{0.2 in } { } \epsilon_{\bf n } \psi^*_{\bf n}(t ' ) \psi_{\bf n}(t ' ) \right\}~ , \nonumber\end{aligned}\ ] ] and the integration along the schwinger - keldysh contour @xmath610 is defined by @xmath611 @xcite . note also that in equation ( [ prob2 ] ) we have explicitly specified the boundary conditions on the functional integral . it is interesting to mention that these boundary conditions are essentially dictated by the topological terms in the action @xmath612 $ ] , which is a general feature of the path - integral formulation of quantum mechanics due to the fact that the quantum theory should have the correct ( semi)classical limit @xcite . making use of the periodicity of the field @xmath606 on the schwinger - keldysh contour , the variational principle @xmath613/\delta \psi^*_{\bf n}(t_{\pm } ) = 0 $ ] indeed leads not only to the euler - lagrange equation @xmath614 which agrees with the heisenberg equation of motion @xmath615 $ ] for the annihilation operators , but also with the appropriate boundary condition @xmath616 . in the same way we find the complex conjugate results if we require that @xmath613/\delta \psi_{\bf n}(t_{\pm } ) = 0 $ ] . substituting the boundary conditions in equation ( [ action ] ) and performing a partial integration , we then finally obtain for the action @xmath617 = \sum_{\bf n } \int_{{\cal c}^t } dt'~ \psi^*_{\bf n}(t ' ) \left ( i\hbar \frac{\partial}{\partial t ' } - \epsilon_{\bf n } \right ) \psi_{\bf n}(t')~,\ ] ] which we for completeness sake also rewrite as @xmath618 \\ & & = \int_{{\cal c}^t } dt ' \int d{\bf x}~ \psi^*({\bf x},t ' ) \left ( i\hbar \frac{\partial}{\partial t ' } + \frac{\hbar^2 \mbox{\boldmath $ \nabla$}^2}{2 m } - v^{\rm ex}({\bf x } ) \right ) \psi({\bf x},t')~. \nonumber\end{aligned}\ ] ] we are now in a position to derive the equation of motion , i.e. , the fokker - planck equation , for the probability distibution @xmath595 $ ] . this is most easily achieved by performing the variable transformation @xmath619 in equation ( [ prob2 ] ) . in this manner the fields @xmath620 and @xmath621 that live on the backward and forward branch of the schwinger - keldysh contour , respectively , are ` projected ' onto the real time axis . moreover , at the same time we effect a separation between the ( semi)classical dynamics described by @xmath622 and the quantum fluctuations determined by @xmath623 . after the transformation we have @xmath624 \\ & & = \int_{\phi({\bf x},t)=\phi({\bf x})}^{\phi^*({\bf x},t)=\phi^*({\bf x } ) } d[\phi^*]d[\phi ] \int d[\xi^*]d[\xi]~ \exp \left\ { \frac{i}{\hbar } s[\phi^*,\phi;\xi^*,\xi ] \right\}~ , \nonumber\end{aligned}\ ] ] with @xmath625 = \sum_{\bf n } \int_{t_0}^t dt'~ \left\ { \phi^*_{\bf n}(t ' ) \left ( i\hbar \frac{\partial}{\partial t ' } - \epsilon_{\bf n } \right ) \xi_{\bf n}(t ' ) \right . \\ & & \hspace*{1.2 in } + \left . \xi^*_{\bf n}(t ' ) \left ( i\hbar \frac{\partial}{\partial t ' } - \epsilon_{\bf n } \right ) \phi_{\bf n}(t ' ) \right\}~. \nonumber\end{aligned}\ ] ] because this action is linear in @xmath626 and @xmath627 , the integration over the quantum fluctuations leads just to a constraint and we find that @xmath628 = \int_{\phi({\bf x},t)=\phi({\bf x})}^{\phi^*({\bf x},t)=\phi^*({\bf x } ) } d[\phi^*]d[\phi]~ \\ & & \times \prod_{\bf n } \delta \left [ \left ( -i \frac{\partial}{\partial t ' } - \frac{\epsilon_{\bf n}}{\hbar } \right ) \phi^*_{\bf n}(t ' ) \cdot \left ( i \frac{\partial}{\partial t ' } - \frac{\epsilon_{\bf n}}{\hbar } \right ) \phi_{\bf n}(t ' ) \right ] \nonumber\end{aligned}\ ] ] or equivalently that @xmath629 = \int d[\phi^*_0]d[\phi_0]~ p[\phi^*_0,\phi_0;t_0 ] \prod_{\bf n } \delta \left ( |\phi_{\bf n } - \phi^{\rm cl}_{\bf n}(t)|^2 \right)~,\ ] ] where we introduced the quantity @xmath630 obeying the semiclassical equation of motion @xmath631 and the initial condition @xmath632 . the latter equation is thus solved by @xmath633 and we conclude from a simple change of variables in equation ( [ prob3 ] ) that for an ideal bose gas @xmath595 = p[\phi^*,\phi;t_0]$ ] , as expected . we also see from equation ( [ prob3 ] ) that the desired equation of motion for @xmath595 $ ] reads @xmath634 = - \left ( \sum_{\bf n } \frac{\partial}{\partial \phi_{\bf n } } \epsilon_{\bf n } \phi_{\bf n } \right ) p[\phi^*,\phi;t ] \\ & & \hspace*{0.8 in } + \left ( \sum_{\bf n } \frac{\partial}{\partial \phi^*_{\bf n } } \epsilon_{\bf n } \phi^*_{\bf n } \right ) p[\phi^*,\phi;t]~. \nonumber\end{aligned}\ ] ] this is indeed the correct fokker - planck equation for an ideal bose gas @xcite . to see that we first consider @xmath635d[\phi]~ \phi_{\bf n } p[\phi^*,\phi;t]$ ] . multiplying equation ( [ fp1 ] ) with @xmath636 and integrating over @xmath637 , we easily find after a partial integration that @xmath638 which precisely corresponds to the equation of motion of the expectation value @xmath639 $ ] in the operator formalism . similarly , we find that @xmath640 in agreement with the result for @xmath641 $ ] . next we consider the average of @xmath642 , for which we immediately obtain @xmath643 we expect this result to be related to the fact that in the operator formalism the occupation numbers @xmath644 are independent of time . although this turns out to be true , to give the precise relation between @xmath645 and @xmath644 is complicated by the fact that at equal times the operators @xmath589 and @xmath588 do not commute . however , the path - integral formulation of quantum mechanics is only capable of calculating time - ordered operator products @xcite . in our case this implies that @xmath645 is the value at @xmath646 of @xmath647 \\ & & = \theta(t , t ' ) { \rm tr } \left [ \hat{\rho}(t_0 ) \hat{\psi}_{\bf n}(t ) \hat{\psi}^{\dagger}_{\bf n}(t ' ) \right ] + \theta(t',t ) { \rm tr } \left [ \hat{\rho}(t_0 ) \hat{\psi}^{\dagger}_{\bf n}(t ' ) \hat{\psi}_{\bf n}(t ) \right]~ , \nonumber\end{aligned}\ ] ] with @xmath648 the time - ordening operator on the schwinger - keldysh contour and @xmath649 the corresponding heaviside function . since the heaviside function is equal to @xmath449 at equal times @xcite , we conclude that @xmath650 and that equation ( [ phisquare ] ) is thus fully consistent with the operator formalism . an intuitive understanding of the relation between @xmath645 and the occupation numbers can be obtained by noting that , since all our manipulations up to now have been exact , we expect @xmath645 to contain both classical and quantum fluctuations . these correspond precisely to the contributions @xmath644 and @xmath449 , respectively . finally we need to discuss the stationary solutions of the fokker - planck equation . it is not difficult to show that any functional that only depends on the amplitudes @xmath642 is a solution . as it stands the fokker - planck equation , therefore , does not lead to a unique equilibrium distribution . this is not surprising , because for an isolated , ideal bose gas there is no mechanism for redistributing the particles over the various energy levels and thus for relaxation towards equilibrium . however , the situation changes when we allow the bosons in the trap to tunnel back and forth to a reservoir at a temperature @xmath117 . the corrections to the fokker - planck equation that are required to describe the physics in this case are considered next . however , to determine these corrections in the most convenient way , we have to slightly generalize the above theory because with the probability distibution @xmath595 $ ] we are only able to study spatial , but not temporal correlations in the bose gas . to study also those we follow the by now well - known procedure in quantum field theory and construct a generating functional @xmath651 $ ] for all ( time - ordered ) correlation functions . it is obtained by performing two steps . first , we introduce the probability distribution @xmath652 $ ] for a bose gas in the presence of the external currents @xmath653 and @xmath654 by adding to the hamiltonian the terms @xmath655 as a result we have @xmath656 = \int_{\psi({\bf x},t)=\phi({\bf x})}^{\psi^*({\bf x},t)=\phi^*({\bf x } ) } d[\psi^*]d[\psi]~ \exp \left\ { \frac{i}{\hbar } s[\psi^*,\psi ] \right\ } \\ & & \hspace*{0.38 in } \times \exp \left\ { i \int_{{\cal c}^t } dt ' \int d{\bf x}~ \left ( \psi({\bf x},t ' ) j^*({\bf x},t ' ) + j({\bf x},t ' ) \psi^*({\bf x},t ' ) \right ) \right\}~. \nonumber\end{aligned}\ ] ] second , we integrate this expression over @xmath637 to obtain the desired generating functional . hence @xmath657 = \int d[\phi^*]d[\phi]~ p_j[\phi^*,\phi;t ] = \int d[\psi^*]d[\psi]~ \exp \left\ { \frac{i}{\hbar } s[\psi^*,\psi ] \right\ } \hspace{0.4 in } \\ & & \hspace*{0.32 in } \times \exp \left\ { i \int_{{\cal c}^{\infty } } dt \int d{\bf x}~ \left ( \psi({\bf x},t ) j^*({\bf x},t ) + j({\bf x},t ) \psi^*({\bf x},t ) \right ) \right\}~. \nonumber\end{aligned}\ ] ] it is important to realize that @xmath651 $ ] is indeed independent of the time @xmath30 because of the fact that @xmath652 $ ] is a probability distribution and thus properly normalized . we are therefore allowed to deform the contour @xmath658 to any closed contour that runs through @xmath586 . since we are in principle interested in all times @xmath659 , the most convenient choice is the countour that runs backward from infinity to @xmath586 and then forwards from @xmath586 to infinity . this contour is denoted by @xmath660 and also called the schwinger - keldysh contour in the following because there is in practice never confusion with the more restricted contour @xmath658 that is required when we consider a probability distribution . with this choice it is also clear that @xmath651 $ ] is a generating functional . indeed , equation ( [ z ] ) shows explicitly that all time - ordered correlation functions can be obtained by functional differentiation with respect to the currents @xmath653 and @xmath654 . we have , for instance , that @xmath661 = \left . \frac{1}{i } \frac{\delta}{\delta j^*({\bf x},t ) } z[j , j^ * ] \right|_{j , j^*=0}\ ] ] and similarly that @xmath662 \\ & & = \left . \frac{1}{i^2 } \frac{\delta^2}{\delta j^*({\bf x},t ) \delta j({\bf x}',t ' ) } z[j , j^ * ] \right|_{j , j^*=0}~. \nonumber\end{aligned}\ ] ] note that the times @xmath30 and @xmath663 always have to be larger or equal to @xmath586 for these identities to be valid . inspired by the caldeira - leggett model for a particle experiencing friction @xcite , we take for the reservoir an ideal gas of @xmath3 bosons in a box with volume @xmath337 . the states in this box are labeled by the momentum @xmath664 and equal to @xmath665 . they have an energy @xmath666 , where @xmath667 accounts for a possible bias between the potential energies of a particle in the center of the trap and a particle in the reservoir . the reservoir is also taken to be sufficiently large that it can be treated in the thermodynamic limit and is in an equilibrium with temperature @xmath117 and chemical potential @xmath60 for times @xmath668 . at @xmath586 it is brought into contact with the trap by means of a tunnel hamiltonian @xmath669 with complex tunneling matrix elements @xmath670 that for simplicity are assumed to be almost constant for momenta @xmath671 smaller that a cutoff @xmath672 but to vanish rapidly for momenta larger than this ultraviolet cutoff . moreover , we consider here only the low - temperature regime in which the thermal de broglie wavelength @xmath673 of the particles obeys @xmath674 , since this is the most appropriate limit for realistic atomic gases . to study the evolution of the combined system for times @xmath659 we thus have to deal with the action @xmath675 = \sum_{\bf n } \int_{{\cal c}^{\infty } } dt~ \psi^*_{\bf n}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu \right ) \psi_{\bf n}(t ) \\ & & + \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \psi^*_{\bf k}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon({\bf k } ) + \mu \right ) \psi_{\bf k}(t)~ , \nonumber \\ & & - \frac{1}{\sqrt{v } } \sum_{\bf n } \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \left ( t_{\bf n}({\bf k } ) \psi_{\bf n}(t ) \psi^*_{\bf k}(t ) + t^*_{\bf n}({\bf k } ) \psi_{\bf k}(t ) \psi^*_{\bf n}(t ) \right ) \nonumber \end{aligned}\ ] ] if we measure all energies relative to the chemical potential and also introduce the complex field @xmath676 for the degrees of freedom of the reservoir . however , we are only interested in the evolution of the bose gas in the trap and therefore only in the time - ordered correlation functions of this part of the system . the corresponding generating functional @xmath677 = \int d[\psi^*]d[\psi ] \int d[\psi_r^*]d[\psi_r]~ \exp \left\ { \frac{i}{\hbar } s[\psi^*,\psi;\psi_r^*,\psi_r ] \right\ } \hspace{0.4 in } \\ & & \times \exp \left\ { i \int_{{\cal c}^{\infty } } dt \int d{\bf x}~ \left ( \psi({\bf x},t ) j^*({\bf x},t ) + j({\bf x},t ) \psi^*({\bf x},t ) \right ) \right\}~. \nonumber\end{aligned}\ ] ] is of the same form as the functional integral in equation ( [ z ] ) , but now with an effective action that is defined by @xmath678 \right\ } \equiv \int d[\psi_r^*]d[\psi_r]~ \exp \left\ { \frac{i}{\hbar } s[\psi^*,\psi;\psi_r^*,\psi_r ] \right\}~.\ ] ] hence , our next task is to integrate out the field @xmath679 , which can be done exactly because it only requires the integration of a gaussian . to familiarize ourselves with the schwinger - keldysh formalism , we perform the gaussian integration here in some detail . in principle , this can of course be done explicitly by making use of the fact that the initial density matrix @xmath680 of the reservoir can be expanded in the projection operators on the coherent states by using the coefficients @xmath681 = \prod_{\bf k } \frac{1}{n({\bf k } ) } e^ { - |\phi_{\bf k}|^2/n({\bf k } ) } ~,\ ] ] with @xmath682 the appropriate bose distribution function and @xmath119 . however , in practice it is much more easy to use a different procedure . it is based on the observation that if we introduce the @xmath440 function on the schwinger - keldysh contour defined by@xmath683 and the green s function @xmath684 obeying @xmath685 the action @xmath686 $ ] can be written as a complete square , or more precisely as the sum of two squares @xmath687 $ ] and @xmath688 $ ] that are given by @xmath689 = \sum_{\bf n } \int_{{\cal c}^{\infty } } dt~ \psi^*_{\bf n}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu \right ) \psi_{\bf n}(t ) \\ & & \hspace*{0.1 in } - \frac{1}{\hbar v } \sum_{\bf n,\bf n ' } \sum_{\bf k } \int_{{\cal c}^{\infty } } dt \int_{{\cal c}^{\infty } } dt'~ \psi^*_{\bf n}(t ) t^*_{\bf n}({\bf k } ) g({\bf k};t , t ' ) t_{\bf n'}({\bf k } ) \psi_{\bf n'}(t ' ) \nonumber \end{aligned}\ ] ] and @xmath690 \\ & & = \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \left ( \psi_{\bf k}^*(t ) - \frac{1}{\hbar\sqrt{v } } \sum_{\bf n } \int_{{\cal c}^{\infty } } dt'~ t^*_{\bf n}({\bf k } ) \psi^*_{\bf n}(t ' ) g({\bf k};t',t ) \right ) \nonumber \\ & & \hspace*{0.7 in } \times \left ( i\hbar \frac{\partial}{\partial t } - \epsilon({\bf k } ) + \mu \right ) \nonumber \\ & & \hspace*{0.7 in } \times \left ( \psi_{\bf k}(t ) - \frac{1}{\hbar\sqrt{v } } \sum_{\bf n } \int_{{\cal c}^{\infty } } dt'~ g({\bf k};t , t ' ) \psi_{\bf n}(t ' ) t_{\bf n}({\bf k } ) \right)~ , \nonumber \end{aligned}\ ] ] respectively . since the first term is independent of the field @xmath679 , we only need to evaluate @xmath691d[\psi_r]~ \exp \left ( i s_2[\psi^*,\psi;\psi_r^*,\psi_r]/\hbar \right)$ ] . performing a shift in the integration variables , we however see that this functional integral is equal to @xmath692d[\psi_r]~ \exp \left\ { \frac{i}{\hbar } \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \psi^*_{\bf k}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon({\bf k } ) + \mu \right ) \psi_{\bf k}(t ) \right\ } \\ = { \rm tr}[\rho_r(t_0 ) ] = 1~. \nonumber\end{aligned}\ ] ] as a result the effective action @xmath693 $ ] is just equal to @xmath687 $ ] , which can be slightly rewritten to read @xmath694 = \sum_{\bf n,\bf n ' } \int_{{\cal c}^{\infty } } dt \int_{{\cal c}^{\infty } } dt'~ \\ & & \hspace*{0.37 in } \times~ \psi^*_{\bf n}(t ) \left\ { \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu \right ) \delta_{\bf n,\bf n ' } \delta(t , t ' ) - \hbar \sigma_{{\bf n},{\bf n}'}(t , t ' ) \right\ } \psi_{\bf n'}(t')~ , \nonumber\end{aligned}\ ] ] with a selfenergy @xmath695 obeying @xmath696 this is our first example of an effective action describing the nonequilibrium dynamics of a bose gas . before we can study its consequences we clearly first need to determine the green s function @xmath684 in terms of which the selfenergy is expressed . although we know that this green s function fulfills equation ( [ greena ] ) , we can not directly solve this equation because we do not know the appropriate boundary condition at @xmath697 . to calculate @xmath684 we therefore have to proceed differently . it is however clear from equation ( [ greena ] ) that @xmath684 is a property of the reservoir and we thus expect that it can somehow be related to a time - ordered correlation function of this reservior . to see that explicitly we consider again the generating functional of these correlation functions , i.e. , @xmath698 \\ & & = \int d[\psi_r^*]d[\psi_r]~ \exp \left\ { \frac{i}{\hbar } \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \psi^*_{\bf k}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon({\bf k } ) + \mu \right ) \psi_{\bf k}(t ) \right\ } \nonumber \\ & & \hspace*{0.9 in } \times \exp \left\ { i \sum_{\bf k } \int_{{\cal c}^{\infty } } dt~ \left ( \psi_{\bf k}(t ) j^*_{\bf k}(t ) + j_{\bf k}(t ) \psi^*_{\bf k}(t ) \right ) \right\}~. \nonumber\end{aligned}\ ] ] it is again a gaussian integral and can thus be evaluated in the same manner as before . the result is now @xmath699 = \exp \left\ { -i \sum_{\bf k } \int_{{\cal c}^{\infty } } dt \int_{{\cal c}^{\infty } } dt'~ j^*_{\bf k}(t ) g({\bf k};t , t ' ) j_{\bf k}(t ' ) \right\}~,\ ] ] which shows by means of equation ( [ corr ] ) that @xmath700~.\ ] ] note first of all that this is indeed consistent with equation ( [ greena ] ) , because the right - hand side obeys @xmath701 \\ & & = i\hbar \delta(t , t ' ) { \rm tr } \left [ \hat{\rho}_r(t_0 ) [ \hat{\psi}_{\bf k}(t),\hat{\psi}^{\dagger}_{\bf k}(t ) ] \right ] \nonumber \\ & & \hspace*{0.97 in } + { \rm tr } \left [ \hat{\rho}_r(t_0 ) t_{{\cal c}^{\infty } } \left ( i\hbar \frac{\partial}{\partial t } \hat{\psi}_{\bf k}(t ) \hat{\psi}^{\dagger}_{\bf k}(t ' ) \right ) \right ] \nonumber \\ & & = i\hbar \delta(t , t ' ) + ( \epsilon({\bf k } ) - \mu ) { \rm tr } \left [ \hat{\rho}_r(t_0 ) t_{{\cal c}^{\infty } } \left ( \hat{\psi}_{\bf k}(t ) \hat{\psi}^{\dagger}_{\bf k}(t ' ) \right ) \right ] \nonumber\end{aligned}\ ] ] in the operator formalism . moreover , from this identification we see that the desired solution fulfilling the appropriate boundary conditions is apparently @xmath702 the specific dependence on the backward and forward branches of the schwinger - keldysh contour is thus solely determined by the heaviside function @xmath649 . as a result it is convenient to decompose the green s function into its analytic pieces @xmath703 and @xmath704 @xcite by means of @xmath705 due to the fact that we are always dealing with time - ordered correlation functions , such a decomposition turns out to be a generic feature of all the functions on the schwinger - keldysh contour that we will encounter in the following @xcite . for a general function @xmath706 , it is however also possible to have @xmath440-function singularities . if that happens the correct decomposition is @xmath707 this more general decomposition is not required here , but will be needed in section [ cf ] when we determine the effective interaction between two atoms in a gas . having obtained the green s function of the reservoir , we can now return to our discussion of the effective action @xmath693 $ ] for the bose gas in the trap . although we have chosen to derive in equation ( [ seff ] ) the effective action for the generating functional @xmath651 $ ] , it is straightforward to show that the effective action for the probability distribution @xmath595 $ ] is obtained by only replacing the contour @xmath708 by @xmath709 . in the following we therefore no longer always specify all the boundary conditions on the time integration , if the precise details of this integration are not important and the discussion applies equally well to both cases . keeping this in mind , we now again perform the transformation @xmath710 to explicitly separate the ( semi)classical dynamics from the effect of fluctuations . it leads in first instance to @xcite @xmath711 = \sum_{\bf n } \int_{t_0 } dt~ \phi^*_{\bf n}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu \right ) \xi_{\bf n}(t ) \\ & & \hspace*{0.45 in } + \sum_{\bf n } \int_{t_0 } dt~ \xi^*_{\bf n}(t ) \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu \right ) \phi_{\bf n}(t ) \nonumber \\ & & \hspace*{0.45 in } - \sum_{\bf n,\bf n ' } \int_{t_0 } dt \int_{t_0 } dt'~ \phi^*_{\bf n}(t ) \hbar \sigma^{(-)}_{{\bf n},{\bf n}'}(t - t ' ) \xi_{\bf n'}(t ' ) \nonumber \\ & & \hspace*{0.45 in } - \sum_{\bf n,\bf n ' } \int_{t_0 } dt \int_{t_0 } dt'~ \xi^*_{\bf n}(t ) \hbar \sigma^{(+)}_{{\bf n},{\bf n}'}(t - t ' ) \phi_{\bf n'}(t ' ) \nonumber \\ & & \hspace*{0.45 in } - \frac{1}{2 } \sum_{{\bf n},{\bf n ' } } \int_{t_0 } dt \int_{t_0 } dt'~ \xi^*_{\bf n}(t ) \hbar \sigma^{k}_{{\bf n},{\bf n'}}(t - t ' ) \xi_{\bf n'}(t')~ , \nonumber\end{aligned}\ ] ] where we introduced the retarded and advanced components of the selfenergy @xmath712 that affect the terms in the action that are linear in @xmath713 and @xmath714 , respectively , and also the keldysh component @xmath715 that is associated with the part quadratic in the fluctuations . the physical content of these various components of the selfenergy is understood most clearly if we now apply again the beautiful functional - integral procedure that is due to hubbard and stratonovich . the basic idea is to write the factor @xmath716 in the integrant of @xmath111d[\phi ] \int d[\xi^*]d[\xi]~ \exp\left ( is^{\rm eff}[\phi^*,\phi;\xi^*,\xi]/\hbar \right)$ ] as a gaussian integral over a complex field @xmath717 . it is equivalent , but in practice more convenient , to just multiply the integrant by a factor @xmath65 that is written as a gaussian integral @xmath718d[\eta]~ \exp\left ( is^{\rm eff}[\eta^*,\eta]/\hbar \right)$ ] with @xmath719 \\ & & = \frac{1}{2 } \sum_{{\bf n},{\bf n } ' } \int_{t_0 } dt \int_{t_0 } dt'~ \left(2 \eta^*_{\bf n}(t ) - \sum_{\bf n '' } \int_{t_0 } dt''~ \xi^*_{\bf n''}(t '' ) \hbar \sigma^k_{{\bf n}'',{\bf n}}(t''-t ) \right ) \nonumber \\ & & \times~ ( \hbar\sigma^k)^{-1}_{{\bf n},{\bf n}'}(t - t ' ) \left(2 \eta_{\bf n'}(t ' ) - \sum_{\bf n '' } \int_{t_0 } dt''~ \hbar \sigma^k_{{\bf n}',{\bf n}''}(t'-t '' ) \xi_{\bf n''}(t '' ) \right ) \nonumber\end{aligned}\ ] ] a complete square . adding this to @xmath720 $ ] the total effective action becomes @xmath721 \\ & & = \sum_{{\bf n},{\bf n } ' } \int_{t_0 } dt \int_{t_0 } dt'~ \phi^*_{\bf n}(t ) \left\ { \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu - \eta^*_{\bf n}(t ) \right ) \delta_{{\bf n},{\bf n } ' } \delta(t - t ' ) \right . \nonumber \\ & & \hspace*{2.57 in } - \left . \raisebox{0.2 in } { } \hbar \sigma^{(-)}_{{\bf n},{\bf n}'}(t - t ' ) \right\ } \xi_{\bf n'}(t ' ) \nonumber \\ & & + \sum_{{\bf n},{\bf n } ' } \int_{t_0 } dt \int_{t_0 } dt'~ \xi^*_{\bf n}(t ) \left\ { \left ( i\hbar \frac{\partial}{\partial t } - \epsilon_{\bf n } + \mu - \eta_{\bf n}(t ) \right ) \delta_{{\bf n},{\bf n } ' } \delta(t - t ' ) \right . \nonumber \\ & & \hspace*{2.5 in } - \left . \raisebox{0.2 in } { } \hbar \sigma^{(+)}_{{\bf n},{\bf n}'}(t - t ' ) \right\ } \phi_{\bf n'}(t ' ) \nonumber \\ & & + 2 \sum_{{\bf n},{\bf n } ' } \int_{t_0 } dt \int_{t_0 } dt'~ \eta^*_{\bf n}(t ) ( \hbar\sigma^k)^{-1}_{{\bf n},{\bf n}'}(t - t ' ) \eta_{\bf n'}(t ' ) \nonumber\end{aligned}\ ] ] and is thus linear in @xmath714 and @xmath713 . integrating over these fluctuations we conclude from this action that the field @xmath722 is constraint to obey the langevin equations @xcite @xmath723 and @xmath724 with gaussian noise terms @xmath725 and @xmath726 that from the last term in the right - hand side of equation ( [ snoise ] ) are seen to have the time correlations @xmath727 in the thermodynamic limit . moreover , in that limit the retarded selfenergy is equal to @xmath728 and the advanced selfenergy can be found from the relation @xmath729 . note that the heaviside functions in these retarded and advanced selfenergies are precisely such that eqs . ( [ lan1 ] ) and ( [ lan2 ] ) are causal , i.e. , the evolution of @xmath722 and @xmath730 depends only on their value at previous times . this is clearly a sensible result and explains why these components of the full selfenergy enter into the terms in the action that are linear in the fluctuations . to understand why the keldysh component enters into the terms that are quadratic in the fluctuations is more complicated and is related to the famous fluctuation - dissipation theorem @xcite . physically , the fluctuation - dissipation theorem ensures in our case that , due to the interactions with the reservoir , the gas in the trap now relaxes towards thermal equilibrium in the limit @xmath731 . let us therefore study this limit in somewhat more detail . however , before we do so we would like to mention for completeness that if we apply the above hubbard - stratonovich transformation to the functional integral for the probability distribution @xmath595$],we immediately find that analogous to equation ( [ prob3 ] ) @xmath732 = \int d[\eta^*]d[\eta]~ p[\eta^*,\eta ] \int d[\phi^*_0]d[\phi_0]~ p[|\phi_0|^2;t_0 ] \\ & & \hspace*{1.9 in } \times \prod_{\bf n } \delta \left ( |\phi_{\bf n } - \phi^{\rm cl}_{\bf n}(t)|^2 \right)~ , \nonumber\end{aligned}\ ] ] where @xmath630 now obeys the langevin equation in equation ( [ lan1 ] ) for a particular realization of the noise @xmath725 and with the initial condition @xmath733 . furthermore , the probability distribution for each realization of the noise is given by @xmath734 \\ & & = \exp \left\ { \frac{2i}{\hbar } \sum_{{\bf n},{\bf n } ' } \int_{t_0}^t dt ' \int_{t_0}^t dt''~ \eta^*_{\bf n}(t ' ) ( \hbar\sigma^k)^{-1}_{{\bf n},{\bf n}'}(t'-t '' ) \eta_{\bf n'}(t '' ) \right\}~. \nonumber\end{aligned}\ ] ] this result shows explicitly that , as expected , the fokker - planck equation associated with the langevin equations in eqs . ( [ lan1 ] ) and ( [ lan2 ] ) is in fact the desired fokker - planck equation for the probability distribution @xmath595 $ ] . after this slight digression , we now return to the long - time behaviour of the gas . in the long - time limit two important simplifications occur . looking at the expresions for the selfenergies , we see that their width as a function of the time difference @xmath735 is at most of order @xmath736 in the low - temperature regime of interest where the thermal de broglie wavelength of the particles obeys @xmath674 . if therefore @xmath737 we are allowed to take the limit @xmath738 . taking this limit means physically that we neglect the initial transients that are due to the precise way in which the contact between the trap and the reservoir is made , and focus on the ` universal ' dynamics that is independent of these details . in addition , at long times the dynamics of the gas is expected to be sufficiently slow that we can neglect the memory effects altogether . finally , we consider here first the case of a reservoir that is so weakly coupled to the gas in the trap that we can treat the coupling with second order perturbation theory . as a result we can also neglect the nondiagonal elements of the selfenergies . in total we are then allowed to put ( but see below ) @xmath739 where the fourier transform of the selfenergies is defined by @xmath740 note that this implies in general that @xmath741 , and in particular therefore that @xmath742 . with these simplifications our langevin equations become @xmath743 and just the complex conjugate equation for @xmath744 . the retarded selfenergy in this equation is given by @xmath745 with @xmath746 the usual notation for the limiting procedure @xmath747 . it clearly has real and imaginary parts that we denote by @xmath748 and @xmath749 , respectively . denoting the cauchy principle value part of an integral by @xmath750 , they obey @xmath751 and @xmath752 the interpretation of these results is quite obvious if we consider the average of the langevin equation , i.e. , @xmath753 which is solved by @xmath754 hence the real part of the retarded selfenergy @xmath748 represents the shift in the energy of state @xmath53 due to the coupling with the reservoir . indeed , equation ( [ shift ] ) agrees precisely with the value of this shift in second order perturbation theory . moreover , the fact that @xmath755 shows that the average rate of decay @xmath756 of the state @xmath53 is equal to @xmath757 , which , together with equation ( [ decay ] ) , exactly reproduces fermi s golden rule . from the equation of motion for the average @xmath758 we can also conclude that the right - hand side of the fokker - planck equation contains the ` streaming ' terms @xmath759 \hspace*{1.5 in } \nonumber \\ + \left ( \sum_{\bf n } \frac{\partial}{\partial \phi^*_{\bf n } } ( \epsilon_{\bf n } + \hbar\sigma^{(-)}_{\bf n } - \mu ) \phi^*_{\bf n } \right ) p[\phi^*,\phi;t]~. \nonumber\end{aligned}\ ] ] however , there must now also be a ` diffusion ' term due to the fact that the average @xmath645 is no longer independent of time since the interactions with the reservoir can lead to changes in the occupation numbers @xmath644 . to obtain the ` diffusion ' term we thus need to determine @xmath760 . to do so we first formally solve the langevin equation by @xmath761 multiplying this with the complex conjugate expression and taking the average , we obtain first of all @xmath762 which shows that @xmath763 moreover , the keldysh component of the selfenergy is given by @xmath764 and therefore obeys @xmath765 with @xmath766 the bose distribution function . this is in fact the fluctuation - dissipation theorem , because it relates the strength of the fluctuations determined by @xmath767 , to the amount of dissipation that is given by @xmath768 . as mentioned previously , the fluctuation - dissipation theorem ensures that the gas relaxes to thermal equilibrium . this we can already see from equation ( [ fluc ] ) , because substitution of the fluctuation - dissipation theorem leads to @xmath769 which tells us that in equilibrium @xmath770 . in fact , we have argued that in general @xmath771 . substituting this identity in equation ( [ fluct ] ) and realizing that @xmath772 due to the energy - conserving @xmath440 function in @xmath756 , we indeed obtain the correct rate equation for the average occupation numbers @xmath773 that might justly be called the quantum boltzmann equation for the gas , because the right - hand side contains precisely the rates for scattering into and out of the reservoir . furthermore , equation ( [ fluc ] ) shows that the ` diffusion ' term in the fokker - planck equation is @xmath774~ , \nonumber\end{aligned}\ ] ] because the first term in the right - hand side of equation ( [ fluc ] ) is also present without the noise and is therefore already accounted for in the ` streaming ' terms . in total we have thus arrived at @xmath775 = - \left ( \sum_{\bf n } \frac{\partial}{\partial \phi_{\bf n } } ( \epsilon_{\bf n } + \hbar\sigma^{(+)}_{\bf n } - \mu ) \phi_{\bf n } \right ) p[\phi^*,\phi;t ] \\ & & \hspace*{0.2 in } + \left ( \sum_{\bf n } \frac{\partial}{\partial \phi^*_{\bf n } } ( \epsilon_{\bf n } + \hbar\sigma^{(-)}_{\bf n } - \mu ) \phi^*_{\bf n } \right ) p[\phi^*,\phi;t ] \nonumber \\ & & \hspace*{0.2 in } - \left ( \frac{1}{2 } \sum_{\bf n } \frac{\partial^2}{\partial \phi^*_{\bf n } \partial \phi_{\bf n } } \hbar \sigma^k_{\bf n } \right ) p[\phi^*,\phi;t]~. \nonumber \end{aligned}\ ] ] using again the fluctuation - dissipation theorem , it is not difficult to show that the stationary solution of this fokker - planck equation is @xmath776 = \prod_{\bf n } \frac{1}{n_{\bf n } + 1/2 } \exp \left\ { - \frac{1}{n_{\bf n } + 1/2 } |\phi_{\bf n}|^2 \right\}~.\ ] ] although this appears to be quite reasonable , it is important to note that this result is not fully consistent because the ` streaming ' terms in equation ( [ fp2 ] ) show that the energies of the states @xmath53 are shifted and equal to @xmath777 . we therefore expect that the equilibrium occupation numbers are equal to the bose distribution evaluated at those energies and not at the unshifted values @xmath19 . the reason for this inconsistency is , of course , that we have taken @xmath778 equal to @xmath779 , which was only justified on the basis of second - order perturbation theory . we now , however , see that to obtain a fully consistent theory we must take instead @xmath780 . this implies that to obtain the retarded selfenergy , we first have to solve for the real part @xmath781 and then substitute this into @xmath782 to obtain the imaginary part . the fluctuation - dissipation theorem then again reads @xmath783 but now with the equilibrium distribution @xmath784 . as a consequence the quantum boltzmann equation and the stationary state of the fokker - planck equation will now also contain these correct equilibrium occupation numbers @xcite . having arrived at the fokker - planck equation of our caldeira - leggett - like toy model , we have now demonstrated the use of essentially all the theoretical tools that are required for a discussion of a weakly interacting bose gas . before we start with this discussion , however , we want to make a final remark about the effect of the nondiagonal elements of the selfenergies . physically , including these nondiagonal elements accounts for the change in the wave functions @xmath53 due to the interaction with the reservoir . it can clearly be neglected if the coupling with the reservoir is sufficiently weak , or more precisely if @xmath785 is much smaller than the energy splitting @xmath786 between the states in the trap . for a quantum dot in solid - state phyisics @xcite this is often the case and for that reason we might call our toy model a bosonic quantum dot . however , in magnetically trapped atomic gases the average interaction energy between the atoms is already slightly below the bose - einstein condensation critical temperature larger than the typical energy splitting and the nondiagonal elements of the selfenergies are very important . such a strong - coupling situation can , of course , also be treated in our caldeira - leggett like model . the main difference is that we need to expand our various fields not in the eigenstates of the trapping potential @xmath53 but in the eigenstates @xmath318 of the nonlocal schrdinger equation @xmath787 \chi'_{\bf n}({\bf x } ' ) \nonumber\end{aligned}\ ] ] where @xmath788 are the new eigenvalues and the retarded selfenergie in coordinate space is @xmath789 . in this new basis the nondiagonal elements of the selfenergies can now be neglected and we find essentially the same results as before . we only need to replace @xmath790 by @xmath788 and @xmath670 by @xmath791 . neglecting the nondiagonal elements in this basis only requires that the real part of the retarded selfenergy is much larger that its imaginary part , which is always fulfilled in our case because @xmath792 . summarizing , the coherent and incoherent dynamics of the gas in the trap is for our toy model quite generally solved by @xmath793 and @xmath794 respectively . in the limit @xmath731 , the average of the annihilation operators @xmath758 thus always vanishes but the average occupation numbers @xmath644 relax to the equilibrium distribution @xmath795 . although this appears to be an immediately obvious result , its importance stems from the fact that it is also true if we tune the potential energy bias @xmath667 such that at low temperatures the groundstate @xmath796 acquires a macroscopic occupation , i.e. , @xmath797 . the gas therefore never shows a spontaneous breaking of the @xmath424 symmetry , in agreement with the notion that we are essentially dealing with an ideal bose gas in the grand - canonical ensemble @xcite . the reason of the absense of the spontaneous symmetry breaking can also be understood from our stationary solution of the fokker - planck equation in equation ( [ stat ] ) , which shows that the probability distribution for @xmath798 is proportional to the boltzmann factor @xmath799 in the degenerate regime of interest . hence , the corresponding landau free energy @xmath800 never shows an instability towards the formation of a nonzero average of @xmath798 due to the fact that @xmath801 can never become less then zero . once we introduce interactions between the atoms in the gas , this picture fully changes . we are now in a position to discuss an interacting bose gas , because in an interacting bose gas , the gas is roughly speaking its own thermal reservoir . the fokker - planck equation therefore turns out to be quite similar to the one presented in equation ( [ fp2 ] ) . to be more precise , however , we should mention that we aim in this section to describe the formation of a condensate in an interacting bose gas under the conditions that have recently been realized in experiments with atomic rubidium , lithium , sodium and hydrogen gases . in these experiments the gas is cooled by means of evaporative cooling . numerical solutions of the boltzmann equation have shown that during evaporative cooling the energy distribution function is well described by an equilibrium distribution with time dependent temperature @xmath802 and chemical potential @xmath803 , that is truncated at high energies due to the evaporation of the highest energetic atoms from the trap @xcite . moreover , in the experiments of interest the densities just above the critical temperature are essentially always such that the gas is in the weak - coupling limit , which implies in this context that the average interaction energy per atom is always much less than the energy splitting of the one - particle states in the harmonic trapping potential . for a trapped atomic gas the appropriate one - particle states are the eigenstates @xmath53 of the external potential @xmath6 . as a result the fokker - planck equation for the interacting gas will now quite generally be given by @xmath804 = - \sum_{{\bf n},{\bf n } ' } \frac{\partial}{\partial \phi_{{\bf n } } } \left\ { \left ( ( \epsilon_{{\bf n } } - \mu(t ) ) \delta_{{\bf n},{\bf n } ' } + \hbar\sigma^{(+)}_{{\bf n},{\bf n } ' } \right ) \phi_{{\bf n } ' } \raisebox{0.3 in } { } \right . \\ & & \hspace*{1.4 in } + \left . \sum_{{\bf m},{\bf m } ' } v^{(+)}_{{\bf n},{\bf m};{\bf n}',{\bf m } ' } \phi^*_{{\bf m } } \phi_{{\bf m } ' } \phi_{{\bf n } ' } \right\ } p[\phi^*,\phi;t ] \nonumber \\ & & + \sum_{{\bf n},{\bf n } ' } \frac{\partial}{\partial \phi^*_{{\bf n } } } \left\ { \left ( ( \epsilon_{{\bf n } } - \mu(t ) ) \delta_{{\bf n},{\bf n } ' } + \hbar\sigma^{(-)}_{{\bf n}',{\bf n } } \right ) \phi^*_{{\bf n } ' } \raisebox{0.3 in } { } \right . \nonumber \\ & & \hspace*{1.4 in } + \left . \sum_{{\bf m},{\bf m } ' } v^{(-)}_{{\bf n}',{\bf m}';{\bf n},{\bf m } } \phi^*_{{\bf n } ' } \phi^*_{{\bf m } ' } \phi_{{\bf m } } \right\ } p[\phi^*,\phi;t ] \nonumber \\ & & - \frac{1}{2 } \sum_{{\bf n},{\bf n } ' } \frac{\partial^2}{\partial \phi_{{\bf n } } \partial \phi^*_{{\bf n } ' } } \left\ { \hbar\sigma^k_{{\bf n},{\bf n } ' } + \sum_{{\bf m},{\bf m } ' } v^k_{{\bf n},{\bf m};{\bf n}',{\bf m } ' } \phi^*_{{\bf m}}\phi_{{\bf m } ' } \right\ } p[\phi^*,\phi;t]~. \nonumber\end{aligned}\ ] ] for this equation to be useful , however , we need to determine the various selfenergies and interactions . to do so , let us for simplicity first consider the normal state of the gas which is usually in the weak - coupling limit @xmath805 for the experiments with atomic alkali gases that are of interest here . in this regime we can neglect the nondiagonal elements of the selfenergies and also the nonlinear terms in the right - hand side of equation ( [ fpinhom ] ) . the fokker - planck equation then reduces to @xmath806 = - \sum_{{\bf n } } \frac{\partial}{\partial \phi_{{\bf n } } } ( \epsilon_{{\bf n } } + \hbar\sigma^{(+)}_{{\bf n } } - \mu(t ) ) \phi_{{\bf n } } p[\phi^*,\phi;t ] \\ & & \hspace*{0.23 in } + \sum_{{\bf n } } \frac{\partial}{\partial \phi^*_{{\bf n } } } ( \epsilon_{{\bf n } } + \hbar\sigma^{(-)}_{{\bf n } } - \mu(t ) ) \phi^*_{{\bf n } } p[\phi^*,\phi;t ] \nonumber \\ & & \hspace*{0.23 in } - \frac{1}{2 } \sum_{{\bf n } } \frac{\partial^2}{\partial \phi^*_{{\bf n } } \partial \phi_{{\bf n } } } \hbar\sigma^k_{{\bf n } } p[\phi^*,\phi;t]~ , \nonumber\end{aligned}\ ] ] with @xmath807 and @xmath808 . this equation looks identical to the one we derived for our caldeira - leggett toy model but describes in fact quite different physics because now the selfenergies are not due to an interaction with a reservoir but due to the interactions between the atoms themselves . this is , however , completely hidden in the selfenergies as we show now explicitly . performing the same manipulations as in sections [ ifd ] and [ ap ] , we find in the many - body t - matrix approximation first of all that @xmath809 and similarly that @xmath810 our next task is therefore the evaluation of the various many - body t - matrix elements @xmath811 that occur in these expressions , which is easily carried out with our experience of the homogeneous gas and by noting that the lippmann - schwinger equation is in fact an operator equation that only needs to be expressed in a new basis . in terms of the appropriate matrix elements of the interatomic interaction , i.e. , @xmath812 the retarded and advanced components of the many - body t matrix thus obey the bethe - salpeter equation @xcite @xmath813 and the optical theorem @xmath814 finally , we also need @xmath815 and the keldysh component @xmath816 notice that the appearance of a @xmath440 function in the last three equations in practice never leads to any problems , because in the experiments with trapped atomic gases the number of particles is always so large that the critical temperature of the gas is much larger than the energy splitting of the trap and the discrete sum over states is well approximated by an integral over a continuous spectrum . making use of the above relations we find that for the states with thermal energies the renormalized energy is given by the pseudopotential result @xmath817 with @xmath818 the density profile of the gas cloud . for the states near the one - particle ground state there is , however , a correction given by @xmath819 and @xmath820 . in the weak - coupling limit , the renormalization of the one - particle energies plays by definition not such an important role and we are much more interested in the imaginary part of the retarded and advanced selfenergies . they are given by @xmath821~. \nonumber \end{aligned}\ ] ] furthermore , the keldysh component of the selfenergies equals @xmath822~. \nonumber \end{aligned}\ ] ] as a result the quantum boltzmann equation for the gas becomes @xmath823 with the rate to scatter out of the state @xmath824 given by @xmath825 and the rate to scatter into this state by @xmath826 as expected from fermi s golden rule . the quantum boltzmann equation found above , in combination with the expressions for the shift in the energy levels , fully describes the dynamics of the gas in the normal state . however , close to the critical temperature when the occupation numbers of the one - particle groundstate @xmath827 start to become large , we are no longer allowed to neglect the nonlinear terms in the fokker - planck equation in equation ( [ fpinhom ] ) . applying again the hartree - fock approximation , which is essentially exact in the weak - coupling limit as we have seen in the equilibrium case , and substituting the ansatz @xmath595 = p_0[\phi^*_{\bf 0},\phi_{\bf 0};t ] p_1[\phi'^*,\phi';t]$ ] into equation ( [ fpinhom ] ) , we obtain for the probability distribution of the condensate @xmath828 \\ & & = - \frac{\partial}{\partial \phi_{\bf 0 } } \left ( \epsilon_{\bf 0 } + \hbar\sigma^{(+)}_{\bf 0 } - \mu(t ) + t^{(+)}_{{\bf 0},{\bf 0};{\bf 0},{\bf 0}}(t ) |\phi_{\bf 0}|^2 \right ) \phi_{\bf 0 } p_0[\phi^*_{\bf 0},\phi_{\bf 0};t ] \nonumber \\ & & + \frac{\partial}{\partial \phi^*_{\bf 0 } } \left ( \epsilon_{\bf 0 } + \hbar\sigma^{(-)}_{\bf 0 } - \mu(t ) + t^{(+)}_{{\bf 0},{\bf 0};{\bf 0},{\bf 0}}(t ) |\phi_{\bf 0}|^2 \right ) \phi^*_{\bf 0 } p_0[\phi^*_{\bf 0},\phi_{\bf 0};t ] \nonumber \\ & & - \frac{1}{2 } \frac{\partial^2}{\partial \phi^*_{\bf 0 } \partial \phi_{\bf 0 } } \hbar \sigma^k_{\bf 0 } p_0[\phi^*_{\bf 0},\phi_{\bf 0};t]~ , \nonumber \end{aligned}\ ] ] where it should be remembered that for consistency reasons the selfenergies need to be evaluated at the energies @xmath829 . in equilibrium the fluctuation - dissipation theorem then reads @xmath830 \propto \exp \left\ { -\beta \left ( \epsilon'_{\bf 0 } - \mu + \frac{t^{(+)}_{{\bf 0},{\bf 0};{\bf 0},{\bf 0}}}{2 } |\phi_{\bf 0}|^2 \right ) |\phi_{\bf 0}|^2 \right\}~. \nonumber\end{aligned}\ ] ] the fokker - planck equation , and hence the boltzmann equation , for the noncondensed part of the gas is the same as in the normal state because for a discrete system the contributions from the condenstate to the selfenergies is already included in our expressions for @xmath831 and needs not be separated out explicitly . this implies of course , that if we evaluate the rates @xmath832 and @xmath833 by replacing the sum over states by an integral over a continuum , we first need to separate the contribution from the condensate to obtain the correct result . in summary we have thus found that in the weak - coupling limit the dynamics of bose - einstein condensation in an external trapping potential is described by the coupled equations in equation ( [ beinhom ] ) and equation ( [ fpcinhom ] ) , i.e. , by a boltzmann equation for the noncondensed part of the gas and a fokker - planck equation for the condensate , respectively . if we apply these equations to a bose gas with a positive scattering length , we essentially arrive at the theory recently put forward by gardiner and coworkers @xcite . indeed , neglecting the energy shifts and assuming that the noncondensed cloud is in equilibrium with a chemical potential @xmath834 , our fokker - planck equation for a @xmath424 invariant probability distribution of the condensate becomes @xmath835 = - \frac{\beta}{4 } \hbar \sigma^k_{\bf 0 } \frac{\partial}{\partial \phi_{\bf 0 } } \left ( \epsilon_{\bf 0 } - \mu + t^{(+)}_{{\bf 0},{\bf 0};{\bf 0},{\bf 0 } } |\phi_{\bf 0}|^2 \right ) \phi_{\bf 0 } p_0[|\phi_{\bf 0}|;t ] \\ & & \hspace*{0.2 in } - \frac{\beta}{4 } \hbar \sigma^k_{\bf 0 } \frac{\partial}{\partial \phi^*_{\bf 0 } } \left ( \epsilon_{\bf 0 } - \mu + t^{(+)}_{{\bf 0},{\bf 0};{\bf 0},{\bf 0 } } |\phi_{\bf 0}|^2 \right ) \phi^*_{\bf 0 } p_0[|\phi_{\bf 0}|;t ] \nonumber \\ & & \hspace*{0.2 in } - \frac{1}{2 } \hbar \sigma^k_{\bf 0 } \frac{\partial^2}{\partial \phi^*_{\bf 0 } \partial \phi_{\bf 0 } } p_0[|\phi_{\bf 0}|;t]~. \nonumber \end{aligned}\ ] ] . the inset shows the evolution of the average condensate number and its fluctuations . ] it is equivalent to the fokker - planck equation for a single - mode laser , since introduction of the dimensionless time @xmath836 , the dimensionless condensate number @xmath837 , and the so - called ` pump parameter ' @xmath838 leads to the fokker - planck equation @xmath839 = \frac{\partial}{\partial i } \left\ { 2(i - a)i + 4i \frac{\partial}{\partial i } \right\ } p[i;\tau]\ ] ] well - known from laser theory @xcite . the typical behaviour of @xmath840 $ ] during bose - einstein condensation in shown in figure 17 . calculating @xmath841 from equation ( [ sml ] ) by making use of the mean - field assumption @xmath842 , precisely reproduces the theory of gardiner _ et al . _ a comparison with the experiments of h .- j . shows that the theory of gardiner and coworkers is unfortunately not fully satisfactory @xcite . the reason for this is presumably that these experiments are really in the strong - coupling limit , where the above theory does not apply . as mentioned previously , the theory is however applicable to gases with a negative scattering length . since we know from our experience with homogeneous gases that the chemical potential will never be larger than @xmath843 in this case , we clearly see from the equilibrium probability distribution @xmath844 $ ] that a condensate in a gas with effectively attractive interactions is only metastable and will ultimately collapse to a dense state after it has overcome an energy barrier , either by thermal fluctuations or by quantum mechanical tunneling . this is clearly in full agreement with the discussion in section [ nsl ] . furthermore , we now also understand how the dynamics of the collapse must be treated . it requires the solution of a stochastic , dissipative nonlinear schrdinger equation , in which the dissipative terms and the noise obey an appropriate fluctuation - dissipation theorem . having arrived at this important result , we are finally left with a discussion of the strong - coupling limit . the difficulty with the strong - coupling limit is that the appropriate one - particle states of the gas are no longer the states @xmath824 of the trapping potential because these states are now strongly coupled by the mean - field interaction . for our caldeira - leggett model this problem could be easily resolved by introducing the states @xmath845 that incorporate the average interaction with the reservoir . however , for a trapped bose gas the same procedure would lead to an expansion into states @xmath846 that parametrically depend on time since the mean - field interaction depends on time also . in principle , we thus have to deal with the situation that not only the occupation numbers vary with time but also the states to which these occupation numbers belong . this results in highly nontrivial dynamics since it is _ a priori _ clear that an adiabatic approximation , in which we neglect the nondiagonal matrix elements @xmath847 , is not valid , because of the lack of a separation of time scales . notice that we were able to circumvent the above problems for a bose gas with effectively attractive interactions due to the fact that on the one hand the growth of the condensate essentially takes place in the weak - coupling regime and that on the other hand the collapse , which certainly occurs in the strong - coupling regime , requires in first instance only knowledge of the collisionless dynamics of the condensate . notwithstanding the above remarks , we can make progress in the strong - coupling limit if we realize that in this limit the typical length scale on which the ( noncondensate ) density of the gas varies is always much larger than the trap size . we are therefore justified in performing a gradient expansion . to do so in a systematic fashion , it is convenient to express the various selfenergies and interactions in the coordinate representation , i.e. , by introducing @xmath848 and similarly @xmath849 in this coordinate representation , our fokker - planck equation acquires in all generality the form @xmath850 = - \int d{\bf x } d{\bf x}'~ \frac{\delta}{\delta \phi({\bf x } ) } \\ & & \hspace*{0.2 in } \left\ { \left ( \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu(t ) \right ) \delta({\bf x}-{\bf x } ' ) + \hbar\sigma^{(+)}({\bf x},{\bf x } ' ) \right ) \phi({\bf x } ' ) \right . \nonumber \\ & & \hspace*{0.4 in } + \left . \int d{\bf y } d{\bf y}'~ v^{(+)}({\bf x},{\bf x}';{\bf y},{\bf y } ' ) \phi^*({\bf y } ) \phi({\bf y } ' ) \phi({\bf x } ' ) \raisebox{0.2 in } { } \right\ } p[\phi^*,\phi;t ] \nonumber \\ & & + \int d{\bf x } d{\bf x}'~ \frac{\delta}{\delta \phi^*({\bf x } ) } \nonumber \\ & & \hspace*{0.2 in } \left\ { \left ( \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu(t ) \right ) \delta({\bf x}-{\bf x } ' ) + \hbar\sigma^{(-)}({\bf x}',{\bf x } ) \right ) \phi^*({\bf x } ' ) \right . \nonumber \\ & & \hspace*{0.4 in } + \left . \int d{\bf y } d{\bf y}'~ v^{(-)}({\bf x}',{\bf x};{\bf y}',{\bf y } ) \phi^*({\bf x } ' ) \phi^*({\bf y } ' ) \phi({\bf y } ) \raisebox{0.2 in } { } \right\ } p[\phi^*,\phi;t ] \nonumber \\ & & - \frac{1}{2 } \int d{\bf x } d{\bf x}'~ \frac{\delta^2}{\delta \phi({\bf x } ) \delta \phi^*({\bf x } ' ) } \left\ { \hbar\sigma^k({\bf x},{\bf x } ' ) \raisebox{0.2 in } { } \right . \nonumber \\ & & \hspace{0.4 in } + \left . \int d{\bf y } d{\bf y}'~ v^k({\bf x},{\bf x}';{\bf y},{\bf y } ' ) \phi^*({\bf y } ) \phi({\bf y } ' ) \right\ } p[\phi^*,\phi;t]~. \nonumber\end{aligned}\ ] ] to study its physical content and to familiarize ourselves with how to perform the desired gradient expansion , we first analyze again the normal state of the gas in which we can neglect the nonlinear terms representing the effect of the condensate . in the normal state the condensate is absent and we are only interested in the behavior of the quantity @xmath851 . from the above fokker - planck equation we see that it obeys the equation of motion @xmath852 introducing now the wigner distribution @xcite by means of @xmath853 and similarly the selfenergies @xmath854 by @xmath855 we find after substituting these relations in lowest nonvanishing order in the gradient @xmath856 the expected quantum boltzmann equation @xcite @xmath857 where the energies of the atoms have to be determined selfconsistently from @xmath858~.\ ] ] moreover , the expressions for the retarded energy @xmath859 , and the collision rates @xmath860 and @xmath861 turn out to be identical to the ones for a homogeneous gas @xcite . we only need to replace the one - particle energies @xmath862 and occupation numbers @xmath863 by their local counterparts @xmath864 and @xmath865 , respectively . notice also that the same substitution has to be made in the bethe - salpeter equation for the many - body t matrix , which implies that the effective interaction now implicitly depends both on the position @xmath51 and the time @xmath30 . as an example , let us consider the gas not too close to the critical temperature . then the effective interaction is well approximated by the two - body t matrix and we simply have @xmath866 as a result the equilibrium solution @xmath867 of the quantum boltzmann equation equals a bose - einstein distribution evaluated at the energy @xmath868 . the equilibrium density profile @xmath869 thus corresponds to a local - density approximation but is indeed very accurate for the atomic gases of interest . apart from describing the relaxation of the gas towards equilibrium , the quantum boltzmann equation can also be used to study the collective modes of the trapped gas above the critical temperature . using the same methods that have recently been applied to inhomogeneous bose gases by zaremba , griffin , and nikuni @xcite , we can for example derive the hydrodynamic equations of motion the gas . moreover , the collisionless modes are obtained from @xmath870 solving this equation is equivalent to the generalized random - phase approximation @xcite . more important is that the above collisionless boltzmann equation can be shown to satisfy the kohn theorem @xcite by substituting the shifted equilibrium profile @xmath871 . for an harmonic external potential the center - of - mass of the gas cloud then precisely follows newton s law @xmath872 , which shows that the gas has three collisionless modes with frequencies that are equal to the three frequencies of the external trapping potential . the full quantum boltzmann equation in equation ( [ qbe ] ) also describes the relaxation of the gas towards equilibrium . it can thus be used to study forced evaporative cooling by adding an appropriate loss term to account for the escaping high - energetic atoms . in agreement with our previous picture we then expect that , if the evaporative cooling is performed slowly , the distribution function @xmath865 can be well approximated by a truncated equilibrium distribution with a time - dependent temperature and chemical potential . as a result , we can use the results from the homogeneous gas to show that in the critical regime the gas again becomes unstable towards bose - einstein condensation . this is the case when @xmath873 for positions near the center of the trap where the density of the gas is highest . if that happens we are no longer allowed to neglect the nonlinear terms in the fokker - planck equation and some corrections are neccesary . applying again the hartee - fock approximation , which is correct sufficiently close to the critical temperature where @xmath874 , we find first of all that the quantum boltzmann equation for the noncondensed part of the gas has the same form as in equation ( [ qbe ] ) . we only have to add the by now familiar mean - field interaction @xmath875 due to the condensate to the renormalized energy @xmath864 . finally , the scattering rates @xmath860 and @xmath861 have to be adjusted in precisely the same way as in the homogeneous case to account for the macroscopic occupation of the condensate @xcite . in addition , the resulting quantum boltzmann equation for the noncondensed atoms is coupled to a fokker - planck equation for the condensate . after a gradient expansion the latter equals @xmath876 \\ & & = - \int d{\bf x}~ \frac{\delta}{\delta \phi_{\bf 0}({\bf x } ) } \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu(t ) + \hbar\sigma^{(+)}({\bf x},{\bf 0 } ) \right . \nonumber \\ & & \hspace{1.22 in } + \left . { } ^{{}^{{}^{{}^{{}^{{}^ { } } } } } } t^{(+)}({\bf 0},0 ) |\phi_{\bf 0}({\bf x})|^2 \right ) \phi_{\bf 0}({\bf x } ) p_0[\phi_{\bf 0}^*,\phi_{\bf 0};t ] \nonumber \\ & & + \int d{\bf x}~ \frac{\delta}{\delta \phi_{\bf 0}^*({\bf x } ) } \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + v^{\rm ex}({\bf x } ) - \mu(t ) + \hbar\sigma^{(-)}({\bf x},{\bf 0 } ) \right . \nonumber \\ & & \hspace{1.22 in } + \left . { } ^{{}^{{}^{{}^{{}^{{}^ { } } } } } } t^{(+)}({\bf 0},0 ) |\phi_{\bf 0}({\bf x})|^2 \right ) \phi_{\bf 0}^*({\bf x } ) p_0[\phi_{\bf 0}^*,\phi_{\bf 0};t ] \nonumber \\ & & - \frac{1}{2 } \int d{\bf x}~ \frac{\delta^2}{\delta \phi_{\bf 0}({\bf x } ) \delta \phi_{\bf 0}^*({\bf x } ) } \hbar\sigma^k({\bf x},{\bf 0 } ) p_0[\phi_{\bf 0}^*,\phi_{\bf 0};t]~ , \nonumber\end{aligned}\ ] ] where @xmath877 denotes @xmath878 . in more detail we have for the real part of the retarded and advanced selfenergies that @xmath879 $ ] and @xmath880 the local energy of a condensate atom . moreover , using our picture of the evaporative cooling of the gas , the imaginary parts follow most conveniently from the fluctuation - dissipation theorem @xmath881 where @xmath882 can in a good approximation be considered as independent of @xmath883 . near equilibrium the fokker - planck equation thus simplifies to @xmath884 \\ & & = - \frac{\beta}{4 } \int d{\bf x}~ \hbar\sigma^k({\bf x},{\bf 0 } ) \frac{\delta}{\delta \phi_{\bf 0}({\bf x } ) } \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + \epsilon'({\bf x},{\bf 0 } ) - \mu \right . \nonumber \\ & & \hspace{1.2 in } + \left . { } ^{{}^{{}^{{}^{{}^{{}^ { } } } } } } t^{(+)}({\bf 0},0 ) |\phi_{\bf 0}({\bf x})|^2 \right ) \phi_{\bf 0}({\bf x } ) p_0[|\phi_{\bf 0}|;t ] \nonumber \\ & & - \frac{\beta}{4 } \int d{\bf x}~ \hbar\sigma^k({\bf x},{\bf 0 } ) \frac{\delta}{\delta \phi_{\bf 0}^*({\bf x } ) } \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + \epsilon'({\bf x},{\bf 0 } ) - \mu \right . \nonumber \\ & & \hspace{1.2 in } + \left . { } ^{{}^{{}^{{}^{{}^{{}^ { } } } } } } t^{(+)}({\bf 0},0 ) |\phi_{\bf 0}({\bf x})|^2 \right ) \phi_{\bf 0}^*({\bf x } ) p_0[|\phi_{\bf 0}|;t ] \nonumber \\ & & - \frac{1}{2 } \int d{\bf x}~ \hbar\sigma^k({\bf x},{\bf 0 } ) \frac{\delta^2}{\delta \phi_{\bf 0}({\bf x } ) \delta \phi_{\bf 0}^*({\bf x } ) } p_0[|\phi_{\bf 0}|;t ] \nonumber\end{aligned}\ ] ] and the probability distribution of the condensate relaxes to the expected stationary state @xmath885 \propto \exp \left\ { - \beta \int d{\bf x}~ \phi_{\bf 0}^*({\bf x } ) \left ( - \frac{\hbar^2\mbox{\boldmath $ \nabla$}^2}{2 m } + \epsilon'({\bf x},{\bf 0 } ) - \mu \right . \right . \\ & & \hspace*{2.0 in } \left . \left . + \frac{t^{(+)}({\bf 0},0)}{2 } |\phi_{\bf 0}({\bf x})|^2 \right ) \phi_{\bf 0}({\bf x } ) \right\}~. \nonumber\end{aligned}\ ] ] from this result we conclude that eqs . ( [ qbe ] ) and ( [ fpsc ] ) give an accurate discription of the dynamics of bose - einstein condensation in the strong - coupling regime . unfortunately , the solution of these coupled equations is not possible analytically and we therefore have to refer to future work for a numerical study and a comparison of this theory with experiment . we here only remark that the semiclassical approximation in this case amounts to the thomas - fermi result for the condensate density @xmath886 such an approximation implicitly assumes , however , that the dominant mechanism for the formation of the condensate is by means of a coherent evolution and not by incoherent collisions . this might be expected not to be very accurate on the basis of the homogeneous result that the second mechanism only diverges logarithmically with the system size @xcite , whereas the first diverges linearly @xcite . nevertheless it seems that only a full solution of equations ( [ qbe ] ) and ( [ fpsc ] ) can settle this important issue . although we have up to now primarily been focussed on the dynamics of bose - einstein condensation itself , it is interesting that we have in fact , similarly as for temperatures above the critical temperature , also arrived at an accurate discription of the collective modes of a condensed gas at nonzero temperatures . since these collective modes are receiving much attention at present , we now briefly want to discuss how such a description emerges from our fokker - planck equation . since most experiments at present are performed in the collisionless limit , we consider only this limit here . it is in principle possible , however , to derive the hydrodynamic equations of motions as well @xcite . let us for simplicitly first assume that the temperature of the gas is sufficiently far below the critical temperature that the many - body t matrix is well approximated by the two - body t matrix , but not so low that the hartree - fock approximation is no longer valid . in this regime we have that @xmath887 and the dynamics of the noncondensed part of the gas is again determined by the collisionless quantum boltzmann equation given in equation ( [ clqbe ] ) , where the total density now consists of two contributions @xmath888 the dynamics of the condensate follows again from the nonlinear schrdinger equation @xmath889 which can be rewritten as the continuity equation @xmath890 together with the josephson relation for the superfluid velocity @xmath891 here , we used that @xmath892 , @xmath893 , and have applied the thomas - fermi approximation , which is appropriate in the strong - coupling limit . it will presumably not come as a surprise that these coupled equations for @xmath865 , @xmath894 , and @xmath895 can be shown to fulfill the kohn theorem exactly . this is an important constraint on any theory for the collective modes of a trapped gas , as stressed recently by zaremba , griffin , and nikuni @xcite . as a first step towards this goal we here consider the time - dependent hartree - fock theory for the collective modes of the bose condensed gas , in which the collisionless boltzmann equation in equation ( [ clqbe ] ) is coupled to the nonlinear schrdinger equation in equation ( [ nlse ] ) . it turns out that a numerical determination of the eigenmodes of the collisionless boltzmann equation is rather difficult . therefore , we have recently proposed a variational approach to solve these coupled equations @xcite . denoting the anisotropic harmonic trapping potential used in the experiments @xcite by @xmath896 , we first of all take for the condensate wave function the gaussian _ ansatz _ @xmath897 with the functions @xmath898 describing the ` breathing ' of the condensate in three independent directions . clearly , this is the anisotropic generalization of the gaussian _ ansatz _ that we used in section [ nsl ] to study the stability of a condensate with negative scattering length . although it is not immediately obvious that such an _ ansatz _ is also appropriate in the strong - coupling regime of interest here , it has been shown by perez - garcia _ et al . _ @xcite that at zero temperature the assumption of a gaussian density profile nevertheless leads to rather accurate results for the frequencies of the collective modes . we therefore anticipate that at nonzero temperatures the same will be the case . next we also need an _ ansatz _ for the distribution function @xmath865 of the thermal cloud . here , we consider the bose distribution @xmath899 where @xmath900 are again three scaling parameters , @xmath901 is the number of noncondensed particles , @xmath902 denotes the equilibrium value of @xmath900 that in general is greater than @xmath65 due to the repulsive interactions between the atoms , and @xmath903 . the functional form of the distribution function is physically motivated by the following reasons . first , by integrating over the momentum @xmath664 we see that the time - dependent density profile of the noncondensed cloud @xmath904 is a time - independent function of @xmath905 , as desired . second , for such a density profile we must have that the local velocity @xmath906 is given by @xmath907 . third , we want the equilibrium distribution to be only anisotropic in coordinate space , and not in momentum space . this explains the appearance of the factor @xmath908 in the exponent . the full lines give the results with a bose distribution _ ansatz _ for the distribution function of the thermal cloud . curves 1 and 2 correspond to the @xmath909 in and out - of - phase modes , respectively . similarly , curves 3 and 4 give the frequencies of the @xmath910 in and out - of - phase modes . the experimental data is also shown . triangles are for a @xmath909 mode and circles for a @xmath910 mode . ] using the ideal gas result for the number of condensate particles and substituting the above _ ansatz _ for the condensate wave function and the wigner distribution function into the nonlinear schrdinger and collisionless boltzmann equations , it is straightforward but somewhat tedious to derive the six coupled equations of motion for the ` breathing ' parameters @xmath898 and @xmath900 . linearizing these equations around the equilibrium solutions @xmath911 and @xmath902 , respectively , we can then also obtain the desired eigenmodes and eigenfrequencies . the results of this calculation for the trap parameters of jin _ et al . _ @xcite are shown in figure 18 together with the experimental data . note that the trap has a disk geometry with @xmath912 and @xmath913 . due to the axial symmetry , the modes of the gas can be labeled by the usual quantum number @xmath4 that describes the angular dependence around the @xmath914-axis . on the whole there appears to be a quite reasonable agreement between this simple theory and experiment . the only exceptions are the two data point almost halfway in between the in and out - of - phase @xmath909 modes . a possible explanation for this discrepancy is that experimentally both modes are excited simultaneously @xcite . to make sure of this , however , we need to determine the oscillator strength of the various modes and it appears that the in - phase mode indeed dominates the response for temperatures larger than about @xmath915 @xcite . with the latter remark we end this course on the use of field - theoretical methods for the study of the equilibrium and nonequilibrium properties of trapped atomic gases . although we have in principle illustrated all the necessary tools for an _ ab initio _ treatment of these new quantum systems , many interesting topics still need to be considered in detail . we already mentioned such topics as fermi - bose mixtures , the dynamics of condensate collapse , the strong - coupling theory for the formation of a condensate in either atomic bose of fermi gases , and the hydrodynamics of single or multicomponent atomic gases . several other important areas of research , which can also be easily addressed within the context of quantum field theory , are atom lasers , the damping of collective modes , the behaviour of spinor condensates , the dynamics of topological excitations such as kinks , vortices and skyrmions , quantum critical phenomena , the optical properties of superfluid gases , and two - dimensional phase transitions . in view of these and many other possibilities that are now experimentally feasible , it appears certain that the physics of degenerate atomic gases will remain very exciting for years to come . 106 e. abers and b.w . lee , phys . rep . * 9c * , 1 ( 1973 ) . k.g . wilson and j. kogut , phys . rep . * 12 * , 75 ( 1974 ) . hohenberg and b.i . halperin , rev . phys . * 49 * , 435 ( 1977 ) . s. sachdev in _ proceedings of the 19th iupap international conference on statistical physics _ , edited by b .- l . hao ( world scientific , singapore , 1996 ) . p.a . lee and t.v . ramakrishnan , rev . phys . * 57 * , 287 ( 1985 ) . anderson , phys . rev . * 109 * , 1492 ( 1958 ) . laughlin in _ the quantum hall effect _ , edited by r.e . prange and s.m . girvin ( springer - verlag , new york , 1990 ) . wen , phys . b * 43 * , 11025 ( 1991 ) mehta , _ random matrices _ ( academic press , new york , 1991 ) . anderson , j.r . ensher , m.r . matthews , c.e . wieman , and e.a . cornell , science * 269 * , 198 ( 1995 ) . bradley , c.a . sackett , j.j . tollett , and r.g . hulet , phys . lett . * 75 * , 1687 ( 1995 ) ; c.c . bradley , c.a . sackett , and r.g . hulet , _ ibid . _ * 78 * , 985 ( 1997 ) . davis , m .- o . mewes , m.r . andrews , n.j . van druten , d.s . durfee , d.m . kurn , and w. ketterle , phys lett . * 75 * , 3969 ( 1995 ) . a. einstein , sitz . ( berlin ) , 3 ( 1925 ) . i.f . silvera and j.t.m . walraven , phys . 44 * , 164 ( 1980 ) . a.l . fetter and j.d . walecka , _ quantum theory of many - particle systems _ ( mcgraw - hill , new york , 1971 ) . note that these authors use the more common notation @xmath916 and @xmath917 for the bosonic creation and annihilation operators . l. mandel and e. wolf , _ optical coherence and quantum optics _ ( cambridge , new york , 1995 ) . negele and h. orland , _ quantum many - particle systems _ ( addison - wesley , new york , 1988 ) . amit , _ field theory , the renormalization group , and critical phenomena _ ( world scientific , singapore , 1984 ) . j. zinn - justin , _ quantum field theory and critical phenomena _ ( oxford , new york , 1989 ) . dyson , phys . rev . * 75 * , 1736 ( 1949 ) . stratonovich , sov . phys * 2 * , 416 ( 1958 ) . j. hubbard , phys . rev . lett , * 3 * , 77 ( 1959 ) . d.g . fried , t.c . killian , l. willmann , d. landhuis , s.c . moss , d. kleppner , and t.j . greytak , phys . * 81 * , 3811 ( 1998 ) . bogoliubov , j. phys . ( u.s.s.r . ) * 11 * , 23 ( 1947 ) pitaevskii , sov . jetp * 13 * , 451 ( 1961 ) and e.p . gross , j. math . phys . * 4 * , 195 ( 1963 ) . jin , j.r . ensher , m.r . matthews , c.e . wieman , and e.a . cornell , phys . lett . * 77 * , 420 ( 1996 ) ; d.s . jin , m.r . matthews , j.r . ensher , c.e . wieman , and e.a . cornell , _ ibid . _ * 78 * , 764 ( 1997 ) . m .- o . mewes , m.r . anderson , n.j . van druten , d.m . kurn , d.s . durfee , c.g . townsend , and w. ketterle , phys . lett . * 77 * , 988 ( 1996 ) . hau , b.d . busch , c. liu , z. dutton , m.m . burns , and j.a . golovchenko , phys . a * 58 * , r54 ( 1998 ) . singh and d.s . rokhsar , phys . lett . * 77 * , 1667 ( 1996 ) . m. edwards , p.a . ruprecht , k. burnett , r.j . dodd , and c.w . clark , phys . 77 * , 1671 ( 1996 ) . fetter , ann . phys . * 70 * , 67 ( 1972 ) . m. lewenstein and l. you , phys . lett . * 77 * , 3489 ( 1996 ) . popov , _ functional integrals in quantum field theory and statistical physics _ ( reidel , dordrecht , 1983 ) and references therein . f. dalfovo , s. giorgini , l. pitaevskii , and s. stringari , rev . mod . * 71 * , 463 ( 1999 ) . hutchinson , e. zaremba , and a. griffin , phys . rev * 78 * , 1842 ( 1997 ) . dodd , m. edwards , c.w . clark , and k. burnett , phys . a * 57 * , r32 ( 1998 ) . b.a . lippmann and j. schwinger , phys . rev . * 79 * , 469 ( 1950 ) . m. bijlsma and h.t.c . stoof , phys . a * 54 * , 5085 ( 1996 ) . h. shi and a. griffin , phys . rep . * 304 * , 1 ( 1998 ) . j. bardeen , l.n . cooper , and j.r . schrieffer , phys . rev . * 108 * , 1175 ( 1957 ) . h. kleinert , forts . phys . * 26 * , 565 ( 1978 ) . abraham , w.i . mcalexander , j.m . gerton , r.g . hulet , r. ct , and a. dalgarno , phys . a * 55 * , r3299 ( 1997 ) . stoof , m. houbiers , c.a . sackett , and r.g . hulet , phys . rev . lett . * 76 * , 10 ( 1996 ) ; m. houbiers , r. ferwerda , h.t.c . stoof , w.i . mcalexander , c.a . sackett , and r.g . hulet , phys . rev . a * 56 * , 4864 ( 1997 ) . for an equilibrium argument at zero temperature see l.d . landau and e.m . lifshitz , _ statistical physics _ ( pergamon , london , 1958 ) and p. nozires , in _ bose - einstein condensation _ , edited by a. griffin , d.w . snoke , and s. stringari , p. 15 ( cambridge , new york , 1995 ) . nonzero temperatures are discussed in h.t.c . stoof , phys . a * 49 * , 4704 ( 1995 ) . m. houbiers and h.t.c . stoof , phys . a * 54 * , 5055 ( 1996 ) . t. bergeman , phys . a * 55 * , 3658 ( 1997 ) . see , for instance , s. coleman in _ aspects of symmetry _ ( cambridge , new york , 1985 ) . freire and d.p . arovas , phys . a * 59 * , 1461 ( 1999 ) . s. stringari , phys . lett . * 77 * , 2360 ( 1996 ) . y. castin and r. dum , phys . 77 * , 5315 ( 1996 ) . perez - garcia , h. michinel , j.i . cirac , m. lewenstein , and p. zoller , phys . lett . * 77 * , 5320 ( 1996 ) . dodd , m. edwards , c.j . williams , c.w . clark , m.j . holland , p.a . ruprecht , and k. burnett , phys . rev . a * 54 * , 661 ( 1996 ) . j. javanainen , phys . rev . a * 54 * , 3722 ( 1996 ) . l. you , w. hoston , and m. lewenstein , phys . a * 55 * , 1581 ( 1997 ) . p. hberg , e.l . surkov , i. tittonen , s. stenholm , m. wilkens , and g.v . shlyapnikov , phys . a * 56 * , r3346 ( 1997 ) . for the anisotropic generalization see , for instance , g. baym and c.j . pethick , phys . lett . * 76 * , 6 ( 1996 ) . fetter , phys . a * 53 * , 4245 ( 1996 ) . the exact result is obtained in p.a . ruprecht , m.j . holland , k. burnett , and m. edwards , phys . a * 51 * , 4704 ( 1995 ) . stoof , j. stat . phys . * 87 * , 1353 ( 1997 ) . for a different calculation of the tunneling rate that however neglects the phase fluctuations of the condensate , see e.v . shuryak , phys . a * 54 * , 3151 ( 1996 ) . sackett , c.c . bradley , m. welling , and r.g . hulet , appl . b * 65 * , 433 ( 1997 ) sackett , h.t.c . stoof , and r.g . hulet , phys . lett . * 80 * , 2031 ( 1998 ) . pitaevskii , phys . a * 221 * , 14 ( 1996 ) . sackett , j.m . gerton , m. welling , and r.g . hulet , phys . * 82 * , 876 ( 1999 ) . anderson , phys . rev . * 112 * , 1900 ( 1958 ) . e. abrahams and t. tsuneto , phys . rev . * 152 * , 416 ( 1966 ) . in principle the effective action for a neutral superconductor contains also a topological term , but this does not affect the final result of the following argument because it only leads to a constant shift in the total number of particles . for more details , see m. stone , int . b * 9 * , 1359 ( 1995 ) and references therein . the magnitude of the frequency gap is actually not accurately described by this time - dependent ginzburg - landau theory , because the action @xmath918 $ ] is implicitly only valid for long wavelengths @xmath919 and low frequencies @xmath920 . this may be compared directly with the results of e.m . wright , d.f . walls , and j.c . garrison , phys . * 77 * , 2158 ( 1996 ) and k. mlmer , phys . a * 58 * , 566 ( 1998 ) . v.v . goldman , i.f . silvera , and a.j . leggett , phys . b * 24 * , 2870 ( 1981 ) . for simplicity we here again neglect a topological term in the action , which essentially plays no role in the following . this is also in agreement with our discussion of the superconductor , where exactly the same topological term was not included in the effective action for the global phase @xcite . thomas , proc . * 23 * , 542 ( 1927 ) and e. fermi , mat . * 6 * , 602 ( 1927 ) . see , however , t .- l . ho and v.b . shenoy , phys . lett * 77 * , 2595 ( 1996 ) . ho , phys . rev . lett . * 81 * , 742 ( 1998 ) . m. houbiers and h.t.c . stoof , phys . a * 59 * , 1556 ( 1999 ) . see , for example , k. huang , _ statistical mechanics _ ( wiley , new york , 1987 ) . j. schwinger , j. math . phys . * 2 * , 407 ( 1961 ) . keldysh , zh . fiz . * 47 * , 1515 ( 1964 ) [ sov . jetp * 20 * , 1018 ( 1965 ) ] . p. danielewicz , ann . ( n.y . ) * 152 * , 239 ( 1984 ) . b. de wit , private communication . p. carruthers and k.s . dy , phys . rev . * 147 * , 214 ( 1966 ) . h. kleinert , _ path integrals in quantum mechanics statistics and polymer physics _ ( world scientific , singapore , 1994 ) . this follows from the requirement that we want to be able to fourier transform the heaviside function . caldeira and a.j . leggett , phys . * 46 * , 211 ( 1981 ) and a.o . caldeira and a.j . leggett , ann . ( n.y . ) * 149 * , 374 ( 1983 ) ; _ ibid . _ * 153 * , 445 ( 1984 ) kadanoff and g. baym , _ quantum statistical mechanics : green s function methods in equilibrium and nonequilibrium problems _ ( addison - wesley , new york , 1962 ) . d.c . langreth and j.w . wilkins , phys . b * 6 * , 3189 ( 1972 ) . in agreement with our previous remarks , the notation @xmath921 implies that the time integration can be either from @xmath586 to @xmath922 or from @xmath586 to @xmath30 . n.g . van kampen , _ stochastic processes in physics and chemistry _ ( north - holland , amsterdam , 1981 ) . d. forster , _ hydrodynamic fluctuations , broken symmetry , and correlation functions _ ( benjamin , reading , 1975 ) . to make a connection with a large body of knowledge in quantum optics , we note that equation ( [ stat ] ) explicitly shows that the probability distribution @xmath595 $ ] does not correspond to a @xmath923 or @xmath924 representation of the density matrix , but to a wigner representation instead . the ` diffusion matrix ' in the fokker - planck equation for @xmath595 $ ] is therefore diagonal and positive , as is explained by c.w . gardiner in _ quantum noise _ ( springer , berlin , 1991 ) , chapter 6 . the same nice feature will also appear in the case of an interacting bose gas . kastner , rev . phys . * 64 * , 849 ( 1992 ) and references therein . see also h.t.c . stoof in _ bose - einstein condensation _ , edited by a. griffin , d.w . snoke , and s. stringari , p. 226 ( cambridge , new york , 1995 ) . luiten , m.w . reynolds , and j.t.m . walraven , phys . rev . a * 53 * , 381 ( 1996 ) . n.p . proukakis and k. burnett , j. res . * 101 * , 457 ( 1996 ) ; n.p . proukakis , k. burnett , and h.t.c . stoof , phys . rev . a * 57 * , 1230 ( 1998 ) . gardiner , p. zoller , r.j . ballagh , and m.j . davis , phys . lett . * 79 * , 1793 ( 1997 ) . h. risken , z. phys . * 186 * , 85 ( 1965 ) ; _ ibid . _ * 191 * , 302 ( 1966 ) . miesner , d.m . stamper - kurn , m.r . andrews , d.s . durfee , s. inouye , and w. ketterle , science * 279 * , 1005 ( 1998 ) . for a recent extension of their theory see c.w . gardiner , m.d . lee , r.j . ballagh , and m.j . davis , and p. zoller , phys . lett . * 81 * , 5266 ( 1998 ) . m. hillery , r.f . oconnell , m.o . scully , and e.p . wigner , phys . rep . * 106 * , 121 ( 1984 ) . d.v . semikoz and i.i . tkachev , phys . lett . * 74 * , 3093 ( 1995 ) . stoof , j. low temp . phys . * 114 * , 11 ( 1999 ) . e. zaremba , a. griffin , and t. nikuni , phys . a * 57 * , 4695 ( 1998 ) . see also a. griffin , w .- c . wu , and s. stringari , phys . 78 * , 1838 ( 1997 ) , g.m . kavoulakis , c.j . pethick , and h. smith , phys . a * 57 * , 2938 ( 1998 ) , t. nikuni and a. griffin , j. low . phys . * 111 * , 793 ( 1998 ) , and v. shenoy and t .- l . ho , phys . rev . lett . * 80 * , 3895 ( 1998 ) . w. kohn , phys . rev . * 123 * , 1242 ( 1961 ) . kirkpatrick and j.r . dorfman , j. low temp . * 58 * , 301 ( 1985 ) ; _ ibid . _ * 58 * , 399 ( 1985 ) . e. zaremba , t. nikuni , and a. griffin , j. low temp . phys . * 116 * , 277 ( 1999 ) . k. damle , s.n . majumdar , and s. sachdev , phys . a * 54 * , 5037 ( 1996 ) . bijlsma and h.t.c . stoof , phys . a * 60 * ( 1999 ) . cornell , private communication .
a fundamental problem in understanding the evolution of galaxy clusters and their constituent galaxies is the importance of dynamical processes such as tidal stripping . theoretical models of cluster evolution have considered collisions , mergers , dynamical friction and tidal stripping in order to explain the galaxy morphology - density relation and the presence of luminous halos around supergiant d galaxies . it is generally accepted that tidal stripping will remove the more loosely bound stars in the outer envelopes of galaxies @xcite . @xcite has argued that the extensive envelopes of cd galaxies consist of this tidal debris moving in the potential well of the cluster . however , the efficiency of tidal stripping is not well understood , because of uncertainties in the velocities , the mean tidal field , the distribution of dark matter , and the orbits of stars in the outer envelopes of the stripped galaxies . typical theoretical estimates indicate that 10 - 20% of the total cluster light should be in the intergalactic medium @xcite and that the amount of icl should increase with time @xcite . @xcite and @xcite have therefore stressed the importance of obtaining reliable measurements of the intracluster light ( icl ) as a direct indicator of the tidal damage suffered by galaxies . attempts to measure the icl date back to the early 1970s , originally being done with photographic plates . more recently ccd surveys have been employed by @xcite , @xcite , @xcite , and @xcite with results ranging from non - detections up to @xmath1% of the total cluster light ( surface brightnesses of @xmath2 mag arcsec@xmath3 ) . even with ccds , detecting the icl directly is notoriously difficult and fraught with uncertainties such as the contributions from overlapping galaxy halos . an independent means of measuring the icl , not subject to the same problems as measuring the extremely low surface brightness diffuse emission is needed . methods using planetary nebulae ( pne ) have been developed and applied to virgo @xcite and fornax @xcite . intergalactic red - giant - branch stars have also been detected in virgo by comparing the luminosity function of point sources in intracluster space with control fields far from any cluster or galaxy @xcite . intracluster sne have also been identified and have extended measurements of the icl out to @xmath4 @xcite . novae typically range in brightness from m@xmath5 to @xmath6 , while the bright end of the pn luminosity function reaches m@xmath7 @xcite and red - giant - branch stars typically reach m@xmath8 . supernovae are much brighter , but also much rarer than novae . novae can therefore be detected more easily in a given cluster , or be observed to greater distances than pne or red - giant stars . the transient nature of novae and their h@xmath9 brightness help eliminate the contamination suffered by pne or red - giant - branch stars due to background emission line objects or unresolved compact galaxies . another appealing property of novae is that the rate of novae per unit mass is high enough such that with a deep enough survey , one can expect to accrue a statistically interesting number of novae in each epoch , thus increasing the accuracy of the icl measurement with each new epoch . these properties give novae the potential to be useful as an independent and accurate measure of the icl in nearby clusters . in spite of this , to our knowledge , no intra - cluster novae have been reported in the literature to date . as an independent check on highly uncertain direct measures of intracluster light , we can use novae to verify the amount of stellar tidal debris in the cluster . this then places constraints on the importance of tidal stripping in the formation and evolution of galaxy clusters . our m81 nova results from the most complete nova survey ever undertaken indicate that novae are preferentially associated with the older , bulge populations in galaxies @xcite . we can use this knowledge to attempt to discern the predominant source of the icl either in bulge - dominated or disk - dominated galaxies . another lesson we have learned from our nova studies is that accurate rates and distributions can only be derived from comprehensive , densely time - sampled surveys @xcite . the data set we use in this paper is diverse and very sparsely sampled . we therefore restrict our investigation to a relative comparison of galaxy versus inter - galaxy novae . our aim is to prove the _ feasibility _ of using novae to perform an accurate assessment of the amount and origin of the icl . we have acquired six epochs of survey data of the central 30 square of the fornax cluster . fornax is nearby and has a fairly high surface density of galaxies in the core , increasing the likelihood of tidal stripping . at the distance of fornax , @xmath10 = 31.35 @xcite , novae will have an average m@xmath11(peak ) @xmath12 ( the brightest novae will have peak brightnesses of m@xmath11 = 21.35 ) , and are easily detectable . novae have been observed by @xcite in virgo cluster galaxies , which are at the same distance as fornax . novae are very bright in h@xmath9 ( b @xmath13 h@xmath14 when the nova erupts , and b @xmath13 h@xmath15 shortly thereafter ) . furthermore , novae remain bright in h@xmath9 for an extended period of time ( months ) , providing a useful means of discovering them . to a limiting magnitude of m@xmath16 = -7.5 ( m@xmath16 = 23.85 at fornax ) the mean h@xmath9 visibility lifetime is 3 months @xcite . assuming a nova rate of 24 yr@xmath17 per 10@xmath18 ( b ) @xcite , a luminosity in galaxies of m@xmath19 in the core of fornax @xcite , a visibility lifetime in h@xmath9 of 3 months , and a conservative completeness factor of 0.5 due to the high surface brightness of galaxy cores , we estimate that at any given epoch there should be @xmath1 novae in eruption ( with visible h@xmath9 emission ) in galaxies in the central 1 deg@xmath20 of the fornax cluster . since we are surveying the central @xmath21 deg@xmath20 we multiply this estimate by @xmath22 and predict we will see 12 novae in any given epoch . the observations that went into this study are a diverse set acquired over the span of more than a decade . the central 30 of fornax , centered on ra 03:38:13.72 , dec -35:32:52.2 ( j2000 ) , was surveyed . this area includes the central dominant ( cd ) galaxy ngc 1399 and three other elliptical galaxies : ngc 1404 , ngc 1387 , and ngc 1389 . this region was imaged twice in 1993 , once in 1995 , for five nights in 1999 , and once each in 2003 and 2004 . the data were acquired during programs for other projects and during engineering time or director s discretionary time . these observations are summarized in table [ tabforobs ] . because of the long - term nature of the data acquisition , there was no uniform observing strategy employed . we started with the tek2048 camera , and when the mosaic ii camera became available ( before the 1999 run ) , we switched to that camera for the remaining epochs . we used standard johnson b- and r - band filters and narrow - band h@xmath9 filters primarily , but for two epochs ( b and c , see below ) we were constrained to using a t1 filter from the washington system ( @xmath23 , fwhm @xmath24 , canterna 1976 ) , which is roughly equivalent to a johnson r filter ( @xmath25 , fwhm @xmath26 ) . the h@xmath9 filter used for the 1993 a epoch was the 6600/75 ctio filter with @xmath27 and @xmath28 which accounts for the velocity of fornax ( 1517 @xmath29 91 km s@xmath17 , grillmair et al . the h@xmath9 filter used for the 1999 and 2003 epochs was the mosaic ii c6009 filter with @xmath30 and @xmath31 . there is no h@xmath9 filter in the mosaic ii set that correctly accounts for the velocity of fornax , so we chose the filter whose central wavelength comes the closest to 6600 . we will discuss the impact of using this filter below . total exposure times were restricted by the gaps available in the schedule for other programs . individual exposure times were chosen to provide at least three frames per epoch so cosmic rays could be removed in the combining process . for the h@xmath9 frames the only exceptions to this are the 1999 g epoch ( 1 exposure ) and the 1999 h epoch ( 2 exposures ) . since cosmic rays could not be adequately removed from these exposures , they were only used for nova confirmation and measurement and not detection ( see below ) . all of the data were acquired using the ctio 4 m telescope with the exception of the b and c epoch data ( taken exclusively through the t1 filter ) which were acquired with the ctio 1.5 m telescope . the data acquired in 1993 and 1995 were taken with the tek2048 chip . the focal ratios of the 4 m and the 1.5 m are such that they produce an identical pixel scale for these images of 043 per pixel . the resulting field of view is 147 and thus required four pointings to cover the central 30 of the cluster . the seeing tabulated in table [ tabforobs ] for these epochs represents the range of seeing across the four pointings . sky flats were acquired in each filter to remove the instrumental response . the data acquired in 1999 , 2003 , and 2004 were taken with the mosaic ii camera on the 4 m telescope . this camera is a mosaic of eight 2048x4096 pixel ccds with a pixel scale of 027 per pixel . the resulting field of view is 368 and thus required only one pointing . sky flats were used to remove the instrumental response for the 1999 epochs , and dome flats were used for the 2003 and 2004 epochs . all data were reduced using the standard tools in iraf @xcite or the * mscred * external iraf package developed by the noao mosaic instrument team . individual exposures for a given epoch were combined to produce a coadded image for the epoch and to remove all but a few cosmic rays . classical novae erupt with amplitudes of up to 20 magnitudes , reaching absolute magnitudes as high as m@xmath32 and fading back to their pre - outburst brightness in a matter of weeks or months @xcite . novae in the fornax cluster are therefore invisible before outburst and fade back to invisibility after outburst on time scales of less than a year . thus , our primary criterion for detecting novae is that they be transient on this timescale . for the data used in this study this means that they must only appear in only one year s images . if a variable object appears in images from more than one year , it is most likely some other kind of variable source : a foreground periodic , or semi - periodic variable , or a background , unresolved active galaxy nucleus ( agn ) . classical novae also exhibit strong h@xmath9 emission in outburst , providing the next criterion : the candidate should be h@xmath9-bright . data from 1993 and 1999 were taken both through h@xmath9 and red continuum filters allowing us to produce h@xmath9 - red continuum subtracted images . these images were used to test candidates from 1993 and 1999 against this criterion . the primary method for implementing these criteria consisted of performing a spatial registration of the different epochs , and then blinking them against each other . we required that any candidate nova be visible on each of the individual exposures for the epoch in which it was detected . this eliminated cosmic ray hits from our candidate list . this method worked well for candidates outside of the galaxies . for candidates in the galaxies , we employed the spatial median filter technique described in @xcite with a box size of 17 pixels . the median subtracted image was then blinked in the same way as the unsubtracted images . for the epochs taken through an h@xmath9 - continuum filter pair , we scaled and subtracted the continuum image from the h@xmath9 image to accentuate the objects with h@xmath9 emission . the a ( 1993 ) and the e , g , and h ( 1999 ) epochs had this filter combination which allowed us to blink the h@xmath9-bright objects in these epochs . figure [ fnx_nov_pos ] shows the locations of the detected nova candidates on an image from the digitized sky survey covering the central 38 of the fornax cluster , corresponding to our survey area . the nova candidates are labeled by number and the galaxy identifications are also indicated . the positions of the novae are listed in table [ tab_fnx_nov_pos ] . since each candidate was discovered using slightly different techniques , each will be discussed in detail below . _ nova 1_. this candidate was discovered in the t1 image from the 1993 epoch b. it was easily compared with the t1 image from the 1995 epoch c which has a deeper frame limit afforded by the longer exposure time ( see table [ tabforobs ] ) . we did a careful spatial registration of the images from other epochs and found that , indeed , this object is a transient variable meeting the first criterion described above . it appears near a faint unresolved object , but is at least 10 times brighter than that object . figure [ fnx_n1 ] shows the three epochs of t1 images of the region ( a , b , and c ) and a subtraction of epoch c from epoch b , which removes the other stars in the field and shows nova 1 as distinct from the faint point source . _ nova 2_. the 1993 a epoch h@xmath9 - t1 image revealed an h@xmath9-bright source near the galaxy ngc 1389 . our deep h@xmath9 image from from the 1999 e epoch allowed us to easily confirm that this source was transient and an excellent nova candidate , as illustrated in figure [ fnx_n2 ] . in addition , we confirmed that the object was not visible in images from any other year . _ nova 3_. in this case we found an h@xmath9-bright source in the h@xmath9 - r image from the 1999 e epoch near the galaxy ngc 1399 as shown in figure [ fnx_n3 ] . we confirmed its transient nature by the fact that it was undetectable in all h@xmath9 and r or t1 filter epochs from other years . because we had 5 closely spaced epochs in 1999 , we were able to generate a light curve in h@xmath9 and b for this candidate ( see below ) and measure its decline rate . _ nova 4_. this candidate is nearly identical to nova 3 in the way it was discovered , although its magnitude is fainter . it is shown in figure [ fnx_n4 ] . the light curve is displayed in the next section and shows an offset between the h@xmath9 and b - band brightness of 2 magnitudes , similar to that shown by the novae in m31 @xcite . due to its faintness , we were unable to derive a decline rate in the b - band for this candidate . it was also not seen in the images from any other year . _ nova 5_. this candidate was discovered in the 2003 i h@xmath9 epoch , which did not have a corresponding red continuum image to allow subtraction . it was compared with the deeper 1999 e epoch which proves that the source is transient , as shown in figure [ fnx_n5 ] . it also does not appear on any continuum ( b or r or t1-band ) image in any other epoch . _ nova 6_. we discovered this candidate in the b - band image from the 1999 epochs after we obtained the 2004 j b - band epoch ( see figure [ fnx_n6 ] ) . in the 1999 e h@xmath9 epoch it fell on a part of the chip that had a readout problem and therefore was missed using the h@xmath9 - red continuum subtraction technique . it was missed in the comparison between the b - band epochs of a ( 1993 ) and e ( 1999 ) because the depth of the 1993 a epoch b - band image was not sufficient for a good comparison . after detecting this candidate in the b - band image from 1999 , we then went back and found the corresponding h@xmath9-bright source in the h@xmath9 - r - band image . we were unable to calibrate this candidate s h@xmath9 magnitude to better than 0.5 magnitudes because there were no calibration sources on the part of the chip with the readout problem . we were able to generate a b - band light curve ( see below ) and measure its decline rate . the h@xmath9 magnitude is highly uncertain for this candidate because of the readout problem mentioned above . nova instrumental magnitudes were derived using daophot @xcite aperture photometry . to account for variable seeing , we set the aperture radii to 1/2 the fwhm of the stellar profile measured from high signal - to - noise , well - isolated stars . photometric calibration was achieved using a variety of secondary standards . for b - band calibration we used the yale / san juan southern proper motion program @xcite which had 7 unsaturated stars in common with our b - band observations in 1999 . this produced a calibration accurate to 0.1 magnitudes . for the calibration of the t1 , and r - band photometry we used the r - band magnitudes for the globular clusters published by @xcite . for the calibration we used measurement apertures that were at least as large as those used to produce the published magnitudes @xcite . since the t1 and the r filter cover a similar wavelength range , we simply forced the t1 magnitudes to the r system . this calibration is also accurate to 0.1 magnitudes . for the h@xmath9 magnitudes , we had no calibrated h@xmath9 sources in our fields and no observations of flux standards in our h@xmath9 filters . we decided , instead , to use the r - band magnitudes from @xcite to calibrate our h@xmath9 magnitudes . this procedure is less than ideal because it introduces several uncertainties . novae are h@xmath9 emission line objects and globular clusters are continuum objects . this has been overcome in the past by using a filter that is broad enough to include most of the h@xmath9 light and then assuming the h@xmath9 light fills the bandpass , i.e. by just quoting a filter magnitude and not a line flux ( see e.g. ciardullo et al . 1989 ) . this will hold for the 1993 a epoch which used an h@xmath9 filter that is correctly redshifted for fornax and a width that includes most of the h@xmath9 light ( nova expansion velocities up to 1700 km s@xmath17 ) . for the other epochs which used an unshifted h@xmath9 filter , most of the h@xmath9 light will fall toward the red edge of the filter bandpass where the transmission is dropping rapidly . without spectra of each nova , the amount of flux lost is unknown . for objects from these filters , there will be an unknown systematic error in the filter magnitudes . this also makes calculating the limiting magnitudes for these epochs uncertain as well . with this in mind , we simply calculated the offset between the r system and our instrumental h@xmath9 magnitudes using the globular clusters , and applied this offset to force our instrumental h@xmath9 magnitudes to the r system . the number of objects used to perform the calibration varied from 13 to 30 . for the 1993 a epoch the external error in h@xmath9 is better than 0.2 magnitudes , while for the other epochs the external h@xmath9 error is such that the magnitudes quoted are overestimates ( the novae are probably brighter ) . table [ tab_fnx_nov_phot ] lists the calibrated magnitudes for each nova and figure [ fnx_lcs ] shows the b and h@xmath9 light curves for the multiply observed novae ( nova 3 , nova 4 , and nova 6 ) . we used error - weighted linear fits to derive a decline rate for nova 3 of 0.183 m@xmath11 day@xmath17 and for nova 6 of 0.205 m@xmath11 day@xmath17 . the fits are are indicated in figure [ fnx_lcs ] by the thin solid lines . the errors quoted in table [ tab_fnx_nov_phot ] and plotted in figure [ fnx_lcs ] are the internal measurement errors and in most cases are smaller than the external calibration errors . could these six candidates be some other kind of variable and not novae ? without spectra , it is impossible to be certain . all of our candidates are transient according to the criterion discussed in the first paragraph of [ novae_det ] , thus making periodic or semi - regular variables unlikely contaminators . four of the six candidates ( novae 2 , 3 , 4 , and 6 ) are h@xmath9-bright and transient , virtually ruling out other possible variables ( see below ) . other types of variables that are transient include background agn , @xmath33-ray bursts ( grbs ) , supernovae ( sne ) , and microlensing events . we must resort to statistical arguments in an attempt to eliminate each of these classifications for novae 1 and 5 . the long - term quasar variability study of @xcite shows that the rms variability of their sample is 0.45 mag in the blue and 0.33 mag in the red ( see their figure 4 ) . both nova 1 and nova 5 are measured at least a magnitude above the plate limit in the red ( t1 for nova 1 and h@xmath9 for nova 5 ) making it unlikely that these could be background agn that have varied on timescales of a year . we also checked the positions of all our novae against the deep _ chandra _ fornax x - ray survey of @xcite and found no coincidences with any x - ray sources . as an additional check we consulted the nasa / ipac extragalactic database ( ned ) and found no coincidences with nova 1 or nova 5 out to a search radius of 30 arcseconds . the total grb rate derived from _ batse _ detections is about 600 yr@xmath17 , of which @xmath040% are dark , or not visible in the optical @xcite . this gives an optical grb rate of close to 1 per day over the entire sky . our survey was roughly 30 arcminutes on a side or 0.25 square degrees , so the a priori probability of detecting a grb in our survey field in any given epoch is @xmath34 . an estimate of the sn rate in the local universe is given by @xcite who gives the rate for all types of sne of @xmath35 sn @xmath36 yr@xmath17 . if we compare this to the nova rate from @xcite of 24 novae @xmath37 yr@xmath17 we see that novae are @xmath38 times more common than sne in the local universe ( using h@xmath39 mpc@xmath17 ) . this makes it more likely that our candidates are novae rather than sne in the fornax cluster that declined substantially before we detected them . it is also very unlikely that a sn in the fornax cluster would go undetected since it would reach a brightness of 12th magnitude and would decay with a timescale of months . that being said , we can see sne 100 times more distant than novae ( sn reach m@xmath40 ) . we must , therefore , consider sne in undetected background galaxies and intracluster sne in background clusters . @xcite did find a background cluster in their deep x - ray survey of fornax , but it is located 2.6 arcminutes from nova 1 and has a diameter of 1.4 arcminutes in the soft band and so it is not likely that nova 1 is a sn associated with this cluster . we can not rule out the possibility that the object next to nova 1 is an unresolved galaxy and that nova 1 is a sn in that galaxy . a spectrum of this object would resolve this issue , but it would require a very large telescope since it is very faint ( r @xmath0 24.5 ) . our deepest b - band image has a plate limit of 26 . no objects are found to this limit at the positions of any of the novae ( with the exception of nova 1 ) . if nova 5 was really a sn in an undetected background galaxy or cluster it would have to be fainter than b of 26 and therefore much fainter than the sn itself . although unlikely , we can not rule this possibility out . to assess the contamination from microlensing events , we will turn to a microlensing survey of m87 in virgo by @xcite . using the wfpc2 camera on _ hst _ every day for a month , this study detected 1 strong microlensing candidate . it s peak magnitude was fainter than 24 and would have been barely detected in our deepest survey epochs . they report that this detection rate is consistent with the sensitivity and coverage of their survey and their microlensing models that included self - lensing from stars in m87 , lensing from the m87 halo , the virgo halo , and foreground lensing from the milky way @xcite . our sensitivity to microlensing events would be considerably less than their survey due to our lower resolution and shallower depth ( our faintest measurement in the b - band is 24.8 , their faint limit is 28.5 assuming a 3@xmath41 detection , see baltz et al . the microlensing models from @xcite show that self - lensing and lensing from the milky way halo provide the smallest contributions to the microlensing rate . the largest microlensing rates come from the virgo halo , however , the rate depends strongly on the concentration of lenses toward the center of m87 where there are more sources to be lensed . this means that our much larger survey area does not drastically increase our sensitivity to microlensing events relative to theirs . this is because even though we include many more lenses ( the fornax cluster halo ) , we only include a factor of a few more sources to be lensed ( the stars of ngc 1399 and the other three galaxies ) which would not make up for the much brighter faint limit and the lower resolution of our survey . from the models of @xcite , we assume that the probability for a microlensing event drops with underlying stellar density . from their observed rate and the relative similarity of m87 and ngc 1399 , we would expect to barely detect one event in ngc 1399 in one month . we can eliminate the microlensing classification for our h@xmath9-bright candidates because microlensing events are not h@xmath9-bright . the two remaining nova candidates , nova 1 and nova 5 , are distant from background stellar light and so the probability of detecting a lensing event in those regions should be much less than predicted in the study by @xcite . admitting that the classification for all our candidates is not iron - clad , these arguments indicate the most probable interpretation is that our candidates are novae . we will assume so for our subsequent analysis . we were fortunate that three of the novae erupted during the five epoch run in 1999 . the two measured decline rates ( see figure [ fnx_lcs ] ) of 0.183 m@xmath11 day@xmath17 for nova 3 and 0.205 m@xmath11 day@xmath17 for nova 6 place both in the fast nova category . if we assume a distance modulus to the fornax cluster of @xmath42 @xcite , then our brightest b magnitude for nova 3 is at m@xmath439.2 . typically , novae reach peak b magnitudes of m@xmath44 , so nova 3 is clearly a brighter - than - average nova ( see , e.g. , figure 5.3 in warner 1995 ) . the flat shape of the h@xmath9 light curve indicates that we have caught this nova near h@xmath9 maximum and during the earliest phases of the b decline . the brightness of nova 3 is consistent with its fast decline rate since decline rate is known to correlate with nova peak brightness @xcite . given the decline rate of nova 6 , it is possible that this nova was at least as bright as nova 3 at maximum , but that we are observing it in its later decline phase . the large uncertainty in the h@xmath9 magnitude for this nova does not allow us to compare its b and h@xmath9 magnitudes . with the uncertainties in the survey h@xmath9 limit mentioned above it is difficult to quantitatively assess the nova populations in the cluster . as stated in [ fnx_intro ] , we predicted 12 novae to have visible h@xmath9 emission down to the limit of m@xmath45 in any given h@xmath9 epoch , yet we detected only 5 novae in our 3 h@xmath9 epochs , a factor of 7 less than we expected . there are several factors that could reduce the number of novae we detected . intracluster novae with velocities that would shift their h@xmath9 emission even farther to the red would lose a significant amount of their flux in the epochs using the unshifted h@xmath9 filter ( epochs e , g , h , and i ) . accounting for the velocity dispersion of the intracluster population and the individual galaxies , it is possible that we are down by at least a factor of 4 by using the unshifted h@xmath9 filter for two of the three years of the survey . we point out that our earlier results indicate that novae are predominantly associated with the bulge population of m81 @xcite . novae , therefore , have the potential to provide some clues to the source of the intracluster light . a long - term , densely time - sampled study is required to properly assess the nova parent population in fornax . since all regions in the survey were sampled with equal frequency , an analysis of the spatial distribution of the fornax novae is worthwhile , though it can not be conclusive . we deliberately limited ourselves to two simple questions about the spatial distribution of novae in the fornax cluster . first , assuming that novae follow the bulge light in galaxies @xcite , how many novae did we miss in the centers of the galaxies ? second , based on the light profiles of the galaxies , how many novae are associated with the cluster potential and not with any galaxy ? to answer these questions , we required the light profiles of the four galaxies in our survey region : ngc 1399 , ngc 1404 , ngc 1389 , and ngc 1387 . since ngc 1399 has an extended halo , we used the study of @xcite for its light profile . they used a combination of wide - field photographic plates and narrower - field ccd data to generate a profile that extends nearly 2000 arcseconds from the center of ngc 1399 . for the other profiles , we used our b - band mosaic image from the 1999 epochs . we measured their profiles using the ellipse task in the stsdas * isophote * package . isophotes were fit on background subtracted images with the background being determined by the region in the chip most distant from the galaxy . the task was given the initial center position , position angle ( pa ) , and ellipticity and allowed to vary these parameters at each semi - major axis position , which was set to grow geometrically and not linearly . this allowed the outer isophotes to contain enough signal to produce reasonable fits . to answer the incompleteness question , we used artificial stars to determine our detection limit for each galaxy . our goal was to produce a simple correction to the number of novae found in the galaxies . we used the artificial stars to ask the question , how close to the nucleus of the galaxy could the nova that was discovered in that galaxy be found ? by comparing this radial limit to the light profile , we know how much light in the galaxy was effectively unsurveyed for novae of the brightness already discovered , and can thus make a correction based on the fraction of the total light unsurveyed . daophot was used to generate a point spread function ( psf ) and then used to add artificial stars around each galaxy . the galaxy images were then median - filtered , and subtracted , and ( as described above ) blinked to see where the artificial novae become undetectable . in all cases this limit was determined by the residual light left by the subtraction of the median - filtered image . thus , the inner profiles of the galaxies strongly affected this detection limit . figure [ fnx_profs ] shows the cumulative light profiles for the four galaxies , normalized to the total light . we set the total light limit at 26 b magnitudes per square arcsecond for the galaxies we measured . the profile of ngc 1404 was contaminated by a very bright star 180 " from the galaxy , so we set a limit at that distance of 25 b magnitudes per square arcsecond . the first thing we notice is the extent of the halo of ngc 1399 . nearly the entire survey area is enclosed by this halo . we will assume that this halo actually traces the inner potential well of the cluster itself as suggested by @xcite . we also see that nova 2 is clearly associated with ngc 1389 : it is close to the half - light radius , but the association of the other novae is not straightforward . nova 4 appears at a radius that encloses over 80% of the light of ngc 1387 , while nova 5 and nova 6 are even farther outside of ngc 1404 . interpreting the nova placement in ngc 1399 is complicated by the extended halo , although it seems clear that nova 3 belongs with that galaxy . the extent of the halo of ngc 1399 means that nova 1 is located just beyond the half - light radius , but clearly it is more properly placed in the cluster rather than in that galaxy ( see figure [ fnx_nov_pos ] ) . @xcite found that the transition from being bound to ngc 1399 and being bound solely to the cluster occurs at a distances between 15 and 55 from the center of ngc 1399 ( see their figure 2 ) . nova 1 is 82 from the center of ngc 1399 and well past this transition . we also placed nova 6 on the ngc 1399 light profile and it appears at a radius that encloses only 65% of the light . to be conservative , we ll assume that 4 of our novae ( 2 , 3 , 4 , and 5 ) are associated with the galaxies in the cluster and 2 ( 1 and 6 ) are not . but how many novae did we miss in the centers of the galaxies ? to generate a correction for missed novae , we look at the fraction of light unsurveyed , as determined by our detection limits and plotted as the vertical dot - dashed lines in figure [ fnx_profs ] . for ngc 1399 this fraction of unsurveyed light is 0.15 , for ngc 1404 it is 0.4 , for ngc 1387 it is 0.3 and for ngc 1389 it is 0.45 , giving a total unsurveyed light fraction of @xmath46 . since we found , on average , 1 nova associated with a galaxy , this would essentially add another @xmath47 novae to our number associated with galaxies , giving a final score of 5.3 novae in and 2 novae outside of galaxies in the cluster . the x - ray study of the fornax cluster by @xcite offers some interesting evidence concerning the association of the novae with the galaxies nearby . figure [ fnx_xray ] shows the x - ray flux levels near ngc 1399 and ngc 1404 generated from a soft band image ( 0.3 - 1.5 kev ) with an exposure time of @xmath045 ksec @xcite . ten evenly spaced contour levels ranging from a flux of 7.7@xmath4810@xmath49 erg cm@xmath3 s@xmath17 arcmin@xmath3 to a flux of 4.9@xmath4810@xmath51 erg cm@xmath3 s@xmath17 arcmin@xmath3 are overplotted on our b - band image from epoch j ( 2004 ) . three of our novae ( nova 3 , nova 5 and nova 6 ) fall in this region and are indicated by small boxes and labeled with their identifications . as pointed out in @xcite , ngc 1404 has a bow - shock on the side facing ngc 1399 due to its motion toward ngc 1399 . the comet - like tail they describe is apparent in the lowest contour of figure [ fnx_xray ] . this figure solidifies the association of nova 6 with intracluster space and not with ngc 1404 because it is well outside the lowest x - ray contours for ngc 1404 and well outside the bow - shock of the x - ray gas in ngc 1404 . using our optical light profiles alone ( see figure [ fnx_profs ] ) , we would associate nova 5 with ngc 1404 . we see from figure [ fnx_xray ] that nova 5 is outside of the bow - shock of ngc 1404 . without a detailed investigation of the dynamics of ngc 1404 , we can not conclusively say that the stars outside the bow - shock are not bound to ngc 1404 , but the placement of nova 5 is suggestive of its membership in the intracluster population . this evidence suggests that we might revise our score to 4.3 novae in galaxies and 3 novae in intracluster space . assuming that novae trace the intracluster light in a galaxy cluster , we calculate that the icl contributes @xmath52% of the total light in the fornax cluster . this is very close to the estimate of 40% by @xcite , who used intergalactic planetary nebulae ( pne ) . this is likely an overestimate of the icl due to contamination by background ly@xmath9-emitting galaxies masquerading as pn @xcite . if we move one nova from the out to the in category ( e.g. , nova 5 ) , we get an icl fraction of @xmath53% of the total light , similar to the icl fraction of 22% estimated for virgo by @xcite from intergalactic pne . if we also assume that one of our intracluster novae is really a background sn ( e.g. , nova 1 ) , then we get an icl fraction of @xmath54% . all these numbers are within the range of other measures of icl . _ our key result is that we have demonstrated for the first time that intergalactic tramp novae very likely exist in the fornax cluster , and that they can be useful tracers of the icl_. clearly , a longer - term , densely time - sampled survey in h@xmath9 and continuum light of this region would allow us to provide a much more accurate measure of the icl in fornax . using an h@xmath9 filter that accounts for both the redshift of fornax and the velocity dispersion of the cluster and member galaxies as well as the range of nova line widths would improve the number of novae detected in each epoch by a factor of up to 7 over the survey reported here . this kind of survey would not only bolster the statistics by adding more novae , but would also provide light curves which can be used to better verify novae and estimate their peak brightnesses . it would also allow us to estimate the nova rates in each galaxy and in the icl . this would be the first step toward assessing the source population of the icl and estimating the effect of the galaxy cluster environment on close binary stars . \1 . intra - cluster novae do exist in the fornax cluster and their distribution is consistent with @xmath016 - 41% of the total light in the cluster being contributed by the icl . the high end of this range is close to the estimate of the icl in fornax using intracluster pne by @xcite . the low end is close to the estimate of the icl in virgo using intracluster pne by @xcite . tramp novae are very interesting and useful as tracers of the icl . their brightness at peak ( m@xmath11 up to @xmath1310 ) allows them to be visible out to the coma cluster with current telescopes . new objects can be accumulated with each observation epoch , increasing the accuracy of the measure of the icl with time . as our understanding of nova parent populations improves , they may become useful for constraining the origin of the tidal debris in clusters . we wish to thank martin bureau , jeremy mould , tim abbott , knut olsen , and chris smith for their support in acquiring fornax cluster survey epochs in 2003 and 2004 . we would also like to thank the anonymous referee for very useful comments on the manuscript . 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 . arnaboldi , m. , et al . 2003 , , 125 , 514 baltz , e. a. , lauer , t. r. , zurek , d. r. , gondolo , p. , shara , m. m. , silk , j. , & zepf , s. e. 2004 , astro - ph/0310845 bernstein , g. m. , nichol , r. c. , tyson , j. a. , ulmer , m. p. , & wittman , d. 1995 , , 110 , 1507 boughn , s. p. , kuhn , j. r. , & uson , j. m. 1990 , iau symp . 139 : the galactic and extragalactic background radiation , 139 , 357 cappellaro , e. 2001 , memorie della societa astronomica italiana , 72 , 863 canterna , r. 1976 , , 81 , 228 ciardullo , r. , jacoby , g. h. , ford , h. c. , & neill , j. d. 1989 , , 339 , 53 ciardullo , r. , shafter , a. w. , ford , h. c. , neill , j. d. , shara , m. m. , & tomaney , a. b. 1990a , , 356 , 472 ciardullo , r. , tamblyn , p. , jacoby , g. h. , ford , h. c. , & williams , r. e. 1990b , , 99 , 1079 ciardullo , r. , feldmeier , j. j. , krelove , k. , jacoby , g. h. , & gronwall , c. 2002 , , 566 , 784 dressler , a. 1984 , , 22 , 185 durrell , p. r. , ciardullo , r. , feldmeier , j. j. , jacoby , g. h. , & sigurdsson , s. 2002 , , 570 , 119 feldmeier , j. j. , ciardullo , r. , & jacoby , g. h. 1998 , , 503 , 109 feldmeier , j. j. , ciardullo , r. , jacoby , g. h. , & durrell , p. r. 2003 , , 145 , 65 ferguson , h. c. 1989 , , 98 , 367 ferguson , h. c. , tanvir , n. r. , & von hippel , t. 1998 , , 391 , 461 gal - yam , a. , maoz , d. , guhathakurta , p. , & filippenko , a. v. 2003 , , 125 , 1087 grillmair , c. j. , freeman , k. c. , bicknell , g. v. , carter , d. , couch , w. j. , sommer - larsen , j. , & taylor , k. 1994 , , 422 , l9 helfand , d. j. , stone , r. p. s. , willman , b. , white , r. l. , becker , r. h. , price , t. , gregg , m. d. , & mcmahon , r. g. 2001 , , 121 , 1872 kissler - patig , m. , grillmair , c. j. , meylan , g. , brodie , j. p. , minniti , d. , & goudfrooij , p. 1999 , , 117 , 1206 maddox , s. j. , efstathiou , g. , sutherland , w. j. , & loveday , j. 1990 , , 243 , 692 merritt , d. 1984 , , 276 , 26 miller , g. e. 1983 , , 268 , 495 murante , g. , et al . 2004 , , 607 , l83 neill , j. d. & shara , m. m. 2004 , 127 , 816 partridge , r. b. 1990 , iau symp . 139 : the galactic and extragalactic background radiation , 139 , 283 platais , i. , et al . 1998 , , 116 , 2556 pritchet , c. j. & van den bergh , s. 1987 , , 318 , 507 richstone , d. o. 1976 , , 204 , 642 saglia , r. p. , kronawitter , a. , gerhard , o. , & bender , r. 2000 , , 119 , 153 shara , m. m. 1981 , apj , 243 , 926 scharf , c. a. , zurek , d. & bureau , m. 2004 , apj , submitted stetson , p. b. 1987 , , 99 , 191 theuns , t. & warren , s. j. 1997 , , 284 , l11 tody , d. 1986 , proc . spie , 627 , 733 vilchez - gomez , r. , pello , r. , & sanahuja , b. 1994 , , 283 , 37 warner , b. 1995 , cataclysmic variable stars , cambridge astrophysics series ( cambridge , new york : cambridge university press ) willman , b. , governato , f. , wadsley , j. & quinn , t. 2004 , , 355 , 159 zhang , b. & meszaros , p. 2004 , astro - ph/0311321 clllllc a & nov-1993 & 4m - ctio & tek2048 & b & 900 & 1.5 - 2.2 + & `` & '' & `` & t1 & 900 & 1.4 - 1.7 + & '' & `` & '' & h@xmath9 & 4500 & 1.4 - 1.9 + b & dec-1993 & 1.5m - ctio & tek2048 & t1 & 5400 & 1.1 - 1.3 + c & oct-1995 & 1.5m - ctio & tek2048 & t1 & 9000 & 1.1 - 1.5 + d & 05-dec-1999 & 4m - ctio & mosaic 8k & b & 4200 & 1.0 + e & 07-dec-1999 & `` & '' & r & 1260 & 0.9 + & `` & '' & `` & h@xmath9 & 7200 & 0.8 + f & 08-dec-1999 & '' & `` & b & 4200 & 1.1 + g & 11-dec-1999 & '' & `` & r & 1260 & 1.2 + & '' & `` & '' & h@xmath9 & 1800 & 1.2 + & `` & '' & `` & b & 1800 & 1.4 + h & 12-dec-1999 & '' & `` & r & 1260 & 1.6 + & '' & `` & '' & h@xmath9 & 3600 & 1.4 + & `` & '' & " & b & 3000 & 1.2 + i & 27-feb-2003 & 4m - ctio & mosaic 8k & h@xmath9 & 6000 & 1.2 + j & 14-jan-2004 & 4m - ctio & mosaic 8k & b & 8100 & 1.8 1 & 03:39:06.2 & -35:23:48 & b & ngc 1399 & 489 + 2 & 03:37:12.1 & -35:44:11 & a & ngc 1389 & 37 + 3 & 03:38:29.0 & -35:29:02 & d - h & ngc 1399 & 121 + 4 & 03:36:55.8 & -35:28:39 & d - h & ngc 1387 & 108 + 5 & 03:39:04.5 & -35:34:45 & i & ngc 1404 & 164 + 6 & 03:39:18.4 & -35:35:55 & d - h & ngc 1404 & 329 rrclrc 1 & 49331.53 & b & t1 & 21.9 & 0.1 + + 2 & 49299.84 & a & h@xmath9 & 21.1 & 0.4 + + 3 & 51519.07 & e & h@xmath9 & 22.1 & 0.1 + & 51523.10 & g & h@xmath9 & 22.0 & 0.1 + & 51525.05 & h & h@xmath9 & 22.1 & 0.1 + + & 51517.05 & d & b & 22.2 & 0.1 + & 51521.04 & f & b & 23.0 & 0.1 + & 51523.13 & g & b & 23.2 & 0.2 + & 51525.10 & h & b & @xmath5523.0 & + + 4 & 51519.07 & e & h@xmath9 & 22.8 & 0.1 + & 51523.10 & g & h@xmath9 & 23.1 & 0.2 + & 51525.05 & h & h@xmath9 & 22.8 & 0.1 + + & 51517.05 & d & b & @xmath5524.9 & + & 51521.04 & f & b & 24.8 & 0.3 + & 51523.13 & g & b & @xmath5523.8 & + & 51525.10 & h & b & @xmath5523.0 & + + 5 & 52697.07 & i & h@xmath9 & 22.8 & 0.1 + + 6 & 51519.07 & e & h@xmath9 & 24.2 & 0.5 + & 51523.10 & g & h@xmath9 & @xmath5523.5 & + & 51525.05 & h & h@xmath9 & @xmath5522.8 & + + & 51517.05 & d & b & 23.4 & 0.1 + & 51521.04 & f & b & 24.2 & 0.1 + & 51523.13 & g & b & @xmath55 24.3 & + & 51525.10 & h & b & @xmath55 22.7 &
the standard big bang theory predicts the existence of @xmath1 neutrinos per flavour in the visible universe ( e.g. , @xcite ) . this is an enormous abundance unrivalled by any other known form of matter , falling second only to the cosmic microwave background ( cmb ) photon . yet , unlike the cmb photon which boasts its first ( serendipitous ) detection in the 1960s and which has since been observed and its properties measured to a high degree of accuracy in a series of airborne / satellite and ground based experiments , the relic neutrino continues to be elusive in the laboratory . the chief reason for this is of course the feebleness of the weak interaction . the smallness of the neutrino mass also makes momentum - transfer - based detection methods highly impractical . at present , the only evidence for the relic neutrino comes from inferences from other cosmological measurements , such as big bang nucleosynthesis ( bbn ) and cmb together with large scale structure ( lss ) data ( e.g. , @xcite ) . nevertheless , it is difficult to accept that these neutrinos will never be detected in a more direct way . in order to design possible direct , scattering - based detection methods , a precise knowledge of the phase space distribution of relic neutrinos is indispensable . in this connection , it is important to note that an oscillation interpretation of the atmospheric and solar neutrino data ( e.g. , @xcite ) implies that at least two of the neutrino mass eigenstates are nonrelativistic today . these neutrinos are subject to gravitational clustering on existing cold dark matter ( cdm ) and baryonic structures , possibly causing the local neutrino number density to depart from the standard value of @xmath2 , and the momentum distribution to deviate from the relativistic fermi dirac function . in this paper , we develop a method that will allow us to predict the phase space distribution of relic neutrinos in our local neighbourhood at earth ( @xmath3 from the galactic centre ) , as well as in outer space . the method systematically takes into account gravitational clustering of relic neutrinos on scales below @xmath4 , and can be applied to the complete range of experimentally and observationally consistent neutrino masses . with these predictions , we determine the precise implications of relic neutrino clustering for future direct search experiments . to this end , we note that in earlier studies of relic neutrino direct detection , the neutrino number density in our local neighbourhood is either assumed to be unrealistically large , or simply left as a free parameter ( e.g. , @xcite ) . with the emergence of the concordance flat @xmath0cdm model as the cosmological model of choice , today we are in a position to compute the relic neutrino phase space distribution within a well defined cosmological framework , and to contemplate again the prospects for their direct detection in a definitive way . our studies here will also be useful for such investigations as relic neutrino absorption @xcite and emission @xcite spectroscopy . the standard procedure for any gravitational clustering investigation is to solve the @xmath5-dimensional vlasov , or collisionless boltzmann , equation using @xmath6-body techniques ( e.g. , @xcite ) . however , these techniques are computationally very expensive and necessarily come with limited resolutions . in the context of the cold+hot dark matter ( chdm ) model , earlier @xmath6-body studies involving neutrinos probe their kinematic effects on structure formation from cluster and galaxy abundances on large scales ( e.g. , ) , to halo properties on small scales @xcite . while the chdm model has fallen out of favour in recent years ( see , however , @xcite ) , it is instructive to note that the halo simulation of @xcite has a formal resolution of only @xmath7 . this is clearly inadequate for our considerations , where the scale of interest is of order @xmath8 . in the context of the flat @xmath0cdm model , singh and ma ( hereafter , sm ) presented a novel approximate method to probe the accretion of neutrinos onto cdm halos at scales below @xmath9 . the salient feature of this study is their use of parametric halo density profiles from high resolution , pure @xmath0cdm simulations as an external input , while the neutrino component is treated as a small perturbation whose clustering depends on the cdm halo profile , but is too small to affect it in return . implementation of this approximation requires the neutrino mass density @xmath10 to be much smaller than its cdm counterpart @xmath11 . on cosmological scales , we know now from lss data that the ratio @xmath12 is at most @xmath13 @xcite . on cluster / galactic scales , neutrino free - streaming ensures that @xmath14 always remains smaller than its cosmological counterpart @xcite . thus the approximation scheme , so far , is sound . furthermore , in order to track the neutrino density fluctuations in the most effortless way , sm employed the linearised vlasov equation instead of its full version . unfortunately , linear methods are known to break down when the density fluctuations reach the order of unity . indeed , in their two trial runs with chdm parameters , the linear results of sm compare favourably with @xmath6-body results of @xcite in the outer part of the halo , where the neutrino overdensity is relatively low . the denser inner parts ( @xmath15 ) , however , show marked disagreement . this discrepancy renders sm s claim that their complete prescription is able to probe neutrino clustering on sub - galactic scales doubtful . in the present investigation , we adopt one of the more attractive features of sm s study , namely , the use of parametric halo profiles as an external input . however , we improve upon their analysis by solving the vlasov equation in its ( almost ) full glory utilising a restricted , @xmath6-@xmath16-body ( pronounced : en - one - body ) method based on the following observation : in the limit @xmath17 and the cdm contribution dominates the total gravitational potential , not only will the cdm halo be gravitationally blind to the neutrinos , the neutrinos themselves will also have negligible gravitational interaction with each other . this allows us to track them one particle at a time in @xmath6 independent simulations , instead of following @xmath6 particles simultaneously in one single run , as in a conventional @xmath6-body study . an obvious advantage of our @xmath6-@xmath16-body technique is that it requires virtually no computing power when compared with a full scale @xmath6-body simulation with the same , large @xmath6 ( @xmath18 ) . it is also less time - consuming since we have done away with the need for a gravity solver ( the core of all @xmath6-body techniques ) . in addition , restricted methods such as ours do not suffer from spurious two body relaxation , and hence do not require the introduction of an artificial softening length that is mandatory in conventional @xmath6-body studies . lastly , we note that restricted methods have been used extensively in the studies of galaxy interactions ( e.g. , @xcite ) , and , when properly motivated , should not be seen as inferior to full scale @xmath6-body techniques . as a closing remark , let us stress again that our purpose here is not to investigate the effects of neutrino mass on cosmology , but rather to address some simple questions such as how many relic neutrinos can we realistically expect to find in this very space we occupy , what kind of energies do they have , where in the universe can we expect to find the highest concentration of relic neutrinos , etc . , given what we know today about cosmology . in this regard , the analysis we present here is most exhaustive . the paper is organised as follows . we begin in section [ cosm ] with an assessment of the current observational constraints on the relic neutrino background . in section [ vlasov ] , we introduce the vlasov equation which is used to track the phase space distribution of the neutrinos . section [ profiles ] contains a brief discussion of the halo density profiles to be employed in our calculations . in section [ onebody ] , we solve the vlasov equation for the halo models using our improved @xmath6-@xmath16-body method for a variety of halo and neutrino masses . in section [ linear ] , we compute for the same halo models and neutrino masses the neutrino overdensities using the linear method of sm and examine its validity . section [ milkyway ] deals exclusively with relic neutrinos in the milky way , and in particular their phase space distribution in our immediate vicinity . we discuss in section [ detection ] the implications of our findings for scattering - based detection methods , and we conclude in section [ conclusion ] . taking as our basis ( i ) the flat @xmath0cdm model with @xmath19 and hubble parameter @xmath20 , ( ii ) neutrino mass splittings inferred from the solar and atmospheric data , @xmath21 , and ( iii ) the invisible @xmath22 decay width from lep which constrains the number of @xmath23 doublet neutrinos to three @xcite , a minimal theory of neutrino clustering is fixed only by the absolute masses of the neutrinos @xmath24 . the current laboratory limit from tritium @xmath25 decay experiments is @xmath26 ( @xmath27 ) @xcite , and should improve to @xmath28 with the upcoming katrin experiment @xcite . cosmology also provides a constraint on @xmath24 . for three degenerate species , an upper bound of @xmath29 ( @xmath27 ) @xcite has been inferred from a combined analysis of the cmb anisotropy from wmap @xcite and galaxy clustering from sdss ( sdss - gal ) @xcite ( or from 2dfgrs @xcite ) , together with an hst prior on the hubble parameter @xcite . ( reference @xcite uses also snia @xcite ) .. ] adding to the fit galaxy bias @xcite and ly@xmath30 forest analyses can tighten the constraint to @xmath31 @xcite ( see also @xcite ) , although the robustness of these additional inputs is still contentious . weak lensing of galaxies @xcite or of the cmb @xcite will provide an alternative probe for the cosmological implications of massive neutrinos . while constraints from cosmology are interesting in their own right , they are also highly model dependent , and degeneracies abound . for instance , if tensors and running of the scalar spectral index are allowed , the last @xmath32 bound relaxes to @xmath33 @xcite . another possibility is an interplay between @xmath24 and @xmath34 , where @xmath34 is the _ effective _ number of thermalised fermionic degrees of freedom present in the radiation - dominated era , such that increasing @xmath34 actually weakens the bound on @xmath24 @xcite . for example , a @xmath35 model receives a cmb+lss+priors constraint of ( i ) @xmath36 , if all six particles are equally massive , ( ii ) @xmath37 , if three are massive and the others exactly massless , and ( iii ) @xmath38 , if only one is massive . currently , @xmath39 is allowed by cmb+lss+priors @xcite . future cmb experiments such as planck will be sensitive to @xmath40 @xcite lastly , we note that bbn prefers @xmath41 ( @xmath27 ) , in the absence of a @xmath42 chemical potential @xmath43 @xcite ( see also @xcite ) . allowing for a nonzero @xmath43 weakens the bounds to @xmath44 for @xmath45 @xcite . there is no lack of candidates in the literature for these extra @xmath46 degrees of freedom . we shall not list them here . what is certain , however , is that they can not take the form of very large chemical potentials in the @xmath47 sectors , since large neutrino mixing inferred from oscillation experiments ensures that @xmath48 , too small to be a significant source of @xmath34 @xcite . in the present analysis , we assume the neutrinos to constitute exactly three thermalised fermionic degrees of freedom , and adopt a conservative mass bound of @xmath49 corresponding to the @xmath50 constraint from the wmap+sdss galaxy cluster analysis @xcite . alternatively , ( [ eq : massbound ] ) may be interpreted as a restrictive bound for models with extra , non - neutrino relativistic particles ( @xmath51 ) , or with a significant running spectral index , as discussed earlier . a system consisting of several types of weakly interacting , self - gravitating particles [ e.g. , cdm plus neutrinos ] may be modelled as a multi - component collisionless gas whose phase space distributions @xmath52 obey the vlasov equation ( e.g. , @xcite ) , @xmath53 the single - particle phase density @xmath54 is defined so that @xmath55 is the number of @xmath56 particles in an infinitesimal phase space volume element . the variables @xmath57 are the comoving distance , its associated conjugate momentum , and the conformal time respectively , with @xmath58 as the scale factor and @xmath59 the mass of the @xmath56th particle species . all temporal and spatial derivatives are taken with respect to comoving coordinates , i.e. , @xmath60 , @xmath61 . in the nonrelativistic , newtonian limit , equation ( [ eq : vlasov ] ) is equivalent to @xmath62 with the poisson equation @xmath63 @xmath64 relating the peculiar gravitational potential @xmath65 to the density fluctuations @xmath66 with respect to the physical mean @xmath67 . the vlasov equation expresses conservation of phase space density @xmath68 along each characteristic @xmath69 given by @xmath70 the complete set of characteristics coming through every point in phase space is thus exactly equivalent to equation ( [ eq : vlasov ] ) . it is generally not possible to follow the whole set of characteristics , but the evolution of the system can still be traced , to some extent , if we follow a sufficiently large but still manageable sample selected from the initial phase space distribution . this forms the basis of particle - based solution methods . a `` first principles '' approach to neutrino clustering requires the simultaneous solution of the vlasov equation ( [ eq : vlasov ] ) [ or , equivalently , the equations for the characteristics ( [ eq : characteristic ] ) ] for both the cdm and the neutrino components . in our treatment , however , we assume only the cdm component @xmath11 contributes to @xmath71 in the poisson equation ( [ eq : poisson ] ) , and @xmath11 to be completely specified by halo density profiles from high resolution @xmath0cdm simulations . we provide in this section further justifications for this approach , as well as a brief discussion on the properties of the halo density profiles to be used in our analysis . it is well known that after they decouple from the cosmic plasma at @xmath72 , light neutrinos ( @xmath73 ) have too much thermal velocity to cluster on small scales via gravitational instability in the early stages of structure formation . accretion onto cdm protoclusters becomes possible only after the neutrino velocity has dropped below the velocity dispersion of the protoclusters . the mean velocity of the unperturbed neutrino distribution has a time dependence of @xmath74 where @xmath75 is the redshift . a typical galaxy cluster has a velocity dispersion of about @xmath76 today ; a typical galaxy , about @xmath77 . thus , for sub - ev neutrinos , clustering on small scales can only have been a @xmath78 event . on the other hand , based on a systematic @xmath6-body study of halo formation in a variety of hierarchical clustering cosmologies , navarro , frenk and white ( hereafter , nfw ) argued in 1996 that density profiles of cdm halos conform to a universal shape , generally independently of the halo mass , the cosmological parameters and the initial conditions @xcite . this so - called nfw profile has a two - parameter functional form @xmath79 where @xmath80 is a characteristic inner radius , and @xmath81 a corresponding inner density . these parameters @xmath80 and @xmath82 are determined by the halo s virial mass @xmath83 and a dimensionless concentration parameter @xmath84 defined as @xmath85 where @xmath86 is the virial radius , within which lies @xmath83 of matter with an average density equal to @xmath87 times the mean matter density @xmath88 at that redshift , i.e. , @xmath89,\end{aligned}\ ] ] where @xmath90 is the present day mean matter density . the factor @xmath87 is usually taken to be the overdensity predicted by the dissipationless spherical top - hat collapse model @xmath91 , which takes on a value of @xmath92 for an einstein de sitter cosmology , while the currently favoured @xmath0cdm model has @xmath93 at @xmath94 . is taken to be the radius @xmath95 , within which the average density is @xmath96 times the _ critical density _ of the universe , irrespective of the cosmological model in hand . ] within this framework , any halo density profile is completely specified by its virial mass and concentration via equations ( [ eq : nfw ] ) to ( [ eq : mvir ] ) . indeed , the nfw profile in its two - parameter form generally gives , for quiet isolated halos , a fit accurate to @xmath97 in the range of radii @xmath98 @xcite . furthermore , nfw argued for a tight correlation between @xmath83 and @xmath84 , such that the mass distribution within a halo is effectively fixed by the halo s virial mass alone . later studies support , to some extent , this conclusion ( e.g. , @xcite ) ; halo concentration correlates with its mass , albeit with a significant scatter . the analysis of @xcite of @xmath99 halos in the mass range @xmath100 reveals a trend ( at @xmath94 ) described by @xmath101 with a @xmath102 spread about the median of @xmath103 for fixed @xmath83 . in addition , for a fixed virial mass , the median concentration parameter exhibits a redshift dependence of @xmath104 between @xmath94 and @xmath105 . in their analysis , sm interpreted the set of equations ( [ eq : nfw ] ) to ( [ eq : cz ] ) as a complete description of an individual halo s evolution in time . while we do not completely disagree with this interpretation , it should be remembered that the relations ( [ eq : cmvir ] ) and ( [ eq : cz ] ) refer only to the behaviour of the statistical mean for _ fixed _ virial masses , and do not actually describe how individual halos accrete mass over time . a better motivation for the equations use comes from the observation that the physical densities of the inner regions of individual isolated halos tend to remain very stable over time ( @xmath106 ) @xcite . this behaviour , as it turns out , can be more or less reproduced by equations ( [ eq : nfw ] ) to ( [ eq : cz ] ) , if we interpret @xmath83 as the halo s virial mass today . merging subhalos tend to affect the main halo s density profile only in its outer region . therefore , provided that neutrino clustering becomes important only after @xmath107 and the halos have had no major mergers since then , we are justified to either use these equations here , or simply model the cdm halo as a static object in physical coordinates , as a first approximation . finally , we note that there exists in the literature a number of other halo profiles ( e.g. , ) , which , in some cases , provide better fits to simulations than does the nfw profile ( [ eq : nfw ] ) . however , these profiles generally differ from ( [ eq : nfw ] ) by less than @xmath108 @xcite so we do not consider them in our study . using the nfw halo density profile ( [ eq : nfw ] ) as an input , we find solutions to the vlasov equation in the limit @xmath17 by solving the equations for the characteristics ( [ eq : characteristic ] ) . we discuss below the basic set - up . technical details can be found in the appendix . we model the cdm distribution as follows . we assume that throughout space is a uniform distribution of cdm . on top of it , sits a spherical nfw halo at the origin . in order that the halo overdensity merges smoothly into the background density , we extend the nfw profile to beyond the virial radius . furthermore , for convenience , we treat the nfw profile as a perturbation @xmath109 , rather than a physical density . this simplification should make very little difference to the final results , since the halo density is always much larger than the background density . the halo s properties and its evolution in time are contained in the set of equations ( [ eq : nfw ] ) to ( [ eq : cz ] ) . for the factor @xmath87 , we take a time - independent value @xmath110 , following and sm . we choose this somewhat uncommon definition so as to facilitate direct comparisons between the results of sm and those of the present study . however , since the choice of @xmath87 affects the profile only through @xmath80 [ see equation ( [ eq : rs ] ) in the appendix ] , we can see immediately that using instead the more common @xmath111 , where @xmath112 with @xmath113 , will make little difference to the outcome . a more important issue is the role played by other cdm structures ( i.e. , other halos and voids ) that should realistically be in the surroundings . in our present scheme , these have all been lumped into one uniform background so as to preserve the spherical symmetry of the problem . in reality , these structures will induce tidal forces and distort the symmetry . however , we expect tidal forces to be important only for clustering in the outer part of the halo where the gravitational potential is low and several halos may compete for the same neutrinos . clustering in the deep potential well of the inner part , on the other hand , should not be seriously affected . for the neutrinos , we take the initial distribution to be the homogeneous and isotropic fermi dirac distribution with no chemical potential , @xmath114 where @xmath115 is the present day neutrino temperature . in principle , the chemical potential need not be exactly zero . in fact , a positive chemical potential @xmath116 should improve , to some extent , the clustering of neutrinos ( as opposed to anti - neutrinos ) by providing more low velocity specimens that cluster more efficiently than do their high velocity counterparts . however , this enhancement is necessarily accompanied by a suppression of clustering in the anti - neutrino sector , for which a negative @xmath117 tends to deplete the low velocity states . currently , the upper bound on @xmath116 is @xmath118 , too small to warrant a detailed investigation into a possible `` clustering asymmetry '' . we simulate initial momentum values in the range @xmath119 , which accounts for more than @xmath120 of the distribution ( [ eq : fermidirac ] ) . the initial spatial positions of the neutrinos range from @xmath121 , to as far as it takes for the fastest particles to land within a distance of @xmath122 from the halo centre at @xmath94 . we consider a sample of three neutrino masses , @xmath123 , consistent with the bound ( [ eq : massbound ] ) , and a range of halo viral masses , @xmath124 , corresponding to halos of the galaxy to the galaxy cluster variety . all simulations model a flat @xmath0cdm cosmology , with the parameters @xmath125 the initial spatial and momentum distribution is divided into small chunks that move under the external potential of the cdm halo , but independently of each other . a low resolution run is first carried out for each set of @xmath126 . all chunks that end up at @xmath94 inside a sphere of radius @xmath122 centred on the halo are traced back to their origin , subdivided into smaller chunks , and then re - simulated . the process is repeated until the inner @xmath127 is resolved . the initial redshift is taken to be @xmath128 . this should be sufficient , since clustering is not expected to be fully under way until @xmath107 . the final neutrino density distribution is constructed from the set of discrete particles via a kernel density estimation method @xcite outlined in the appendix , with a maximum smoothing length of @xmath129 in the inner @xmath130 of the halo . = 15.7 cm = 15.7 cm the basic results of our @xmath6-@xmath16-body simulations are presented in figure [ fig : overdensities ] , which shows the neutrino overdensities @xmath131 for various sets of @xmath132 . a companion figure , figure [ fig : nuvscdm ] , shows the same results expressed in terms of the mass density ratio @xmath133 normalised to the background mean @xmath134 . the essential features of the curves in figures [ fig : overdensities ] and [ fig : nuvscdm ] can be understood in terms of neutrino free - streaming , which causes the @xmath131 curves to flatten out at small radii , and the mass density ratio @xmath133 to drop substantially below the background mean . ( the latter feature also provides a justification for our @xmath6-@xmath16-body method . ) both @xmath131 and @xmath14 approach their respective cosmic mean of @xmath16 and @xmath134 at large radii . similar behaviours have also been observed in the chdm simulations of reference @xcite . = 15.7 cm = 15.0 cm naturally , clustering improves with increasing neutrino and/or halo masses . in figure [ fig : halomass ] , we plot the neutrino overdensities at @xmath135 for various neutrino masses as a function of the halo virial mass @xmath83 . a similar plot is constructed in figure [ fig : msquare ] , with the neutrino mass @xmath24 as the independent variable . in the former , the quantity @xmath136 is seen to be roughly proportional to @xmath83 for a fixed radius and a fixed neutrino mass . the @xmath136 versus @xmath24 trend in figure [ fig : msquare ] is more difficult to quantify . a roughly @xmath137 dependence can be discerned for some fixed halo masses at some fixed radii . other combinations , however , display noticeably different behaviours . whatever these dependences are , it is interesting to note that they are never shallower than @xmath138 , or steeper than @xmath139 . as we shall see later , an @xmath137 and an @xmath140 dependence for the overdensities can both be motivated from theory . the former can be derived from the linearised vlasov equation ( cf . section [ linear ] ) , while the latter follows naturally from phase space considerations ( or the so - called tremaine gunn bound , cf . section [ milkyway ] ) . lastly , in order to test the robustness of our results , we ( i ) push the initial redshift of the simulations back to @xmath141 , ( ii ) vary the cosmological parameters within their allowed ranges , and ( iii ) alter the time dependences of some of the halo parameters . in all cases , we find , as expected , the heavier masses @xmath126 to suffer more from these variations . for our heaviest set , @xmath142 , the neutrino overdensity changes by about @xmath143 to @xmath144 at small radii , and some @xmath145 to @xmath146 at @xmath147 . the gap narrows with smaller neutrino and halo masses . for galaxy size halos ( @xmath148 ) , the density variations with respect to ( i ) , ( ii ) and ( iii ) are no more than @xmath97 everywhere . thus our simulation results are generally quite robust . the linear approximation is often used in the literature to find approximate solutions to the vlasov equation . in the nonrelativistic limit , the pioneering work of gilbert @xcite has , over the years , been applied to the study of the pure hot dark matter ( hdm ) model @xcite , as well as in the analysis of hdm accretion onto nonadiabatic seeds such as cosmic strings @xcite and onto cdm halos . the procedure consists of first switching to a new time variable @xmath149 , and then fourier transforming the @xmath150-dependent functions , @xmath151 , \qquad \hat{\phi } ( \bm{k},s ) \equiv \mathcal{f}[\phi ( \bm{x},s)],\ ] ] to obtain a new differential equation in fourier space , @xmath152 where @xmath153 is the convolution product . the equation is said to be linearised when one makes the replacement @xmath154 where @xmath155 is the dirac delta function , so that the convolution product becomes a simple scalar product @xmath156 . replacement ( [ eq : replacement ] ) is valid as long as the condition @xmath157 holds . in practice , however , the quantity @xmath158 is somewhat cumbersome to compute , so the `` rule of thumb '' regarding the linear approximation is to abandon it as soon as the spatial density fluctuation @xmath159 , defined in ( [ eq : fluctuations ] ) , exceeds order unity , as emphasised in @xcite . nonetheless , the linear theory has been time and again used beyond this putative limit . we shall also apply it to our case , to gain physical insight as well as to see how it compares with @xmath6-@xmath16-body simulations . equation ( [ eq : vlasovfourier ] ) together with the replacement ( [ eq : replacement ] ) has a very simple solution , @xmath160 where @xmath161 , @xmath162 is some initial time , and we take the initial phase space distribution to be isotropic and homogeneous in space , i.e. , @xmath163 . the neutrino number density per fourier mode relative to the mean density is obtained by integrating ( [ eq : fsolution ] ) over momenta @xmath164 , @xmath165,\end{aligned}\ ] ] with @xmath166 the correct form for @xmath167 should be the relativistic fermi dirac function ( [ eq : fermidirac ] ) , which gives for @xmath168 a series representation @xcite @xmath169 where @xmath170 is the riemann zeta function . however , in order to simplify calculations and/or to gain physical insight , other forms of @xmath167 have also appeared in the literature . some are listed in table [ table : f(q ) ] , along with their corresponding @xmath168 . the solution to the poisson equation ( [ eq : poisson ] ) in fourier space is @xmath171 substituting this into ( [ eq : nudensity ] ) and using the definition @xmath172 , we obtain for the neutrino density fluctuations @xmath173.\ ] ] this is the `` master equation '' for the linear approach . we solve equation ( [ eq : perturbation ] ) numerically for a variety of neutrino and halo masses . the results are presented in figure [ fig : overdensities ] , alongside their @xmath6-@xmath16-body counterparts . @lll distribution & @xmath167 & @xmath168 + relativistic fd & @xmath174^{-1}$ ] & @xmath175 + mb & @xmath176 & @xmath177 + @xmath178 distribution & @xmath179 & @xmath180 + before comparing the two approaches , let us first study the linear approximation for its own sake . consider the master equation ( [ eq : perturbation ] ) . in the limit @xmath168 grows much faster than @xmath181 and @xmath182 , i.e. , @xmath183 equation ( [ eq : perturbation ] ) may be solved by asymptotic expansion . the resulting approximate solution looks somewhat messy at first sight , @xmath184 , \end{aligned}\ ] ] but may be rendered into a physically transparent form if we first cross out the second term , which is well justified since the initial @xmath58 and @xmath185 should always be much smaller than the final ones , and then rewrite the expression as @xmath186 here , @xmath187 is but the free - streaming wave vector , defined as @xmath188 and we identify @xmath189 as the neutrino s characteristic thermal speed . the functional form of equation ( [ eq : asymptoticsolution ] ) already tells us something very interesting ; large fourier modes in the neutrino density fluctuations are suppressed by a factor proportional to @xmath190 relative to their cdm counterparts . this is clearly a manifestation of free - streaming , which is responsible for inhibiting the growth of structures on scales below @xmath191 . furthermore , @xmath192 has an @xmath137 dependence through @xmath187 , meaning that , at small scales , a neutrino twice as heavy as another is able to cluster four times more efficiently . this @xmath137 dependence is reflected , approximately , by both our linear and @xmath6-@xmath16-body results in figure [ fig : overdensities ] , and is particularly pronounced at small radii . that equation ( [ eq : asymptoticsolution ] ) is a solution of ( [ eq : perturbation ] ) is contingent upon the satisfaction of the condition ( [ eq : asymptoticcondition ] ) , which requires , for a fixed neutrino mass , @xmath193 to be larger than some nominal @xmath194 determined by the rates of change of the scale factor @xmath58 and of the cdm perturbations @xmath195 . the rate of change of @xmath58 is a simple and well defined function of the cosmological model . the growth rate of @xmath182 , on the other hand , is usually more complicated . however , because our halos are practically static in physical coordinates , this rate ( in comoving fourier space ) can only be at most of the order of the universal expansion rate ( which comes in through the conversion factor @xmath58 when we switch the halo profile from physical to comoving coordinates ) . thus the condition ( [ eq : asymptoticcondition ] ) is roughly equivalent to @xmath196 where @xmath197 is the hubble expansion parameter at time @xmath198 . since @xmath199 at most times , we see that equation ( [ eq : asymptoticsolution ] ) is indeed applicable to all @xmath193 modes larger than the free - streaming wave vector @xmath187 . unfortunately , the opposite @xmath200 limit has no simple approximate solution because of the complicated dependence of the scale factor @xmath58 on the new time variable @xmath198 . however , we find the following formula to give a decent fit to the solution of ( [ eq : perturbation ] ) for a wide range of @xmath193 , @xmath201 with @xmath202 typically , @xmath203 , such that for @xmath200 , the growth of @xmath192 approximately matches that of @xmath185 . therefore , equation ( [ eq : fitting ] ) is roughly equivalent to @xmath204 in real space , with @xmath205 $ ] acting like a normalised filter function with window width @xmath206 , that gently smears out the neutrino density contrasts on scales below @xmath207 relative to their cdm counterparts . we shall be using equation ( [ eq : nufilter ] ) again in section [ target ] . comparing the linear results from this section with @xmath6-@xmath16-body simulations from section [ onebody ] , it is immediately clear in figure [ fig : overdensities ] that the former systematically underestimates the neutrino overdensities over the whole range of neutrino and halo masses considered in this study . the discrepancy is most prominent in the dense , inner regions ( @xmath208 ) , and worsens as we increase ( i ) the neutrino mass @xmath24 , and ( ii ) the halo mass @xmath83 . the worst case corresponds to when both @xmath24 and @xmath83 are large ; the case of @xmath209 , for example , sees the @xmath6-@xmath16-body and the linear overdensities differ by a factor of about six . for smaller neutrino and halo masses , concordance between the two approaches improves as we move to larger radii . indeed , for @xmath210 , complete agreement is seen throughout the region of interest . upon closer inspection , one finds that the linear theory ceases to be a faithful approximation once the neutrino overdensity reaches a value of about three or four . this is of course fully consistent with the standard lore that perturbative methods fail once the perturbations exceed unity and nonlinear effects set in . can the linear approximation be salvaged ? yes , provided we impose a great deal of smoothing . in the case of @xmath209 , for example , the overdensities computed from the two different approaches can be reconciled with each other if we smooth them both over a scale of roughly @xmath211 . such a large smoothing length will render the linear method completely useless for the study of neutrino clustering on sub - galactic scales ( unless of course the neutrino mass is so small that the overdensity does not exceed unity by much anyway ) . but the method can still be useful for obtaining quick estimates of @xmath131 on larger scales for absorption and emission spectroscopy calculations ( cf . section [ target ] ) . finally , we note that the neutrino overdensities in figure 2 of sm are at odds with our linear results in figure [ fig : overdensities ] . this discrepancy can not be ameliorated by simply supposing that sm have normalised their neutrino densities for _ three _ flavours to the _ one _ flavour average @xmath2 , since this normalisation will render some of their results specifically , where @xmath212unphysical . only a normalisation to _ three _ flavours can give these results a physical meaning , but at the expense of incompatibility with our figure [ fig : overdensities ] , as well as with sm s own figure 3 . we therefore conclude that sm s results as presented in their figure 2 are erroneous . in this section , we consider relic neutrino clustering in the milky way . we compute explicitly the number of neutrinos and their distribution in momentum space in our local neighbourhood at earth ( @xmath213 from the galactic centre ) . this information is essential for any direct search experiment . we perform this calculation using the @xmath6-@xmath16-body method of section [ onebody ] , but with a few modifications to the external potential . firstly , we note that the central region of the milky way ( @xmath214 ) is dominated by baryonic matter in the form of a disk , a bulge , and possibly a rapidly rotating bar . each of these components has its own distinct density profile . furthermore , in the standard theory of hierarchical galaxy formation , baryons and dark matter are initially well mixed , and collapse together to form halos via gravitational instability . galactic structures arise when the baryons cool and fall out of the original halo towards the centre . as the baryons condense , their gravitational forces tend to pull the dark matter inward , thereby distorting the inner @xmath215 of the original halo profile ( e.g. , @xcite ) . this kind of modification to the mass distribution is important for us at @xmath216 . fortunately for our calculation , gravitating neutrinos do not distinguish between halo and baryons . therefore , it suffices to use simply the total mass distribution inferred from observational data ( e.g. , rotation curves , satellite kinematics , etc . ) , without any detailed modelling of the individual components . what is still missing , however , is the redshift dependence of the mass distribution . unfortunately , we have not been able to find in the literature any simple parametric form for this dependence . however , mass modelling of the milky way @xcite suggests that certain observationally acceptable bulge+disk+halo models are indeed consistent with the aforementioned theory of baryonic compression , and can be traced back to halos _ originally _ of the nfw form by supposing that the system has undergone a phase of adiabatic contraction @xcite . thus , for our investigation , it is probably fair to think of the nfw profile as the initial mass distribution , and the evolution as a smooth transition from this initial distribution to the present day one . instead of modelling this transition , however , our strategy here is to conduct two series of simulations , one for the present day mass distribution of the milky way ( mwnow ) which we assume to be static , and one for the nfw halo ( nfwhalo ) that would have been there , had baryon compression not taken place . the real neutrino overdensity should then lie somewhere between these two extremes . for the nfwhalo run , we use the parameters @xmath217 and @xmath218 . these numbers are taken from the paper of klypin , zhao and somerville ( hereafter , kzs ) @xcite , from their `` favoured '' mass model of the milky way . note that we are not using the @xmath84-@xmath83 relation ( [ eq : cmvir ] ) , which , as we discussed before , is only a statistical trend . however , the concentration parameter @xmath84 should still carry a redshift dependence _ la _ equation ( [ eq : cz ] ) , in order to reproduce the correct time dependence of the density profile . for mwnow , we adopt the total mass distribution ( halo+disk+bulge ) presented in figure 3 of kzs , and fit it approximately to a power law from @xmath121 to @xmath219 , @xmath220 where @xmath221 means the total mass contained within a radius @xmath222 . we assume this fit to hold for the region inside a _ physical _ radius of @xmath219 at all times , i.e. , @xmath223 . the region outside of this @xmath219 sphere is not affected by baryonic compression according to the kzs mass model ( cf . their figure 7 ) so we adopt the original nfw density profile outwardly from @xmath219 . thus , schematically , we have @xmath224 where @xmath225 , @xmath226 is the mass contained in an nfw halo at radius @xmath222 [ or @xmath227 in equation ( [ eq : nfwmass ] ) ] , and @xmath228 for the parameters used in this analysis . = 15.7 cm our milky way simulation results for four neutrino mass @xmath229 are displayed in figure [ fig : milkyway ] . the shaded region in each plot corresponds to a possible range of overdensities at @xmath94 . at first glance , it may seem unphysical that the apparently static mwnow potential ( in _ physical coordinates _ ) should capture so many neutrinos . to resolve this `` paradox '' , one must remember that neutrino clustering is studied in the context of an expanding universe ; the ( unbound ) neutrino thermal velocity decreases with time [ equation ( [ eq : fdspeed ] ) ] , thus causing them to be more readily captured . equivalently , in _ comoving coordinates _ , it is easy to see that while the neutrino conjugate momentum ( [ eq : comoving ] ) does not redshift , the mwnow potential well shrinks in size and deepens with time . in each scenario we studied , the final momentum distribution at @xmath216 is almost isotropic , with a zero mean radial velocity @xmath230 , and second velocity moments that satisfy approximately the relation @xmath231 ( cf . table [ table : moments ] ) . for this reason , we plot the smoothed , or coarse - grained , phase space densities @xmath232 only as functions of the absolute velocity ( cf . figure [ fig : momentum ] ) . @l|l|lll|lll & @xmath131 & @xmath233 & @xmath234 & @xmath235 & @xmath236 & @xmath237 & @xmath238 + relativistic fermi dirac & 1 & @xmath239 & @xmath240 & @xmath241 & @xmath242 & @xmath243 & @xmath244 + nfwhalo , @xmath245 & 12 & @xmath246 & @xmath247 & @xmath248 & @xmath249 & @xmath250 & @xmath251 + nfwhalo , @xmath252 & 6.4 & @xmath246 & @xmath253 & @xmath254 & @xmath255 & @xmath256 & @xmath257 + nfwhalo , @xmath258 & 3.1 & @xmath246 & @xmath259 & @xmath260 & @xmath261 & @xmath262 & @xmath263 + nfwhalo , @xmath264 & 1.4 & @xmath246 & @xmath259 & @xmath265 & @xmath266 & @xmath267 & @xmath263 + mwnow , @xmath245 & 20 & @xmath246 & @xmath268 & @xmath269 & @xmath270 & @xmath271 & @xmath272 + mwnow , @xmath252 & 10 & @xmath246 & @xmath273 & @xmath268 & @xmath274 & @xmath275 & @xmath271 + mwnow , @xmath258 & 4.4 & @xmath246 & @xmath276 & @xmath277 & @xmath278 & @xmath279 & @xmath275 + mwnow , @xmath264 & 1.6 & @xmath246 & @xmath259 & @xmath280 & @xmath281 & @xmath262 & @xmath263 + = 15.7 cm as expected , the coarse - grained distribution @xmath232 for the case with the highest overdensity ( mwnow , @xmath245 ) resembles the original fermi dirac spectrum the least , while @xmath282 for the case with the lowest overdensity ( nfwhalo , @xmath264 ) is almost fermi dirac - like . all spectra share the feature that they are flat at low momenta , with a common value of @xmath283 . the turning point for each distribution coincides approximately with the `` escape momentum '' @xmath284 ( i.e. , @xmath24 times the escape velocity @xmath285 ) for the system concerned , beyond which the phase space density falls off rapidly , until it matches again the fermi dirac function at the very high momentum end of the spectrum . deviation from the original fermi dirac spectrum is therefore most severe around @xmath284 . the maximum value of @xmath282 is a little less than @xmath286 . this is consistent with the requirement that the final coarse - grained density must not exceed the maximal value of the initial fine - grained distribution , @xmath287 @xcite . for neutrinos , @xmath288 has a value of @xmath289 at @xmath290 . thus , our @xmath282 not only satisfies but completely saturates the bound at low momenta up to @xmath284 , forming a kind of semi - degenerate state that can only be made denser by filling in states above @xmath284 . however , since neutrinos with momenta above @xmath284 do not become gravitationally bound to the galaxy / halo , these high momentum states are much less likely to be fully occupied . this explains @xmath282 s rapid drop beyond @xmath284 . also , the hottest neutrinos are not significantly affected by the galaxy / halo s gravitational forces . therefore the very high end of the momentum distribution remains more or less fermi dirac - like . finally , we note that because the filling of phase space happens from bottom up , the mean momenta for the least clustered cases tend to be lower than the fermi dirac value @xmath291 , in contrast with the nave expectation that clustering is necessarily accompanied by an increase in the neutrinos average kinetic energy . it is interesting to compare our results with nominal bounds from phase space arguments . by demanding the final coarse - grained distribution to be always less than the maximum of the original fine - grained distribution , tremaine and gunn argued in 1979 that if neutrinos alone are to constitute the dark matter of a galactic halo , their mass must be larger than @xmath292 , assuming that the halo has a maxwellian phase space distribution as motivated by the theory of violent relaxation @xcite . a modern version of this bound , in which the assumption about the phase space distribution is relaxed and which allows for contribution to the total gravitational potential from more than one form of matter , can be found in reference @xcite . the revised mass bound may be written , alternatively , in the form of a constraint on the overdensity , which reads @xmath293 for the neutrino masses @xmath294 , this expression evaluates to @xmath295 and @xmath296 for mwnow and nfwhalo , respectively , at @xmath216 . at first sight , some of our numerical results seem to have completely violated the bound ( [ eq : kullbound ] ) . but this can not be , since we have seen explicitly that all of our final coarse - grained distributions satisfy perfectly the constraint @xmath287 . furthermore , an upper bound of 0.25 on the overdensity for a @xmath297 neutrino is obviously nonsensical . the answer , as the astute reader would have figured , lies in accounting . in the derivation of ( [ eq : kullbound ] ) , a semi - degenerate distribution has been summed only up to the momentum state corresponding to the escape velocity of the system . neutrinos with higher momenta that could be hovering around in the vicinity have been completely ignored . in contrast , in our calculations , it is of no concern to us whether or not the relic neutrinos actually form bound states with the galaxy / halo . therefore it is more appropriate for us to sum every neutrino in sight , rather than imposing a cut - off at @xmath298 . however , if we had imposed such a cut - off , one can easily see from figure [ fig : momentum ] that our overdensities would have just saturated the bound ( [ eq : kullbound ] ) , so there is no conflict . nonetheless , this illustrates how nominal bounds such as ( [ eq : kullbound ] ) must be used with care . = 15.7 cm before we conclude this section , let us note that for an nfw halo , the bound ( [ eq : kullbound ] ) can be written as @xmath299^{3/2 } \nonumber \\ & \simeq & 0.58 \times \left(\frac{m_{\nu}}{\rm ev } \right)^3 \left [ \left(\frac{m_{\rm vir}}{10^{12 } h^{-1 } m_{\odot } } \right ) \left(\frac{h^{-1 } { \rm mpc}}{r } \right ) \frac{\ln(1+r / r_s)}{g(c ) } \right]^{3/2},\end{aligned}\ ] ] where the function @xmath300 is defined in equation ( [ eq : gc ] ) in the appendix . figure [ fig : tremainegunn ] shows the bound as a function of radius for the various halo and neutrino masses considered in section [ onebody ] , along with the overdensities obtained from @xmath6-@xmath16-body simulations . we find the limit ( [ eq : kullbound ] ) to be saturated for the lightest halo and neutrino masses ( and also some apparent violation of the bound due to different accounting ) . this explains why some of the overdensities exhibit an almost @xmath140 dependence ( cf . figure [ fig : msquare ] ) . on the other hand , the heaviest @xmath126 set is short of the bound by at least one order of magnitude . this is also consistent with expectations : heavier halo and neutrino masses give a higher escape momentum @xmath284 , and higher momentum states are more difficult to fill up to a semi - degenerate level , since there are less particles in these states in the original fermi dirac distribution to begin with . in this section , we determine the implications of our clustering results for the direct detection of relic neutrinos , in contrast with the purely cosmological inferences discussed in section [ cosm ] . we consider various proposed detection methods based on scattering processes , involving the relic neutrinos both as a beam and as a target . in particular , we shall discuss ( i ) coherent elastic scattering of the relic neutrino flux off target matter in a terrestrial detector ( section [ flux ] ) , as well as ( ii ) the scattering of extremely energetic particles ( accelerator beams or cosmic rays ) off the relic neutrinos as a target ( section [ target ] ) . the low average momentum @xmath301 of the relic neutrinos ( cf . table [ table : moments ] ) corresponds to a ( reduced ) de broglie wavelength of macroscopic dimension , @xmath302 @xmath303 ( cf . table [ table : debroglie ] ) . therefore , one may envisage scattering processes in which many target atoms act coherently @xcite over a macroscopic volume @xmath304 , so that the reaction rate for elastic scattering becomes proportional to the square of the number of target atoms in that volume . compared to the case where the neutrinos are elastically scattered coherently only on the individual nuclei of the target , the rate in this case is enhanced by a huge factor of @xmath305 where @xmath306 is the avogadro constant , @xmath307 is the atomic mass , and @xmath308 is the mass density of the target material .. however , one needs a large lepton asymmetry for a non - negligible effect . ] @l|l|l|l|l & @xmath309 & @xmath310 & @xmath311 & @xmath312 + relativistic fermi dirac & 1 & @xmath313 ev & @xmath314 cm & see ( [ eq : fdspeed ] ) + nfwhalo , @xmath315 & 12 & @xmath316 ev & @xmath317 cm & @xmath318 + nfwhalo , @xmath319 & 6.4 & @xmath320 ev & @xmath321 cm & @xmath322 + nfwhalo , @xmath323 & 3.1 & @xmath324 ev & @xmath325 cm & @xmath326 + nfwhalo , @xmath327 & 1.4 & @xmath328 ev & @xmath329 cm & @xmath330 + mwnow , @xmath315 & 20 & @xmath331 ev & @xmath332 cm & @xmath333 + mwnow , @xmath319 & 10 & @xmath334 ev & @xmath335 cm & @xmath336 + mwnow , @xmath323 & 4.4 & @xmath337 ev & @xmath314 cm & @xmath338 + mwnow , @xmath327 & 1.6 & @xmath339 ev & @xmath340 cm & @xmath341 + by exploiting the above coherence effect , a practical detection scheme for the local relic neutrino flux is based on the fact that a test body of density @xmath308 at earth will experience a neutrino wind force through random neutrino scattering events , leading to an acceleration @xcite @xmath342 where @xmath343 is the elastic neutrino nucleon cross section , and @xmath344 is the mean velocity of the relic neutrinos in the rest system of the detector . here , @xmath345 denotes the velocity of the earth through the milky way . expression ( [ eq : accel ] ) is valid as long as the radius @xmath346 of the target is smaller than the reduced de broglie wavelength @xmath347 of the relic neutrinos . furthermore , it applies only to dirac neutrinos . for majorana neutrinos , the acceleration is suppressed , in comparison with ( [ eq : accel ] ) , by a factor of @xmath348 for an unpolarised target , or @xmath349 for a polarised one . a target size much larger than @xmath350 can be exploited , while avoiding destructive interference , by using foam - like @xcite or laminated @xcite materials . alternatively , grains of size @xmath351 could be randomly embedded ( with spacing @xmath352 ) in a low density host material @xcite . to digest these estimates , we note that the smallest measurable acceleration at present is @xmath353 , using conventional cavendish - type torsion balances . possible improvements with currently available technology to a sensitivity of @xmath354 have been proposed @xcite . however , such a sensitivity is still off the prediction ( [ eq : accel ] ) by at least three orders of magnitude , as an inspection of the currently allowed range of local relic neutrino overdensities in table [ table : debroglie ] reveals . therefore , we conclude that an observation of this effect will not be possible within the upcoming decade , but can still be envisaged in the foreseeable future ( thirty to forty years according to reference @xcite , exploiting advances in nanotechnology ) , as long as our known light neutrinos are dirac particles . should they turn out , in the meantime , to be majorana particles , flux detection via mechanical forces will be a real challenge . let us note finally that the background contribution to the acceleration ( [ eq : accel ] ) from the solar @xmath355 neutrinos [ @xmath356 , @xmath357 ( e.g. , @xcite ) ] , @xmath358 @xcite , may be rejected by directionality . the background from weakly interacting massive particles ( wimps @xmath359 , with mass @xmath360 ) @xcite , @xmath361 should they be the main constituent of the galactic dark matter with mass density @xmath362 at @xmath363 , can be neglected as soon as the wimp nucleon cross section @xmath364 is smaller than @xmath365 @xmath366 . this should be well established by the time relic neutrino direct detection becomes a reality . note also that neutrinos produced in the earth s atmosphere do not contribute appreciably to @xmath367 because of their small flux ; for @xmath368 , the integrated flux is @xmath369 ( e.g. , @xcite ) . let us consider next the idea to take advantage of the fact that , for center - of - mass ( c.m . ) energies below the @xmath370- and @xmath22-resonances , the weak interaction cross sections grow rapidly with energy . one may then contemplate the possibility to exploit a flux of extremely energetic particles either from accelerator beams or from cosmic rays for scattering on the relic neutrinos . @l|c|r|r|r|r|r||r accelerator & @xmath6 & @xmath307 & @xmath22 & @xmath371 [ tev ] & @xmath372 [ km ] & @xmath373 [ a ] & @xmath374}$ ] + & @xmath375 & 1 & 1 & @xmath376 & @xmath377 & @xmath378 & @xmath379 yr@xmath380 + lhc & & & & & & & + & pb & 208 & 82 & @xmath381 & @xmath377 & @xmath382 & @xmath383 yr@xmath380 + & @xmath375 & 1 & 1 & @xmath384 & @xmath385 & @xmath386 & @xmath387 yr@xmath380 + vlhc & & & & & & & + & pb & 208 & 82 & @xmath388 & @xmath385 & @xmath389 & @xmath390 yr@xmath380 + ulhc & @xmath375 & 1 & 1 & @xmath391 & @xmath392 & @xmath393 & @xmath143 yr@xmath380 + we start with a discussion of the prospects to detect interactions with the relic neutrinos at forthcoming accelerator beams such as the lhc @xcite and the vlhc @xcite , with beam energies @xmath371 ranging from @xmath394 for the lhc running with protons , up to several thousands of tev for the heavy ion option at the vlhc ( cf . table [ table : accelerators ] ) . at these beams , the attainable momentum transfers and c.m . energies , @xmath395 are so small that the cross sections for their interactions with the relic neutrinos are enhanced by a factor @xmath396 due to coherent elastic scattering over the size of the nucleus @xcite , and grow linearly with the beam energy , @xmath397 this leads to a scattering rate @xcite @xmath398 & \simeq & { 2\times 10^{-8 } \left ( \frac{n_\nu}{\bar n_\nu}\right ) \left ( \frac{m_{\nu}}{{\rm ev}}\right ) \frac{a^2}{z}\ , \left ( \frac{e_n}{{\rm 10\ tev}}\right)\ , \left ( \frac{l}{100\ { \rm km}}\right)\ , \left ( \frac{i}{0.1\ { \rm a}}\right)\ , \ { \rm yr}^{-1 } } , \nonumber\end{aligned}\ ] ] for a beam of particles @xmath399 , with charge @xmath400 , length @xmath372 and current @xmath373 . in view of the currently allowed range of local relic neutrino overdensities displayed in table [ table : debroglie ] , and the beam parameters of the next generation of accelerators summarised in table [ table : accelerators ] , the expected rate ( [ eq : acc_rate ] ) is clearly too small to give rise to an observable effect in the foreseeable future . the extremely energetic @xmath401 lead beam at the lhc will have less than @xmath402 interactions per year with relic neutrinos , for @xmath403 and , correspondingly , @xmath404 . even a lead acceleration option for the vlhc , with @xmath405 ( or , equivalently , @xmath406 per nucleon ) and a current @xmath407 ( a hundredth of the nominal current of the @xmath375 running mode ) will give less than @xmath408 events per year for the most optimistic neutrino mass scenario . thus , there is little hope , in the foreseeable future , to detect relic neutrino using terrestrial accelerator beams . let us nevertheless dream about the far future , in which an ultimate large hadron collider ( ulhc ) exists and is able to accelerate protons to energies above @xmath409 between the electroweak scale @xmath410 and the scale of grand unification @xmath411 . the intermediate scale is exploited in many schemes of supersymmetry breaking ( e.g. , @xcite ) and in seesaw mechanisms for neutrino masses @xcite . ] in a ring of ultimate circumference @xmath412 around the earth , thus leading to an interaction rate of more than one event per year ( cf . table [ table : accelerators ] ) . even under these most optimistic circumstances , is it possible to reliably detect these interactions ? clearly , elastic scattering of the beam particles with the relic neutrinos one of the contributions to the rate ( [ eq : acc_rate])will be next to impossible to detect because of the small momentum transfers involved ( @xmath413 at @xmath414 ) . a very promising alternative is to consider again a heavy ion beam , and to exploit the contribution of the inverse beta decay reaction , @xmath415 to the rate ( [ eq : acc_rate ] ) . this reaction changes the charge of the nucleus , causing it to follow an extraordinary trajectory and finally to exit the machine such that it becomes susceptible to detection @xcite . a detection of this reaction would also clearly demonstrate that a neutrino was involved in the scattering . in the meantime , until the ulhc has been constructed , target detection of the relic neutrinos has to rely on extremely energetic cosmic rays . in fact , cosmic rays with up energies up to @xmath416 have been seen by air shower observatories . the corresponding c.m . energies are @xmath417 when scattering off the relic neutrinos . these energies are not too far from the @xmath370- and @xmath22-resonances , at which the electroweak cross sections become sizeable . indeed , it was pointed out long ago by weiler @xcite ( for earlier suggestions , see @xcite ) that the resonant annihilation of extremely energetic cosmic neutrinos ( eec@xmath418 ) with relic anti - neutrinos ( and vice versa ) into @xmath22-bosons appears to be a unique process having sensitivity to the relic neutrinos . on resonance , @xmath419 the associated cross section is enhanced by several orders of magnitude , @xmath420 leading to a `` short '' mean free path @xmath421 which is _ only _ about @xmath422 times the hubble distance . neglecting cosmic evolution effects , this corresponds to an annihilation probability for eec@xmath418 from cosmological distances on the relic neutrinos of @xmath423 . the signatures of annihilation are ( i ) absorption dips @xcite ( see also @xcite ) in the eec@xmath418 spectrum at the resonant energies , and ( ii ) emission features @xcite ( @xmath22-bursts ) as protons ( or photons ) with energies spanning a decade or more above the predicted greisen zatsepin kuzmin ( gzk ) cutoff at @xmath424 ev @xcite . this is the energy beyond which the cmb is absorbing to nucleons due to resonant photopion production .- bursts with the mysterious cosmic rays observed above @xmath425 is a controversial possibility @xcite . ] the possibility to confirm the existence of relic neutrinos within the next decade from a measurement of the aforementioned absorption dips in the eec@xmath418 flux was recently investigated in @xcite . presently planned neutrino detectors ( pierre auger observatory @xcite , icecube @xcite , anita @xcite , euso @xcite , owl @xcite , and salsa @xcite ) operating in the energy regime above @xmath426 appear to be sensitive enough to lead us , within the next decade , into an era of relic neutrino absorption spectroscopy , provided that the flux of the eec@xmath418 at the resonant energies is close to current observational bounds and the neutrino mass is sufficiently large , @xmath427 . in this case , the associated @xmath22-bursts must also be seen as post - gzk events at the planned cosmic ray detectors ( pierre auger observatory , euso , and owl ) . = 15.7 cm what are the implications of relic neutrino clustering for absorption and emission spectroscopy ? firstly , absorption spectroscopy is predominantly sensitive to the relic neutrino background at early times , with the depths of the absorption dips determined largely by the higher number densities at large redshifts ( @xmath428 ) . since neutrinos do not cluster significantly until after @xmath429 , clustering at recent times can only show up as secondary dips with such minimal widths in energy @xcite that they do not seem likely to be resolved by planned observatories . on the other hand , emission spectroscopy is directly sensitive to the relic neutrino content of the local universe ( @xmath430 ) . however , since the neutrino density contrasts approximately track those of the underlying cdm above the neutrino free - streaming scale @xmath206 ( cf . section [ linear ] ) , it is clear that there can not be a substantial neutrino overdensity over the whole gzk volume ( @xmath431 ) . indeed , if we take the linear fitting formula ( [ eq : nufilter ] ) , and apply it to the local cdm distribution inferred from peculiar velocity measurements ( with a @xmath4 smoothing ) , one can see in figure [ fig : local ] that the estimated neutrino overdensity is always @xmath432 . hence the overall emission rate can not be significantly enhanced by gravitational clustering . or we could imagine doing `` relic neutrino tomography '' of the local universe , by exploiting the fact that there are several galaxy clusters ( @xmath433 ) , such as virgo ( distance @xmath434 ) and centaurus ( @xmath435 ) , within the gzk zone with significant neutrino clustering ( cf . figure [ fig : overdensities ] ) . one could conceivably search for directional dependences in the emission events as a signature of eec@xmath418 annihilating on relic anti - neutrinos ( and vice versa ) . for example , agasa has an angular resolution of @xmath436 @xcite . this is already sufficient to resolve the internal structures of , say , the virgo cluster ( distance @xmath434 , @xmath437 ) which spans some @xmath438 across the sky . using our @xmath6-@xmath16-body clustering results in figure [ fig : overdensities ] , the average neutrino overdensity along the line of sight towards and up to virgo is estimated to be @xmath439 and @xmath440 for @xmath441 and @xmath297 respectively , given an angular resolution of @xmath436 . the corresponding increases in the number of events coming from the direction of the virgo cluster relative to the unclustered case , assuming an isotropic distribution of eec@xmath418 sources , are given roughly by the same numbers , since protons originating from @xmath434 away arrive at earth approximately unattenuated . the numbers improve to @xmath442 and @xmath443 respectively with a finer @xmath444 angular resolution . note that our estimates here are generally a factor of few higher than the predictions of sm . this is expected , because the linear method adopted in their analysis can not account for additional clustering from nonlinear effects , as we demonstrated in section [ linear ] . we have conducted a systematic and exhaustive study of the gravitational clustering of big bang relic neutrinos onto existing cdm and baryonic structures within the flat @xmath0cdm model , with the aim of understanding their clustering properties on galactic and sub - galactic scales for the purpose of designing possible scattering - based detection methods . our main computational tools are ( i ) a restricted , @xmath6-@xmath16-body method ( section [ onebody ] ) , in which we neglect the gravitational interaction between the neutrinos and treat them as test particles moving in an external potential generated by the cdm / baryonic structures , and ( ii ) a semi - analytical , linear technique ( section [ linear ] ) , which requires additional assumptions about the neutrino phase space distribution . in both cases , the cdm / baryonic gravitational potentials are calculated from parametric halo density profiles from high resolution @xmath6-body studies @xcite ( section [ profiles ] ) and/or from realistic mass distributions reconstructed from observational data ( e.g. , @xcite ) . using these two computational techniques , we track the relic neutrinos accretion onto cdm halos ranging from the galaxy to the galaxy cluster variety ( @xmath445 ) , and determine the neutrino number densities on scales @xmath446 for neutrino masses satisfying current constraints ( [ eq : massbound ] ) from cmb and lss ( figures [ fig : overdensities ] and [ fig : nuvscdm ] ) . because we can simulate only a finite set of halo and neutrino parameters , we provide also additional plots illustrating the approximate dependences of the neutrino overdensities on the halo and neutrino masses ( figures [ fig : halomass ] and [ fig : msquare ] ) . these can be used for interpolation between simulation results . furthermore , we find that the linear technique systematically underestimates the neutrino overdensities over the whole range of halo and neutrino masses considered in this study ( figure [ fig : overdensities ] ) . reconciliation with @xmath6-@xmath16-body simulations can only be achieved if we impose a smoothing scale of @xmath447 , or if the overdensity is no more than three or four . we therefore conclude that the linear theory does not generally constitute a faithful approximation to the vlasov equation in the study of neutrino clustering on galactic and sub - galactic scales ( @xmath448 ) . however , it may still be useful for finding the minimum effects of neutrino clustering in other contexts not considered in this work ( e.g. , the nonlinear matter power spectrum @xcite ) . next we apply our @xmath6-@xmath16-body method to calculate the relic neutrino number density in the milky way ( section [ milkyway ] ) , especially their phase space distribution in our local neighbourhood at earth @xmath216 , taking also into account contributions to the total gravitational potential from the galactic bulge and disk . we find a maximum overdensity of @xmath449 per neutrino flavour in our immediate vicinity , provided that the neutrino mass is at its current upper limit of @xmath450 ( table [ table : moments ] and figure [ fig : milkyway ] ) . for neutrino masses less than @xmath297 , the expected overdensity from gravitational clustering is less than two . the associated coarse - grained momentum spectra show varying degrees of deviation from the relativistic fermi dirac function , but share a common feature that they are semi - degenerate , with phase space density @xmath451 , up to the momentum state corresponding to the escape velocity from the milky way at @xmath216 ( figure [ fig : momentum ] ) . this means that the neutrino number densities we have calculated here for @xmath216 are already the _ highest possible _ , given the neutrino mass , without violating phase space constraints ( e.g. , @xcite ) . in order to attain even higher densities , one must now appeal to non - standard theories ( e.g. , @xcite ) . in terms of scattering - based detection possibilities , this meager enhancement in the neutrino number density in the milky way from gravitational clustering means that relic neutrinos are still far from being detected in fully earthbound laboratory experiments . for flux detection methods based on coherent elastic scattering of relic neutrinos off target matter in a terrestrial detector ( section [ flux ] ) , a positive detection could be thirty to forty years away @xcite , provided that light neutrinos are dirac particles . for light majorana neutrinos , another @xmath452 times more sensitivity would be required in the detector for a positive signal . target detection methods using accelerator beams ( section [ accelerator ] ) seem equally hopeless , unless the accelerator is the size of the earth and operates at an energy of @xmath453 ( table [ table : accelerators ] ) . meanwhile , target detection using extremely energetic cosmic neutrinos ( eec@xmath418 , @xmath454 ) remains the only viable means to confirm the existence of big bang relic neutrinos within the next decade or so ( section [ uhecr ] ) . resonant annihilation of eec@xmath418 on relic neutrinos can be revealed as absorption dips in the eec@xmath418 flux ( e.g. , @xcite ) , or as emission features in the @xmath22-decay products . however , since absorption spectroscopy is largely insensitive to late time ( @xmath78 ) relic neutrino clustering , our findings here have little impact on the conclusions of @xcite . on the other hand , emission spectroscopy is sensitive to the relic neutrino content of the local gzk zone , @xmath455 . while we find no significant large scale clustering within @xmath456 ( figure [ fig : local ] ) and therefore no significant enhancement in the overall emission rates , it is still conceivable to exploit the considerable neutrino overdensities in nearby galaxy clusters , and search for directional dependences in the post - gzk emission events . for the virgo cluster , for example , we estimate the event rate from the central @xmath457 region to be @xmath442 and @xmath443 times the unclustered rate for neutrino mass @xmath441 and @xmath297 respectively , assuming an isotropic distribution of eec@xmath418 sources . [ our estimates differ by a factor of a few from the predictions of because their linear technique fails to account for additional clustering from nonlinear effects , as we have demonstrated in this study ( section [ linear ] ) ] . planned observatories such as the pierre auger observatory @xcite , euso @xcite and owl @xcite will have sufficient angular resolution to , in principle , see this enhancement . however , considering the rapidly improving constraints on both the eec@xmath418 flux and neutrino masses , it remains to be seen if the enhancement can indeed be observed with enough statistical significance . we would like to thank david cerdeo , pasquale di bari , birgit eberle , jan hamann , oleg lebedev , helmut mais , and thomas reiter for useful discussions . we give in this appendix a detailed discussion of the techniques used in our @xmath6-@xmath16-body simulations . following @xcite , we begin with a lagrangian @xmath458 $ ] for a test particle ( neutrino ) with mass @xmath24 and peculiar velocity @xmath459 moving in a spherically symmetric gravitational potential well @xmath460 generated by the cdm halo . because of spherical symmetry , the motion of the test particle is confined to a plane . this allows us to switch to polar coordinates @xmath461 to obtain @xmath462,\ ] ] and eliminate two superfluous variables in the process . the canonical momenta conjugate to @xmath222 and @xmath463 are @xmath464 and @xmath465 respectively , leading to the hamiltonian @xmath466 + a m_{\nu } \phi(r,\tau),\ ] ] and hence hamilton s equations @xmath467 where the last equality expresses the conservation of conjugate comoving angular momentum @xmath468 . for computational purposes , it is convenient to rewrite the set of differential equations ( [ eq : hamilton ] ) in terms of redshift @xmath75 and the normalised quantities @xmath469 and @xmath470 , thus rendering ( [ eq : hamilton ] ) into the form @xmath471 where @xmath472 and @xmath473 are , respectively , the scale factor and the expansion rate , the latter of which is determined by the present day hubble constant @xmath474 and the cosmological model in hand . the gravitational potential @xmath475 obeys the poisson equation @xmath476 since we are treating the cdm halo as a density perturbation sitting on top of a uniform background , i.e. , @xmath477 , we have the following ( partial ) solution : @xmath478 where @xmath227 is the physical mass contained in a sphere of radius @xmath222 . for an nfw halo , @xmath479 the meanings of these various symbols are explained in section [ profiles ] . the number of neutrinos at some initial time @xmath480 in the interval @xmath481 is defined to be @xmath482 we assume the initial phase space distribution to be isotropic and homogeneous , i.e. , @xmath483 . this initial homogeneity implies that the ensemble s subsequent evolution under a spherically symmetric potential will depend only on the initial radial distance @xmath222 , radial momentum @xmath484 , and transverse momentum @xmath485 . thus we may rewrite the phase space volume element as @xmath486 with @xmath487 , @xmath488 $ ] , @xmath489 , and @xmath490 , @xmath491 , @xmath492 . in an ideal and perfectly deterministic calculation , one would track the motion of every particle for every occupied combination of @xmath222 , @xmath484 and @xmath493 . this is obviously impossible in practice ; one must therefore resort to simulating only a representative set of particles and endow each with a `` weight '' according to the phase space region from which the particle is drawn . we show below how this weight is calculated .- body simulations @xcite . the representative point particles are drawn randomly from the initial phase space distribution . ] suppose that a point particle initially at @xmath494 is representative for all particles in the initial phase space interval @xmath495 . the weight carried by this representative particle is defined as @xmath496 where @xmath497 means summing over all @xmath463 , @xmath71 and @xmath498 , which is simple : @xmath499 . the spatial integral over @xmath222 is also readily calculable . the remaining phase space integral @xmath500 can be solved by way of the parameterisation @xmath501 , and @xmath502 , and hence @xmath503 thus the weight carried by a point particle centred on and representing the interval @xmath504 is given by @xmath505 where @xmath506 is a dimensionless momentum variable . we construct the neutrino density profiles from the discrete outputs of our @xmath6-@xmath16-body simulations using the kernel method of @xcite and @xcite , which we reproduce here for completeness . the number density corresponding to a set of particles is estimated to be @xmath507,\ ] ] where @xmath508 is the weight carried by the @xmath56th particle , @xmath509 is the window width , and @xmath510 is a normalised kernel such as the gaussian kernel , @xmath511 in order to obtain a spherically symmetric profile , we smear every particle around the surface of a sphere with radius @xmath512 centred on @xmath121 . the spherically symmetric density estimate is then @xmath513 with @xmath514 substituting for the gaussian kernel , we get @xmath515.\ ] ] the optimal choice of @xmath509 generally depends on the underlying function @xmath516 @xcite . in our analysis , however , a constant @xmath509 is adopted for simplicity . our momentum distributions in section [ milkyway ] are also constructed with this kernel method . 99 b. eberle , a. ringwald , l. song and t. j. weiler , phys . d * 70 * ( 2004 ) 023007 [ arxiv : hep - ph/0401203 ] . d. fargion , b. mele and a. salis , astrophys . j. * 517 * ( 1999 ) 725 [ arxiv : astro - ph/9710029 ] . t. j. weiler , astropart . * 11 * ( 1999 ) 303 [ arxiv : hep - ph/9710431 ] . s. yoshida , g. sigl and s. j. lee , phys . * 81 * ( 1998 ) 5505 [ arxiv : hep - ph/9808324 ] . z. fodor , s. d. katz and a. ringwald , phys . * 88 * ( 2002 ) 171101 [ arxiv : hep - ph/0105064 ] . z. fodor , s. d. katz and a. ringwald , jhep * 0206 * ( 2002 ) 046 [ arxiv : hep - ph/0203198 ] . a. d. dolgov , s. h. hansen , s. pastor , s. t. petcov , g. g. raffelt and d. v. semikoz , nucl . b * 632 * ( 2002 ) 363 [ arxiv : hep - ph/0201287 ] . y. y. y. wong , phys . d * 66 * ( 2002 ) 025015 [ arxiv : hep - ph/0203180 ] . c. hagmann , in _ cosmo98 : proceedings of the conference on particle physics and the early universe _ , edited by d. o. caldwell ( aip , woodbury ny , 1999 ) , p.460 [ aip conf . * 478 * ( 1998 ) 460 ; arxiv : astro - ph/9902102 ] . j. bernstein , m. ruderman and g. feinberg , phys . rev . * 132 * ( 1963 ) 1227 . b. p. konstantinov and g. e. kocharov , j. exp . * 19 * ( 1964 ) 992 . r. cowsik , y. pal and s. n. tandon , phys . * 13 * ( 1964 ) 265 . t. hara and h. sato , prog . * 64 * ( 1980 ) 1089 . t. hara and h. sato , prog . phys . * 65 * ( 1981 ) 477 . p. gondolo , g. gelmini and s. sarkar , nucl . b * 392 * ( 1993 ) 111 [ arxiv : hep - ph/9209236 ] . e. roulet , phys . d * 47 * ( 1993 ) 5247 . s. yoshida , h. y. dai , c. c. jui and p. sommers , astrophys . j. * 479 * ( 1997 ) 547 [ arxiv : astro - ph/9608186 ] . k. greisen , phys . * 16 * ( 1966 ) 748 . g. t. zatsepin and v. a. kuzmin , jetp lett . * 4 * ( 1966 ) 78 [ pisma zh . fiz . * 4 * ( 1966 ) 114 ] . o. e. kalashev , v. a. kuzmin , d. v. semikoz and g. sigl , phys . d * 65 * ( 2002 ) 103003 [ arxiv : hep - ph/0112351 ] . d. s. gorbunov , p. g. tinyakov and s. v. troitsky , astropart . * 18 * ( 2003 ) 463 [ arxiv : astro - ph/0206385 ] . d. v. semikoz and g. sigl , jcap * 0404 * ( 2004 ) 003 [ arxiv : hep - ph/0309328 ] . g. gelmini , g. varieschi and t. weiler , arxiv : hep - ph/0404272 . saltdome shower array , p. gorham , d. saltzberg , a. odian , d. williams , d. besson , g. frichter and s. tantawi , nucl . instrum . a * 490 * ( 2002 ) 476 [ arxiv : hep - ex/0108027 ] . t. reiter , unpublished notes .
a broad range of diffusive processes on networks , from consensus formation @xcite to current flows on electric circuits @xcite , can be modeled by random walks or , equivalently , by markov chains . their unbiased exploration of the underlying structure also makes them popular tools for designing algorithms for , e.g. , navigation and search on networks @xcite , defining central nodes in a given network @xcite , community detection @xcite , and respondent - driven sampling @xcite . in tandem with these applications , the impact of network structure on dynamics of random walks , including the hitting time , mixing time , and stationary density has been extensively studied . however , recent studies identified limitations of the classical network paradigm , where dynamics is modeled by a dynamical process on a static underlying structure . in a broad range of empirical systems , evidence suggests instead that dynamics presents non - trivial correlations between events and long - tailed interevent time distributions @xcite . these observations are incompatible with the poissonian statistics implicitly assumed in stochastic models , therefore calling for richer models for temporal networks @xcite . an important practical question concerns the impact of the temporality of a network on diffusion . in order to address this question , a first approach consists in simulating random walks on real or synthesized data of time - stamped event sequences , and to compare their dynamical properties to those of properly defined null models @xcite . a second approach consists in studying analytically the properties of random walks on specific models of temporal networks . in most studies , however , network structure changes at regular time intervals , and transitions between networks at different times are independent @xcite or markovian @xcite . in the present work , we follow the latter path to analytically model diffusion under the non - poissonian nature of interevent times . previous studies proposed modeling temporal networks as stochastic sequences of events obeying a prescribed distribution of interevent times attached on each link @xcite . a random walk process , called the active random walk , was then defined as a renewal process ; i.e. , after a walker arrives at a node , the interevent times attached to all links incident to the node are reinitialized @xcite . despite its non - markovianity owing to the fact that the rate at which an event takes place depends on the time of the previous event , this stochastic process can be described by a generalized master equation , and some of its properties , such as the stationary density @xcite and the relaxation time @xcite , were analytically solved . in this work , we will first derive an analytical expression for the mean recurrence time for the active random walk . then , we consider a different non - renewal process , called the passive random walk , in which interevent times are reset only on the links traversed at each jump . we will use the fact that the passive random walk shows a stronger non - markovianity than the active random walk does because passive random walkers remember their past trajectories to some extent . finally , we will perform numerical simulations to test our analytical predictions and compare the stationary density and mean recurrence time between the two types of walks . we consider an undirected network on @xmath0 nodes . we denote the set of nodes by @xmath1 and the set of links by @xmath2 . we often refer to this network as the aggregate network because it is considered as the aggregation of the temporal network , which we introduce in the following , across time . stochastic temporal networks add a time dimension to the aggregated networks by assigning a random interevent time @xmath3 to each link @xmath4 in a renewal manner @xcite . for a chain network of three nodes , a hypothetical sequence of interevent times is depicted in fig . [ fig : schematic passive ] . the interevent time , denoted by @xmath5 , @xmath6 , @xmath7 , for link @xmath8 in fig . [ fig : schematic passive ] , is the interval between two consecutive activation events of the link . we denote the probability density function ( pdf ) of @xmath3 by @xmath9 . we assume that the mean of @xmath3 is finite . link activation is an instantaneous event , and we assume that the random walker moves from the current node to the neighboring node through the activated link whenever it is possible . the example shown in fig . [ fig : schematic passive ] indicates that , if the walker starts from node 1 at @xmath10 , it moves to node 2 at @xmath11 . then , it moves back to node 1 at @xmath12 . after coming back to node 2 at @xmath13 , the walker transits to node 3 at @xmath14 . some remarks are in order . first , some activation events , such as those immediately following the interevent times @xmath15 and @xmath16 , may not be used by a walker . second , although the walker s movement as illustrated in fig . [ fig : schematic passive ] may look deterministic , it is stochastic because the interevent times are random variables . third , the probability that more than one links are simultaneously activated is equal to zero because the model is defined in continuous time . fourth , the walker is assumed not to have any internal waiting time . therefore , the walker instantaneously moves upon the activation of a link incident to the walker s location . fifth , the continuous - time nature of our model makes it different from the previously proposed discrete - time random walk defined on the so - called activity driven model @xcite . a main difference between the present model and those considered in ref . @xcite is that the former generates interevent times from a given pdf , which lends itself to an analytical study of the process , whereas the latter uses interevent times observed in real data ( and their randomizations ) , more appropriate for the study of empirical data . we analyze two versions of random walks on stochastic temporal networks @xcite . the first version is the so - called active random walk . in the active random walk , when the random walker arrives at a node , it reinitializes the interevent times on all links , which makes the process renewal . in this case , the waiting time , i.e. , the time for which a walker waits on a node before the link appears , is equivalent to the interevent time @xcite . in the example shown in fig . [ fig : schematic active ] , once the random walker moves from node 1 to node 2 at @xmath11 , the new interevent times , denoted by @xmath17 and @xmath18 , are independently drawn . because @xmath19 , the walker moves to node 3 at @xmath20 . it should be noted that interevent times for link @xmath21 before @xmath11 do not matter for the movement of the walker in question , reflecting the renewal nature of the active random walk . phenomena in which a node starts something in response to an external input or the arrival to the node might be represented by the active walk . possible examples include the broadcasting of gossips @xcite and random surfing on the world wide web by users . in the latter case , a user that has arrived at a webpage may stay there looking at the content . then , after a randomly distributed amount of time , the user may decide to jump to a different webpage , which is selected with equal probability among the hyperlinks that the original page has . the second version is the so - called passive random walk , which does not assume the reinitialization at all links . when the random walker moves to a neighbor through link @xmath8 , a new interevent time is drawn only for @xmath8 . in fact , the example shown in fig . [ fig : schematic passive ] corresponds to the passive random walk . transmission of infectious diseases or information may be better described with the passive than active random walk because infection of a node does not seem to instantaneously affect interevent times on other links @xcite . in other words , to a first - order approximation , the inter - contact time on any link @xmath8 can be regarded to obey a given distribution and is not altered by whether an infection occurs upon the contact . then , a pathogen or piece of information starting from an initial node may travel to other nodes via the passive random walk . the complexity of the passive random walk lies in its non - renewal nature . in other words , transition rates of the passive random walk depend on the trajectory that the walker has taken such that we have to account for the entire trajectory of the random walker to accurately evaluate its behavior . it should be noted that for exponentially distributed interevent times the active and passive random walks are identical and reduce to the usual continuous - time random walk on the aggregate ( i.e. , static ) network . the walker starts at node 1 at @xmath10 . the link activation is represented by horizontal bars . @xmath22 ( @xmath23 ) represents the @xmath24th interevent time on link @xmath8 . the random walker follows the path indicated by the thick lines.,width=453 ] and follows the path indicated by the thick lines . note that @xmath18 is not the second interevent time on link @xmath21 . it is a realization of the interevent time drawn at @xmath11.,width=453 ] the steady state of the active random walk was derived through a master equation approach in ref . @xcite . in this section , we first review these results ( secs . [ sub : formulate active ] and [ sub : steady state active ] ) . then , we derive the mean recurrence time for the active random walk ( sec . [ sub : mean recurrence time active ] ) . the main results when the interevent times of different links are distributed according to a common pdf are shown in eqs . and for the steady state and mean recurrence time , respectively . we denote the probability that the random walker is located at node @xmath24 ( @xmath25 ) at time @xmath26 by @xmath27 . the normalization is given by @xmath28 . the rate at which the walker arrives at node @xmath29 from node @xmath24 at time @xmath26 is denoted by @xmath30 . the transition rate for a single walker to move from @xmath24 to @xmath29 at time @xmath26 is given by @xmath31 . the master equation that governs the random walk is given by @xmath32\\\ & = \sum_{j;(ji ) \in e } \left [ r_{i \gets j}(t)p_j(t ) - r_{j \gets i}(t)p_i(t ) \right ] . \notag\end{aligned}\ ] ] if the underlying static network is connected , which we assume in the following , the random walk is mixing and the stationary density , denoted by @xmath33 , is obtained if we set @xmath34 for all nodes @xmath24 . for exponentially distributed interevent times , the transition rate is given by @xmath35 where @xmath36 denotes the mean of @xmath37 , the interevent time on link @xmath38 . by combining eq . and @xmath39 , we conclude that the steady state is the uniform distribution @xcite . for arbitrary interevent time distributions , we can not usually calculate @xmath40 . in addition , @xmath40 may not be symmetric with respect to @xmath24 and @xmath29 so that the steady state may deviate from the uniform distribution . to calculate the steady state in this case , we define @xmath41 as the rate at which the walker transits from @xmath24 to @xmath29 after time @xmath26 has elapsed since the walker arrived at @xmath24 . this event happens when , link @xmath38 is activated at time @xmath26 and any other link @xmath42 , where @xmath43 , has not been activated by @xmath26 . because all @xmath44 s , with the case @xmath45 included , are reinitialized at the arrival of the walker at node @xmath24 , we obtain @xmath46 we use eq . to derive the master equation for the active random walk . the rate at which the random walker reaches node @xmath29 from an adjacent node @xmath24 at time @xmath26 satisfies @xmath47 where @xmath48 is the rate at which the walker arrives at node @xmath24 at time @xmath26 from an arbitrary neighbor , @xmath49 are initially chosen weights on the links satisfying @xmath50 and @xmath51 is dirac s delta function . a detailed proof for eq . is found in ref . @xcite . by substituting eq . in eq . , we obtain @xmath52 dt'\ ] ] for any @xmath53 . we define @xmath54 i.e. , the pdf of the time to transit from node @xmath24 to somewhere . in other words , @xmath55 is the pdf of @xmath56 , which is the first time at which a link incident to @xmath24 is activated . by integrating eq . and abbreviating @xmath57 which is the probability to remain at @xmath24 for a time longer than @xmath26 , we obtain @xmath58 for any @xmath53 . the derivation of eq . is shown in the appendix . the steady state of the active random walk is evaluated via the laplace transform of eqs . and , expansion of the exponential , and application of the final value theorem @xcite . here we briefly present a slightly modified derivation of the steady state . we take the laplace transform of eq . to obtain @xmath59 here , @xmath60 is the laplace transform of @xmath61 , and parallel definitions are applied to @xmath62 and @xmath63 . equation implies @xmath64 according to the final value theorem , the steady state probability for node @xmath24 , denoted by @xmath65 , is given by @xmath66 . by combining eqs . and , we obtain @xmath67 where @xmath68 is the rate at which the random walker arrives at node @xmath29 in the steady state . to calculate @xmath69 , we define the @xmath70 matrices @xmath71 and @xmath72 , where @xmath73 is the laplace transform of matrix @xmath74 . we transform eq . into the laplace space to obtain @xmath75 where @xmath76 and @xmath77 . we multiply both sides of eq . with @xmath78 and obtain @xmath79 + s\bm p(0).\ ] ] by taking the limit @xmath80 on both sides of eq . , we find that vector @xmath69 is the dominant eigenvector of @xmath81 , i.e. , @xmath82 suppose that the random walker just arrived at a node . @xmath81 contains the probabilities to make a transition from one node to another in one step , @xmath83 contains the probabilities to do so in two steps , and so on . equation implies that @xmath84 is proportional to the steady state of the discrete - time random walk on the aggregate network with transition probabilities defined by @xmath81 . it should be noted that @xmath84 is unique up to the scaling factor because the aggregate network has been assumed to be connected . to obtain @xmath85 in eq . , we weight the steady state vector of the discrete - time random walk with the mean time for which the walker stays at the node . when interevent times for different links are identically distributed according to @xmath86 , eq . is reduced to @xmath87 where @xmath88 is the degree of node @xmath24 . by combining eqs . and , we obtain @xmath89 which is consistent with the fact that the steady state of the simple random walk on an arbitrary static undirected network is proportional to the degree @xcite . by substituting eq . in eq . and using @xmath90 , we obtain @xmath91 where @xmath92 s are i.i.d . copies of the interevent time . it should be noted that @xmath93^{d_i } dt\ ] ] depends solely on the degree of a node . therefore , the steady state depends only on the node s degree . if the interevent time is exponentially distributed , we obtain @xmath94 so that the steady state is the uniform distribution . this is consistent with our previous argument in sec . [ sub : formulate active ] , where we derived this fact directly from the master equation [ see eq . ] . otherwise , @xmath95 is not necessarily proportional to @xmath96 such that the steady state may not be the uniform distribution . let @xmath97 be the recurrence time , i.e. , the time at which a random walker starting at @xmath24 returns to @xmath24 for the first time . we denote the pdf of @xmath97 by @xmath98 . our goal in this section is to determine the mean recurrence time given by @xmath99 the hopping rate of the walker generally depends on the time already spent at a node . therefore , we confine ourselves to the recurrence time since the walker has just arrived at node @xmath24 . we will adapt the derivation of the mean recurrence time for discrete - time random walks on static undirected networks @xcite to the case of the active random walk . denote by @xmath100 the probability that the random walker is located at node @xmath24 at time @xmath26 given it started at node @xmath24 . we obtain @xmath101 the first term on the right - hand side of eq . governs the case in which the random walker has not left @xmath24 until time @xmath26 . the second term accounts for the walker that has left @xmath24 at least once . by transforming eq . into the laplace space , we obtain @xmath102 equation implies @xmath103 where @xmath104 equation implies that the mean recurrence time of node @xmath24 is the first moment of @xmath98 . therefore , the mean recurrence time is derived by the laplace transform of @xmath98 as follows : @xmath105 by substituting eq . in eq . , we obtain @xmath106\left [ p^*_i + sr_{ii}(s ) \right ] + s \hat{\phi}_i(s ) \left [ r_{ii}(s)+sr_{ii}'(s ) \right]}{\left [ p^*_i + sr_{ii}(s ) \right]^2 } \bigg|_{s=0}.\ ] ] application of the final value theorem yields @xmath107 by using the rule of lhospital , we obtain @xmath108 furthermore , we recall that @xmath109 [ see eq . ] , so that the tail of @xmath110 decreases fast enough to imply @xmath111 by substituting eqs . , , , , and in eq . , we obtain @xmath112 in particular , if interevent times on different links are identically distributed , the combination of eqs . and yields @xmath113 it should be noted that for discrete - time random walks on undirected static networks the mean recurrence time is also inversely proportional to the node s degree @xcite . in this section , we evaluate the steady state and mean recurrence time of the passive random walk . the main results when the interevent times of different links are distributed according to a common pdf are shown in eqs . and for the steady state and mean recurrence time , respectively . the difference from the active random walk is that the move of a walker does not reinitialize interevent times except on the link used for that jump . when the passive random walker arrives at a node , links incident to the node have thus already been inactive for some random time . therefore , in contrast to the case of the active random walk , the intervent time and the waiting time @xmath114 ( @xmath115 ) , i.e. , the time until link @xmath8 is activated since the random walker arrives at a node incident to @xmath8 , are different in general . they are equivalent only when the interevent time distribution is exponential , corresponding to the fact that the poisson process is memoryless . otherwise , the waiting time depends on the time at which the random walker arrives at a node . this distinction is important because the waiting time plays a direct role in diffusion on networks , whereas the interevent time plays only an indirect role . when the arrival of a walker on a node and the activation of an link are independent processes , it is known that the waiting time distribution @xmath116 is given in terms of the interevent time distribution @xmath9 by @xmath117 and that the average waiting time depends on the variance of the interevent time , a property called the waiting time paradox or bus paradox in queuing theory @xcite . in general , the time at which the random walker jumps to an adjacent node depends on the previous trajectory of the walker . however , we assume that eq . holds true in the following analysis . in the following , we perform an approximation in order to analytically evaluate the waiting time , steady state , and mean recurrence time of the passive random walk . to this end , we neglect the trajectory of the random walker except for its last and current positions denoted by @xmath118 and @xmath24 , respectively . more precisely , we retain the last activation time of the link through which the walker arrived at @xmath24 and suppose that all the other links were activated at random times in the past . the waiting time for link @xmath42 is given by the interevent time distribution @xmath119 . the waiting times for all the other links incident to @xmath24 are approximately distributed according to eq . . the pdf of the time when the walker transits from node @xmath24 to node @xmath29 given that it arrived at node @xmath24 from node @xmath118 is approximated by @xmath120 \int_{t}^{\infty } \psi_{(ik)}(r ) dr & ( j \neq k),\\ \psi_{(ij)}(t ) \prod_{\ell \neq j , k;(i\ell ) \in e } \left [ \int_{t}^{\infty } \rho_{(i\ell)}(r ) dr \right ] & ( j = k ) . \end{cases}\ ] ] equation suggests that the trajectory of the random walker impacts the order of link activation . in particular , when the interevent time obeys a long - tailed distribution , a transition from node @xmath24 back to node @xmath118 is more likely than that from @xmath24 to another node . it should be noted that the probability of transition does not depend on the destination node in the case of the active random walk . it should be also noted that eq . is exact for exponentially distributed interevent times . similarly to eq . , we obtain @xmath121 + p_{j \gets i}(0 ) \delta(t),\ ] ] where the initial condition satisfies eq . . similarly to eq . , we define @xmath122 which is the approximate probability that the walker stays at node @xmath29 for time longer than @xmath26 given that it arrived from node @xmath24 . by substituting eq . in eq . , using eq . , and performing a calculation similar to the derivation of eq . in the appendix , we obtain @xmath123\ ] ] for any @xmath124 . equations and govern the dynamics of the passive random walk . to calculate the approximate steady state distribution , we proceed similarly to the case of the active random walk . by taking the laplace transform of eq . , we obtain @xmath125 + p_{j \gets i}(0 ) . \label{eq : hatq_ji(s)}\ ] ] in terms of the vectors , eq . is written as @xmath126 where @xmath127 is the laplace transform of the @xmath128 matrix ( @xmath129 is the number of links in the aggregate network ) given by @xmath130 @xmath131 is the laplace transform of the @xmath132-dimensional column vector @xmath133 , and @xmath134 is the laplace transform of the @xmath132-dimensional column vector @xmath135 . in eq . , we define @xmath136 for @xmath137 because such a transition is impossible . by multiplying both sides of eq . by @xmath78 and letting @xmath138 , we obtain @xmath139 where @xmath140 and @xmath141 . to determine the steady state @xmath142 , we transform eq . to the laplace space and multiply both sides by @xmath78 to obtain @xmath143 by setting @xmath138 in eq . , we obtain @xmath144 finally , eq . implies @xmath145 therefore , @xmath146 is the ( approximate ) mean time for which a walker arriving at @xmath24 from node @xmath29 waits before moving to a neighbor . in summary , the steady state is approximately given by @xmath147 where @xmath148 is the solution of eq . , and the normalization is given by @xmath90 . denote the components of @xmath149 by @xmath150 for @xmath151 . when interevent times for different links are identically distributed according to @xmath86 , which we assume in this section , we obtain @xmath152 the first equality in eq . follows from the fact that @xmath153 , indicating that the walker moved from @xmath29 to @xmath24 and then from @xmath154 to @xmath118 , if and only if @xmath155 . the second equality follows from the assumption that @xmath156 for any @xmath4 . equation indicates that @xmath149 is a doubly stochastic matrix . therefore , the solution of eq . is given by @xmath157 , where @xmath158 represents the @xmath132-dimensional column vector whose all elements are equal to unity . by using eq . , we obtain @xmath159 this result is the same as that for the active random walk with identical interevent time distributions [ see eq . ] . to evaluate the right - hand side of eq . , we use @xmath160 \left [ \int_t^{\infty } \rho(t')dt ' \right]^{d_i-1 } dt,\end{aligned}\ ] ] where @xmath161 s are i.i.d . copies of the waiting time distributed according to a common pdf @xmath162 . by substituting @xmath163 which is derived by the combination of the definition of @xmath162 [ see eq . ] and the leibniz integral rule , in eq . , we obtain @xmath164 \left [ \int_t^{\infty } \rho(t')dt ' \right]^{d_i-1 } dt\notag\\ = & \langle \tau \rangle - ( d_i-1)\left\langle \min_{k = 1 , \ldots , d_i-1}\ { \tau , \tilde{\tau}_{k } \ } \right\rangle.\end{aligned}\ ] ] in the last equality in eq . , we used integration by parts . equation implies @xmath165 therefore , the mean time that the passive random walker spends at a node before transiting to a neighboring node does not depend on @xmath86 except for the dependence on @xmath166 . by combining eqs . , , , and and using @xmath90 , we obtain @xmath167 unlike for the active random walk [ see eq . ] , the ( approximated ) steady state of the passive random walk is the uniform distribution for any @xmath86 and network structure . this is also consistent with the case when interevent times obey the exponential distribution , for which we derived the steady state in sec . [ sub : formulate active ] directly from the master equation . to evaluate the mean recurrence time for the passive random walk , we denote by @xmath168 the time at which a random walker leaving node @xmath24 returns to @xmath24 through link @xmath169 for the first time . the pdf of @xmath168 is denoted by @xmath170 . we also define @xmath171 as the probability that the random walker is located at node @xmath24 at time @xmath26 , given that it arrived at @xmath24 from @xmath29 at time 0 . it should be noted that bayes rule results in @xmath172 the pdf of the first recurrence time satisfies @xmath173 here , @xmath110 denotes the probability that the walker resides at node @xmath24 for time longer than @xmath26 and is given by @xmath174 owing to bayes rule . by converting eq . to the laplace space , we obtain @xmath175 to evaluate eq . , we resort to a mean - field ansatz given by @xmath176 , where @xmath177 . in other words , we neglect from which node the walker returns to @xmath24 . this approximation may be accurate if the interevent times are identically distributed for different links , which we assume hereafter . under this approximation , eq . is reduced to @xmath178 by following the same steps as in eqs . , we obtain @xmath179 by substituting eqs . and in eq . , we obtain @xmath180 it should be noted that the mean recurrence time is also inversely proportional to the degree for the active random walk [ see eq . ] . it should be also noted that @xmath181 is independent of @xmath86 for the passive random walk except for the factor @xmath182 , but not for the active random walk . to illustrate the theoretical results derived in secs . [ sec : active ] and [ sec : passive ] , we analyze examples in this section . we assume that the interevent times are identically distributed for all links according to @xmath86 , which is either a power - law or weibull distribution . ( solid line ) , the power - law distribution given by eq . with @xmath183 ( dotted line ) , and the weibull distribution given by eq . with @xmath184 and @xmath185 ( dashed line).,width=453 ] consider the case in which all interevent times follow a power - law distribution given by @xmath186 where @xmath187 , corresponding to the assumption @xmath188 . we plot the pdfs of the power - law by the dotted line in fig . [ fig : distr ] . in fact , many real data show @xmath189 or 1.5 @xcite , which apparently contradicts our choice @xmath190 . here , for simplicity we assume @xmath190 to investigate the effect of long - tailed distributions on the random walk on temporal networks . for the active random walk , the pdf of the transition time is calculated from the substitution of eq . in eq . as follows : @xmath191 the probability to make a transition from node @xmath24 to an adjacent node @xmath29 [ see eq . ] is given by @xmath192 the mean time for which the random walker stays at node @xmath24 before moving to a neighbor [ see eq . ] is given by @xmath193 for the passive random walk , we substitute eq . in eq . to obtain the pdf of the waiting time as follows : @xmath194 it should be noted that the mean waiting time diverges if @xmath195 . however , the following results are valid for all @xmath190 for the following reason . the link walked through in the last jump has a finite mean waiting time because of resetting . this fact guarantees that the mean time to the next jump is always finite . therefore , the steady state and the mean recurrence time safely exist . by substituting eq in eq . , we obtain @xmath196 the approximate probability to make a transition from node @xmath24 to a neighbor @xmath29 conditioned that the random walker reached node @xmath24 through link @xmath42 [ see eq . ] is given by @xmath197 to illustrate the difference between the active and passive random walks , we consider the network composed of three nodes shown in figs . [ fig : schematic passive ] and [ fig : schematic active ] . for the active random walk , eq . leads to @xmath198 these transition probabilities are the same as those in the case of identically distributed exponential interevent times . in contrast , for the passive random walk , eq . leads to @xmath199 because @xmath200 , we obtain @xmath201 . therefore , the passive random walker tends to travel on the link that the walker has used the last time . in particular , when @xmath202 , with @xmath203 , we obtain @xmath204 . therefore , a random walker starting at node 1 , for example , will be trapped between nodes 1 and 2 for a long time before transiting to node 3 . nodes generated by the barabsi - albert model @xcite . the interevent time is assumed to obey the power - law distribution with exponent @xmath205 for all links . ( a ) steady state of the active ( circles ) and passive ( triangles ) random walks . ( b ) mean time to return to an initial node for the active and passive random walks . the results for the case in which the interevent time obeys the exponential distribution are omitted because they are indistinguishable from the results for the passive random walk ( triangles ) . the lines represent the theoretical estimates . the nodes are sorted in ascending order of the degree.,width=377 ] for the active random walk , the mean time to stay at node @xmath24 , i.e. , eq . , is reduced to @xmath206 where @xmath92 s are i.i.d . copies of the interevent time . by substituting eq . in eq . , we obtain @xmath207 where subscript `` @xmath208 '' here and in the following corresponds to the active random walk . equation implies that the steady state is not uniform , which is in contrast to the case of exponentially distributed interevent times . in particular , @xmath209 is large for nodes with small degrees , which is opposite to the case of the discrete - time simple random walk in undirected networks for which the steady state is proportional to the node s degree @xcite . in contrast , for the passive random walk , the steady state is the uniform distribution [ see eq . ] . to test our theory , we carried out numerical simulations . we set @xmath183 and used the barabsi - albert scale - free network @xcite with @xmath210 nodes . the two parameters in the barabsi - albert model were set to @xmath211 . we used the same single realization of the aggregate network for both active and passive random walks . we calculated the steady state as the average time for which the walker spent on each node between @xmath212 and @xmath213 . the choice @xmath212 is to exclude the transient . we initially placed the walker at one of the two nodes initially created in the barabsi - albert algorithm , each with probability @xmath214 . the passive random walk was simulated as follows . for each link , we stored the sum of realized interevent times . as soon as the random walker arrived at a node , a new interevent time was drawn for the link just used by the walker . for each of the remaining links incident to the node , interevent times were drawn until the sum of realized interevent times exceeded the time of arrival of the random walker . then , we selected the link @xmath8 with the smallest sum of realized interevent times . finally , the random walker jumped to a new node through @xmath8 . this procedure was repeated until the final time was reached . the numerically obtained steady state probability of each node is shown in fig . [ fig : powid](a ) for the active ( circles ) and passive ( triangles ) random walks . the nodes on the horizontal axis are shown in the ascending order of the degree . the numerical results are accurately predicted by the theory ( lines ) . in particular , the approximations made for analyzing the passive random walk do not cause a notable discrepancy between the numerical and theoretical results . next , we examine the mean recurrence time . for the active random walk , we substitute eqs . and in eq . to obtain @xmath215 for the passive random walk , by substituting @xmath216 in eq . , we obtain @xmath217 where subscript `` @xmath218 '' corresponds to the passive random walk . in the numerical simulations of the passive random walk , we assume that the walker arrived at the starting node @xmath24 from each neighbor of @xmath24 with the same probability at @xmath10 . to mimic the steady state of the stochastic temporal network , we assumed that the initial interevent time obeys the waiting time distribution [ see eq . ] for all links except for link @xmath169 from which the random walker arrived at @xmath10 . the initial interevent time of link @xmath169 is drawn from @xmath86 . for each starting node @xmath24 , we averaged the recurrence time over @xmath219 realizations of the random walk . numerically obtained mean recurrence times on the same scale - free network as that used in fig . [ fig : powid](a ) is shown in fig . [ fig : powid](b ) . the theory ( lines ) accurately matches the numerical results ( symbols ) . we also confirmed that the numerical results when @xmath86 is the exponential distribution with the same mean ( i.e. , @xmath220 ; shown in fig . [ fig : distr ] by the solid line ) completely overlap with those for the passive random walk when @xmath86 is the power - law distribution [ hence not shown in fig . [ fig : powid](b ) ] . figure [ fig : powid](b ) also indicates that , for each node , the active random walk realizes a smaller mean recurrence time than the passive random walk does . and we used the same realization of the scale - free network as in fig . [ fig : powid ] . ( a ) steady state . ( b ) mean recurrence time . the lines represent the theory . the nodes are sorted in ascending order of the degree.,width=377 ] for the power - law distribution of interevent times , the steady state probability of a node decreases with the degree for the active random walk , and the mean recurrence time at each node is larger for the passive than active random walk ( sec . [ sub : power ] ) . however , these results are not universal . to show this , we consider the case in which the interevent time obeys the weibull distribution given by @xmath221 where @xmath222 and @xmath223 are parameters . the weibull distribution with @xmath184 and @xmath185 , which yields @xmath220 , is shown by the dashed line in fig . [ fig : distr ] . it should be noted that the tail of the distribution is shorter than that of the exponential distribution with the same mean ( solid line ) . we start by illustrating the dynamics of the passive random walk on the three - node network . refer to fig . [ fig : schematic passive ] for a schematic . by combining @xmath224 with eq . , we obtain @xmath225 independent of the @xmath226 value . for @xmath227 , the interevent times are exponentially distributed such that @xmath228 . for @xmath229 , we obtain @xmath230 such that the random walker tends to alternate between two nodes , which is similar to the dynamics when @xmath86 is the power - law distribution ( sec . [ sub : power ] ) . for @xmath231 , we obtain @xmath232 such that the random walker tends to avoid traveling on the same link in consecutive transitions . for the rest of this section , we focus on the case @xmath184 . the numerical results for the weibull distribution with @xmath184 , @xmath185 , and the same scale - free network as that used in the previous section are shown in fig . [ fig : weibid ] . the steady state probability for the active random walk increases with the degree [ circles in fig . [ fig : weibid](a ) ] . in fact , a direct calculation of eq . for the weibull distribution yields @xmath233 , as shown by the lines overlapping with the circles in fig . [ fig : weibid](a ) . this result is in contrast to the case of the power - law distribution of interevent times , for which @xmath65 decreases with @xmath88 [ see eq . ] . for the passive random walk , the steady state obeys the uniform distribution ( triangles ) , which is consistent with the theory . the mean recurrence time under the weibull distribution of interevent times is shown in fig . [ fig : weibid](b ) . the results for the passive random walk ( triangles ) are indistinguishable from those for the power - law and exponential distributions , which is consistent with the theory . we also find that , for any node , the mean recurrence time is larger for the active than passive random walk . this result is opposite to that for the power - law distribution of interevent times . we studied two models of random walks on stochastic temporal networks . our main findings are summarized as follows . first , the steady state for the passive random walk with identically distributed interevent times on links is uniform for any network and distribution of interevent times . second , for the active random walk , the steady state probability decreases and increases with the degree for the power - law and weibull distribution of interevent times , respectively . third , the mean recurrence time for both types of walks is inversely proportional to the node s degree . fourth , the mean recurrence time for the passive random walk does not depend on the distribution of interevent times . fifth , the active random walk produces smaller mean recurrence times for each node than the passive walk does when the interevent time obeys the power - law distribution . in contrast , the mean recurrence times are larger for the active random walk than the passive random walk when the interevent time obeys the weibull distribution . the present result that the mean recurrence time is inversely proportional to the node s degree is consistent with that in ref . in particular , both studies conclude that the distribution of interevent times does not affect the mean recurrence time ( squares and diamonds in fig . 7 in ref . we reached this conclusion by explicit derivation of the mean recurrence time . in contrast , we consider that the strength of ref . @xcite in this respect lies in numerically showing the universality of this result across different data sets . it should be noted that a discrete - time simple random walk on a different temporal network model yields different results ; the mean recurrence time decreases but is not inversely proportional to the degree @xcite . the passive random model induces a correlated random walk . interesting connections of the present study may be made to seminal work on correlated random walk on lattices @xcite and to recent work modeling empirical pathways on networks by second - order markov processes @xcite . pursuing connection to anomalous diffusion on lattices @xcite may be also interesting . it may be also interesting to explore the case in which the mean interevent time diverges , which would make the walker to linger at a node for extremely long time . l.s . acknowledges the support provided through erato , jst and daad . r.l . acknowledges financial support from fnrs , from the arc `` mining and optimization of big data models '' , and from the eu project optimizr . this paper presents research results of the belgian network dysco funded by the interuniversity attraction poles programme . n.m . acknowledges the support provided through crest , jst . this work was partly supported by a bilateral joint research project between jsps , japan , and f.r.s .- fnrs , belgium . by substituting eq . in eq . , we obtain @xmath234 for any @xmath124 and @xmath235 . by integrating eq . , we obtain @xmath236 dt ' . \label{app : int}\ ] ] by applying the leibniz integral rule , i.e. , @xmath237 by setting @xmath238 , @xmath239 , and @xmath240 , we obtain @xmath241 to evaluate the second term on the right - hand side of eq . , we use the definition of @xmath110 [ see eq . ] to obtain @xmath242 by using @xmath243 and substituting eq . in eq . , we obtain @xmath244 by substituting eq . in eq . , we obtain @xmath245 to evaluate the right - hand side of eq . , we have to evaluate @xmath246 this task needs carefulness because @xmath247 behaves like @xmath248 around @xmath10 . by using the initial value theorem , we obtain @xmath249 because @xmath250 , we obtain @xmath251 where the last equality follows from eq . . the laplace transform of eq . yields @xmath252 using the dominated convergence theorem , we obtain @xmath253 the use of the dominated convergence theorem is justified because @xmath254 and @xmath255 by combining eqs . , , , and , we obtain @xmath256 therefore , by evaluating the right - hand side of eq . , we obtain @xmath257 which is equivalent to eq . . 39ifxundefined [ 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 * * , ( ) in @noop _ _ ( , ) , pp . @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) in @noop _ _ , , ( , ) , pp . in @noop _ _ ( , ) , pp . in @noop _ _ , ( , , ) pp . @noop * * , ( ) @noop * * , ( ) @noop ( ) @noop _ _ ( , , ) @noop * * , ( ) @noop ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * ( ) @noop _ _ ( , , )
the main goal of this paper is to develop a general theory of flag bicolorings and the related concepts of coverings and pseudo - orientations . the idea of consistent colorings of the flags of a map with two colors has appeared previously in literature in different contexts . for instance , when the automorphism group of a map which is a polytope has two orbits on the flags , we may color the flags with two colors in such a way that flags in different orbits have different colors . these are called 2-orbit maps in @xcite , but this investigation requires the extra property that the maps be abstract . specific @xmath0-colorings of flags are equivalent to the concept of @xmath1-_compatible _ maps introduced in @xcite , where @xmath1 is a class of @xmath0-orbit maps . this concept , generalized to hypermaps is called @xmath1-_conservative _ in @xcite . in all these instances , the colorings are used as a tool to work with automorphisms of maps . although some of the motivations behind the bicoloring of flags come from maps admitting symmetries , the ideas can be applied to general maps . in this work we shall not make assumptions on the automorphism groups of the maps in consideration . maps admitting a bicoloring of flags in which adjacent flags have different colors are precisely maps on orientable surfaces ( see proposition [ oble012 ] ) . although the other types of bicolorings of flags that we consider in this paper do not have a known topological equivalence , some results on orientability do translate directly to analogous results on bicolorability . in this sense we may think of bicolorings of flags as a generalization of orientability of maps . the paper is organized as follows . sections [ sec : maps ] , [ sec : ori ] , [ sec : col ] introduce the concepts of maps , pseudo - orientations and bicolorings , respectively . a relationship between pseudo - orientability and bicolorings is provided in section [ sec : colori ] . section [ sec : op ] explores the impact on bicolorability of operations on maps . in section [ sec : doubles ] , we discuss a natural double cover of every map which is not bicolorable . the set of bicolorings of any map can be given a natural group structure . in section [ sec : everygroup ] we determine , for each possible group , which surfaces admit maps with the given group of bicolorings . finally , in section [ sec : mpx ] , we generalize some of the results about maps to higher dimensional structures . a map @xmath2 is , first and foremost , an embedding of a graph ( or pseudograph ) on a ( compact , connected ) surface so that the components of the complement of the embedding ( called _ faces _ ) are topologically open disks . we can , for example , regard the cube as an embedding of the graph @xmath3 on the sphere . the graph @xmath3 is an example of a graph which is _ bipartite _ , i.e. , its vertices can be colored with two colors so that every edge joins vertices of opposite colors . when speaking about maps , we will use the term _ vertex - bipartite _ to describe a map @xmath2 whose underlying graph is bipartite . similarly , we will call @xmath2 _ face - bipartite _ provided its faces can be colored with two colors so that each edge separates faces of opposite colors . we will call @xmath2 _ edge - bipartite _ provided that the edges can be colored with two colors so that edges which are consecutive around a face ( and hence around a vertex ) have different colors . to look more closely at the structure of a map , we find the following subdivision useful : choose a point in the interior of each face to call its _ center _ and a point in the relative interior of each edge to be its _ midpoint_. draw dotted lines to connect each face - center with each incidence with the surrounding vertices and edge - midpoints . the original edges and these dotted lines divide the surface into triangles called _ flags_. figure [ fig : cubeflags ] shows the subdivision of the cube into flags . if a face meets a vertex ( or an edge ) more than once , we emphasize that the face center is connected to each incidence each appearance of the vertex or midpoint . for instance , consider the map @xmath4 shown in the left of figure [ fig : m4 ] . this map has only one face , an octagon , four edges and only one vertex . nonetheless , the dissection into flags draws 16 dotted lines , dividing the octagon , and the map , into 16 flags , as shown on the right . , title="fig:",height=113 ] , title="fig:",height=113 ] each flag corresponds to a mutual incidence of face , edge , and vertex , though different flags may correspond to the same triple . let @xmath5 be the set of flags . then let @xmath6 be the permutations on @xmath5 which match each flag @xmath7 with its three immediate neighbors , as in figure [ fig : flags ] . in that figure , we see that @xmath7 and @xmath8 are adjacent along a face - center - to - edge - midpoint line . thus @xmath7 and @xmath9 differ only in their incidences to a vertex , a 0-dimensional face of @xmath2 . similarly , @xmath7 and @xmath10 differ only in their incidences to an edge , a 1-dimensional face , while @xmath7 and @xmath11 differ only in their incidences to a 2-dimensional face . notice from figure [ fig : flags ] that the flag @xmath12-adjacent to @xmath8 is also @xmath13-adjacent to @xmath14 . in other words , as permutations on @xmath15 and @xmath12 commute . we can take a slightly more abstract point of view by defining a map to be a pair @xmath16)$ ] where @xmath5 is a set of things called _ flags _ , the @xmath17 s and @xmath18 are fixed point free permutations of order 2 on @xmath5 , the _ connection group _ @xmath19 is transitive on @xmath5 , and @xmath13 and @xmath12 commute . this @xmath20 is often called the _ monodromy group _ of the map . we can then think of vertices in @xmath2 as orbits of @xmath21 in @xmath20 . similarly , edges correspond to orbits of @xmath22 and faces to orbits of @xmath23 . we will use the word _ kite _ or _ corner _ for the area within a face where two consecutive edges meet . more formally , a kite is the union of two flags which are @xmath24-adjacent . if @xmath2 and @xmath25 are maps on surfaces @xmath26 and @xmath27 , a _ projection _ from @xmath25 to @xmath2 is a function @xmath28 mapping @xmath27 to @xmath26 which is locally a homeomorphism at all points of @xmath25 except perhaps at vertices and/or face - centers , and which sends faces to faces , edges to edges , and vertices to vertices . in combinatorial terms , if @xmath29)$ ] and @xmath30)$ ] , a projection of @xmath25 to @xmath2 is a function @xmath28 mapping @xmath31 to @xmath5 such that @xmath32 for all @xmath33 . we call such an @xmath25 a _ cover _ of @xmath2 . notice that if @xmath25 is a cover of @xmath2 , and the pre - image of some one flag in @xmath2 has size @xmath0 , then the projection @xmath28 is @xmath0-to-1 onto every flag of @xmath2 . we say then that @xmath25 is a _ @xmath0-fold _ cover of @xmath2 . we call a map _ orientable _ provided that the surface on which it is embedded is itself orientable . we check that a map is orientable by giving it a _ face orientation _ ; this is an assignment of a circular arrow to each face of @xmath2 such that at every edge , the arrows on the faces joined by the edge point along the edge in _ opposite _ directions , as in figure [ fig : faceori ] . we can define a _ vertex orientation _ similarly , and it is clear that @xmath2 has a vertex - orientation if and only if it has a face - orientation , and this happens if and only if @xmath2 is orientable . the word _ pseudo - orientation _ has been used in two different ways . we will use the term _ vertex pseudo - orientation _ ( vpso for short ) for what is called in @xcite simply a pseudo - orientation . here , we mean an assignment of one circular arrow to each vertex so that at each edge the two arrows cross in the _ same _ direction , as in figure [ fig : pso ] . if we visualize a gear wheel at each vertex so that cogs on the wheels of adjacent vertices mesh , the map is vpso if we can turn one wheel , causing all wheels to turn at the same time . in @xcite , the idea is used to make an important distinction about @xmath0-fold rotary covers of a rotary map @xmath2 for which the branching is totally ramified at vertices . if @xmath2 is non - orientable , @xmath0 can be larger than 2 only if @xmath2 is vertex pseudo - orientable . similarly , a _ face pseudo - orientation _ ( or fpso ) is an assignment of one circular arrow to each face , as in figure [ fig : pso ] , so that at each edge , the arrows in adjacent faces flow along the edge in the _ same _ direction . we can simplify this by orienting each edge so that the cycle of edges around each face is consistently oriented . in @xcite this is called a pseudo - orientation , and the result in that paper is that the _ dart graph _ of @xmath2 is connected if and only if @xmath2 is not face pseudo - orientable . finally , we may define an _ edge pseudo - orientation _ ( or epso ) to be an orientation of the edges such that in every face the direction of the arrows in adjacent edges flows in the _ same _ direction ( either into or out of the face ) as in figure [ fig : pso ] . in this paper , the word ` coloring ' will be used to describe what might be more fully notated as a ` consistent flag 2-coloring ' . if @xmath34 is any subset of @xmath35 , then an @xmath34-coloring of a map @xmath2 is a function @xmath36 such that for every flag @xmath7 , if @xmath37 , then @xmath38 , while if @xmath39 , then @xmath40 . to say that in another way , @xmath41 figure [ fig : cubecols ] shows a @xmath42-coloring , a @xmath43-coloring , and an @xmath44-coloring of the cube . the reader should check that the colorings extend unambiguously to the unseen faces of the cube as well . let us first notice that if @xmath45 is an @xmath34-coloring of @xmath2 , then so is @xmath46 , and these are , in fact , the only @xmath34-colorings of the map . give the name _ sub - edge _ to the segments forming sides of the flags . these are of three types : type 0 : : : midpoint - of - edge to face - center , type 1 : : : face - center to vertex , type 2 : : : vertex to midpoint - of - edge . each flag @xmath7 is adjacent to flag @xmath47 along their mutual sub - edges of type @xmath48 . if @xmath2 is @xmath34-colorable , therefore , the sub - edges bounding each color are the sub - edges of types in @xmath34 . note that when considering maps in class @xmath49 as defined in @xcite , the subset @xmath34 is used to indicate that @xmath48-adjacent flags belong to the _ same _ orbit this means that maps in class @xmath49 admit an @xmath50-coloring . in the context of colorings it turns out to be more practical to define @xmath34-coloring requiring @xmath48-adjacent flags to be a different color rather than requiring @xmath48-adjacent flags to be the same color if and only if @xmath51 . let @xmath52 stand for the symmetric difference operation : @xmath53 . let @xmath54 be the power set of @xmath44 . it is well - known that @xmath54 is a group under @xmath52 and it is an elementary abelian group of order eight whose identity element is @xmath55 . [ delta ] if a map @xmath2 admits an @xmath34-coloring and a @xmath56-coloring , then it admits an @xmath57-coloring . proof : it is easy to verify that the sum of an @xmath34-coloring and a @xmath56-coloring is an @xmath57-coloring ' '' '' from this proposition we see that , for each map @xmath2 , the sets @xmath34 for which @xmath2 has an @xmath34-coloring form a set @xmath58 which is a subgroup of @xmath54 . whenever @xmath34 is a subset of size 1 , @xmath34-colorable maps have a nice characterization . [ vbip ] a map @xmath2 is vertex - bipartite if and only if @xmath2 has a @xmath42-coloring . proof : given such a coloring , all the flags incident with one vertex will be the same color , while those at an adjacent vertex will be the other color . this is a bipartite coloring of the vertices . conversely , given a bipartition of the vertices , color all flags incident with black vertices with color 1 and all those incident with white vertices with color 0 . this is then a @xmath42-coloring of @xmath2 . ' '' '' similar proofs lead us to these : [ fbip ] a map @xmath2 is face - bipartite if and only if @xmath2 has a @xmath59-coloring . ( _ i.e. _ a @xmath59-coloring corresponds to a bipartite coloring of the faces of @xmath2 . ) [ ebip ] a map @xmath2 is edge - bipartite if and only if @xmath2 has a @xmath60-coloring . ( _ i.e. _ a @xmath60-coloring corresponds to a bipartite coloring of the edges of @xmath2 . ) consider a _ @xmath61 of flags in a map @xmath2 . we abbreviate that @xmath62 , where @xmath63 is the string @xmath64 . write @xmath65 for @xmath66 . we call a cycle @xmath62 _ @xmath34-consistent _ if the number of occurrences in @xmath63 of indices which are in @xmath34 is even . if @xmath62 is @xmath34-consistent , this says that , at least along the cycle , we can color flags from two colors so that flags which are @xmath48-connected for @xmath67 are the same color and those for which @xmath51 are not . a cycle which is not @xmath34-consistent is _ @xmath34-inconsistent_. it follows , then that @xmath2 is @xmath34-colorable if and only if every cycle in @xmath2 is @xmath34-consistent . consider this quite general fact : for any function @xmath68 from a set @xmath69 to @xmath70 , extend @xmath68 to the power set of @xmath69 by @xmath71 for each subset @xmath72 of @xmath69 . then for any subsets @xmath73 of @xmath69 , it is clear that @xmath74 . [ lemma13 ] for any subsets @xmath75 of @xmath44 , if @xmath76 , then every cycle in @xmath77 is consistent either for exactly one of @xmath78 or for all three . proof : fix a cycle @xmath62 of @xmath2 , and define @xmath68 on @xmath44 with @xmath79 being the parity of the number of occurrences in @xmath63 of @xmath17 ; furthermore , consider @xmath68 extended to @xmath54 . the cycle is @xmath34-consistent , then , provided that @xmath80 . now consider @xmath81 . these are elements of @xmath70 , and so we can rephrase that in this form : @xmath82 . the number of zeros in @xmath83 must be 1 or 3 ; the conclusion follows directly . ' '' '' we will have use for this lemma in section [ sec : doubles ] . a natural operator in topology is the _ connected sum _ of two surfaces , which is defined to be the surface formed by cutting a small disk from each of the two surfaces and attaching them along the newly created border . we extend this definition to maps by causing the attachment sets to be the boundaries of two suitably chosen faces . formally , let @xmath29)$ ] and @xmath30)$ ] be maps and suppose @xmath84 and @xmath85 are flags of these maps for which their respective faces @xmath86 and @xmath87 have the same number @xmath0 of sides . furthermore , suppose that no flag of @xmath86 is @xmath12-connected to any of the flags of @xmath86 , and similarly for @xmath87 . then we define the _ connected sum _ @xmath88 with respect to @xmath89 to be the map whose flag set is @xmath90 , with connections @xmath91 , where @xmath92 and @xmath93 are the restrictions of @xmath94to this set , and @xmath95 is the same except that , for each @xmath96 is @xmath95-connected to @xmath97 and @xmath98 is @xmath95-connected to @xmath99 . because of the restriction on the faces , these flags are both in the new flag set , and it is easy to check that @xmath100 . the following figure illustrates this construction : [ th : oplus ] suppose @xmath2 and @xmath25 are @xmath34-colorable maps for some @xmath101 and suppose @xmath88 exists for some flags @xmath84 and @xmath85 . then @xmath88 is @xmath34-colorable . proof : choose an @xmath34-coloring for @xmath2 and one for @xmath25 in such a way that @xmath84 and @xmath85 have opposite colors if @xmath102 and the same color if @xmath103 . then the induced coloring provides the desired coloring on @xmath104 . ' '' '' as we ll see in the following theorem , it is almost always the case that the group @xmath105 is equal to @xmath106 . this can fail , roughly , when for some @xmath107 , @xmath108 is @xmath34-colorable but @xmath2 is not , or similarly for @xmath25 . [ t : plus ] suppose @xmath2 and @xmath25 are maps and suppose @xmath88 exists for some flags @xmath84 and @xmath85 . furthermore , suppose that for each @xmath109 there exists an @xmath34-inconsistent cycle that does nt include any of the flags of face @xmath86 , and similarly for @xmath25 . then @xmath110 . proof : for @xmath107 , if @xmath111 , then @xmath112 by proposition [ th : oplus ] . if @xmath113 , then without loss of generality , assume that @xmath2 is not @xmath34-colorable . then @xmath2 has cycles which are @xmath34-inconsistent , and by the hypothesis , at least one of them includes none of the flags in @xmath86 . this cycle , then , occurs in @xmath88 , and is still @xmath34-inconsistent . thus @xmath114 . ' '' '' we now offer and prove a series of propositions connecting the ideas of orientations and colorings . first , we show a collection of easy facts to embody our belief that colorings generalize orientations . we then generalize these facts in theorem [ thm : super ] . [ oble012 ] a map is orientable if and only if it has an @xmath44-coloring . proof : if @xmath2 has an @xmath44-coloring , orient each face so that , along each edge , the orientation points from the flag with color 0 to the flag with color 1 . this is consistent within the face and faces that meet along an edge have orientations which meet correctly . conversely , given the orientation , assign colors so that along each edge within a face , the orientation points from the flag with color 0 to the flag with color 1 . this causes colors to alternate within each face . because the orientation opposes the orientation in each adjacent face , the colors alternate at each edge , giving an @xmath44-coloring . ' '' '' similar arguments give us the following two propositions : [ th : fpso01 ] a map @xmath2 has a face pseudo - orientation if and only if @xmath2 admits a @xmath115-coloring . [ vpso12 ] a map @xmath2 has a vertex pseudo - orientation if and only if @xmath2 admits a @xmath43-coloring . our aim in this subsection is to generalize the definition of pseudo - orientability and the previous results . for any subset @xmath34 of @xmath44 ( except @xmath55 ) , we form the map @xmath116 whose faces are the regions formed by conjoining flags which are connected by each @xmath17 for which @xmath67 . for example , if @xmath2 is the cube , then @xmath117 and @xmath118 are shown below . and @xmath118 ] notice that the boundaries of the faces in @xmath120 are the sub - edges of types that are _ in _ @xmath34 . thus @xmath2 is @xmath34-colorable if and only if @xmath120 is face - bipartite . however , the map @xmath120 can be constructed even if @xmath2 is not @xmath34-colorable . we call @xmath2 _ @xmath34-pseudo - orientable _ provided that @xmath121 is face pseudo - orientable . if this happens , we can orient the edges of @xmath69 so that around each face of @xmath69 , the arrows all point the same way . we say a map is @xmath55-pseudo - orientable when it is orientable . it is clear that @xmath122 and so @xmath59-pso is equivalent to fpso . observation [ th : fpso01 ] shows that this happens exactly when @xmath2 is @xmath115-colorable . similarly , @xmath2 is @xmath42-pso when it is vpso , and this happens exactly when @xmath2 is @xmath43-colorable . surprisingly , this generalizes completely : [ thm : super ] let @xmath2 be a map . then @xmath2 is @xmath34-pseudo - orientable if and only if @xmath2 is @xmath123-colorable . proof : first , recall that the sub - edges of @xmath2 are of three different types : type 0 : : : midpoint - of - edge to face - center , type 1 : : : face - center to vertex , type 2 : : : vertex to midpoint - of - edge . furthermore , let us think of the sub - edges as directed . call the directions above the ` standard ' orientations for the sub - edges , and the reverse of these are the non - standard orientations . in order to form @xmath121 , we delete or ignore sub - edges of each type @xmath124 , so we only consider sub - edges of type in @xmath34 . suppose @xmath2 is already @xmath123-colored . then in @xmath120 , both flags around each sub - edge have the same color . then we may assign this as the color of the sub - edge . now give every sub - edge with color 0 the standard orientation , and each sub - edge with color 1 the non - standard direction . if @xmath34 has 2 elements , then consecutive sub - edges around a face of @xmath69 of different types are the same color , while those of the same type are opposite colors . this causes the directions to be consistent about a face . on the other hand , if @xmath34 has only one element , then the sub - edges around a face of @xmath69 alternate in color and are all of the same type ; again , this forces the orientation to be consistent . the following examples show how this works : 1 . if @xmath125 , then by definition the map is @xmath34-pseudo - orientable if and only if it is orientable , and this happens if and only if it is @xmath44-colorable . 2 . if @xmath126 , then @xmath127 , which is face pseudo - orientable if and only if it is @xmath115-colorable . in this case , we orient black type 2 sub - edges from vertex to midpoint - of - edge but white type 2 sub - edges from midpoint - of - edge to vertex . if @xmath128 , we orient black type 1 sub - edges from vertex to face - center and black type 2 sub - edges from midpoint - of - edge to vertex . we orient the opposite colors in the opposite direction : + 4 . if @xmath129 , the @xmath55-coloration colors all sub - edges the same and so the faces of @xmath69 , the flags of @xmath2 , are all oriented consistently . every map is @xmath55-colorable and every map is @xmath44-pso . + for the converse we may use the orientation to color in the way prescribed to match the previous list . ' '' '' we summarize here the results of previous sections relating to the question of which maps have @xmath34-colorings for a given @xmath34 . [ cols=">,<,^",options="header " , ] proof : the medial of any map always admits a @xmath59-coloring ( face - bipartition ) , since we have two types of faces : those that surround vertices of @xmath2 , and those that surround face - centers of @xmath2 . for the first row , if the map is edge - bipartite , then we may color each vertex of the medial with the color corresponding to the color of the edge it was the midpoint of and vice - versa . for the second , note that an edge of the medial corresponds to a kite of the original . finally , the medial maps lies on the same surface as the original . ' '' '' given a map @xmath29)$ ] and a subset @xmath34 of @xmath35 , we define the _ @xmath34-double _ of @xmath2 , denoted @xmath130 , to be the map @xmath131)$ ] , where @xmath132 here , we write `` @xmath133 '' for @xmath134 . notice that the function sending @xmath133 to @xmath7 is a projection of @xmath25 onto @xmath2 , and so @xmath130 is a covering of @xmath2 . also notice that the function which sends @xmath133 to @xmath135 , is an @xmath34-coloring of @xmath130 . now , if @xmath2 has an @xmath34-coloring , then @xmath130 consists of two disjoint copies of @xmath2 ; in this case we discard one copy and say that @xmath130 is isomorphic to @xmath2 . if @xmath2 does not have an @xmath34-coloring then @xmath130 truly is a double cover ( i.e. , a 2-fold topological covering which may or may not be branched at vertices and/or face - centers ) of @xmath2 . in this case , for every cycle @xmath62 in @xmath2 , if it is @xmath34-consistent , then @xmath136 , and @xmath137 , and so the cycle is covered by two cycles of the same length . and if @xmath62 is @xmath34-inconsistent , then @xmath138 , and @xmath139 . then the cycle @xmath62 is covered by a single cycle of twice the length of @xmath62 . in that case the covering cycle is @xmath140 , which is consistent for any subset of @xmath44 . we have remarked above that the map @xmath130 always has an @xmath34-coloring . moreover , we claim , it is universal minimal in the sense of the following proposition : [ th : mindub ] if @xmath25 is any map that is a covering of @xmath2 and which is @xmath34-colorable , then @xmath25 must itself cover @xmath130 . proof : let @xmath45 be an @xmath34-coloring of @xmath25 and let @xmath28 be a projection of @xmath25 onto @xmath2 . define @xmath141 mapping @xmath25 onto @xmath130 to send the flag @xmath7 of @xmath25 to the flag @xmath142 of @xmath130 . because @xmath45 is an i - coloring of @xmath25 , this @xmath141 is a projection , as required . ' '' '' an example of a double covering is @xmath143 when @xmath2 is a non - orientable map . because @xmath2 does not have an @xmath44-coloring , @xmath25 is truly a double cover of it , and since @xmath25 then _ does _ have an @xmath44-coloring , it must be orientable and so @xmath25 is the orientable twofold cover ( some say the _ canonical _ cover ) of @xmath2 . another example of the double construction in the literature comes from sherk s 1962 paper @xcite . at that time few non - toroidal chiral maps were known . sherk constructed an infinite family of such maps on an infinite number of different surfaces , using a simple and clever technique . starting with a chiral map @xmath2 on the torus , with triangles as faces and six of them incident to each vertex , he made a new map @xmath25 in the following way . for each vertex @xmath144 of @xmath2 , he created vertices @xmath145 ; for each edge @xmath146 , he made edges @xmath147 and @xmath148 ; and each triangular face @xmath149 $ ] became a hexagonal face @xmath150 $ ] . this map has hexagonal faces , still meeting 6 at a vertex . thus if @xmath2 had @xmath151 vertices , @xmath152 faces and @xmath153 edges , then @xmath25 has @xmath152 vertices , @xmath152 faces and @xmath154 edges . it follows that @xmath25 must lie on the orientable surface of characteristic @xmath155 , i.e. , of genus @xmath156 . from our point of view , we can construct @xmath25 as @xmath157 . as a final example , the reader might like to verify that if @xmath158 is the tetrahedron then @xmath159 is the orientable map with hexagonal faces , six of them around each vertex , shown in figure [ fig:1yt ] . , height=170 ] we present here some of the facts about @xmath34-doubles : [ th : jup ] if @xmath2 has a @xmath34-coloring then so does any @xmath160 . proof : in fact , by definition _ any _ covering of @xmath2 inherits an @xmath34-coloring from that of @xmath2 . ' '' '' [ t : idj ] if @xmath2 has a @xmath56-coloring then any @xmath130 is isomorphic to @xmath161 . proof : for any map @xmath2 , @xmath130 is the smallest covering of @xmath2 which admits an @xmath34-coloring . if @xmath2 has a @xmath56-coloring , then @xmath130 has a @xmath56-coloring as well . thus , by proposition [ delta ] , @xmath130 admits an @xmath162-coloring , and whether @xmath2 has an @xmath34-coloring or not , this covering of @xmath2 must be minimal of those admitting an @xmath162-coloring . thus @xmath163 . ' '' '' the following two results summarize the examples given above . if @xmath2 is non - orientable , then @xmath164 is isomorphic to the orientable double covering of @xmath2 . we say that a map @xmath25 is the _ sherk double cover _ of a non - vertex - bipartite map @xmath2 whenever @xmath25 can be obtained from @xmath2 by the above procedure used by sherk to obtain the chiral maps with hexagonal faces . if @xmath2 is not vertex - bipartite , then @xmath157 is isomorphic to its sherk double cover . note that @xmath42 , @xmath60 , @xmath59 generate @xmath54 under @xmath52 , and so if @xmath2 is any map , then @xmath165 has @xmath166 , and so @xmath25 is an un - double - able map in our context . [ th : dubgp ] for any map @xmath2 and any subset @xmath34 of @xmath44 , @xmath167 proof : this is obvious if @xmath168 . if not , it is clear that @xmath169 is contained in @xmath170 , and so we need only to show that the opposite containment holds . this is equivalent to showing that if @xmath171 and @xmath172 , then @xmath173 ( see proposition [ t : idj ] ) . let @xmath174 and note that , if @xmath175 , then @xmath25 is a @xmath56-colorable double cover of @xmath2 ; by proposition [ th : mindub ] , @xmath25 must be isomorphic to @xmath176 . because @xmath172 , there is some cycle @xmath62 which is @xmath56-inconsistent . if any such cycle is @xmath34-consistent , then in @xmath25 , the cycle @xmath177 has the same length as @xmath62 , and so is also @xmath56-inconsistent , contradicting @xmath175 . thus every cycle which is @xmath56-inconsistent is also @xmath34-inconsistent . a similar argument in @xmath178 shows that every cycle in @xmath2 which is @xmath34-inconsistent is also @xmath56-inconsistent . therefore every cycle is either consistent for both , or inconsistent for both . by lemma [ lemma13 ] , every cycle in @xmath2 is @xmath76-consistent , and then @xmath179 , as required . ' '' '' we want to generalize the result from @xcite which says , loosely speaking , that a reflexible map is a 2-fold orientable cover of a non - orientable reflexible map if and only if its rotation group contains an involutory element which conjugates each of the generators to its inverse . to re - phrase that in our context , a reflexible map @xmath25 which has an @xmath44-coloring is an @xmath44-double of some reflexible map @xmath2 which does not have such a coloring if and only if the subgroup of @xmath180 which preserves colors in the coloring contains an involutory element which conjugates each of the generators ( of the color - preserving group ) to its inverse . if @xmath25 is @xmath34-colorable , let @xmath45 be either of the @xmath34-colorings of @xmath25 and define @xmath181 to be the subgroup of @xmath182 consisting of all @xmath183 such that @xmath184 for all @xmath185 . because @xmath25 is @xmath34-colorable , this group has index 2 , and so is normal , in @xmath182 . note that theorem [ t : isdouble ] generalizes the result from @xcite not only to all @xmath34-colorings , but also to maps with no assumption on their automorphism group . [ t : isdouble ] a map @xmath186)$ ] is @xmath130 for some non-@xmath34-colorable map @xmath187)$ ] if and only if @xmath25 is @xmath34-colorable and there exists @xmath51 and @xmath188 of order 2 such that for every element @xmath189 , @xmath190 . * proof * : first , assume @xmath174 with @xmath2 a non-@xmath34-colorable map . we will prove the existence of @xmath183 . the group @xmath191 has index 2 in @xmath182 , and there is an isomorphism @xmath192 which is the letter - to - letter projection , @xmath193 whenever the word @xmath194 is in @xmath191 and so has an even number of indices in @xmath34 . because @xmath77 is not @xmath34-colorable , @xmath34 is not empty . so consider any @xmath51 and let @xmath195 . note that @xmath183 is an involution since @xmath7 is an isomorphism . this @xmath183 is not @xmath17 , since @xmath196 ; it is , instead , expressible as some longer product of @xmath197 s . then for any @xmath198 we have @xmath199 , while @xmath200 , since @xmath201 is in @xmath202 , and @xmath7 acts there as letter - to - letter projection . since @xmath7 is an isomorphism and hence one - to - one , we conclude that @xmath203 . now assume we have @xmath188 of order 2 and @xmath204 such that for every element @xmath205 , @xmath190 . in particular , with @xmath206 , we have @xmath207 . from this we see that @xmath208 . we then construct @xmath2 by identifying flags in the following consistent way : let @xmath209 , so that @xmath210 is an involution . we shall identify flag @xmath185 with @xmath211 . because @xmath210 is an involution , this identification is unambiguous , and each flag is identified with exactly one other . since @xmath183 preserves coloring and @xmath17 changes it , we are identifying a white flag with a black flag . these equivalence classes are the flags of @xmath2 . then we have to define the connectors @xmath212 and @xmath213 of @xmath2 . notice that for every @xmath214 we have @xmath215 , so @xmath210 commutes with all of @xmath191 . but @xmath191 has index 2 , and @xmath216 so @xmath210 commutes with all of @xmath182 . then we define @xmath217)s_j = [ f r_j , f u r_j = f r_j u]$ ] . the @xmath218 s are well defined and we can easily see @xmath219 and @xmath213 commute , since @xmath13 commutes with @xmath12 . to see that @xmath25 is the @xmath34-double cover of @xmath2 , identify each black flag @xmath7 of @xmath25 with the flag @xmath220_1 $ ] of @xmath130 , and each white flag @xmath7 with the flag @xmath220_0 $ ] . ' '' '' in order to use this theorem we would , in theory , need to prove the existence of @xmath183 and observe that under conjugation , @xmath183 acts in the same way as some @xmath17 on _ every _ element of @xmath221 . but in practice , we only need to find @xmath183 that acts in such a way for the _ generators _ of @xmath221 . the element @xmath17 must not be in @xmath202 and so we will refer to it as an _ external generator_. table i below summarizes information about each subset @xmath34 of @xmath44 . the second column ( adapted from @xcite ) gives generators for @xmath221 as a subgroup of @xmath182 . the third column gives the @xmath222 which are not in @xmath202 , and the last column shows the actions on the generators of @xmath202 under conjugation by this external generator . the purpose of this section is to prove that every subgroup @xmath223 of @xmath54 occurs on every surface on which it _ can _ appear . to clarify that statement , we call a group which includes @xmath44 an _ orientable _ subgroup of @xmath54 , and subgroups which do not contain @xmath44 are _ non - orientable_. so we want @xmath223 to appear on @xmath224 when they are both orientable or both non - orientable . this is almost true , as explained by the following theorem : [ thm : everygroup ] for every subgroup @xmath223 of @xmath54 and every surface @xmath224 , there is a map @xmath2 on @xmath224 such that @xmath225 if and only if @xmath224 and @xmath223 are both orientable or both non - orientable , with the following exceptions : 1 . the group @xmath226 does not appear on the sphere . 2 . the group @xmath227 does not appear on the projective plane . 3 . the group @xmath228 does not appear on the projective plane . [ allgp ] the proof of this claim is contained in and scattered through the remaining parts of this section . we will consider the 16 possible groups subgroups of @xmath54 individually and in dual pairs . such a proof would appear to require 16 constructions , but we can use lemma [ di ] and proposition [ th : jup ] for some simplification . moreover , the insertion techniques that we will introduce and the connected sum construction of subsection [ sec : connectedsum ] allow even more simplification . we consider the non - orientable cases first ; then a simple argument , using the results of section [ sec : doubles ] , takes care of the orientable ones . consider any triangulation of a non - orientable surface @xmath224 . refine the triangulation slightly by placing one new vertex inside some triangular face and connect it to the three vertices on its boundary so as to have one vertex of degree 3 . call this map @xmath2 . because of the triangular faces and the the degree-3 vertex , the cheat sheet of section [ sec : colori ] shows us that none of the sets of size 1 or 2 are in the group , and because @xmath224 is non - orientable , it can not contain @xmath44 either . thus @xmath58 must be the trivial group . in this subsection , we introduce the technique of _ insertion_. we shall prove that any map @xmath2 can be easily modified to produce a map which is face or vertex bipartite and also face or vertex pseudo - orientable . let us assume , for example , that we wish to modify @xmath2 to construct a vertex - bipartite map . color the vertices of @xmath2 at random . for the edges for which both vertices are the same color , add a vertex ( of degree 2 ) to split the edge in two . the same idea can be applied if we wish vertex pseudo - orientability . we may assign orientations to the vertices at random and then split the edges whose endpoints do nt match into two edges . call this process _ vertex - insertion_. this can also be used to turn a map which _ is _ vertex - bipartite or vpso into one that is _ not_. note that vertex - insertion does not modify the status of either face - bipartiteness of fpso of @xmath2 . dually , we may replace an edge by a face bounded by two edges to get either face - bipartition or fpso . the following picture illustrates this for face pseudo - orientation : we begin with an arbitrary map on a surface @xmath224 , and an arbitrary orientation of each face . if there is an edge , as in figure [ fig : otopso]a , where the assigned arrows flow in opposite directions along the edge , replace the edge with a face bounded by a pair of edges as shown in [ fig : otopso]b . orient the resulting lune with an arrow as shown . this gives a map on the same surface with one fewer edge for which the given arrows fail to give a face pseudo - orientation . continuing this process eventually gives us a map on @xmath224 admitting a face pseudo - orientation . let s call this process _ face - insertion_. here , we used it to produce a fpso map on any surface . we can use it equally well to turn any map into a face - bipartite map on the same surface . as with vertex insertion , this same process can also be used to convert a map which _ is _ face bipartite or fpso to one which is _ not_. for many groups we can use these two operations to produce maps with that group on any non - orientable surface . we will show one of these in detail and summarize the rest . consider the non - orientable group @xmath229 . for any non - orientable surface @xmath224 , we choose any map on @xmath224 , and , using face - insertion as above , construct a map which is fpso . now we must be certain that the group of the map is no bigger than @xmath223 . the other cosets of @xmath223 in @xmath54 are @xmath230 . by inserting one vertex on one edge , if necessary , we can cause the map to have a face with odd degree , eliminating @xmath42 and @xmath43 and hence all @xmath231 from the group . the non - orientability of the surface eliminates @xmath35 and hence all of @xmath232 . thus the final form @xmath2 of the map has @xmath233 , as required . notice that the group @xmath234 is then @xmath235 . we produce maps having most of the other non - orientable groups in a similar way . starting with an arbitrary map on a non - orientable surface we perform insertions in the following ways : * use face - insertion to force fpso and vertex - insertion to force vpso . the result is a map whose group is @xmath236 . * use face - insertion to force fpso and then vertex - insertion to force it to be vertex - bipartite . the resulting map has group @xmath237 . its dual has group @xmath238 . * use face - insertion to force the map to be face - bipartite and vertex - insertion to force it to be vertex - bipartite . the resulting map has group @xmath239 . * use face - insertion to force face - bipartiteness and vertex - insertion in order to force some face to have an odd number of sides . the group must be just @xmath240 . its dual then has group @xmath241 . these constructions take care of all non - orientable groups except for two : @xmath242 and @xmath243 . in order to cope with the two non - orientable exceptional groups , we apply the technique of connected sums of maps , developed in subsection [ sec : connectedsum ] , as well as by giving constructions of parameterized families of suitable maps in order to have explicit constructions . for each group @xmath244 with @xmath245 and @xmath246 , there exists a map @xmath2 on every non - orientable surface such that all the following hold : * @xmath225 . * there is a face @xmath68 with exactly 2 sides and exactly 2 vertices . * @xmath247 . that is , the flags of @xmath2 that are not in @xmath68 are _ not _ @xmath34-colorable for every @xmath248 . proof : let @xmath224 be an arbitrary non - orientable surface . to construct a map on @xmath224 with the desired properties , first consider any map for which no face shares a vertex with itself . because the surface is non - orientable , we may choose an arbitrarily fine subdivision for which there is a mbius band of faces which does not use all the faces or edges . then apply the operations described in subsection [ insert ] to make the map @xmath249-colorable for each @xmath249 in @xmath223 . this is possible since any group @xmath223 with 4 elements with @xmath245 must contain two of the following : vertex - bipartite , face - bipartite , fpso or vpso , with the other non trivial element being the symmetric difference of the other two . now consider an edge that is not in any of the faces of the chosen mbius band and convert it into three edges , as with face - insertion but with three edges instead of two . this operation does _ not _ change the status of vertex - bipartition , face - bipartition , vpso or fpso as required . now pick either of the two newly created faces and call it @xmath68 . since we still have a mbius strip , @xmath250 , but every @xmath251 is in @xmath252 . therefore , @xmath247 . ' '' '' for @xmath253 and @xmath254 we have that for every surface ( except the projective plane ) there exists a map @xmath2 on that surface with @xmath255 . proof : let @xmath224 be a surface with non - orientable genus @xmath256 . consider maps as in the previous lemma @xmath257 and @xmath258 such that @xmath257 lies on a surface with non - orientable genus @xmath259 and @xmath258 lies on the projective plane and such that : * if @xmath254 , then @xmath260 * if @xmath261 , then @xmath262 then @xmath263 ( along the 2-sided faces as in the previous lemma ) lives on the non - orientable surface with genus @xmath0 and @xmath264 ( see theorem [ t : plus ] ) . ' '' '' we give explicit constructions for these last two groups : let @xmath265 , and consider the map @xmath266 where @xmath267 . this map has @xmath268 faces , each one a square . they are arranged in an @xmath269 rectangle . the top @xmath270 edges are identified with the bottom @xmath270 directly and orientably . each vertical edge is identified with the one diametrically across from it ; the first @xmath0 orientably , the rest non - orientably . figure [ fig : g573 ] shows @xmath271 . , height=226 ] because vertical and horizontal edges alternate around every face and every vertex , this map is always edge - bipartite , i.e.,@xmath60-colorable , and lies on the non - orientable surface of genus @xmath272 . ( @xmath273 , then , is on the klein bottle ) if @xmath274 is odd and @xmath275 , then the map is neither face - bipartite nor vertex - bipartite , and so its group is @xmath227 . for @xmath276 consider the maps shown in figure [ fig : one ] : in general we have a rectangle one unit wide , having any height . the maps shown in figure [ fig : one ] have characteristic 0 and 1 respectively . in the left picture , by making the strip long enough , we may switch enough pairs of adjacent labels on the right side ( for example , switch labels 3 and 4 on the right ) to make any even genus non - orientable surface . similarly , for the picture on the right we may switch enough pairs of adjacent labels in a sufficiently long strip to construct any odd genus non - orientable surface . to see that the group is indeed @xmath277 , we see that neither map is face - bipartite , since some faces are glued to themselves , or face pseudo - orientable , since both contain faces that are glued to themselves in an orientable way . similarly , neither are vertex - bipartite or vertex pseudo - orientable . this concludes the proof for the non - orientable case . if the subgroup @xmath223 contains @xmath44 , then it is generated by @xmath44 and some subgroup @xmath278 not containing @xmath44 . each orientable surface @xmath224 is the orientable twofold cover of some non - orientable surface @xmath279 . by the preceding subsections , @xmath279 has a map @xmath280 whose group is @xmath278 ( unless s is the sphere , @xmath279 is the projective plane and @xmath223 is @xmath281 ) . then @xmath282 is a map on @xmath224 whose group is @xmath223 . by proposition [ th : dubgp ] , @xmath58 is exactly @xmath223 . thus , with the one exception , for each subgroup @xmath223 containing @xmath44 , every orientable surface admits a map whose group is @xmath223 . ' '' '' all that is left to prove now is that the three exceptions mentioned in the theorem are in fact exceptions namely that there is no map on the specified surfaces with the specified group . we shall prove this first for the group @xmath226 on the sphere . [ l : nosphere ] there is no map on the sphere with group @xmath226 . proof : suppose @xmath2 is one such map . since @xmath283 , the underlying graph of @xmath2 is not bipartite , which means it has an odd cycle @xmath284 . by jordan s closed curve theorem , this divides the sphere into two connected components . choose either and erase every edge and vertex exterior to it in order to form the deleted part into a single face whose boundary is @xmath284 . this gives a new map @xmath280 , also on the sphere . since @xmath285 , every face of @xmath2 must have an even number of sides . therefore @xmath280 has exactly one face with an odd number of sides . then the underlying graph of @xmath286 has exactly one vertex of odd degree , a well - known impossibility . ' '' '' no map @xmath2 on the projective plane has @xmath287 or @xmath288 . proof : if such a map were to exist , its orientable double cover would violate lemma [ l : nosphere ] . ' '' '' the proof of theorem [ allgp ] is now complete . no map on a non - orientable surface has an @xmath44-coloring . in all other cases , for any subset @xmath34 of @xmath44 , and any surface @xmath224 , there is a map @xmath2 on @xmath224 which is @xmath34-colorable . to generalize the results of this paper to higher - dimensional objects , we use the idea of a _ maniplex _ , first introduced in @xcite . we summarize the definitions here : an _ @xmath274-maniplex _ is a pair @xmath289)$ ] , where @xmath5 is a set of things called _ flags _ , and each @xmath17 is a fixed - point - free involution on @xmath5 such that @xmath290 is transitive on @xmath5 , and for every @xmath48 , @xmath291 such that @xmath292 , the permutations @xmath17 and @xmath197 commute and are disjoint . maniplexes clearly generalize maps and slightly generalize ( the flag - graphs of ) abstract polytopes . in particular , every map is a 2-maniplex and every 2-maniplex is a map . the language of maniplexes is the language of polytopes : an @xmath48-face is an orbit under the group @xmath293 , generated by all of the @xmath197 s except @xmath17 . a 0-face is a vertex , a 1-face is an edge , an @xmath274-face is a _ facet_. the definitions of _ coloring , projection , cover _ generalize easily . there is no notion of surface or manifold which applies to maniplexes ; nevertheless , we describe a maniplex which has an @xmath294-coloring as being orientable . in view of theorem [ thm : super ] , one may define a maniplex to be @xmath34-pseudo - orientable if and only if it is @xmath295-colorable . we may give a definition of fpso that is more in the spirit of our intuition : first , we need each facet to be orientable and second , orientations can be assigned to the facets so that they agree on every subfacet . it is easy to see that this is equivalent to an @xmath296-coloring of @xmath2 . the idea of an @xmath34-coloring , for any @xmath107 , comes forward easily , as does the fact that the set of @xmath34 s for which a given maniplex @xmath2 is @xmath34-colorable forms a subgroup @xmath58 of the power set @xmath54 of @xmath44 under @xmath52 . we can define @xmath48-face bipartiteness : an @xmath48-face is a connected component under all of the @xmath197 s except @xmath17 . make these the vertices of a graph , and join two of them by an edge is some flag of one is @xmath17 adjacent to some flag of the other . if this graph is bipartite , we say that @xmath2 is @xmath48-face bipartite . then , it is easy to show then that @xmath2 is @xmath48-face bipartite if an only if it is @xmath297-colorable . we can define connected sum of two @xmath274-maniplexes , removing isomorphic facets , one from each maniplex and adjusting the @xmath298 connections to form the sum . the theorem that the sum has any coloring common to both still holds . the operators generalize with a little care . we form @xmath299 from @xmath2 by replacing @xmath12 with the product @xmath300 . we form @xmath301 from @xmath2 by reversing the order of the generators . and then @xmath302 . similar facts hold about how colorings of @xmath2 are inherited by its direct derivates . the medial operation is more difficult to generalize . the construction of @xmath34-doubles is straightforward to generalize , and all of the facts about the doubles do as well . the fact that maps have rank 3 plays no role in any of these proofs . again , because there is no notion equivalent to that of ` surface ' , there is no natural way to generalize the results of section [ sec : everygroup ] . the constructions in section [ sec : everygroup ] lead to quite general maps , although some of them may be somewhat degenerate in the viewpoint of other previous work . it may be the case that more exceptions in theorem [ allgp ] are needed if we ask the extra requirement that the maps are polyhedral maps ( that is , the intersection of the closure of two distinct faces is either empty , a single vertex , or a single edge . see @xcite ) . however , this is beyond the scope of this paper . in previous sections we considered colorings of the flags of maps insisting that we use only two colors . in doing so , we ensure that if a given flag is @xmath48-adjacent to another flag with the same color then all flags are the same color as their @xmath48-adjacent flags . consistent colorings with @xmath0 colors can be defined following the idea of @xcite and @xcite , that is , each flag is assigned a color in @xmath303 with the restriction that for every @xmath304 and @xmath305 , the color of the @xmath48-adjacent flag to a flag @xmath7 colored @xmath291 depends on @xmath291 but not on @xmath7 . when @xmath0-coloring a map consistently for @xmath306 then it may not be true that a flag is @xmath48-adjacent to another flag with the same color if and only if each flag is @xmath48-adjacent to another flag with the same color . for example , we can color the flags of the pentagons of a pentagonal prism with color @xmath307 , the flags of the edges shared by two squares with color @xmath308 , and the remaining flags with color @xmath309 , as in figure [ fig:3orb ] . the reader can easily verify that this is a consistent 3-coloring , however , flags colored @xmath307 are @xmath307-adjacent to flags colored @xmath307 , and flags colored @xmath308 are @xmath307-adjacent to flags colored @xmath309 . it is unlikely that the concept of pseudo - orientability admits a generalization that preserves the connections between 2-colorings and pseudo - orientations of maps . however , @xmath0-colorings may still keep an interesting relation with @xmath0-fold covers of maps and maniplexes as well . * acknowledgements : * the authors , particularly the last , would like to thank the organizers of the workshop on abstract polytopes ( this set of organizers overlaps the set of authors ) for bringing us together . without that meeting , this research might not have happened . the first author was partially supported by agencija za raziskovalno dejavnost republike slovenije under the research project mu - prom/11 - 010 . the second and fourth authors were partially supported by papiit mexico under grant in112512 . the third author was supported supported by dgapa postdoctoral scholarship and additionally by papiit grant ib100312 .
meson - antimeson mixing has traditionally been of importance because it is sensitive to heavy degrees of freedom that propagate in the underlying mixing amplitudes . estimates of the charm quark and top quark mass scales were inferred from the observation of mixing in the @xmath4 and @xmath5 systems , respectively , before these particles were discovered directly . this success has motivated attempts to indirectly detect new physics ( np ) signals by comparing the observed meson mixing with predictions of the standard model ( sm ) . mixing in the kaon sector has historically placed stringent constraints on the parameter space of theories beyond the sm and provides an essential hurdle that must be passed in the construction of models with np . however , anticipated breakthroughs from the b - factories and the tevatron collider have not been borne out the large mixing signal in the @xmath5 and @xmath6 systems is successfully described in terms of the sm alone ( although the parameter spaces of various np models have become increasingly constrained ) . short of awaiting lhcb and the construction of a super - b facility , there is one remaining example for possibly observing indirect signs of np in meson mixing , the @xmath1 flavor oscillations . in this case , the sm mixing rate is sufficiently small that the np component might be able to compete @xcite . there has been a flurry of recent experimental activity regarding the detection of @xmath1-@xmath7 mixing @xcite , which marks the first time flavor changing neutral currents ( fcnc ) have been observed in the charged @xmath8 quark sector . with the potential window to discern large np effects in the charm sector @xcite and the anticipated improved accuracy for future mixing measurements , the motivation for a comprehensive up - to - date theoretical analysis of new physics contributions to @xmath3 meson mixing is compelling . the heightened interest in @xmath1-@xmath7 mixing started with the almost simultaneous observations by the babar @xcite and belle @xcite collaborations of nonzero mixing signals at about the per cent level , , @xmath9 , @xmath10 and @xmath11 are standard and are given in eqs . ( [ xy]),([y - defs ] ) . ] & & y_d = ( 0.97 0.44 0.31 ) 10 ^ -2 , + & & y_d^(cp ) = ( 1.31 0.32 0.25)10 ^ -2 . this was soon followed by the announcement by the belle collaboration of mixing measurements from the dalitz plot analyses of @xmath12 @xcite , x_d = ( 0.80 0.29 0.17)10 ^ -2 , y_d = ( 0.33 0.24 0.15)10 ^ -2 . a preliminary fit to the current database , @xmath13 . these are essentially unchanged from the hfag preliminary results given above and used in our analysis ; the difference will not affect our numerical results . ] by the heavy flavor averaging group ( hfag ) gives @xcite [ hfag ] & & x_d = 8.7^+3.0_-3.4 10 ^ -3 , y_d = ( 6.62.1 ) 10 ^ -3 . since this paper addresses the issue of the mass splitting induced by mixing , our primary concern is with the signal for @xmath14 , seen here to be a 2.4 sigma effect . this is below the generally accepted threshold for `` evidence '' and is more in the nature of a `` hint '' . however , we note that a 2.4 sigma effect will automatically have a non - zero lower bound at @xmath15 confidence - level . for the sake of reference , we cite the one - sigma window for the hfag value of @xmath14 , 5.4 10 ^ -3 < x_d < 11.7 10 ^ -3 , [ xbnds ] or equivalently for @xmath16 itself , 8.7 10 ^ -15 < m_d < 1.9 10 ^ -14 . [ mbnds ] let us briefly describe our strategy for dealing with the above hfag values in light of both sm and np contributions . we shall argue in sect . iii that the sm predictions , although indeed compatible with the observed range of values for the @xmath3 mixing parameters , contain significant hadronic uncertainties . moreover , we do not know the relative _ phase _ between the sm contribution and that from any np model , so that @xmath14 will lie between the extreme limiting cases of constructive and destructive interference . in addition , since the observation of @xmath3 mixing is new , the measurements will fluctuate with future refinements in the analyses and as more data is collected . to best deal with these realities , we will present our results by displaying a given np prediction as a pure np signal ( _ i.e. _ as if there were no sm component ) and for comparison , display curves of constant @xmath14 for the five values x_d = 15.0 10 ^ -3 , 11.7 10 ^ -3 , 8.0 10 ^ -3 , 5.0 10 ^ -3 , 3.0 10 ^ -3 . [ xrange ] this ( approximately hfag 2@xmath17 ) range reveals the sensitivity of @xmath14 to variations in the underlying np parameter space . we will then show the present constraints placed on the np model parameter space , by assuming that the np contribution can not exceed the @xmath18 upper bound on @xmath19 this procedure mirrors that which is traditionally employed in obtaining bounds on np from @xmath4-@xmath20 mixing . @xmath1-@xmath7 mixing at the observed level is much larger than the quark - level ( ` short - distance ' ) sm prediction @xcite but is in qualitative accord with hadron - level ( ` long - distance ' ) sm expectations . however , because the latter are beset with hadronic uncertainties , it can not be rigorously concluded that only sm physics is being detected . in this paper , we will consider a broad menu of np possibilities . as the operation of the lhc looms near , the number of potentially viable np models has never been greater . our organizational approach to analyzing these is to address np models with : 1 . extra fermions ( sect . iv ) a : fourth generation b : heavy vector - like quarks ( 1 ) : @xmath21 singlet quarks ( 2 ) : @xmath22 singlet quarks c : little higgs models 2 . extra gauge bosons ( sect . v ) a : generic z models b : family symmetries c : left - right symmetric model d : alternate left - right models from e@xmath23 theories e : vector leptoquark bosons 3 . extra scalars ( sect . vi ) a : flavor conserving two - higgs - doublet models b : flavor changing neutral higgs models c : scalar leptoquark bosons d : higgsless models 4 . extra space dimensions ( sect . vii ) a : universal extra dimensions b : split fermion models c : warped geometries 5 . extra symmetries ( sect . viii ) a : minimal supersymmetric standard model b : quark - squark alignment models c : supersymmetry with r - parity violation d : split supersymmetry in the above , we have chosen to consider only supersymmetry in sect . viii due to its extensive literature and to cover other extended symmetries elsewhere in the paper . any np degree of freedom will generally be associated with a generic heavy mass scale @xmath24 , at which the np interaction will be most naturally described . at the scale @xmath25 of the charm mass , this description will have been modified by the effects of qcd . these should not be neglected , so we perform our np analyses at one - loop level for the strong interactions . the theoretical background for this is presented in sect . finally , in order to place the np discussion within its proper context , it makes sense to first review sm charm mixing . this is done in sect . the remainder of the paper then amounts to considering charm mixing with lots of ` extras ' . the paper concludes in sect . ix with a summary of our findings . let us first review some formal aspects of charm mixing . the mixing arises from @xmath26 interactions that generate off - diagonal terms in the mass matrix for @xmath1 and @xmath27 mesons . the expansion of the off - diagonal terms in the neutral @xmath3 mass matrix to second order in the weak interaction is [ m12 ] ( m - ) _ 21 = 0bar | h_w^|c|=2 | d^0 + _ n 0bar | h_w^|c|=1 | n n | h_w^|c|=1 | d^0 m_d - e_n+i , where @xmath28 and @xmath29 are the effective @xmath26 and @xmath30 hamiltonians . the off - diagonal mass - matrix terms induce mass eigenstates @xmath31 and @xmath32 that are superpositions of the flavor eigenstates @xmath1 and @xmath27 , @xmath33 where @xmath34 . the key quantities in @xmath1 mixing are the mass and width differences , @xmath35 or equivalently their dimensionless equivalents , x_d , y_d , [ xy ] where @xmath36 is the average width of the two neutral @xmath3 meson mass eigenstates . two quantities , @xmath11 and @xmath37 , which are actually measured in most experimental determinations of @xmath38 , are defined as y_d^(cp ) & & ( _ + - _ -)/ ( _ + + _ - ) = y_d - x_d(-a_prod ) , + y_d & & y_d _ k - x_d _ k , [ y - defs ] where the transition rates @xmath39 pertain to decay into final states of definite cp , @xmath40 is the so - called production asymmetry of @xmath1 and @xmath41 ( giving the relative weight of @xmath1 and @xmath2 in the sample ) and @xmath42 is the strong phase difference between the cabibbo favored and double cabibbo suppressed amplitudes @xcite . the quantities @xmath43 and @xmath44 account for the presence of cp violation in @xmath1-@xmath7 mixing , with @xmath43 being related to the @xmath45 parameters of eq . ( [ definition1 ] ) as @xmath46 and @xmath44 a cp - violating phase of @xmath47 ( if one neglects direct cp - violation ) @xcite . in practice , @xmath48 is measured by comparing decays of @xmath1 into a state of definite cp , such as @xmath49 , to decays of @xmath1 into a final state which is not a cp - eigenstate ( such as @xmath50 ) whereas @xmath51 is extracted from a time - dependent analysis of the @xmath52 transition @xcite . the states @xmath53 allow for effects of cp violation . however , cp violation in @xmath1 mixing is negligible in the standard model and there is no evidence for it experimentally @xcite . many new physics scenarios contain new phases which can induce sizable cp violation in the @xmath3 meson sector . nonetheless , a thorough investigation of such effects is beyond the scope of the present paper . therefore , we shall work in the limit of cp invariance ( so that @xmath54 ) for the remainder of this paper . throughout , our phase convention will be @xmath55 then @xmath56 become the cp eigenstates @xmath57 with @xmath58 . keeping in mind the neglect of cp - violation and also the phase convention of eq . ( [ phase ] ) , we relate the mixing quantities @xmath14 and @xmath9 to the mixing matrix as @xmath59 \ \ , \nonumber \\ & & y_{\rm d } = \frac{1}{2 m_{\rm d}\gamma_{\rm d}}\ , { \rm im}\ , \langle \d0bar | \,i\ { \rm d}^4 x\ , t \big\ { { \cal h}^{|\delta c|=1}_w ( x)\ , { \cal h}^{|\delta c|=1}_w(0 ) \big\ } | d^0 \rangle \ \ , \end{aligned}\ ] ] where @xmath60 is the weak hamiltonian density for @xmath30 transitions and @xmath61 denotes the time - ordered product . there is no contribution to @xmath9 from the local @xmath26 term , as it has no absorptive part . new physics contributions to @xmath9 have already been addressed in ref . @xcite , so the primary thrust of this paper will be to focus on @xmath14 . the next step , in sect . ii , is to expand the time - ordered product of eq . ( [ ope ] ) in local operators of increasing dimension ( higher dimension operators being suppressed by powers of @xmath62 ) . though the particles present in models with new physics may not be produced in charm quark decays , their effects can nonetheless be seen in the form of effective operators generated by the exchanges of these new particles . even without specifying the form of these new interactions , we know that their effect is to introduce several @xmath26 effective operators built out of the sm degrees of freedom . by integrating out new degrees of freedom associated with new interactions at a scale @xmath24 , we are left with an effective hamiltonian written in the form of a series of operators of increasing dimension . operator power counting then tells us the most important contributions are given by the operators of the lowest possible dimension , @xmath63 in this case . this means that they must contain only quark degrees of freedom . realizing this , we can write the complete basis of these effective operators , which can be done most conveniently in terms of chiral quark fields , [ seriesofoperators ] f | h_np | i = g _ i=1 c_i ( ) f | q_i | i ( ) , where the prefactor @xmath64 has the dimension of inverse - squared mass , the @xmath65 are dimensionless wilson coefficients , operators as @xmath66 and those for @xmath67 operators as @xmath68 . ] and the @xmath69 are the effective operators : l q_1 = ( _ l _ c_l ) ( _ l ^c_l ) , + q_2 = ( _ l _ c_l ) ( _ r ^c_r ) , + q_3 = ( _ l c_r ) ( _ r c_l ) , + q_4 = ( _ r c_l ) ( _ r c_l ) , l q_5 = ( _ r _ c_l ) ( _ r ^ c_l ) , + q_6 = ( _ r _ c_r ) ( _ r ^c_r ) , + q_7 = ( _ l c_r ) ( _ l c_r ) , + q_8 = ( _ l _ c_r ) ( _ l ^ c_r ) . [ setofoperators ] in total , there are eight possible operator structures that exhaust the list of possible independent contributions to @xmath26 transitions . since these operators are generated at the scale @xmath24 where the new physics is integrated out , a non - trivial operator mixing can occur when we take into account renormalization group running of these operators between the scales @xmath24 and @xmath70 , with @xmath70 being the scale where the hadronic matrix elements are computed . we shall work at the renormalization scale @xmath71 gev . this evolution is determined by solving the rg equations obeyed by the wilson coefficients , [ anomeq ] c ( ) = ^t c ( ) , where @xmath72 represents the matrix of anomalous dimensions of the operators in eq . ( [ setofoperators ] ) ( note the transposition ) . ( [ anomeq ] ) can be solved by transforming to the basis where the transpose of the anomalous dimension matrix is diagonal , integrating , and then transferring back to the original basis @xmath73 . at leading order , we have c ( ) = u ( , m ) c(m ) , where @xmath74 is the evolution matrix , obtained from eq . ( [ anomeq ] ) by u ( _ 1,_2 ) = v _ d v^-1 . in the above , @xmath75 is the vector containing the diagonal elements of the diagonalized transposed matrix of the anomalous dimensions @xmath76 , @xmath77 is the matrix that diagonalizes @xmath78 and r(_1,_2 ) . for completeness , we display the matrix of anomalous dimensions at leading - order ( lo ) in qcd @xcite , = ( c c c c c c c c 6- & 0 & 0 & 0 & 0 & 0 & 0 & 0 + 0 & & 12 & 0 & 0 & 0 & 0 & 0 + 0 & 0 & -6 n_c+ & 0 & 0 & 0 & 0 & 0 + 0 & 0 & 0 & 6 - 6 n_c+ & - & 0 & 0 & 0 + 0 & 0 & 0 & -24- & 6 + 2 n_c - & 0 & 0 & 0 + 0 & 0 & 0 & 0 & 0 & 6- & 0 & 0 + 0 & 0 & 0 & 0 & 0 & 0 & 6 - 6 n_c+ & - + 0 & 0 & 0 & 0 & 0 & 0 & -24- & 6 + 2 n_c - + ) we note that ref . @xcite also includes the next - to - leading order ( nlo ) expressions for the elements in the anomalous dimensions matrix . however , we perform our calculations at lo here since the nlo corrections to the matching conditions in the various models of new physics have generally not been computed . due to the relatively simple structure of @xmath79 , one can easily write the evolution of each wilson coefficient in eq . ( [ seriesofoperators ] ) from the new physics scale @xmath80 down to the hadronic scale @xmath70 , taking into account quark thresholds . corresponding to each of the eight operators @xmath81 ( @xmath82 ) is an rg factor @xmath83 . the first of these , @xmath84 , is given explicitly by r_1(,m)&= & ( ) ^2/7 ( ) ^6/23 ( ) ^6/25 . [ wilson ] and the rest can be expressed in terms of @xmath84 as l r_2(,m ) = [ r_1(,m)]^1/2 , + r_3(,m ) = [ r_1(,m)]^-4 , + r_4(,m ) = [ r_1(,m)]^(1 + ) /6 , + r_5(,m ) = [ r_1(,m)]^(1 - ) /6 , l + r_6(,m ) = r_1(,m ) , + r_7(,m ) = r_4(,m ) , + r_8(,m ) = r_5(,m ) . the rg factors are generally only weakly dependent on the np scale @xmath24 since it is taken to be larger than the top quark mass , @xmath85 , and the evolution of @xmath86 is slow at these high mass scales . in table [ tab : lo ] , we display numerical values for the @xmath83 with @xmath87 tev and @xmath88 gev . here , we compute @xmath86 using the one - loop evolution and matching expressions for perturbative consistency with the rg evolution of the effective hamiltonian . c||c|c|c|c|c @xmath89 & @xmath90 & @xmath91 & @xmath92 & @xmath93 & @xmath94 + @xmath95 & @xmath96 & @xmath97 & @xmath98 & @xmath99 & @xmath100 + @xmath101 & @xmath102 & @xmath103 & @xmath104 & @xmath105 & @xmath106 + .05 in we will need to evaluate the @xmath1-to-@xmath7 matrix elements of the eight dimension - six basis operators . in general , this implies eight non - perturbative parameters that would have to be evaluated by means of qcd sum rules or on the lattice . we choose those parameters ( denoted by @xmath107 ) as follows , @xmath108 where latexmath:[$\langle q_i \rangle \equiv \langle { \bar d}^0 @xmath3 meson decay constant . by and large , the compensatory @xmath110-factors @xmath111 are unknown , except in vacuum saturation and in the heavy quark limit ; there , one has @xmath112 . since most of the matrix elements in eq . ( [ me ] ) are not known , we will need something more manageable in order to obtain numerical results . the usual approach to computing matrix elements is to employ the vacuum saturation approximation . however , because some of the @xmath110-parameters are known , we would like to introduce a ` modified vacuum saturation ' ( mvs ) , where all matrix elements in eq . ( [ me ] ) are written in terms of ( known ) matrix elements of @xmath113 and @xmath114 matrix elements @xmath115 and @xmath116 , @xmath117 where we denote @xmath118 as the number of colors and , as in ref . @xcite , define [ bbar ] |b_d^(s ) b_d^(s ) as well as [ eta ] . in our numerical work , we take @xmath119 , which is the most recent result from the quenched lattice calculation @xcite , and use the cleo - c determination @xmath120 @xcite . we urge the lattice community to perform an evaluation of the @xmath121 parameters defined in eq . ( [ me ] ) for the full operator set relevant to @xmath3 meson mixing . theoretical predictions of @xmath14 and @xmath9 within the standard model span several orders of magnitude . roughly , there are two approaches , neither of which give very reliable results because @xmath25 is in some sense intermediate between the heavy - quark and light - quark limits . consider , for example , @xmath122 as given in eq . ( [ ope ] ) @xmath123 to utilize this relation , one inserts intermediate states between the @xmath124 weak hamiltonian densities @xmath125 . this can be done using either quark or hadron degrees of freedom . let us consider each of these possibilities in turn . the `` inclusive '' ( or quark - level ) approach is based on the operator product expansion ( ope ) . in the @xmath126 limit , where @xmath127 is a scale characteristic of the strong interactions , @xmath128 and @xmath129 can be expanded in terms of matrix elements of local operators @xcite of increasing dimensions suppressed by powers of inverse charm quark mass . . ] an instructive example concerns a recent analysis of the leading dimension @xmath130 case @xcite in which the width difference @xmath9 is calculated in terms of quarks ( _ cf _ fig . [ fig : loop ] ) and the mass difference @xmath14 is then found from dispersion relations . the calculation is carried out as an expansion in qcd , including leading - order @xmath131 and next - to - leading order @xmath132 contributions with @xmath133 here , lo and nlo denote only the corrections at that order and not the full quantity computed to that order . because @xmath134 , @xmath38 experiences no @xmath135-quark contribution .- quark contribution to @xmath16 ; its numerical contribution is subleading ( @xmath136 whereas @xmath137 ) . ] this leaves only @xmath138 , @xmath139 and @xmath140 intermediate states contributing to the mixing diagram of fig . [ fig : loop ] . taking @xmath141 , the mixing loop - functions will depend on @xmath142 . table [ tab : qkmix ] examines in detail the loop - functions for @xmath122 and shows the results of carrying out an expansion in powers of @xmath143 . we see that the contributions of the individual intermediate states in the mixing diagram are _ not _ intrinsically small in fact , they begin to contribute at @xmath144 . however , flavor cancellations remove all contributions through @xmath145 for @xmath122 , so the net result is @xmath146 . charm mixing clearly experiences a remarkable gim suppression ! the corresponding result for @xmath16 turns out to be @xmath145 . summarizing , the leading dependences in @xmath143 for the dimension six contributions are y_d^(lo ) ~z^3x_d^(lo ) ~z^2y_d^(nlo ) ~z^2x_d^(nlo ) ~z^2 . [ order ] .flavor cancellations in @xmath122 . [ cols="^,^,^,^",options="header " , ] .05 in we recommend further experimental study of this subject on two fronts . first , of course , is the need to reduce error bars in the measured values of @xmath9 and especially @xmath14 . equally important is continuing the search for evidence of cp violation in mixing for the @xmath1 system . cp violation provides an interesting contrast with @xmath1 mixing because it provides an independent arena for competition between the sm and np signals . there is especially room for improvement in the sm analysis of charm cp violation , and work on this is underway . the work of e.g. was supported in part by the u.s . national science foundation under grant phy0555304 , j.h . was supported by the u.s . department of energy under contract de - ac02 - 76sf00515 , s.p . was supported by the u.s . department of energy under contract de - fg02 - 04er41291 and a.p . was supported in part by the u.s . national science foundation under career award phy0547794 , and by the u.s . department of energy under contract de - fg02 - 96er41005 . j.h . would like to thank the theoretical physics group at fermilab for their hospitality and e.g. , j.h . , and a.p . thank the high energy physics group at the university of hawaii for their hospitality while part of this work was performed . we would like to thank c.f . berger , s. chivukula , b. dobrescu , j. donoghue , k.c . kong , b. lillie , e. lunghi , e. simmons and t. tait for discussions related to this work and especially t. browder , t. rizzo , and x. tata for a careful reading of the manuscript . here , we collect the formulae used throughout the manuscript to compute the contributions to @xmath1-@xmath147 mixing in the various new physics models . the loop functions , first calculated by inami lim , apply for several np scenarios discussed in the text . for a contribution from two internal quarks of the same flavor in the box diagram , the loop function is @xmath148 -{3x^3\over 2(1-x)^3}\ln x\ \ , \ ] ] and for two quarks of different flavors , @xmath149 + ( x_i \leftrightarrow x_j ) \right . \nonumber \\ & & \left . \hspace{5.0 cm } - { 3\over 4(1-x_i)(1-x_j ) } \right ) \ \ .\end{aligned}\ ] ] the loop functions with one @xmath155 and one sm @xmath156 boson and with two @xmath155 being exchanged in the box diagram are @xmath157\ , \nonumber \\ a_{\rm wh}(x , y ) & = & 2x^2\left[{1\over ( x - y)(1-x)}+{y\ln y\over ( x - y)^2(1-y ) } + { ( x^2-y)\ln x\over ( x - y)^2(1-x)^2}\right . \nonumber\\ & & \left . -{1\over 4}\left ( { x\over ( x - y)(1-x ) } + { y^2\ln y\over ( 1-y)(x - y)^2 } + { x(x+xy-2y)\ln x\over ( x - y)^2(1-x)^2}\right)\right]\ \ .\end{aligned}\ ] ] the coupling of two zero - mode fermions to the @xmath159 gauge boson kk state , @xmath160 , relative to the sm coupling strength is @xmath161 \int_{e^{-\pi kr_c}}^1 dz\ , z^{(1 - 2c_f)}{j_1(x_nz)+\alpha_ny_1(x_nz ) \over |j_1(x_n)+\alpha_ny_1(x_n)|}\ \ , \ ] ] where the roots @xmath162 for the gauge boson kk spectrum are given by @xmath163 = 0\ \ , \ ] ] and @xmath164 is defined by @xmath165 # 1 # 2 # 3 int . . phys . * # 1 * , # 2 ( # 3 ) # 1 # 2 # 3 mod . # 1 * , # 2 ( # 3 ) # 1 # 2 # 3 nucl . # 1 * , # 2 ( # 3 ) # 1 # 2 # 3 phys . # 1 * , # 2 ( # 3 ) # 1 # 2 # 3 phys . rep . * # 1 * , # 2 ( # 3 ) # 1 # 2 # 3 phys . * # 1 * , # 2 ( # 3 ) # 1 # 2 # 3 phys . lett . * # 1 * , # 2 ( # 3 ) # 1 # 2 # 3 rev . # 1 * , # 2 ( # 3 ) # 1 # 2 # 3 z. phys . * # 1 * , # 2 ( # 3 ) a. datta and d. kumbhakar , z. phys . c * 27 * , 515 ( 1985 ) . the babar collaboration , b. aubert , , ` search for @xmath168 mixing using doubly flavor tagged semileptonic decay modes , ' arxiv:0705.0704 [ hep - ex ] . g. burdman and i. shipsey , ann . nucl . part . sci . * 53 * , 431 ( 2003 ) [ arxiv : hep - ph/0310076 ] ; g. burdman , e. golowich , j.l . hewett and s. pakvasa , phys . d * 52 * , 6383 ( 1995 ) [ arxiv : hep - ph/9502329 ] ; j.l . hewett , t. takeuchi and s.d . thomas , in _ electroweak symmetry breaking and beyond the standard model _ ed . by t. barklow , , world scientific , arxiv : hep - ph/9603391 ; s. bianco , f. l. fabbri , d. benson and i. bigi , riv . nuovo cim . * 26n7 * , 1 ( 2003 ) [ arxiv : hep - ex/0309021 ] ; a. a. petrov , int . j. mod . a * 21 * , 5686 ( 2006 ) [ arxiv : hep - ph/0611361 ] ; a. a. petrov , _ in the proceedings of flavor physics and cp violation ( fpcp 2003 ) , paris , france , 3 - 6 jun 2003 , pp mec05 _ [ arxiv : hep - ph/0311371 ] ; h. n. nelson , in _ proc . of the 19th symp . on photon and lepton interactions at high energy lp99 jaros and m.e . peskin , arxiv : hep - ex/9908021 . heavy flavor averaging group , http://www.slac.stanford.edu/xorg/hfag/charm/index.html ; d. asner , ` charm physics ' , talk at the _ workshop on flavor in the era of the lhc _ , cern ( 3/26 - 28/07 ) ; b. peterson , talk given at the _ charm 2007 workshop _ , cornell university , ithaca , ny , august 2007 . e. golowich and a.a . petrov , phys . b * 625 * , 53 ( 2005 ) [ arxiv : hep - ph/0506185 ] . a. f. falk , y. nir and a. a. petrov , jhep * 9912 * , 019 ( 1999 ) [ arxiv : hep - ph/9911369 ] . s. bergmann , y. grossman , z. ligeti , y. nir and a.a . petrov , phys . b * 486 * , 418 ( 2000 ) [ arxiv : hep - ph/0005181 ] . y. nir , ` lessons from babar and belle measurements of @xmath169 mixing parameters , ' arxiv : hep - ph/0703235 . particle data group ( w .- m . yao et al . ) , ` review of particle physics ' , j.phys.g33:1-1232,2006 . e. golowich , s. pakvasa and a.a . petrov , phys . * 98 * , 181808 - 1 ( 2007 ) [ arxiv : hep - ph/0610039 ] . m. ciuchini , e. franco , v. lubicz , g. martinelli , i. scimemi and l. silvestrini , nucl . b * 523 * , 501 ( 1998 ) [ arxiv : hep - ph/9711402 ] . r. gupta , t. bhattacharya and s. r. sharpe , phys . d * 55 * , 4036 ( 1997 ) [ arxiv : hep - lat/9611023 ] . m. artuso _ et al . _ [ cleo collaboration ] , phys . * 95 * , 251801 ( 2005 ) [ arxiv : hep - ex/0508057 ] . bigi and n.g . uraltsev , nucl . b * 592 * , 92 ( 2001 ) [ arxiv : hep - ph/0005089 ] . falk , y. grossman , z. ligeti and a.a . petrov , phys . d * 65 * , 054034 ( 2002 ) [ arxiv : hep - ph/0110317 ] . e. golowich , a.a . petrov and g. yeghiyan ( in progress ) . e. golowich and a. a. petrov , phys . b * 427 * , 172 ( 1998 ) [ arxiv : hep - ph/9802291 ] . donoghue , e. golowich , b.r . holstein and j. trampetic , phys . d * 33 * , 179 ( 1986 ) . l. wolfenstein , b164 170 1985 . f. buccella , m. lusignoli , g. miele , a. pugliese and p. santorelli , phys . d * 51 * , 3478 ( 1995 ) [ arxiv : hep - ph/9411286 ] . falk , y. grossman , z. ligeti , y. nir and a.a . petrov , phys . d * 69 * , 114021 ( 2004 ) [ arxiv : hep - ph/0402204 ] . m.e . peskin and t. takeuchi , d * 46 * , 381 ( 1992 ) . lep electroweak working group , phys . rept . * 427 * , 257 ( 2006 ) [ arxiv : hep - ex/0509008 ] . see , for example , h.j . he , n. polonsky and s.f . su , phys . d * 64 * , 053004 ( 2001 ) [ arxiv : hep - ph/0102144 ] . chanowitz , m.a . furman and i. hinchliffe , phys . b * 78 * , 285 ( 1978 ) ; nucl . b * 153 * , 402 ( 1979 ) . babu , x.g . he , x. li and s. pakvasa , phys . b * 205 * , 540 ( 1988 ) . f. gursey , p. ramond and p. sikivie , phys . b * 60 * , 177 ( 1976 ) . j.l . hewett and t.g . rizzo , phys . * 183 * , 193 ( 1989 ) . n. arkani - hamed , a.g . cohen , e. katz and a.e . nelson , jhep * 0207 * , 034 ( 2002 ) [ arxiv : hep - ph/0206021 ] . n. arkani - hamed , a.g . cohen , e. katz , a.e . nelson , t. gregoire and j.g . wacker , jhep * 0208 * , 021 ( 2002 ) [ arxiv : hep - ph/0206020 ] . bjorken , s. pakvasa and s.f . tuan , phys . d * 66 * , 053008 ( 2002 ) [ arxiv : hep - ph/0206116 ] . n. arkani - hamed , a.g . cohen and h. georgi , phys . b * 513 * , 232 ( 2001 ) [ arxiv : hep - ph/0105239 ] ; for a review , see , m. perelstein , prog . part . * 58 * , 247 ( 2007 ) [ arxiv : hep - ph/0512128 ] . j. y. lee , jhep * 0412 * , 065 ( 2004 ) [ arxiv : hep - ph/0408362 ] . c. h. chen , c. q. geng and t. c. yuan , ` @xmath3-@xmath170 mixing and rare d decays in the littlest higgs model with non - unitarity matrix , ' arxiv:0704.0601 [ hep - ph ] . c. csaki , j. hubisz , g. d. kribs , p. meade and j. terning , phys . d * 67 * , 115002 ( 2003 ) [ arxiv : hep - ph/0211124 ] ; j. l. hewett , f. j. petriello and t. g. rizzo , jhep * 0310 * , 062 ( 2003 ) [ arxiv : hep - ph/0211218 ] ; m. c. chen and s. dawson , phys . d * 70 * , 015003 ( 2004 ) [ arxiv : hep - ph/0311032 ] . c. csaki , j. hubisz , g. d. kribs , p. meade and j. terning , phys . d * 68 * , 035009 ( 2003 ) [ arxiv : hep - ph/0303236 ] . h. c. cheng and i. low , jhep * 0309 * , 051 ( 2003 ) [ arxiv : hep - ph/0308199 ] ; h. c. cheng and i. low , jhep * 0408 * , 061 ( 2004 ) [ arxiv : hep - ph/0405243 ] . c. t. hill and r. j. hill , arxiv : hep - ph/0701044 ; c. t. hill and r. j. hill , arxiv:0705.0697 [ hep - ph ] . j. hubisz , s. j. lee and g. paz , jhep * 0606 * , 041 ( 2006 ) [ arxiv : hep - ph/0512169 ] ; m. blanke , a. j. buras , a. poschenrieder , c. tarantino , s. uhlig and a. weiler , jhep * 0612 * , 003 ( 2006 ) [ arxiv : hep - ph/0605214 ] . m. blanke , a. j. buras , s. recksiegel , c. tarantino and s. uhlig , ` littlest higgs model with t - parity confronting the new data on @xmath169 mixing , ' arxiv : hep - ph/0703254 . see , for example , p. savard , `` searches for extra dimensions and new gauge bosons at the tevatron , '' talk given at the _ xxxiii international conference on high energy physics _ , 26 july-2 august 2006 , moscow , russia ; t. adams , `` searches for new phenomena with lepton final states at the tevatron , '' talk given at _ rencontres de moriond electroweak interactions and unified theories 2007 _ , la thuile , italy 10 - 17 march 2007 ; a. abulencia _ et al . _ [ cdf collaboration ] , ` search for @xmath171 boson decaying to electron - neutrino pairs in @xmath172 collisions at @xmath173 tev , ' arxiv : hep - ex/0611022 . t. g. rizzo , z phenomenology and the lhc , arxiv : hep - ph/0610104 . p. langacker and m. plumacher , phys . d * 62 * , 013006 ( 2000 ) [ arxiv : hep - ph/0001204 ] . a. arhrib , k. cheung , c. w. chiang and t. c. yuan , phys . d * 73 * , 075015 ( 2006 ) [ arxiv : hep - ph/0602175 ] . n. g. deshpande , e. keith and t. g. rizzo , phys . lett . * 70 * , 3189 ( 1993 ) [ arxiv : hep - ph/9211310 ] . k. kiers , m. assis and a. a. petrov , phys . rev . d * 71 * , 115015 ( 2005 ) [ arxiv : hep - ph/0503115 ] . g. beall , m. bander and a. soni , phys . lett . * 48 * , 848 ( 1982 ) . d. chang , r. n. mohapatra and m. k. parida , phys . d * 30 * , 1052 ( 1984 ) . f. i. olness and m. e. ebel , phys . d * 30 * , 1034 ( 1984 ) ; p. langacker and s. uma sankar , phys . d * 40 * , 1569 ( 1989 ) . t. g. rizzo , phys . d * 50 * , 325 ( 1994 ) [ arxiv : hep - ph/9311265 ] . z. chacko , h. s. goh and r. harnik , phys . lett . * 96 * , 231802 ( 2006 ) [ arxiv : hep - ph/0506256 ] ; h. s. goh and s. su , phys . d * 75 * , 075010 ( 2007 ) [ arxiv : hep - ph/0611015 ] . k. s. babu , x. g. he and e. ma , phys . rev . d * 36 * , 878 ( 1987 ) . e. ma , mod . lett . a * 3 * , 319 ( 1988 ) . b. abbott _ et al . _ [ d0 collaboration ] , phys . lett . * 83 * , 2896 ( 1999 ) [ arxiv : hep - ex/9904023 ] . s. davidson , d. c. bailey and b. a. campbell , z. phys . c * 61 * , 613 ( 1994 ) [ arxiv : hep - ph/9309310 ] . m. leurer , phys . d * 50 * , 536 ( 1994 ) [ arxiv : hep - ph/9312341 ] . for a summary , see , j. l. hewett and t. g. rizzo , phys . d * 56 * , 5709 ( 1997 ) [ arxiv : hep - ph/9703337 ] . see , for example , b. grinstein and m. b. wise , phys . b * 201 * , 274 ( 1988 ) ; t. g. rizzo , phys . d * 38 * , 820 ( 1988 ) ; j.l . hewett , phys . lett . * 70 * , 1045 ( 1993 ) [ arxiv : hep - ph/9211256 ] ; v.d . barger , m. s. berger and r. j. n. phillips , phys . * 70 * , 1368 ( 1993 ) [ arxiv : hep - ph/9211260 ] . m. misiak _ et al . _ , phys . lett . * 98 * , 022002 ( 2007 ) [ arxiv : hep - ph/0609232 ] . barger , j.l . hewett and r.j.n . phillips , phys . d * 41 * , 3421 ( 1990 ) . g. isidori and a. retico , jhep * 0111 * , 001 ( 2001 ) [ arxiv : hep - ph/0110121 ] ; a.j . buras , p.h . chankowski , j. rosiek and l. slawianowska , nucl . b * 619 * , 434 ( 2001 ) [ arxiv : hep - ph/0107048 ] . s. pakvasa and h. sugawara , 73b 61 1978 . cheng and m. sher , phys . d * 35 * , 3484 ( 1987 ) . l. j. hall and s. weinberg , phys . d * 48 * , 979 ( 1993 ) [ arxiv : hep - ph/9303241 ] . d. acosta _ et al . _ [ cdf collaboration ] , phys . rev . d * 72 * , 051107 ( 2005 ) [ arxiv : hep - ex/0506074 ] ; a. abulencia _ et al . _ [ cdf collaboration ] , phys . rev . d * 73 * , 051102 ( 2006 ) [ arxiv : hep - ex/0512055 ] . r. s. chivukula , n. j. evans and e. h. simmons , phys . d * 66 * , 035008 ( 2002 ) [ arxiv : hep - ph/0204193 ] . n. arkani - hamed , a. g. cohen , t. gregoire and j. g. wacker , jhep * 0208 * , 020 ( 2002 ) [ arxiv : hep - ph/0202089 ] . c. csaki , c. grojean , l. pilo and j. terning , phys . lett . * 92 * , 101802 ( 2004 ) [ arxiv : hep - ph/0308038 ] . c. csaki , c. grojean , h. murayama , l. pilo and j. terning , phys . d * 69 * , 055006 ( 2004 ) [ arxiv : hep - ph/0305237 ] . y. nomura , jhep * 0311 * , 050 ( 2003 ) [ arxiv : hep - ph/0309189 ] ; r. barbieri , a. pomarol and r. rattazzi , phys . b * 591 * , 141 ( 2004 ) [ arxiv : hep - ph/0310285 ] ; h. davoudiasl , j. l. hewett , b. lillie and t. g. rizzo , phys . d * 70 * , 015006 ( 2004 ) [ arxiv : hep - ph/0312193 ] ; g. burdman and y. nomura , phys . d * 69 * , 115013 ( 2004 ) [ arxiv : hep - ph/0312247 ] ; g. cacciapaglia , c. csaki , c. grojean and j. terning , phys . d * 70 * , 075014 ( 2004 ) [ arxiv : hep - ph/0401160 ] ; h. davoudiasl , j. l. hewett , b. lillie and t. g. rizzo , jhep * 0405 * , 015 ( 2004 ) [ arxiv : hep - ph/0403300 ] . c. csaki , c. grojean , j. hubisz , y. shirman and j. terning , phys . d * 70 * , 015012 ( 2004 ) [ arxiv : hep - ph/0310355 ] . for a review of tev - scale extra dimensional models and their experimental consequences , see for example , j. hewett and m. spiropulu , ann . nucl . part . sci . * 52 * , 397 ( 2002 ) [ arxiv : hep - ph/0205106 ] ; t. g. rizzo , _ in the proceedings of 32nd slac summer institute on particle physics ( ssi 2004 ) : natures greatest puzzles , menlo park , california , 2 - 13 aug 2004 , pp l013 _ [ arxiv : hep - ph/0409309 ] . i. antoniadis , phys . b * 246 * , 377 ( 1990 ) ; j. d. lykken , phys . d * 54 * , 3693 ( 1996 ) [ arxiv : hep - th/9603133 ] ; i. antoniadis and m. quiros , phys . b * 392 * , 61 ( 1997 ) [ arxiv : hep - th/9609209 ] . t. appelquist , h. c. cheng and b. a. dobrescu , phys . d * 64 * , 035002 ( 2001 ) [ arxiv : hep - ph/0012100 ] . g. servant and t. m. p. tait , nucl . b * 650 * , 391 ( 2003 ) [ arxiv : hep - ph/0206071 ] , new j. phys . * 4 * , 99 ( 2002 ) [ arxiv : hep - ph/0209262 ] ; h. c. cheng , j. l. feng and k. t. matchev , phys . lett . * 89 * , 211301 ( 2002 ) [ arxiv : hep - ph/0207125 ] . c. lin , search for universal extra dimensions in the multi - lepton channel at the tevatron , cdf thesis , fermilab - pub-05 - 572-e . h. c. cheng , k. t. matchev and m. schmaltz , phys . d * 66 * , 036005 ( 2002 ) [ arxiv : hep - ph/0204342 ] . a. j. buras , m. spranger and a. weiler , nucl . b * 660 * , 225 ( 2003 ) [ arxiv : hep - ph/0212143 ] . n. arkani - hamed , y. grossman and m. schmaltz , phys . d * 61 * , 115004 ( 2000 ) [ arxiv : hep - ph/9909411 ] . e. a. mirabelli and m. schmaltz , phys . d * 61 * , 113011 ( 2000 ) [ arxiv : hep - ph/9912265 ] ; g. c. branco , a. de gouvea and m. n. rebelo , phys . b * 506 * , 115 ( 2001 ) [ arxiv : hep - ph/0012289 ] ; d. e. kaplan and t. m. p. tait , jhep * 0111 * , 051 ( 2001 ) [ arxiv : hep - ph/0110126 ] ; y. grossman and g. perez , phys . d * 67 * , 015011 ( 2003 ) [ arxiv : hep - ph/0210053 ] ; c. biggio , f. feruglio , i. masina and m. perez - victoria , nucl . b * 677 * , 451 ( 2004 ) [ arxiv : hep - ph/0305129 ] ; b. lillie , jhep * 0312 * , 030 ( 2003 ) [ arxiv : hep - ph/0308091 ] ; y. grossman , r. harnik , g. perez , m. d. schwartz and z. surujon , phys . d * 71 * , 056007 ( 2005 ) [ arxiv : hep - ph/0407260 ] ; g. perez and t. volansky , phys . d * 72 * , 103522 ( 2005 ) [ arxiv : hep - ph/0505222 ] . a. delgado , a. pomarol and m. quiros , jhep * 0001 * , 030 ( 2000 ) [ arxiv : hep - ph/9911252 ] ; w. f. chang and j. n. ng , jhep * 0212 * , 077 ( 2002 ) [ arxiv : hep - ph/0210414 ] . b. lillie and j. l. hewett , phys . d * 68 * , 116002 ( 2003 ) [ arxiv : hep - ph/0306193 ] . l. randall and r. sundrum , phys . * 83 * , 3370 ( 1999 ) [ arxiv : hep - ph/9905221 ] . s. j. huber and q. shafi , phys . b * 498 * , 256 ( 2001 ) [ arxiv : hep - ph/0010195 ] ; s. j. huber , nucl . b * 666 * , 269 ( 2003 ) [ arxiv : hep - ph/0303183 ] ; g. burdman , phys . b * 590 * , 86 ( 2004 ) [ arxiv : hep - ph/0310144 ] ; k. agashe , g. perez and a. soni , phys . rev . lett . * 93 * , 201804 ( 2004 ) [ arxiv : hep - ph/0406101 ] . k. agashe , g. perez and a. soni , phys . d * 71 * , 016002 ( 2005 ) [ arxiv : hep - ph/0408134 ] . h. davoudiasl , j. l. hewett and t. g. rizzo , phys . b * 473 * , 43 ( 2000 ) [ arxiv : hep - ph/9911262 ] ; a. pomarol , phys . b * 486 * , 153 ( 2000 ) [ arxiv : hep - ph/9911294 ] . y. grossman and m. neubert , phys . b * 474 * , 361 ( 2000 ) [ arxiv : hep - ph/9912408 ] ; t. gherghetta and a. pomarol , nucl . b * 586 * , 141 ( 2000 ) [ arxiv : hep - ph/0003129 ] . h. davoudiasl , j. l. hewett and t. g. rizzo , phys . d * 63 * , 075004 ( 2001 ) [ arxiv : hep - ph/0006041 ] . s. j. huber and q. shafi , phys . d * 63 * , 045010 ( 2001 ) [ arxiv : hep - ph/0005286 ] ; s. j. huber , c. a. lee and q. shafi , phys . b * 531 * , 112 ( 2002 ) [ arxiv : hep - ph/0111465 ] ; c. csaki , j. erlich and j. terning , phys . rev . d * 66 * , 064021 ( 2002 ) [ arxiv : hep - ph/0203034 ] ; j. l. hewett , f. j. petriello and t. g. rizzo , jhep * 0209 * , 030 ( 2002 ) [ arxiv : hep - ph/0203091 ] . k. agashe , a. delgado , m. j. may and r. sundrum , jhep * 0308 * , 050 ( 2003 ) [ arxiv : hep - ph/0308036 ] . k. agashe , r. contino and a. pomarol , nucl . phys . b * 719 * , 165 ( 2005 ) [ arxiv : hep - ph/0412089 ] ; k. agashe , a. e. blechman and f. petriello , phys . d * 74 * , 053011 ( 2006 ) [ arxiv : hep - ph/0606021 ] . h. davoudiasl and a. soni , arxiv:0705.0151 [ hep - ph ] . j. r. ellis and d. v. nanopoulos , phys . b * 110 * , 44 ( 1982 ) . h. p. nilles , phys . rept . * 110 * , 1 ( 1984 ) . h. georgi , phys . b * 169 * , 231 ( 1986 ) . l. j. hall , v. a. kostelecky and s. raby , nucl . b * 267 * , 415 ( 1986 ) . for a comprehensive study of fcnc effects in supersymmetry see , s. bertolini , f. borzumati , a. masiero and g. ridolfi , nucl . b * 353 * , 591 ( 1991 ) ; f. gabbiani , e. gabrielli , a. masiero and l. silvestrini , nucl . b * 477 * , 321 ( 1996 ) [ arxiv : hep - ph/9604387 ] ; i. i. y. bigi , f. gabbiani and a. masiero , z. phys . c * 48 * , 633 ( 1990 ) . m. ciuchini , e. franco , d. guadagnoli , v. lubicz , v. porretti and l. silvestrini , jhep * 0609 * , 013 ( 2006 ) [ arxiv : hep - ph/0606197 ] . j. a. bagger , k. t. matchev and r. j. zhang , phys . b * 412 * , 77 ( 1997 ) [ arxiv : hep - ph/9707225 ] . m. ciuchini , e. franco , d. guadagnoli , v. lubicz , m. pierini , v. porretti and l. silvestrini , d - anti - d mixing and new physics : general considerations and constraints on the mssm , arxiv : hep - ph/0703204 . y. nir and n. seiberg , phys . b * 309 * , 337 ( 1993 ) [ arxiv : hep - ph/9304307 ] . y. nir and g. raz , phys . d * 66 * , 035007 ( 2002 ) [ arxiv : hep - ph/0206064 ] . see , for example , l. e. ibanez and g. g. ross , nucl . b * 368 * , 3 ( 1992 ) . k. agashe and m. graesser , phys . d * 54 * , 4445 ( 1996 ) [ arxiv : hep - ph/9510439 ] . for a recent summary of constraints on the r - parity violating couplings , see , h. baer and x. tata , _ weak scale supersymmetry : from superfields to scattering events _ ( cambridge university press , cambridge , england 2006 ) . see , also , b. c. allanach , a. dedes and h. k. dreiner , phys . d * 60 * , 075014 ( 1999 ) [ arxiv : hep - ph/9906209 ] ; g. bhattacharyya , ` a brief review of r - parity - violating couplings , ' arxiv : hep - ph/9709395 , and references therein . n. arkani - hamed and s. dimopoulos , jhep * 0506 * , 073 ( 2005 ) [ arxiv : hep - th/0405159 ] ; g. f. giudice and a. romanino , nucl . b * 699 * , 65 ( 2004 ) [ erratum - ibid . b * 706 * , 65 ( 2005 ) ] [ arxiv : hep - ph/0406088 ] .
nasa s _ kepler _ mission has discovered that many of the planetary systems around solar - type stars contain multiple planets ( kepler multis , * ? ? ? * ; * ? ? ? * ; * ? ? ? planets detected in these systems likely possess low mutual inclinations ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and are spaced closely . in multiple systems that contain high number of planets ( 4 - 6 ) , the spacing is so tight that there may be of order @xmath1 planets fitted inside the orbit of mercury . systems where a single planet transit ( kepler single ) and where multiple planets transit ( kepler multis ) may share the same intrinsic architecture but only differ in transit geometry . however , a few statistical studies have shown that this is not the case . there appears to be an excess in the number of singles over what one expects by extrapolating from the kepler multis @xcite , suggesting that many of the singles are intrinsically singles . even the multis may contain different sorts of systems . a direct probe of the underlying population is provided by transit - time - variations ( ttv ) . @xcite probed the presence of close - companions ( transiting or not ) around kepler candidates using ttv signals . they found a striking correlation between the fraction of planets that show detectable ttvs and their transit multiplicity . systems where four or more planets transit enjoy four times higher ttv fraction than kepler singles do , and about twice as high as those with two or three transiting planets . this led them to propose that there are at least two different classes of kepler systems , one closely packed and one sparsely populated ( also see * ? ? ? besides from planet spacing , orbital eccentricity and inclinations also differ between the two classes . compared to kepler multis , the transit durations of kepler singles appear to be longer , requiring their orbits to have higher eccentricities ( wu , unpublished ) ; study of stellar obliquity suggests that kepler singles may lie on orbits more misaligned with the stellar spin than the multis @xcite . both lend to the picture that kepler multis are dynamically cold systems , while singles are dynamically hotter . what could have produced such a dichotomy ? @xcite explored the possibility of dynamical instability . they boosted the masses of observed triple planetary systems by about a factor of 10 and found that these systems could become destabilized within 10 billion years . this led them to hypothesize that kepler singles may be the aftermath of dynamical instability . in this study , we provide much more definite arguments to this hypothesis by investigating the spacing distribution among the closely packed multi systems ( 4 planets and more ) . we find that , adopting realistic planetary masses and ages , kepler high multiple systems are packed close to the boundary of stability . this proximity has to be explained . unless the formation mechanism is somehow capable of producing this proximity , the most likely explanation is that the primordial systems had a range of spacing , but the ones that were packed too closely have been sculpted away by the subsequent billions of years of evolution . in this hypothesis , kepler singles and lower multiples are the descendants of kepler multis . if this true , we may be able to recover the primordial spacing of planetary systems . here , we set the scene by reviewing some previous results on stability in multiple planetary systems . johannes kepler ( 1571 - 1630 ) , after whom the _ kepler _ mission is named , initially believed that the spacing of planets in the solar system is limited by packing perfect solids ( polygons with identical faces ) . he discarded this idea after developing his famous kepler laws . however , the question still remains , how tight can you pack planets together without affecting dynamical instability . over the past few decades , some theoretical progress have been made in answering this question , and in some special cases the answers are known exactly . one usually measure spacing between every two planets in a unit called the mutual hill sphere @xmath4 where @xmath5 , @xmath6 are the semi - major axis and mass of the two planets , and @xmath7 the mass of the central star . in this unit , planet spacing can be expressed as a dimensionless number @xmath8 @xmath9 we first focus only on orbits that are circular and coplanar . a well known result , derived by @xcite based on overlapping of first order mean - motion resonances , is that a test particle ( @xmath10 ) perturbed by a planet ( @xmath11 ) is subject to orbital chaos if it lies too close to the planet , @xmath12 where @xmath13 is the planet mass ratio and the numerical coefficient is adopted from @xcite . in @xmath8 value , the critical spacing is @xmath14 this result is little modified if the test particle is instead substituted by a planet @xcite . the dependence of @xmath15 on @xmath16 ( albeit weak ) indicates that hill sphere is not necessarily the best unit to measure stability . however , we adopt it here for convenience . there is also an important result on hill stability ( stability against orbit crossing ) , popularized in astronomy by @xcite . as a result of total energy and angular momentum conservation , two initially circular , coplanar planets can avoid close encounters at all times if they are separated by more than @xmath17 . these two criteria meet at @xmath18 and their physical connection is discussed in @xcite . for systems where there are more than 2 planets , there is yet no theoretical understanding for the origin of dynamical instability ( see attempts by , e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? whatever information we have on how tight one can pack planets in these cases are empirical results from numerical simulations @xcite . these we summarize below . generally speaking , systems with more planets ( @xmath19 ) require larger @xmath8 spacing to stay stable . to compare the various numerical studies , we scale the physical timescale by the orbital period of the inner most planet in the system , @xmath20 ( also see * ? ? ? we also suppress error - bars in the fitting results but caution that there are intrinsically large scatters . * dependence on @xmath21 : two - planet systems can remain indefinitely stable as long as they are above a critical spacing . this may not hold for large - n systems . it appears that a larger spacing does not guarantee eternal peace but only delays the time of instability ( we return to this point in [ subsec : extrapolate ] ) . numerically , the critical spacing at which an average system undergoes their first encounter at time @xmath21 can be modelled as @xcite , @xmath22 though the alternative scalings @xmath23 and @xmath24 $ ] have been adopted by @xcite and @xcite respectively . the numerical coefficients @xmath25 and @xmath26 depend on system parameters like the number of planets , their masses , and orbital eccentricities and inclinations . two results most relevant to our case are @xmath27 for systems of @xmath28 planets all with @xmath29 @xcite , and @xmath30 for @xmath31 planet systems with @xmath32 @xcite . both scalings are obtained for circular , coplanar systems with a uniform planetary mass and a uniform k - spacing . * dependence on @xmath19 : the most comprehensive study to date on this issue is that by @xcite . they found , for @xmath33 , and @xmath34 , the critical separation for a system of circular , coplanar planets goes as @xmath35 this results roughly recovers equation when @xmath36 . the initial rise of @xmath15 with @xmath19 is intuitively understood as the presence of more planetary perturbers exerts more strain on the delicate dynamical systems . the subsequent flattening of @xmath15 with @xmath19 is , however , surprising and not understood . one guess is that dynamical instability is most affected by close neighbours , and the presence of far - away neighbours , once beyond a certain separation , do not matter . for instance , a system of 8 neptunes will space a period range of @xmath37 . with such a large spacing , newly added planets have no chance of participating in any mmrs with the pre - existing planets , and therefore should not adversely affect the stability of the system . * dependence on @xmath16 : it is found that less massive planets , with their smaller hill spheres , need be separated by slightly higher values of @xmath8 . both @xcite and @xcite found that , for every decade of drop in @xmath16 ( within the range of @xmath38 to @xmath39 ) , the critical separation @xmath8 should increase by @xmath40 . * eccentricity / inclination : if the planetary orbits are dynamically hot , more elbow room is needed to maintain stability ( e.g. * ? ? ? * ) . @xcite showed that , at low eccentricities , increasing the average eccentricity by @xmath41 raises @xmath42 by @xmath43 . we know of no study that explores the impact of mutual inclination separately , so we provide some illustrations of this below . returning to our problem at hand a system of 8 flat , circular planets , with masses of @xmath44 , and an inner orbit of @xmath45 au , may remain stable for over 1 gyrs ( or @xmath46 ) , if @xmath47 ( extrapolating from equation [ eq : tauk ] ) . this valueis broadly consistent with the spacing observed in tight kepler systems . however , a more refined approach is required . to compare theoretical results to kepler systems , we need to overcome a few restrictions of previous studies . first , in almost all previous numerical simulations , planets in a given systems are assigned identical masses and identical spacing , in contrast to realistic systems . inhomogeneities in mass and spacing may impact results significantly , e.g. , they will introduce large scatter in stability lifetimes , among systems that have comparable bulk parameters . it is therefore necessary to study system stability in a statistical sense previous investigations have studied only a small number of cases ( typically fewer than a hundred systems ) . we accomplish this by integrating a much larger number of systems ( @xmath48 ) . in addition , we wish to study the impacts of small eccentricity and inclination . lastly , our simulations run for a factor of @xmath49 longer than previous works , up to @xmath50 years . although still shorter than the typical system ages ( a giga - year ) , they provide a better indicator for long - term stability . we generate artificial planetary systems as follows : each system contains @xmath51 planets orbiting around a 1 solar mass star , on coplanar and circular orbits . the number @xmath51 is chosen for a number of reasons : this is the largest number of transiting planets in a multiple system ( e.g. * ? ? ? * ) ; it is close enough to the plateau in equation , the results may be considered valid for all systems with more than @xmath51 planets ; it is computationally feasible . the mass of the host star is taken to be @xmath52 , while the mass of the planets are drawn from a uniform distribution between @xmath53 , consistent with the range of masses observed in kepler systems @xcite . the latter assumption is robust as we know the stability threshold only depends weakly on the planet masses ( [ subsec : previous ] ) . the inner most planet is placed at @xmath45 au . we then determine the spacing , in unit of mutual hill sphere , for an adjacent pair of planets by drawing from a normal distribution with mean @xmath54 and variance @xmath55 , @xmath56 the initial arguments of periapsis and the mean motions are drawn from a uniform distribution in @xmath57 $ ] . we adopt the wisdom - holman integration scheme @xcite , as implemented in the swifter package by david e. kaufmann , an improved version of the original swift package by levison and duncan . the post - newtonian correction is added as an additional potential of the form @xmath58 , and we adopt a time - step of @xmath59 years . integrations are discontinued either when two planets approached one another within one mutual hill radius ( and therefore are destined for a close encounter ) , or when a preset time limit ( usually @xmath60 years ) is reached . we record the stability time of each system , and analyze how it depends on various parameters . of particular interest is the survival probability at a given time . we integrate , up to a time of @xmath60 years , a total of 8,000 circular , coplanar systems ( @xmath61 ) with @xmath62 and @xmath63 . in addition , we simulate 120 such systems , within the range @xmath62 and @xmath64 , up to a time of @xmath50 years . we report the survival probability , the fraction of systems that remain stable up to a time @xmath65 , as a function of planet spacing . overall , a wider @xmath54 leads to a higher survival probability ( fig . [ fig : figure3 ] ) , however , the dispersion in @xmath8 ( @xmath55 ) also impacts stability . for the same @xmath66 , systems with a greater @xmath8 dispersion amongst its pairs are less stable . in other words , a pair of planets with a smaller - than - average @xmath8 have a destabilizing effect on the whole system that out - weighs any corresponding stabilizing effect afforded by a planet pair with a larger - than - average @xmath8 . we find , empirically , that our results can be succinctly summarized using an effective spacing , @xmath67 we arrive at this combination by noticing that , in our simulations , increasing @xmath68 by 1.0 has the same effect on stability as decreasing @xmath66 by @xmath69 . so for systems with a varied spacing , @xmath70 acts as an equivalence of @xmath8 in systems with a uniform spacing . such a combination can also explain the results in @xcite where they made some preliminary explorations on the impact of a @xmath8-dispersion . we can now fit the survival probability reported in fig . [ fig : figure4 ] as a function of @xmath70 ( the functional form being somewhat arbitrary ) @xmath71^{2}}\ , , } \label{eq : sk}\ ] ] for @xmath72 and @xmath73 otherwise . here the dimensionless time @xmath20 , and @xmath74 is the orbital period of the inner most planet ( @xmath2 days in our case ) . we now define a threshold spacing , @xmath75 , as the smallest @xmath70 necessary to achieve a survival probability of @xmath76 at time @xmath65 . this spacing has a similar functional form to that presented in @xcite , but with different numerical coefficients , @xmath77 we compare our critical spacing with those from previous works in fig . [ fig : kcollection ] . our results , with a spread in @xmath8 and in planetary masses , have to be characterized by @xmath78 . this differs from the definition of @xmath15 for other groups , which typically adopt a uniform @xmath8 and mass . however , results broadly agree within error - bars ( as given by * ? ? ? * ) but ours tend to fall below . in other words , at the same spacing ( measured by @xmath70 in our cases ) , inhomogeneous systems ( with mass and spacing dispersions ) remain stable slightly longer than homogeneous systems do . we note that homogenous systems , with their uniform period ratios , tend to be strongly affected by the presence of mean - motion resonances ( * ? ? ? * ; * ? ? ? * see , e.g. ) . this may partially explain the difference . a more important difference between homogenous and inhomogeneous systems is that the former may reach eternal stability at a finite spacing ( as suggested by * ? ? ? * ) , while we do not observe this for the latter ( see [ subsec : extrapolate ] ) . from now on , we consider that our results ( equation [ eq : k50 ] ) supercedes equations - . our scaling is valid for 4 orders of magnitude in @xmath21 ( @xmath79 to @xmath80 , fig . [ fig : figure4 ] ) . in subsequent sessions , we assume that it remains valid for another one order of magnitude , to @xmath81 ( or @xmath82 1 gyrs ) . at this epoch , the critical spacing is @xmath83 . ( with @xmath65 ranging from @xmath84 to @xmath50 yrs ) , for systems with different @xmath54 . each point represents results from some @xmath85 systems , with its horizontal error bars representing the width in @xmath54 and vertical error bars accounting for poisson error in numbers . the solid lines show the aggregated results for a distribution of @xmath55 . the latter is taken to be a gaussian distribution with a mean of @xmath86 and a dispersion of @xmath43 , with the exception of the black curve where @xmath64 . it appears that @xmath54 is a reasonable indicator for system stability . however , system stability also depends on the dispersion in @xmath8 . as the two curves ( dashed and dotted , both for @xmath87 yrs but each with a different @xmath55 ) show , systems with larger dispersion tend to be more unstable . so simply measuring the mean does not reveal the whole picture.,scaledwidth=49.0% ] but with the results plotted against @xmath88 . this combination of parameters is found to describe the numerical results succinctly . different solid lines represent equation , evaluated at their respective times . dashed curve shows the extrapolation of equation to @xmath89 yrs ( or @xmath81 ) . , scaledwidth=49.0% ] spacing required to maintain stability over time @xmath21 ( x - axis , in logarithm ) , as obtained by different groups . the @xcite results ( in red , equation [ eq : tauk ] , with vertical error bars ) are for @xmath29 ; the @xcite results ( in blue , equation [ eq : tauk2 ] ) applies to @xmath90 ; we insert @xmath91 into the fitting formula of @xcite , and modify their expression to suit present notation , to obtain the magenta line . all these simulations are of uniform mass and uniform spacing and therefore are not statistical in nature . our results , @xmath78 ( eq . [ eq : k50 ] ) , are plotted as a black curve . the horizontal extent of each curve also delineates the respective time range probed by different studies the present study probes up to@xmath33 , ten times longer than previous ones . our critical spacing are broadly consistent , though somewhat smaller , than those of previous studies . , scaledwidth=45.0% ] here , we explore quantitatively how eccentricity and inclination affect the critical spacing . we generate 12,000 artificial 7-planet systems with a normal distribution in spacing , @xmath92 , @xmath93 , and a rayleigh distribution in eccentricity and inclination . the scale factors are @xmath94 and @xmath95 , e.g. , @xmath96 we explore the range @xmath97 $ ] , and @xmath98 $ ] ( in radian , corresponding to a range of @xmath99 $ ] ) . the initial longitudes of ascending nodes are randomly distributed between @xmath73 and @xmath100 . given the extra dimension in orbital parameters , it is only feasible to integrate these systems up to a time of one million years ( @xmath101 ) . the survival probability for different types of orbits are shown in fig . [ fig : figure5 ] . individually , mutual inclination and initial eccentricity each has a strong destabilizing effect on stability . for the range of parameters that we explore , the critical spacing can be characterized as @xmath102 where @xmath103 is the corresponding value for circular , coplanar systems for the same @xmath21 ( eq . [ eq : k50 ] ) . an eccentricity dispersion of a mere one percent requires the spacing to expand by one hill sphere , for @xmath104 . in relative terms , mutual inclination is less destructive to stability , by about a factor of @xmath105 . one can also scale the above result in a natural unit of the problem , the hill eccentricity , @xmath106 . for our parameters , ( @xmath107 ) , @xmath108 . the expression for critical spacing , re - cast in these units , is , @xmath109 this leads to a surprisingly geometrical interpretation for our result . for every unit of eccentricity rise ( in unit of @xmath110 ) , the average spacing between two adjacent planets ( @xmath8 ) have to rise by @xmath111 in order to maintain stability ; however , such a rise in eccentricity also reduces the closest approach ( measured from the apo - apsis of the inner planet to the periapsis of the outer one ) by @xmath112 . so a system of mildly eccentric planets behaves similarly as a circular system that have the same closest approach . we do not have a ready explanation for this result . in the following , we assume that the survival stability obtained at these shorter integrations can be extrapolated to longer times , in the form of equation . as a function of orbital eccentricity , at time @xmath113 myrs ( @xmath101 ) . different curves stand for different mutual inclinations . for comparison , we overplot in dots the result of @xcite for coplanar systems . , scaledwidth=49.0% ] we do not yet understand the origin for the scaling in equation . so it may be prudent to ask : can the scaling be extrapolated to as long as one billion years ? for instance , in two - planet systems , it is known that beyond certain spacing the system can remain eternally stable . if high - n planetary system exhibits the same behaviour , equation will break down at a certain spacing . lacking analytical understandings , we turn to numerical simulations for insight . @xcite experimented with equal - mass , constant - spacing 3-planet ( or 5-planet ) systems and found that as planet spacing reaches @xmath114 ( @xmath115 for 5-planet ) , the stability time , clocked at @xmath60 dynamical time at this point , starts to lift off and increases with @xmath8 much more steeply than eq . . it appears that this equation is violated beyond @xmath60 dynamical times , and indefinite stability at large spacing appears possible . however , our results here suggest the opposite . we have integrated non - equal - mass , non - uniform spacing 7-planet systems for up to @xmath80 dynamical times , while still observing that equation holds well . there is no unusual transition that occurs at @xmath60 dynamical times . we suspect that our heterogeneous set - up ( a dispersion of mass and spacing ) opens up a more complex phase - space . such dynamical system may not exhibit a single characteristic timescale , as the more restrictive cases in @xcite . in which case , it may be reasonable to extrapolate eq . by one order of magnitude , to 1 gyrs ( @xmath116 dynamical time ) . previous studies @xcite have identified that mean motion resonances tend to be disruptive for dynamical stability . this is also demonstrated in more detail in @xcite where they showed that , when the positions for kepler-11 planets are shifted away from their current values , they may encounter mean motion resonances and become destabilized . here , using the 8000 circular , coplanar systems that we have produced , we perform a statistical investigation on the effect of resonances on stability ( fig . [ fig : mmr ] ) . it appears that , by @xmath60 yrs , systems that contain near resonant pairs are preferentially destroyed , compared to neighbouring systems . this is an order unity effect for first - order resonances , and weaker , albeit still visible , for second order resonances . similar carving is also seen in our eccentric , inclined ensembles . amongst kepler s multi - planet systems , there has been an observed tendency for planets to ` pile up ' just wide of mean motion resonances @xcite . a number of mechanisms have been proposed for this asymmetry , including resonant repulsion @xcite , stochastic migration @xcite , adiabatic mass accretion @xcite , and planetesimal interactions @xcite . we observe that ( inset in fig . [ fig : mmr ] ) , for circular , coplanar systems , systems with period ratios just inward of mmrs are more severely carved relative to those on the far side of resonances . the asymmetry is more pronounced in slightly eccentric , inclined systems . this asymmetric carving may present another explanation for the observed systems . more investigation is planned in the future . yrs ) the locations of the first order ( dashed red lines ) and the second order resonances ( dashed blue lines ) coincide with reduced survival rates , showing that mean - motion resonances tend to de - stabilize systems . the little inset at the top - right focuses on regions of width @xmath117 around 3 first order mmrs ( 2:1 , 3:2 , 4:3 ) . there is a definite asymmetry : systems on the near side of resonances are preferentially depleted relative to those on the far side . the lower panel is the observed period ratios among all kepler systems . these also show a relative deficit of systems near these resonances in the near - side . the similarity between the two is intriguing.,scaledwidth=45.0% ] observational data from the space telescope was retrieved from the nasa exoplanet archive in june 19 , 2014 . all planet candidates with a koi disposition of `` false positive '' were removed . we are only concerned with high - multiple systems ( @xmath118 planets ) and our sample include 4 six - planet systems , 18 five - planet systems and 55 four - planet systems . there are a total of @xmath119 adjacent pairs , with @xmath120 pairs in @xmath121 and @xmath122 systems alone . to compare against theoretical results in [ sec : tight ] , we require the intrinsic @xmath8-distribution ( as opposed to the apparent one ) of kepler s multi - systems . however , such a distribution is not straightforward to obtain , due to three factors . first , the @xmath8 parameter depends on planetary masses , which are not known for vast majority of planets . secondly , the presence of non - transiting ( ` missing ' ) planets , either because they are too small to be detectable , or because they are unfortunately inclined relative to our line - of - sight , will cause the observed @xmath8 to be artificially inflated . thirdly , one must account for selection effects . the koi sample is severely incomplete for planets with period outward of @xmath123 days @xcite . systems that have tighter - than - average spacing will be preferentially picked up as high multiple systems , while those with larger spacings will not . this tends to skew the apparent @xmath8-distribution toward smaller @xmath8 values in these systems . we overcome the first hurdle by adopting the mass - radius relationship as obtained for planets that have measured masses . recent studies using both transit timing variation @xcite and radial velocity @xcite techniques have shown that planet masses scales approximately linearly with planet radii , or @xmath124 . this applies to planets with sizes of a few earth radii , the vast majority of our sample , and recent studies show that this relation may even extend to planets as small as @xmath125 @xcite . this relation differs from the one that applies to solar system planets @xcite . we plot the resulting apparent @xmath8-distribution in fig . [ fig : figure1 ] . there appears to be a sharp rise in the number of pairs above @xmath126 , and a sharp drop beyond @xmath127 , with some @xmath128 of pairs spaced between @xmath129 . in addition , the apparent @xmath8-distribution for @xmath130 systems and that of @xmath131 systems are statistically different ( ks test returns a value of @xmath132 ) , with the former peaking at @xmath133 while the latter around @xmath134 . it appears that even 4p systems may be heterogeneous , containing both tight spacing systems and large spacing systems . this prompts us to focus our attention exclusively to @xmath121 and @xmath122 systems in subsequent sections . lastly , we notice that our adopted mass - radius relation leads to a sharper peak in @xmath8-distribution , when compared to other power laws . -distribution of high - multiple kepler systems are shown as solid histograms for the 4p and 5p+6p systems . the mass - radius relation is assumed to be @xmath135 . according to ks test , the two histograms are statistically different , with the 5p+ systems more tightly spaced . the dotted curve shows the @xmath136 systems when we adopt instead @xmath137 , a scaling obtained for solar system planets @xcite . this leads to a more diffuse @xmath8-distribution . , scaledwidth=49.5% ] to overcome the second and the third hurdles , it is necessary to `` observe '' synthetic populations to extract their apparent @xmath8-distributions . we now describe our procedure in constructing and observing these synthetic populations . each planetary system orbit around a central star that is sun - like , and have planets whose radii are distributed based on the determination in @xcite , namely , with equal probability per logarithmic bin between @xmath113 and @xmath138 , and a linearly decreasing probability between @xmath139 to @xmath140 . the mass of each planet is then determined using our default mass - radius expression . the orbital inclination , measured relative to a fixed plane , satisfies a rayleigh distribution with scale parameter @xmath95 ( eq . [ eq : probe ] ) . planets are placed in circular orbits , and all euler angles are uniformly distributed . the period of the inner - most planet in each system is drawn from a rayleigh distribution with a scale parameter of 5 days . this distribution is a good representation for our sample of planetary systems , but our conclusion is not sensitive to this particular choice . to assign periods to the remaining planets , we assign a @xmath8 value that satisfies equation with a mean @xmath54 and a variance @xmath55 . each system is assumed to extend out to well beyond the observing boundary . with these choices , our average system , with @xmath141 days and @xmath142 , contain roughly 7 planets ( ranging from 6 - 10 planets ) inward of @xmath143 days . for the entire ensemble , the period distribution is flat in logarithmic space , consistent with observational determinations @xcite . we then identify planets that transit ( impact parameter smaller than the stellar radius ) . not every planet that transits may be identified by the kepler pipeline . we correct for this by adopting the detection completeness as a function of planet size and orbital period , as obtained in fig . 2 of @xcite . this effectively limit the detection sphere to periods inward of @xmath123 days , and miss some small planets . our procedure therefore accounts for both the missing planets and the selection bias . the @xmath8-distribution from these detections are then compared against that in fig . [ fig : figure1 ] . we find that the best fit , using only 5p+ systems , is obtained when @xmath144 . this corresponds to an effective spacing of @xmath145 . as is shown in fig . [ fig:3 ] , the observed @xmath8-distribution is always broader than its intrinsic counterpart . the broadening is caused by intervening planets missed by transit surveys : our simulations show that for low inclinations ( @xmath146 ) , @xmath147 of the apparent 5p and 6p systems contain one or more missing planets hidden in - between the transiting planets . moreover , the apparent tight spacing in high multiples is not a result of selection bias , namely , only systems that have intrinsically the tightest distribution can fit inside the detection sphere ( @xmath123 days ) and be observed as a high - multiple system ( see [ sec : discussion ] ) . we can translate the @xmath8-distribution to a more intuitive unit , the number of planets within a decade of semi - major axis , typically located from 0.1 to 1.0 au for our case . this number is marked as planets per sma decade in multiple figures . the 5p+ systems contain between 6 to 10 planets per sma decade , with a mean of @xmath115 . contrast this with only 4 planets per decade in the terrestrial region of the solar system . our determination of the intrinsic @xmath8-distribution is not affected by our choice of inclination dispersion , perhaps a bit paradoxically . a higher dispersion means more planets in a given system escape detection in a transit survey , and this should have increased the average @xmath8 value . however , this also disqualifies such a system from our high multiple sample . as a result , the intrinsic @xmath8-distribution we obtain , by focussing on high multiples only , is not much affected . , for the 5p+ systems fail to account for some of the more widely spaced 4p pairs . here , we adopt @xmath148 the @xmath8-distribution does not vary much when the inclination dispersion is varied . , scaledwidth=49.0% ] gyro - chronology studies , the dating of stars using their stellar rotation periods , reveal that stars in the kepler field , at least ones that have measurable rotation periods , have typical ages of a couple gyrs @xcite . this motivates us to assume a default age of 1 gyrs for all kepler systems . this may be an underestimate in many cases . older stars are much less magnetically active and therefore less variable . they also rotate with longer periods . as a result , they may not be selected for gyro - chronology studies and results from the latter may be biased towards younger stars . fortunately , an older age simply strengthens the argument we present below . if all kepler high multi systems are circular , coplanar , equation predicts that @xmath76 of the planetary systems can remain stable to beyond @xmath149 yrs ( @xmath150 ) if @xmath151 . this is surprisingly close to our best fit for kepler high multis , @xmath145 . the proximity is even more striking if one allows for just a small amount of eccentricity and non - coplanarity in the systems . adopting the measurements by @xcite for ttv pairs , @xmath152 , equation requires that @xmath153 . this observation , that kepler high multis are perched at the cliff of instability ( see also * ? ? ? * ; * ? ? ? * ) , as we suggest below , is most likely the result of continuous sculpting . this suggestion is the main thesis of this work . our intrinsic distribution exhibits a sharp fall - off at larger @xmath8 ( see fig . [ fig:3 ] ) , similar to the observed population . we believe this fall - off is genuine , not a result of observational bias . when synthetically observing populations of planets with a range of @xmath54 and @xmath55 , we find that even systems with spacing as large as @xmath154 ( fixing @xmath155 ) can be observed as 5p+ only @xmath156 less frequently than our best fit ( @xmath157 ) . their presence should therefore be detectable . the lack of such wide spacing system in the observed high multis is therefore likely genuine . the formation of planets themselves is not yet understood and different theories have been proposed @xcite . however , it looks likely that they were made in the first few myrs . this is corroborated by the presence of hydrogen envelopes on these low - mass planets , for both neptunes ( e.g. * ? ? ? * ) and super - earths @xcite . how does such a formation time jibe with the orbital spacing of the observed systems ? the nascent disk can weed out ( and possibly reform ) systems that are unstable within a few million years ( @xmath158 , eq . [ eq : k50 ] ) , but it should not discriminate among systems that have spacing wider than this . on the other hand , while it is natural that only systems with spacing wide enough to be stable over their lifetimes can be observed , it is striking how the observed systems skirt the edge of stability . this chain of thought then leads us to propose the following scenario . planetary systems were initially formed with a range of @xmath159 . this distribution is continuously sculpted by dynamical instability . so by 1 gyrs , the remaining population will exhibit a sharp cut - off at the border of stability . we call this the sculpting hypothesis. under this hypothesis , it is possible to reconstruct the initial planet population , after we posit a number of assumptions , some more sound than others . we assume that the all systems have the same small @xmath8-dispersion as we observe in today s high multis , @xmath160 . since orbital spacing in a system likely reflects its formation environment , we expect the spacings to be similar within a given system . we further assume that all systems at birth have the same dispersion in eccentricity as the currently observed ttv sample , @xmath161 , and a comparable dispersion in inclination . to reverse the effect of carving , we augment the current planet population by three populations that are similar in number but shifted to a smaller @xmath70 by @xmath162 , for the three logarithmic time intervals between @xmath163 and @xmath164 yrs . the value @xmath165 , according to our numerical experiments , is the range of spacing that is carved away by the passing of each logarithmic interval . and the initial @xmath163 yrs is the putative lifetime of our nascent disks . the resulting initial population is illustrated in fig . [ fig : figure6 ] . in this model , by 1 gyrs of age , roughly @xmath128 of the original systems have been disrupted . this fraction could rise or drop depending on what we assume for the initial @xmath70 distribution . , namely , the observed ( histogram ) and the underlying ( green curve ) spacing distributions for 5p and 6p systems . the dashed curves , from bottom to top , represent our reconstruction of the earlier spacing distributions at , @xmath50 , @xmath60 and @xmath166 years , respectively . see text for details . the primordial population ( posited to be the one at @xmath166 yrs ) are in average more tightly spaced , and contain roughly 4 times more systems than those remain today . , scaledwidth=49.0% ] what happen to systems that experience dynamical instability , in the interval since their birth and today ? once two planets are excited to crossing orbits , they will physically collide with each other after a short time , as determined by the area of the planet subtended on the sky , @xmath167 this expression assumes that the planets have acquired significant eccentricities and inclinations before impact , and that there is little gravitational focussing . both assumptions are born out by our numerical simulations . in particular , the latter assumption is reasonble because at collision , their relative velocities can be higher than their surface escape velocities ( @xmath168 , see , e.g. * ? ? ? * ) , since the keplerian orbital speed is of order @xmath169 at the relevant distance . such a collision will likely be destructive , as opposed to conglomerative @xcite . some planets may be lost and massive debris may be produced . the former will increase the spacing within the system , while the latter will tend to cool down the orbit and prevent further instabilities from happening . at the moment , we are incapable of predicting the outcome of such a collision , but speculate that this may explain the abundant low - multiple systems one observes in the field . we speculate briefly on the physical environment that can give rise to the compact spacing of systems . if our sculpting hypothesis proves correct , this is not restricted to only the high multiple systems , but likely applies to all close - in planetary systems . the orbital spacings between terrestrial planets are wide , with @xmath170 . these values have been reproduced by a number of simulations where these planets conglomerate out of a sea of smaller bodies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in these simulations where there is zero or little dissipation acting on the planetary orbits , the proto - planets can stir each other up to of order their surface escape velocities . their orbits , being dynamically hot , require larger spacing to avoid further disruption . this reasoning is supported by simulations of @xcite , which we discuss below . starting from a population of mars - sized proto - planets , but exerting an artificial dissipation on the planet eccentricity with timescale @xmath171 , @xcite showed that , terrestrial - like systems ( @xmath172 ) can be produced if @xmath171 is relatively weak ( more than a million times the orbital period ) , and closely - packed systems result if damping is strong . when @xmath173 orbital time , in particular , the simulations produce systems that are spaced between @xmath28 to @xmath174 much like the initial population that we inferred in fig . [ fig : figure6 ] . at @xmath45 au , this corresponds to a damping time of @xmath175 yrs . in systems , the primordial disk is likely more massive than the terrestrial system studied by @xcite , so the requisite e - damping time may be even shorter . to put this timescale into context , the expected e - damping time from type - i migration in a minimum mass solar nebula is @xmath176 yrs @xcite . so if our sculpting hypothesis is correct , we infer that the formation environment of the planets have to be moderately dissipative , with an eccentricity damping timescale that is longer than the type - i damping timescale , but much shorter than the million - year lifetime of the nascent disks . we can look at our results in a different way . the observed spacing could be used to constrain the values of eccentricity and mutual inclination in ( and only in ) these high multiple systems . inverting equation , adopting @xmath177 ( required at 1 gyrs ) and @xmath178 ( observed at 1 gyrs ) , we find the following independent limits , @xmath179 although this constraint is obtained for our choice of mass - radius relationship , it likely applies even when the latter deviates from our choice . similar constraints on orbital shapes were also obtained by @xcite for a specific planetary system , kepler-11 . if @xmath180 , as characteristic of 2-body relaxation in disks @xcite , we would obtain @xmath181 and @xmath182 . these are indeed very cold dynamical systems . how do these constraints compare to what we have found via other channels ? most previous studies have not isolated high - multiple systems from the general catalogue and their results are therefore not directly comparable , though a number of studies of planet mutual inclinations have indicated low values ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? there are only a few studies on planet eccentricities . studying the distribution of transit durations in kepler candidates have led @xcite to conclude that @xmath183 . however , this result is heavily plagued by uncertainties in stellar radius determinations and is likely an over - estimate . ttv studies @xcite , on the other hand , have led to a much smaller eccentricity dispersion ( @xmath161 ) , for a sub - population of planet pairs that are near mean - motion resonances . in fact , this small but non - zero dispersion partially motivates us in claiming that the observed systems lie on the edge of instability . ttv studies also reported the existence of another sub - population where the eccentricities are substantially higher ( @xmath184 ) , but these likely occur exclusively in low - multiple systems . the architecture of kepler multiple systems contains important pieces of clue for planet formation . in this work , we show that high - multiple kepler systems have spacings that narrowly afford them dynamical stability in the giga - year timescale . this leads us to suggest that many more planetary systems were initially formed with even smaller spacing , but they have been since whittled down by dynamical instability , possibly producing some of the low - multiple kepler systems we see today . if this is true , the final formation stage of these planets proceed in highly dissipative environment , with eccentricity damping timescales much shorter than the disk lifetimes . along the way , our numerical explorations have uncovered a number of interesting features in the stability of high - n systems . consistent with previous investigations , we find that system stability time , quantified here as the time by which @xmath76 of the initial systems remain stable , rises sharply with the spacing . when the spacings in a given system are not uniform but exhibit a spread , one can easily quantify the effect by adopting an effective spacing , @xmath185 . we give fitting formula for stability time when the planetary orbits are slightly eccentric and/or mutually inclined . eccentricity , even at the level @xmath186 , affects critical spacing significantly . as a result , we are able to place a stringent limit on the eccentricity of high - multiple kepler systems , @xmath181 . we find that systems that contain mean - motion - resonance pairs are preferentially destabilized . a large number of both theoretical and observational questions remain unanswered . what provides the strong eccentricity dissipation in planet formation ? what is the origin of dynamical instability in high - multiple systems ? what are the final outcomes in these dynamically unstable systems ? how do these outcomes compare against the observed systems ? we thank the anonymous referee for a meticulous and thoughful set of comments . bp thanks nserc for an undergraduate summer fellowship , and nikhil mahajan for assistances . yw acknowledges the nserc funding , as well as conversations with yoram lithwick , doug lin ( who pointed out that johannes kepler did ponder about the planet packing problem ) and scott tremaine .
transition metal dichalcogenides ( tmds ) such as molybdenum disulfide ( mos@xmath0 ) have layered structures , where the atoms within a tmd monolayer form a covalently bonded planar network , and the interaction between these layers is a weak , van der waals interaction.@xcite a monolayer of mos@xmath0 consists of a layer of molybdenum atoms sandwiched between two layers of sulfur atoms . monolayers can be exfoliated through micro - mechanical cleavage , similar to graphene or boron nitride.@xcite unlike graphene ( a metal ) , or boron nitride ( an insulator ) , mos@xmath0 is a semiconductor . moreover , whereas bulk mos@xmath0 has an indirect band gap ( 1.2 ev ) , monolayer mos@xmath0 has a direct band gap ( @xmath21.8 - 1.9 ev ) , and shows a strong optical absorption and luminescence.@xcite at present mos@xmath0 , and tmds in general , are vehemently pursued as promising materials for applications in electronics and optoelectronics.@xcite contacting mos@xmath0 to metal electrodes proves to be a problem ; it tends to produce unexpectedly high interface resistances , indicative of a high schottky barrier at the interface.@xcite a high barrier could be caused by strong interface bonding creating interface states that pin the fermi level,@xcite or by weak bonding creating a potential step due to pauli repulsion at the interface.@xcite the nature of the interaction at the mos@xmath0/metal interface is far from trivial . on the one hand , one could argue that , as mos@xmath0 has no dangling bonds at its surface , its interaction with metal substrates should be weak and van - der - waals - like . on the other hand , many metal species form ( di)chalcogenide compounds,@xcite and when adsorbing mos@xmath0 onto a metal substrate , there could be a competition between the metal surface and the mo atoms for interacting with the sulfur atoms at the interface . in that case , not only the mos@xmath0/metal bonding would be a much stronger chemical bonding , but also the structure and electronic structure of the mos@xmath0 adsorbate could be significantly perturbed . in this paper we explore the adsorption of mos@xmath0 on a variety of metal substrates by means of first - principles density functional theory ( dft ) calculations , following up on work briefly reported in a short paper.@xcite previous dft studies have concentrated foremost on the schottky barrier formed at mos@xmath0/metal interfaces using the local density approximation ( lda).@xcite lda gives a reasonable description of the adsorption of graphene and _ h_-bn on metal surfaces , but such results can not be generalized to other systems , as it is known that lda often leads to an unrealistic overbinding.@xcite other studies have used a generalized gradient approximation ( gga ) functional , such as pbe,@xcite which apparently works well for tmds adsorbed on metals,@xcite although it generally gives bad results for weakly bonded systems.@xcite here we focus on the interface interaction and its implications for the structure and electronic structure of the mos@xmath0 adsorbate and the schottky barrier . we choose a wide range of metal substrates : the ( 111 ) surfaces of al , ni , cu , pd , ag , pt and au , and the ( 0001 ) surfaces of mg , ti , and co , which are expected to have a wide range of interaction strength with the adsorbate . as the interface interaction can vary from weak ( physisorption ) to strong ( chemisorption ) , it is a priori not clear which dft functional describes such bonding . we test and compare results obtained with a van der waals functional , designed to describe weak , van der waals , interactions,@xcite to results obtained with gga and lda functionals , which are conventionally used to describe chemical bonding . we assess the importance of van der waals interactions for the interface interaction , and evaluate its effect on the structure and electronic structure of the mos@xmath0 adsorbant . we consider the situation where a mos@xmath0 layer is adsorbed as a whole on a metal substrate , making it more likely that the integrity of the mos@xmath0 layer is preserved in the adsorption process . if the mos@xmath0/metal interaction is not too strong , and the mos@xmath0 and metal surface lattices are not matched , the interface structure is likely to be incommensurable . in a supercell calculation one is forced to approximate such a structure by a commensurable one . previous calculations have used small supercells , where in some cases appreciable artificial strain is generated because of the mismatch between the mos@xmath0 and the metal surface lattices.@xcite we apply a strategy for choosing supercells such that the artificial strain is minimal , and test the influence of strain on the electronic properties of the interface . this paper is organized as follows . section [ sec : calculations ] describes the dft calculations , comparing different functionals in sec . [ sec : functionals ] and discussing the effect of lattice mismatch in sec . [ sec : mismatch ] . results are discussed in sec . [ sec : results ] , with the metal / mos@xmath0 interaction in sec . [ sec : bonding ] and its effects on the interface potential step and the schottky barrier in sec . [ sec : sb ] . strong chemisorption is discussed in more detail in sec . [ sec : ti ] , and a summary and the conclusions are presented in sec . [ sec : conclusions ] . we calculate ground - state energies and optimize geometries at the density functional theory ( dft ) level , using projector - augmented waves ( paws ) as implemented in the vasp code.@xcite the plane - wave kinetic - energy cutoff is set at 400 ev . the surface brillouin zone is integrated with the methfessel - paxton technique using a smearing parameter of 0.05 ev,@xcite and a @xmath3-point sampling grid with a spacing of 0.01 @xmath4 . the mos@xmath0/metal interface is modeled as a slab of 4 - 6 layers of metal atoms with one or two layers of mos@xmath0 adsorbed on one side and a vacuum region of @xmath212 . the in - plane supercell is chosen such as to minimize the mismatch between the mos@xmath0 and metal lattices , which is discussed in more detail in sec . [ sec : mismatch ] . a dipole correction is applied to avoid spurious interactions between periodic images of the slab.@xcite we allow the positions of the atoms to relax until the force on each atom is smaller than 0.01 ev@xmath4 , except for the bottom layer of metal atoms , whose positions are kept fixed . the electronic self - consistency criterion is set to 10@xmath5 ev . it is well known that commonly used dft exchange - correlation functionals , based upon lda@xcite or gga,@xcite give decent descriptions of covalent and ionic bonding , but they may fail for weakly bonded systems , as such functionals do not contain a description of van der waals interactions . for example , gga functionals such as pw91 or pbe,@xcite do not capture the bonding between _ h_-bn or graphene layers , nor that between _ h_-bn or graphene and transition metal(111 ) surfaces.@xcite a priori we do nt know how important van der waals interactions are in the bonding between mos@xmath0 and a metal surface . in sec . [ sec : functionals ] we compare results obtained using a van der waals density functional ( vdw - df),@xcite with results obtained with gga and lda functionals . structure with corresponding plane - averaged electrostatic potential @xmath6 . @xmath7 is defined as the difference between the work function on the metal side and on the mos@xmath0 side . ] one way of visualizing bonding at a mos@xmath0/metal interface is by the electron density difference @xmath8 where @xmath9 , @xmath10 , and @xmath11 are the electron densities of mos@xmath0 adsorbed on the metal , of the metal surface and of the free standing mos@xmath0 , respectively . the system as a whole is neutral , and @xmath12 is localized around the metal / mos@xmath0 interface , i.e. @xmath13 for @xmath14 sufficiently far from the interface . solving the poisson equation with @xmath15 as source then gives a potential step across the interface @xmath16 here @xmath17 is the direction normal to the interface , @xmath18 is the interface area , and @xmath19 is the difference between the asymptotic values of the potential left and right of the interface . figure [ fig : deltav ] also illustrates an alternative expression for the potential step @xmath20 where @xmath21 , @xmath22 are the work functions of the clean metal surface , and of the metal surface covered by mos@xmath0 , respectively . a practical way of obtaining work functions from dft calculations is to track the plane - averaged electrostatic ( hartree ) potential @xmath6 into the vacuum , see fig . [ fig : deltav ] , where typically the asymptotic value is reached with a few from the surface . in converged calculations the expressions of eqs . [ eq : sb2 ] and [ eq : sb3 ] give results that are with a few mev of one another . the interface potential step plays an important role in determining the schottky barrier , i.e. , the potential barrier for charge carrier transport from a metal into a semiconductor . the schottky barrier for electrons can be written as @xmath23 with @xmath24 the fermi level and @xmath25 the electron affinity of mos@xmath0 . there are several ways to extract the schottky barrier height from mos@xmath0/metal slab calculations . one could determine @xmath26 by measuring @xmath27 in the band structure or in the density of states of the mos@xmath0/metal slab , as in refs . and . in order to identify the mos@xmath0 related states , one needs to calculate the amplitudes of the projections of the wave functions of the slab on the mos@xmath0 layer . there is always some arbitrariness involved in such a projection if the adsorbate and the substrate are in close connection . in addition , identification of states belonging to the adsorbate is possible only if its electronic structure is not significantly perturbed in the adsorption process , which is only the case if the adsorbate is ( weakly ) physisorbed on the substrate.@xcite in practice we find that this procedure for obtaining the schottky barrier height at mos@xmath0/metal interfaces is not sufficiently accurate when applied to the ( projected ) density of states , and of practical use only when applied to the ( projected ) band structure of a small supercell . in contrast , starting from the right - hand side of eq . [ eq : sb ] , @xmath21 , @xmath25 and @xmath7 are easily obtained in separate calculations on the clean metal surface , the free - standing mos@xmath0 layer , and the mos@xmath0/metal slab , respectively . of course , if the mos@xmath0 electronic structure is very strongly perturbed by adsorption , one has to reconsider the definition of the schottky barrier , see sec . [ sec : ti ] . in the following the potential step @xmath19 is used to characterize the mos@xmath0/metal interface , along with the binding energy and the structure . an alternative way of locating the conduction band edge @xmath25 in a mos@xmath0/metal slab calculation , without having to resort to wave function projections , is by aligning the core levels of the mo or s atoms in the slab with the corresponding core levels in free - standing mos@xmath0 . it allows us to compare the densities of states of free - standing and adsorbed mos@xmath0 , see sec . [ sec : sb ] . in principle , this procedure also allows for calculating the schottky barrier height . as the method described in the previous paragraph , this only makes sense if the mos@xmath0 electronic structure is not perturbed too strongly by the adsorption . materials such as graphite , _ h_-bn , and mos@xmath0have a layered structure , where the atoms within one layer form strong covalent bonds , but the interaction between the layers consists of weak , van der waals , forces . common gga functionals , such as pbe,@xcite lack a description of van der waals interactions , which results in a severe underestimation of the interlayer binding energy in graphite and _ h_-bn , and an overestimation of the interlayer bonding distance.@xcite similar problems are encountered when graphene or _ adsorbed on a metal substrate.@xcite the lda functional also lacks a description of van der waals interactions , but it , somewhat fortuitously , gives reasonable binding energies and geometries for graphite , _ h_-bn , and for the adsorption of these materials on metals.@xcite in general however , the lda functional tends to overestimate binding energies , which is regularly accompanied by an underestimation of the bonding distance . [ sec : functionals ] ccccccccccccc & & & & + + & @xmath28( ) & @xmath7(ev ) & @xmath29(ev ) & @xmath28( ) & @xmath7(ev ) & @xmath29(ev ) & @xmath28( ) & @xmath7(ev ) & @xmath29(ev ) & @xmath28( ) & @xmath7(ev ) & @xmath29(ev ) + lda & 2.6 & 0.54 & @xmath300.27 & 2.5 & 0.10 & @xmath300.33 & 2.2 & 0.50 & @xmath300.69 & 2.4 & 0.85 & @xmath300.43 + pbe & 3.3 & 0.38 & @xmath300.02 & 2.8 & 0.10 & @xmath300.08 & 2.3 & 0.34 & @xmath300.25 & 2.6 & 0.66 & @xmath300.11 + vdw - df & 2.9 & 0.41 & @xmath300.33 & 2.8 & 0.11 & @xmath300.35 & 2.4 & 0.30 & @xmath300.54 & 2.6 & 0.71 & @xmath300.43 + [ tab : functionals ] many of these problems are mitigated when using vdw - dfs,@xcite which , for instance , describe the bonding in graphite very well.@xcite the exchange - correlation energy in a vdw - df takes the form @xmath31 where @xmath32 , @xmath33 and @xmath34 describe the exchange part , and the local and nonlocal electron - electron correlations , respectively . for @xmath34 we use the vdw kernel developed by dion _ _ @xcite and for @xmath33 the correlation part of the lda functional . for the exchange part @xmath32 we use the optb88 functional.@xcite the opt88-vdw - df has given good results for binding energies and geometries of graphite , _ h_-bn , and the adsorption of these materials on metals.@xcite in the following we test the gga / pbe , lda , and opt88-vdw - df functionals for the adsorption of mos@xmath0 on metals . as test cases we use the 4d and 5d metals ag , au , pd , and pt . we place a mos@xmath0 monolayer on top of the ( 111 ) surface of these metals , choosing a @xmath35 in - plane mos@xmath0 unit cell on top of a @xmath36 ( 111 ) surface cell . the in - plane mos@xmath0 lattice parameters are kept at their optimized values for a free - standing layer , and the in - plane metal lattice parameter is adapted accordingly . the size of the adaption is maximal for au , where it results in a compression of the in - plane au lattice by 4.2% . the effects of this artificial strain are discussed in the next section . figure [ fig : binding](a ) shows the binding curves of mos@xmath0on au and pt(111 ) for the three functionals . the binding energy is defined as the total energy per mos@xmath0formula unit of the metal / mos@xmath0slab minus the total energies of the clean metal slab and the free - standing mos@xmath0layer , as a function of the distance @xmath37 between the top layer of metal atoms and the bottom layer of sulfur atoms . for mos@xmath0 on au(111 ) , pbe gives virtually no bonding , and opt88-vdw - df gives a sizable binding energy . the opt88-vdw - df result suggests that mos@xmath0 is physisorbed on au(111 ) , with van der waals interactions playing the decisive role in the bonding . pbe does not capture this at all . lda gives a equilibrium binding distance that is 0.3 smaller , and an equilibrium binding energy that is 32% larger . for mos@xmath0 on pt(111 ) all three functionals give equilibrium bonding distances that are shorter than for mos@xmath0 on au(111 ) , and a bonding that is stronger , which suggests that mos@xmath0 may be weakly chemisorbed on pt(111 ) . pbe and opt88-vdw - df give a similar equilibrium distance , although pbe captures only 26% of the binding energy , indicating that van der waals interactions still play an important role here . lda gives a similar binding energy as opt88-vdw - df , but an equilibrium binding distance that is 0.2 smaller . table [ tab : functionals ] shows the equilibrium binding distances and energies obtained with the three functionals for mos@xmath0on au , ag , pd , and pt(111 ) . treating the results for opt88-vdw - df as a benchmark , pbe is seen to severely underestimate binding energies , whereas lda gives quite reasonable binding energies . lda however gives binding distances that are up to 0.3 shorter than those obtained with opt88-vdw - df , in particular for cases where the bonding is weak , such as au and ag . in contrast , pbe gives binding distances that are similar to those obtained with opt88-vdw - df , except for au , where pbe essentially fails to give any significant bonding . potential steps @xmath7 as a function of the distance @xmath37 between the top layer of metal atoms and the bottom layer of sulfur atoms , calculated according to eq . [ eq : sb3 ] , are shown in fig . [ fig : binding](b ) for au and pt . the curves for the pbe and the opt88-vdw - df functionals are within 0.05 ev of one another in the range @xmath38 - 3 , whereas lda gives a potential step that is 0.10 - 0.15 ev higher . in view of the considerable differences in the binding curves for these three functionals , the differences in the potential steps are remarkably small . this is true for all metal substrates listed in table [ tab : functionals ] . in ref . the main contribution to the potential step in the adsorption of _ h_-bn on metal substrates was attributed to pauli repulsion . this can be modeled by an electron density that is obtained by anti - symmetrizing the product of the metal and the adsorbate wave functions . as long as these wave functions do not strongly depend on the functional , the electron density and the potential step are also relatively insensitive to the functional used . this is unlike the total energy , which for a given electron density is very dependent on the functional . for the potential step to be accurate it is however important to obtain the proper equilibrium binding distance.@xcite the absolute values of the binding energies given in table [ tab : functionals ] are much smaller than what one expects to find for true chemical bonding . the differences between the values obtained with pbe and opt88-vdw - df indicate that van der waals interactions play a significant role in the bonding . with such a weak metal / adsorbate bonding it is unlikely that the metal substrate can enforce its lattice periodicity onto the mos@xmath0 overlayer . therefore , a metal / mos@xmath0 interface very likely becomes incommensurable if the metal / mos@xmath0 lattice mismatch is substantial . in electronic structure calculations one is forced to use commensurable structures to model incommensurable systems . obviously care must be taken to ensure that the artificial strain introduced this way , does not alter the electronic structure in an unrealistic way . based upon previous experience , we expect that modifying the in - plane lattice constant of a close - packed metal surface by a few percent only affects its electronic properties mildly.@xcite in contrast , changing the lattice parameter of mos@xmath0 by just one percent already alters the band gap by @xmath20.1 ev , and changes it from direct to indirect . a larger change in the lattice parameter has an even more dramatic effect . applying a tensile strain of @xmath39% to mos@xmath0 reduces the band gap by @xmath40 ev.@xcite monolayer in a @xmath41@xmath42@xmath41 cell , where the direct band gap appears at @xmath43 ; ( b ) band structure of mos@xmath0/au(111 ) with the in - plane au lattice compressed by 4.2% to match the mos@xmath0 lattice ; the blue color indicates the weight of a projection of the wave functions on the mos@xmath0 sites ; ( c ) as ( b ) but with the mos@xmath0 lattice stretched by 4.2% to match the au(111 ) lattice.,scaledwidth=50.0% ] as an example , the pbe optimized in - plane lattice parameters of mos@xmath0 and au(111 ) are 3.19 and 2.88 . placing a @xmath44 mos@xmath0 cell on top of 2@xmath422 au(111 ) surface cell then leads to a lattice mismatch of 4.2% . figure [ ref : strain](b ) shows the electronic band structure of mos@xmath0/au(111 ) where the in - plane au(111 ) is compressed by 4.2% to match the lattice parameter of mos@xmath0 . as the interaction between mos@xmath0 and the au surface is relatively small , it is not surprising to see that the band structure of adsorbed mos@xmath0 resembles that of free - standing mos@xmath0 , shown in fig . [ ref : strain](a ) . note that in the @xmath45 mos@xmath0 cell the bands are folded such that the direct band gap appears at the @xmath43 point . the work function of clean au(111 ) is changed by only 0.08 ev by the 4.2% compression of its lattice . for comparison , [ ref : strain](c ) shows the band structure of mos@xmath0/au(111 ) when mos@xmath0 is stretched by 4.2% to match the au(111 ) lattice . clearly the band structure of mos@xmath0 is now changed significantly . it no longer shows a direct band gap at @xmath43 , but an indirect band gap , and the size of the band gap is reduced to @xmath21 ev , which is consistent with previous studies.@xcite the schottky barrier for electrons ( the energy difference between the bottom of the conduction band and the fermi level ) , which is a sizable 0.7 ev in fig . [ ref : strain](b ) , is reduced to zero in fig . [ ref : strain](c ) as in ref . . the latter is clearly unphysical : one would not expect a high work - function metal such as au to form a barrierless contact for electrons . indeed experimentally au is found to form a substantial schottky barrier with mos@xmath0.@xcite in the following we base the in - plane lattice constant of the mos@xmath0/metal slab on the optimized values of free - standing mos@xmath0 , which are 3.13 , 3.18 , and 3.19 for the lda , optb88-vdw - df , and pbe functionals , respectively . experimentally reported bulk mos@xmath0 lattice constants are in the range 3.13 - 3.16 ,@xcite suggesting that the lda result may be more accurate and both pbe and the vdw functional are overestimating the lattice constant somewhat . in making a commensurable structure we adapt the metal to the mos@xmath0 lattice . to minimize the artificial strain that is introduced by this adaptation , we construct in - plane supercells following the procedure of ref . . we denote a basis vector of a mos@xmath0 supercell by @xmath46 , with @xmath47 the basis vectors of the primitive cell , and @xmath48 integers . similarly , @xmath49 is a basis vector of a metal surface supercell , with @xmath50 the basis vectors of the primitive cell , and @xmath51 integers . we search for a set of values for @xmath52,@xmath53,@xmath54 , and @xmath55 , such that the difference between the mos@xmath0 and the metal supercell basis vectors is less than a margin @xmath56 , @xmath57 we then rotate the mos@xmath0 lattice by an angle @xmath58 such , that the directions of the @xmath59 and @xmath60 vectors coincide . because of the symmetry of the lattice the second basis vector of the supercell is easily obtained by a 120@xmath61 rotation , @xmath62 . the commonly used surface science notation of this supercell is a @xmath63 mos@xmath0 lattice on top of a @xmath64 metal lattice , where @xmath65 and @xmath66 . . in - plane supercell defined by the mos@xmath0 lattice vector @xmath67 , where @xmath46 and the metal lattice vector @xmath49 . @xmath56 gives the mismatch between the mos@xmath0 and metal lattices , eq . [ eq : sb6 ] ( pbe values ) . [ cols="<,^,^,^,^",options="header " , ] [ tab : structures ] the displacements are quite large for the mos@xmath0/ti(0001 ) structure , indicating that there is a significant distortion of the geometries of both the mos@xmath0 adsorbate and the ti surface , which strongly suggests that mos@xmath0 is chemisorbed on ti . at the opposite end of the scale we find mos@xmath0/au(111 ) , where the atomic displacements are small , indicating that here we are in the physisorption regime . the behavior of the other metal substrates is in between these two extreme cases but more to the physisorption side . the 3d transition metals co , ni , cu and the low work function simple metal mg show somewhat larger distortions than the 4d and 5d metals pd , ag , pt and the simple metal al . table [ tab : deltav ] gives the interface potential steps @xmath19 created by the adsorption of mos@xmath0 on a metal substrate . this potential step strongly influences the schottky barrier at metal / mos@xmath0 contacts , see eq . [ eq : sb ] , and as such it plays an important role in the physics of mos@xmath0 semiconductor devices . the potential steps can be divided into two groups , i.e. , positive @xmath19 for metals with a high work function , and negative @xmath19 for low work function metals . a positive @xmath19 means that adsorption of mos@xmath0 effectively lowers the work function of the substrate . the mos@xmath0 layer has no intrinsic dipole moment perpendicular to the layer that could create such a potential step . so the work function lowering is a purely electronic effect that results from the displacement of surface electron density into the metal by physisorption of the adsorbate . lccccccc & & + + & @xmath68 & @xmath69 & @xmath70 & @xmath71 & @xmath69 & @xmath70 + mg & @xmath72 & @xmath73 & @xmath74 & 3.96 & @xmath75 & @xmath76 + al & @xmath77 & @xmath78 & @xmath79 & 4.20 & @xmath80 & @xmath81 + ag & @xmath82 & @xmath83 & @xmath84 & 4.82 & @xmath85 & @xmath86 + ti & @xmath87 & @xmath88 & 0.53@xmath89 & 4.80 & @xmath90 & 0.54@xmath89 + cu & @xmath91 & @xmath92 & @xmath93 & 5.10 & 0.39 & @xmath86 + au & @xmath94 & @xmath95 & @xmath96 & 5.58 & @xmath97 & @xmath98 + pd & @xmath99 & @xmath92 & @xmath100 & 5.48 & @xmath101 & @xmath102 + pt & @xmath103 & @xmath104 & @xmath105 & 5.96 & @xmath96 & @xmath106 + co & @xmath107 & @xmath108 & @xmath109 & 5.42 & @xmath110 & @xmath111 + ni & @xmath112 & @xmath113 & @xmath114 & 5.40 & @xmath115 & @xmath116 + @xmath89 see sec . [ sec : ti ] . [ tab : deltav ] this effect is known as the push - back effect or the pillow effect , which is a general phenomenon observed in the physisorption of closed - shell atoms , molecules , and layers on metal substrates . in ref . we have developed a quantitative model for this effect , based upon an anti - symmetrization of the product of the metal and adsorbate wave functions . when an adsorbate approaches a metal surface , the wave functions of the two systems overlap . pauli exchange repulsion between these states leads to a spatial redistribution of the electron density , in particular to a decrease of the density in the overlap region . since the metal wave functions are usually more extended and more easily deformable than those of the adsorbate , the net result of this redistribution is that electrons are pushed back into the metal , which effectively lowers the work function . in the adsorption of graphene and _ h_-bnon high work function metals we found potentials steps of up to 1 - 2 ev . the potential steps for mos@xmath0 adsorbed on the same metals are generally smaller , and more typically around 0.3 - 0.4 ev . the wave functions of first - row elements ( b , c , n ) are compact and not easily deformable , as compared to the wave functions of the metal substrate . the effect of pauli repulsion in the metal / adsorbate overlap region is then very asymmetric . it is foremost the metal electron density that is deformed , i.e. pushed back , which gives a large work function lowering . if the adsorbate contains heavier elements , such as mos@xmath0 , the effect of pauli repulsion is more symmetric , i.e. , both the metal and the adsorbate electrons are pushed out of the overlap region in a more symmetric way . this gives a smaller effect on the work function . note that if the effect of pauli repulsion would be completely symmetric , the work function would be unchanged . /al(111 ) slab ; ( red ) the sum of the densities of states of free - standing mos@xmath0 and of the al(111 ) slab . the densities of states are aligned by aligning the mo 4s core levels , and the al 2p core levels . ] low work function metals experience an increase of the work function upon adsorption of mos@xmath0 , i.e. a negative @xmath19 , which indicates a net transfer of electrons from the metal to the mos@xmath0 adsorbate . as mos@xmath0 is a semiconductor it can only receive electrons in its conduction band . therefore , for low work function metal substrates one expects the fermi level to be in the conduction band of mos@xmath0 . analysis of the electronic structure of the mos@xmath0/metal slab however shows that this is not the case . the interaction between mos@xmath0 and the metal at the interface leads to interface states with energies in the mos@xmath0 band gap . that seems obvious if mos@xmath0 is chemisorbed onto the substrate , as in the case of mos@xmath0/ti(0001 ) , which we will discuss in the next section . of the mos@xmath0/al(111 ) interface at the equilibrium distance @xmath117 , and ( b ) at a distance @xmath118 ; ( c , d ) the corresponding band structures ; the blue color indicates the weight of a projection of the wave functions on the mos@xmath0 sites . ] somewhat surprisingly , a significant density of interface states also forms if the interaction between mos@xmath0 and the metal substrate is relatively weak . for example , as discussed in the previous section , the interaction between mos@xmath0 and al(111 ) can be classified as physisorption , see fig . [ fig : physchem ] . nevertheless , states with energies inside the mos@xmath0 band gap are formed at the mos@xmath0/al(111 ) interface , as is immediately obvious when comparing the density of states of the interface with that of free - standing mos@xmath0 , see fig . [ fig : dos_al ] . the density of these interface states is not extremely high , yet sufficiently high to pin the fermi level below the mos@xmath0 conduction band , as demonstrated by fig . [ fig : dos_al ] . the density of interface states increases with increasing mos@xmath0/metal interaction , but even for physisorption it seems sufficiently high to prevent the fermi level from reaching the mos@xmath0 conduction band . one can prove that these interface states are indeed responsible for pinning the fermi level by artificially enlarging the distance between the mos@xmath0 layer and the metal surface . this breaks the direct mos@xmath0/metal contact that is responsible for the formation of interface states . in the absence of interface states , the fermi level is at the bottom of the mos@xmath0 conduction band , see fig . [ fig : alfinal](d ) , which is what one would expect if the work function of the metal @xmath21 is smaller than the electron affinity of mos@xmath0 @xmath119 , see eq . [ eq : sb ] . a transfer of electrons between the metal and the mos@xmath0 overlayer then yields a charge distribution that can be associated with a simple interface dipole , see fig . [ fig : alfinal](b ) . in contrast , if interface states are formed , the band structure of adsorbed mos@xmath0 is perturbed , see fig . [ fig : alfinal](c ) , and the pattern of the charge distribution at the interface is much more complicated , as shown in fig . [ fig : alfinal](a ) . in that case , the interface states pin the fermi level below the bottom of the mos@xmath0 conduction band , see fig . [ fig : dos_al ] . schottky barrier heights ( sbhs ) for electrons , calculated according to eq . [ eq : sb ] , are also listed in table [ tab : deltav ] . the functional causes some uncertainty , as the work functions of the clean metal surfaces obtained with the opt88-vdw - df functional tend to be somewhat higher than those obtained with the pbe functional . lda in general gives even higher work functions , so opt88-vdw - df gives work functions that are in between those of pbe and lda.@xcite note that the interface potential steps @xmath19 do not depend strongly on the functional . as the opt88-vdw - df functional also gives a larger electron affinity for mos@xmath0 , the schottky barrier @xmath26 according to eq . [ eq : sb ] , also does not depend strongly on the functional . we see that the sbh decreases with decreasing metal work function but does not go to zero . instead it goes through a minimum for cu and ag , and then increases again for low work function metals like al and mg . as discussed above , this phenomenon is caused by interface states . the only way to get rid of such states is to break the direct interaction between mos@xmath0 and the metal substrate . . discusses a practical way of doing this by inserting an atomic layer between the metal surface and the mos@xmath0 layer . if this intermediate layer is purely van der waals - bonded to mos@xmath0 , no gap states are formed at its interface with mos@xmath0 . in addition the intermediate layer should be transparent to electrons , such that the interface resistance is not dramatically increased . a monolayer of _ h_-bn or graphene satisfy these criteria.@xcite adsorbed on ti(0001 ) . ] as discussed in sec . [ sec : bonding ] , mos@xmath0 is chemisorbed in ti(0001 ) . the binding energy and the equilibrium bonding distance of mos@xmath0 on ti(0001 ) does not seem to be qualitatively different from those for mos@xmath0 on metal substrates such as co(0001 ) or ni(111 ) , see table [ tab : energies ] and fig . [ fig : physchem ] . however , the structural deformation of the mos@xmath0 layer adsorbed on ti(0001 ) is much larger than that of mos@xmath0 on other metals , see table [ tab : structures ] . this deformation is illustrated in fig . [ fig : ti_structure ] . atoms of the bottom sulfur layer make a bond with ti atoms of the top layer of the substrate , where several of these metal atoms are pulled up from the substrate . the mos@xmath0 and the ti(0001 ) lattices do not fit very well ; one needs a @xmath120 mos@xmath0 supercell on top of a @xmath121 ti(0001 ) supercell to get a mismatch below 1% , see table [ tab : supercell ] . the result therefore is a mos@xmath0/ti(0001 ) interface that contains a substantial local strain , which explains why the binding energy is not very large , despite the bonding being chemisorption . the potential step @xmath19 at the mos@xmath0/ti(0001 ) interface is negative , in contrast to the potential step at the mos@xmath0/ag(111 ) interface for instance , which is positive , despite the fact that the work functions of ti and ag are very similar , see table [ tab : deltav ] . we argued that physisorption should lead to a positive potential step because of the pauli repulsion effect , and indeed mos@xmath0 is physisorbed on ag(111 ) . chemisorption , as in the case of mos@xmath0 on ti(0001 ) , leads to a more drastic reorganization of the charge distribution at the interface , because of the formation of new chemical bonds . upon the formation of these bonds there is apparently a net displacement of electronic density towards the sulfur atoms , which is not unreasonable as sulfur is more electronegative than ti . this displacement results in an increase of the work function , i.e. , a negative @xmath19 . one expects that chemisorption also leads to a strong perturbation of the electronic structure of the adsorbate . figure [ fig : dos_ti ] shows the density of states ( dos ) of a mos@xmath0 bilayer projected on the individual mos@xmath0 layers . the dos of the first ( chemisorbed ) layer is indeed strongly perturbed as compared to the dos of a free - standing mos@xmath0 layer . the mos@xmath0 wave functions strongly hybridize with those of the underlying ti substrate , and the resulting hybridized states give a non - zero dos for energies all through the mos@xmath0 band gap . it is sometimes argued that such interface states promote having a good ( ohmic ) mos@xmath0/metal contact.@xcite one could however also argue that chemisorption is harmful to obtaining a good contact , because it damages the integrity of the mos@xmath0 layer . in ref . it is found that mos@xmath0/ag gives a better contact than mos@xmath0/ti , due to a much smoother interface in the former case , suggesting to prefer physisorption over chemisorption . layer of mos@xmath0 adsorbed on ti(0001 ) and on the second mos@xmath0 layer , respectively . ] it is not possible to define a schottky barrier for a single mos@xmath0 layer adsorbed on ti(0001 ) . chemisorption affects the electronic structure of mos@xmath0 to such an extend that its semiconducting character is lost . it is however possible to define a schottky barrier for a second mos@xmath0 layer that is adsorbed on the first layer . the first and second mos@xmath0 layer are bonded by a van der waals interaction , which does not perturb the electronic structure of the second layer significantly . indeed the dos of the second mos@xmath0 layer is quite similar to that of a free - standing mos@xmath0 layer with a clear band gap , see fig . [ fig : dos_ti ] . the height of the schottky barrier to the second layer is a sizable 0.54 ev . this mainly results from the fact that the first adsorbed mos@xmath0 layer effectively increases the work function of the ti substrate , see table [ tab : deltav ] . the size of the schottky barrier indicates that it is not possible to make an ohmic contact to pristine ( undoped ) mos@xmath0 with ti.@xcite in this paper , we explore the adsorption of mos@xmath0 on a range of metal substrates by means of first - principles dft calculations . the ( 111 ) surfaces of al , ni , cu , pd , ag , pt and au , and the ( 0001 ) surfaces of mg , co and ti cover a range of metals with different interaction strengths , allowing for a systematic study of the metal - mos@xmath0 interface . we compare the results obtained with different dft functionals . in many cases the gga / pbe density functional only captures a small part of the binding energy of mos@xmath0 on a metal substrate , as compared to the opt88-vdw - df van der waals density functional , which indicates the importance of van der waals interactions in the interface bonding . nevertheless , the equilibrium binding distances obtained with both functionals are generally very similar , and so are the interface potential steps and schottky barrier heights . exceptions are cases for which van der waals interactions essentially describe the whole bonding , where pbe fails completely . lda tends to overbind , leading to shorter binding distances and larger interface potential steps . the interface structure that results from adsorbing an mos@xmath0 layer on a metal surface will be incommensurable in most cases , as the two lattices have a mismatch . we investigate the effects of the artificial strain introduced by approximating the lattice using a commensurable supercell . we conclude that these effects are moderate provided the mos@xmath0 lattice parameter is kept at its optimized value , and the metal lattice is strained . large lattice mismatches should however be avoided , and straining the mos@xmath0 lattice can lead to very unphysical results.@xcite of the metal substrates studied ti is the one on which mos@xmath0 is clearly chemisorbed . adsorption of mos@xmath0 on ti(0001 ) is accompanied by a clear structural deformation of the ti surface and of the mos@xmath0 overlayer , due to the formation of bonds between the surface ti atoms and the sulfur atoms at the interface . formation of these interface bonds significantly alters the electronic structure of the mos@xmath0 adsorbate . in particular , the interface states fill up the band gap of mos@xmath0 , which makes defining a schottky barrier for this layer meaningless . however , for a second , unperturbed , adsorbed mos@xmath0 layer a schottky barrier of 0.54 ev can be extracted . mos@xmath0 is physisorbed on au(111 ) , where the bonding is almost completely due to van der waals interactions , and the structure and electronic structure of mos@xmath0 are hardly perturbed by the adsorption . the properties of mos@xmath0 adsorbed on other metal substrates fall in the range between the two extreme cases ( ti and au ) , without the possibility of drawing a clear dividing line , as has been done for the adsorption of graphene or _ h_-bn on metal substrates.@xcite experiments have focused foremost on schottky barrier heights . transport measurements on multilayer mos@xmath0 devices generally yield small numbers for the schottky barrier heights , i.e. 0.03 - 0.2 ev , for different metals,@xcite whereas photoemission , photoconduction , and scanning tunneling spectroscopy give higher values 0.2 - 0.9 ev.@xcite it has been suggested that the mos@xmath0 samples used in devices is quite defective and inhomogeneous , such that the position of the fermi level does not reflect an intrinsic property of mos@xmath0 or of the mos@xmath0/metal contact,@xcite which obstructs a comparison to calculated results . our results for the 4d and 5d metals ag , au , pd , and pt agree qualitatively with those reported in previous calculations,@xcite provided the mos@xmath0 lattice is not stretched.@xcite quantitatively , the reported schottky barrier heights for these metals are @xmath122 ev larger than our results . these calculations were based upon the lda functional , which tends to overbind , and to overestimate the metal work functions.@xcite compressing the metal lattice , which is sometimes required to accommodate a lattice mismatch in a small supercell , does not help either , as that gives an even higher work function.@xcite the same is likely true for simple metals such as al and in.@xcite the strong interaction we find for ti is also found in lda calculations.@xcite in those calculations the lattice mismatch used was large , however , which can alter the interface interactions . the overall picture emerging from these calculations is that mos@xmath0 interacts strongly with the early transition metals , where it is clearly chemisorbed . the interaction with the late transition metals is much weaker , where the 3d metals interact stronger than the 4d and 5d metals . interacts rather weakly with the simple metals , but the interaction increases for very low work function metals . in all but the strongly chemisorbed case , van der waals forces play an important role in the interface interactions . in case the interface interaction is weak ( physisorption ) the interface potential step can be understood as resulting from pauli repulsion , which effectively decreases the substrate work function . the schottky barrier is then simply calculated from the modified work function . strong interaction ( chemisorption ) leads to the formation of bonds between the substrate metal atoms and the adsorbate sulfur atoms . it increases the substrate work function if the electronegativity of the adsorbate is higher than that of the metal . if the mos@xmath0 layer is chemisorbed , its electronic structure is perturbed to an extend that a schottky barrier can not be defined . however , a second adsorbed mos@xmath0 layer then shows the characteristics of a single unperturbed layer . we acknowledge taher amlaki and deniz ak@xmath123r for fruitful discussions . this work is part of the research program of the foundation for fundamental research on matter ( fom ) , which is part of the netherlands organization for scientific research ( nwo ) . the use of supercomputer facilities was sponsored by the physical sciences division ( ew ) of nwo . 67ifxundefined [ 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 10.1038/nchem.1589 [ * * , ( ) ] link:\doibase 10.1021/cr300263a [ * * , ( ) ] link:\doibase 10.1038/nature12385 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.136805 [ * * , ( ) ] link:\doibase 10.1038/nnano.2012.193 [ * * , ( ) ] link:\doibase 10.1021/ar500274q [ * * , ( ) ] link:\doibase 10.1021/nn303513c [ * * , ( ) ] link:\doibase 10.1021/nl303583v [ * * , ( ) ] link:\doibase 10.1021/nl4010157 [ * * , ( ) ] link:\doibase 10.1038/srep01634 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4895767 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4866340 [ * * , ( ) ] link:\doibase 10.1103/physrevb.91.161304 [ * * , ( ) ] link:\doibase 10.1103/physrevb.90.085415 [ * * , ( ) ] link:\doibase 10.1103/physrevb.90.201411 [ * * , ( ) ] link:\doibase 10.1021/jp212558p [ * * , ( ) ] link:\doibase dx.doi.org/10.1063/1.4893875 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.156802 [ * * , ( ) ] link:\doibase 10.1021/nl303909f [ * * , ( ) ] link:\doibase 10.1103/physrevx.4.031005 [ * * , ( ) ] link:\doibase 10.1021/nl403465v [ * * , ( ) ] link:\doibase 10.1039/c4tc01794c [ * * , ( ) ] link:\doibase 10.1103/physrevb.89.245403 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.92.246401 [ * * , ( ) ] link:\doibase 10.1103/physrevb.76.125112 [ * * , ( ) ] link:\doibase 10.1103/physrevb.83.195131 [ * * , ( ) ] link:\doibase 10.1103/physrevb.47.558 [ * * , ( ) ] link:\doibase 10.1103/physrevb.49.16223 [ * * , ( ) ] link:\doibase 10.1103/physrevb.54.11169 [ * * , ( ) ] link:\doibase 10.1103/physrevb.59.1758 [ * * , ( ) ] link:\doibase 10.1103/physrevb.40.3616 [ * * , ( ) ] link:\doibase 10.1103/physrevb.46.16067 [ * * , ( ) ] link:\doibase 10.1103/physrevb.23.5048 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.77.3865 [ * * , ( ) ] link:\doibase 10.1103/physrevb.76.073103 [ * * , ( ) ] link:\doibase 10.1103/physrevb.90.155448 [ * * , ( ) ] link:\doibase 10.1021/acsami.5b02782 [ * * , ( ) ] link:\doibase 10.1103/physrevb.84.195414 [ * * , ( ) ] link:\doibase 10.1103/physrevb.78.045409 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.186102 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.156401 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.085405 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4807855 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.026803 [ * * , ( ) ] link:\doibase 10.1103/physrevb.79.195425 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.235502 [ * * , ( ) ] link:\doibase 10.1103/physrevb.89.121103 [ * * , ( ) ] link:\doibase 10.1103/physrevb.87.155304 [ * * , ( ) ] link:\doibase 10.1103/physrevb.85.235407 [ * * , ( ) ] link:\doibase 10.1103/physrevb.85.033305 [ * * , ( ) ] link:\doibase 10.1103/physrevb.88.075409 [ * * , ( ) ] link:\doibase 10.1103/physrevb.88.245309 [ * * , ( ) ] link:\doibase 10.1021/nl4014748 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.156802 [ * * , ( ) ] link:\doibase 10.1103/physrevb.36.1647 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/j.susc.2005.10.042 [ * * , ( ) ] http://stacks.iop.org/0022-3727/1/i=7/a=416 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0022-0248(72)90129-7 [ * * , ( ) ] link:\doibase 10.1103/physrevb.64.235305 [ * * , ( ) ] link:\doibase 10.1103/physrevb.88.085318 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3524232 [ * * , ( ) ] link:\doibase 10.1021/nn506567r [ * * , ( ) ] link:\doibase 10.1109/led.2014.2313340 [ * * , ( ) ] link:\doibase 10.1021/am506921y [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3696045 [ * * , ( ) ] link:\doibase 10.1021/nn500044q [ * * , ( ) ] link:\doibase 10.1021/nl5047736 [ * * , ( ) ]
entanglement is the property of states of multipartite quantum systems that is the most important resource for quantum information processing @xcite . for this reason , one of the most important tasks of quantum information science is to establish entanglement at long distances in quantum networks , and to optimize final entanglement and probability of success . quantum networks @xcite play a key role in quantum information processing . here we limit our attention to those networks in which quantum states can be prepared initially and shared . that is , entanglement can be generated between neighboring or , at least , not - too - widely - separated nodes ( or stations ) . there are two instances in which the above - mentioned tasks become obviously relevant . on one hand , one can consider macroscopic quantum communication networks , such as cryptographic networks , or more generally quantum communication nets @xcite , or distributed quantum computation @xcite involving arbitrary nodes of the network . the second instance concerns microscopic or mesoscopic networks that could constitute architectures of quantum computers ( cf . ref . ) . despite enormous progress in experimental techniques ( cf . ref . and references therein ) , it is in principle a very hard task to establish entanglement at large distances due to decoherence and attenuation effects . two remedies for this problem have been proposed : * _ quantum repeaters_. this concept has been developed for 1d quantum communication chains @xcite . although the simple entanglement swapping @xcite procedure can lead to quantum communication at large distances ( see fig . [ fig : swapping ] ) , for imperfect resources , the performance of such communication chains decays exponentially with the distance ( i.e. the number of repeaters ) . however , one can use more sophisticated quantum repeater protocols , which use purification and swapping methods that lead to polynomial decay only . * _ entanglement percolation_. recently , our collaborators , together with two of us , proposed using networks in which properties of the connectivity of the network enable the establishment of , and determine the probability of , entanglement on large distances . in refs . and we considered in particular _ pure - state networks _ on regular lattices , where the nearest - neighbor ( nn ) nodes share a non - maximally entangled pair of qubits , or more generally qudits ( an entangled bond ) . we searched for local operations and classical communication ( locc ) protocols that lead to establishment of entanglement between remote nodes of the network . entanglement swapping and 1d repeater . circles are qubits . heavy lines represent pure non - maximally - entangled states . loops represent the entanglement swapping measurement on a pair of qubits . a ) two states @xmath0 and @xmath1 of the form specified in ( [ stateab ] ) . after the operation , qubits @xmath2 and @xmath3 may be in an entangled , mixed state . b ) swapping as the first step in a repeater . c ) a 1d chain of repeaters . ] at present , only a handful of studies that explicitly apply percolation theory to problems in quantum information have appeared in the literature . these articles , comprising refs . and as well as applications to cluster states , are reviewed in ref . . our results leading to the present work can be summarized as follows : * for 1d chains we proved that even optimal locc strategies only allow the establishment of entanglement between distant nodes that decays exponentially with both the distance , as well as the quality of entanglement of nn bonds . * in 2d and higher dimensions , the possibilities for protocols are greatly expanded . the most straightforward , naive protocol the one we use as a baseline to evaluate other protocols is the one we term _ classical entanglement percolation _ ( cep ) , although it does involve obviously some quantum operations . this protocol begins with converting ( using locc ) each of the entangled bonds into a singlet ( _ i.e. _ maximally entangled state ) with probability @xmath4 ( the so - called singlet conversion probability ( scp ) @xcite ) . after the conversion , each of the parties ( nodes ) knows obviously which of the bonds are now perfectly entangled . using classical communications , the parties establish whether there exists an infinite percolating cluster ( or one spanning the lattice ) and who belongs to it . then , by performing a series of entanglement swappings , it is possible to propagate entanglement between any two ( widely ) separated nodes that belong to the percolating cluster . we call this scheme `` classical '' because it essentially maps the problem onto a classical bond percolation problem @xcite , and its success or failure is equivalent to the success or failure of bond percolation in the same lattice . namely , if @xmath5 , where @xmath6 is the lattice - dependent critical percolation threshold ( or , in other words , critical open - bond density ) , then entanglement between any two remote nodes can be established with probability @xmath7 , which asymptotically does not depend on the distance . * at the same time that we introduced cep , we presented several schemes that went beyond the simple application of singlet conversion everywhere followed by entanglement swapping along a path . we call these _ quantum entanglement percolation _ ( qep ) protocols because they use some kind of quantum pre - processing for instance quantum measurements to transform one percolation problem to another one . in these protocols cep is also used , but is preceded by application of certain locc , which remove and replace bonds in the network lattice , resulting in a new lattice geometry with fundamentally different long - range properties . the pre - processing in qep may greatly improve the possibility of entanglement percolation , either by reducing @xmath6 , or by increasing @xmath8 . in this paper we present a more systematic and thorough study of qeps based on lattice transformations . in particular , we explore the entire parameter space with the addition of monte carlo and series - expansion methods . although general principles remain to be found , we do discover certain rules and regularities governing such strategies ( see sec . [ sec : examples ] ) . the paper is organized as follows . in sec . [ sec : models ] we formulate and describe the models and fix the notation . section [ sec : nets_and_perc ] discusses quantum networks and percolation , whereas sec . [ sec : transformations ] deals with loccs used for lattice transformations . our main results are presented in sec . [ sec : examples ] , which contains descriptions of five examples of transformations enhancing cep : i ) the transformation of the kagom lattice to the square lattice , ii ) the transformation of the double - bond honeycomb ( hexagonal ) lattice to the triangular lattice , iii ) the transformation of the square lattice to two decoupled copies of the square lattice , iv ) the transformation of the bowtie lattice to the decoupled triangular and square lattices , and v ) the transformation of a triangular lattice with two different degrees of entanglement into the decoupled square and triangular lattices . the cases ii ) , iii ) , and v ) were discussed already in ref . , but we present here more general and stronger results . we conclude in sec . [ sec : conclusions ] . over the past few decades , classical percolation theory has seen the development of a number of quite sophisticated and powerful methods , which we have adapted to the questions at hand . in order to demonstrate the supremacy of qep over cep in each of the cases i)-v ) we employ these methods combined with the methods of quantum information theory . an overview of these methods is presented in the appendices : appendix [ sec : majorization ] presents the necessary facts from majorization theory and singlet conversion protocols @xcite , appendix [ sec : montecarlo ] explains some details of our monte carlo ( mc ) simulations , and finally appendix [ sec : series ] deals with series expansions . in this section , we describe the classes of quantum networks and communication protocols that we investigated . the networks we consider consist , prior to application of communication protocols , of a collection of qubits partitioned into pairs , each pair being prepared in an identical pure state @xmath9 . with appropriate choice of bases , any such state can be written @xmath10 where the schmidt coefficients @xmath11 satisfy @xmath12 and @xmath13 . we identify these pairs with bonds on a two - dimensional lattice or edges on a graph , with two spatially - separated qubits , one occupying each end of the bond . at regular positions , a small set of these qubits are arranged near enough to one another to allow measurements on any subset . such a set of qubits then constitutes a vertex ( or node or site ) which is incident to each edge that contributes a qubit to the vertex . this defines a lattice to which we apply methods of statistical physics . in particular , many results on the possibility of long - range entanglement are described using percolation theory @xcite , and depend only on the graph structure of the system . given a lattice prepared as described above , we search for the protocols consisting of local operations and classical communication ( locc ) that yield the maximum probability of achieving entanglement between two nodes separated by an arbitrarily large distance . the answer depends on the single parameter @xmath14 characterizing the state @xmath15 . even with a small palette of possible operations , finding the globally optimal solution is a difficult task . instead , we search for promising classes of protocols . the simplest protocol consists of attempting to convert the state associated with each bond to a singlet via the `` procrustean method '' of entanglement concentration , which is the optimal strategy at the level of a single bond @xcite . this conversion succeeds with probability @xmath16 ( see appendix [ sec : majorization ] . ) , while failure leaves the pair in a state with no entanglement . in this way the system is described exactly by a bond percolation process with open - bond density @xmath4 . if there is a path of open ( maximally entangled ) bonds connecting two nodes , a sequence of entanglement swapping measurements , one at each intermediate node , is then applied in order to entangle the first and last node . this protocol is the simplest example of classical entanglement percolation . here we review a few fundamental ideas in percolation theory necessary to analyze cep . the nodes in the lattice can be partitioned into sets such that each node within a set is connected to each other node in the set via a path of open bonds . such a set of nodes is called an open cluster , or sometimes simply a cluster . the central fact of percolation theory is that percolation processes on most commonly - studied lattices in dimension @xmath17 and higher exhibit a continuous phase transition as the bond density passes through a critical value @xmath6 . for @xmath5 there exists with probability one a unique ( for the lattices we study here ) cluster of infinite mass ( number of nodes ) , while for @xmath18 all clusters are finite with probability one . it follows that improving an locc protocol to obtain a small change in @xmath4 can have a dramatic effect on the probability of long - range entanglement . in the supercritical phase , long - range entanglement is possible , while in the subcritical phase it is not possible . serving as the order parameter is the density of the infinite cluster @xmath19 which we define via @xmath20,\ ] ] the probability that a fixed node @xmath21 ( say , the node at the origin ) is in the the infinite cluster @xmath22 . when referring to @xmath23 on a specific lattice , we use a symbol representing the lattice as a subscript . the probability that two selected nodes are members of the same cluster decays roughly exponentially in their separation distance to an asymptotic value of @xmath24 . ( the length scale of decay is the correlation length @xmath25 . ) this means , for the problem at hand , that in cep the probability that information can be propagated between two nodes is asymptotically @xmath24 . it was proven in refs . and that cep is not the optimal strategy for establishing long - distance entanglement . the demonstration is based on applying certain locc prior to cep . all of these pre - processing locc act on pairs of qubits , and they either transform the state on a given bond , or they replace two adjacent pure - state bonds by one , in general , mixed state bond . there are essentially three types of generalized measurement used : * _ singlet conversion _ the optimal locc singlet protocol @xcite is used in three situations . in cep with single - bond lattices we apply it directly to the state ( 1 ) , which results in the singlet conversion probability @xmath16 . if the protocol is successful , the bond that is converted to the singlet can be used for entanglement propagation ( swapping ) , otherwise it is useless . in cep with a double - bond lattice we apply it directly to the two copies of the state ( 1 ) , which lives in @xmath26 , and has schmidt coefficients @xmath27 , _ ergo _ the singlet conversion probability is @xmath28 finally , in all of the qep strategies that we study , we also apply singlet conversion to any remaining untouched bonds after the lattice transformation . * _ entanglement swapping _ this protocol @xcite , illustrated in fig . [ fig : swapping]a , consists of performing the so called bell measurement on a pair of qubits ( b and c ) in a node , i.e. a von neumann measurement in a basis of 4 maximally entangled orthonormal states ( in the computational basis ) . it allows conversion of a pair of adjacent singlets into a singlet connecting end points ( a and d ) with probability 1 , i.e. allows for perfect entanglement propagation in a connected cluster of singlets . at the same time , when applied to a pair of imperfect states ( 1 ) , it produces a mixed state , which , amazingly , has the average singlet conversion probability equal to @xmath16 . unfortunately , this effect can not be iterated . when applied to the mixed states , entanglement swapping reduces the singlet conversion probability . this last point places a significant constraint on our choice of lattice transformations . + we use entanglement swapping in both cep and qep , but the two uses have very different effects . in the the case of cep , entanglement swapping is used to locally move entanglement between neighboring nodes , an operation that is repeated in hopes of transporting entanglement over long distances ( this is , roughly speaking , a brute - force method ) . but with qep , before attempting to transfer entanglement , we search for a way to selectively apply entanglement swappings to alter the geometry of the lattice and hence its long - range connectivity properties as given by percolation theory . the goal of this paper and future work is to enumerate the rich possibilities and point a way towards a general description of qep . * _ worst case entanglement _ finally , in refs . and the worst case protocol was used , which maximizes minimal entanglement over all measurement outcomes . this protocol consists also of bell measurement , but the basis is computational for one qubit , and corresponding to eigenstates of @xmath29 for the second . when applied to qubits , it produces a mixed state with the property that for all measurement outcomes , the resulting pure states have the same singlet conversion probability . we will not use this protocol here . in order to describe the lattice transformations involved in qep , we need a more general formulation of the lattice than the one given in sec . [ networksbipartite ] for cep . it is useful to define the percolation processes a bit more precisely . we begin with a graph , that is , a set of edges ( bonds ) @xmath30 and a set of vertices ( nodes ) @xmath31 . we consider embedding the graph in @xmath32 in order to treat geometric properties . in fact , the important properties do nt change if we force the vertices to occupy points in @xmath33 . the configuration of open and closed bonds can be described by a probability sample space @xmath34 , with points @xmath35 , where @xmath36 takes the values @xmath37 and @xmath38 . we allow each bond to be open with a different probability , _ i.e. _ @xmath39 . the appropriate measure is a product measure on @xmath40 with marginal probabilities defined by @xmath41 , @xmath42 . to allow all transformations possible via locc , we take the graph to be complete that is , all possible edges @xmath43 with @xmath44 are present ( in fact , sometimes we need double bonds , as well . ) for percolation properties , any graph with fewer edges can be identified with this complete graph by setting the appropriate @xmath45 to zero . ( color online ) a ) mapping a quantum network to a percolation problem . qubits are represented by small circles . each vertex ( node ) contains six qubits . b ) using entanglement swapping to transform the lattice structure . the two red ( grey ) bonds in a ) are replaced by the red bond in b ) . this process can be continued to produce a double - bond hexagonal lattice . ] now we describe the correspondence of entanglement on the physical system of qubits to this percolation formulation . we denote the set of the indices of all qubits by @xmath21 . each qubit @xmath46 is assigned to a vertex @xmath47 , with , in general , multiple qubits assigned to each vertex . for example , a portion of a network that can be described by a triangular lattice is shown in figure fig . [ fig : show_qep]a , where vertex @xmath48 contains the six circled qubits . for every pair of qubits @xmath49 with @xmath50 , we denote the reduced state by @xmath51 , and by @xmath52 the singlet conversion probability ( scp ) , that is , the probability of conversion of @xmath51 to a singlet , maximized over all possible measurements . then we assign @xmath53 for every @xmath54 . initially , all reduced bipartite states have either @xmath55 for separable states , or @xmath56 for the prepared , pure , partially entangled states . in fig . [ fig : show_qep]a the edges with @xmath57 are shown in black , while edges with @xmath55 are simply absent in the diagram . two bonds are colored red ( grey ) only to show that they will be replaced by another bond ( in the sense of altering the scp ) via entanglement swapping . for qep , a successful preprocessing , entanglement swapping operation such as that depicted in fig . [ fig : swapping ] , alters the percolation process by setting @xmath58 and @xmath59 and @xmath60 . likewise , in the example in fig . [ fig : show_qep ] , an entanglement swapping measurement on qubits @xmath61 and @xmath62 sets @xmath63 , @xmath64 , and @xmath65 . each of the examples discussed below conforms to this description . we consider five examples of lattice transformations in this paper , each exhibiting a different combination of features , with implications for the analysis of protocols . in particular , two transformations convert one lattice to another , while the other three convert a lattice into two decoupled lattices . four transformations involve lattices with single bonds , while the fifth involves double bonds . quantities arising in the analysis , such as @xmath19 , are analytic about @xmath66 . in fact , the lowest - order term in the expansion about @xmath66 is typically much larger than the remaining terms even relatively far from @xmath66 , with the result that most of the interesting crossover behavior in comparing protocols occurs for smaller values of @xmath4 . but this behavior does not appear when using the distillation procedure for double bonds that leads to ( [ fourqubits ] ) . the reason is that distillation produces a saturation point for @xmath19 smaller than @xmath66 . this results in a much stronger difference between classical and quantum protocols in the high - density regime than does singlet conversion on single bonds ( two - qubit pure states ) . four of the transformations produce a smaller critical density on at least one of the resulting lattices , which gives the most pronounced advantage in the regime near the critical density . the fifth example shows that on some lattices where a particular qep strategy is not advantageous , allowing bonds of different strengths @xmath4 and @xmath67 can produce regions in the phase space @xmath68 where qep is indeed advantageous . finally , consider comparing cep in which singlet conversion is applied to each bond separately , with qep that results in a single transformed lattice . in every such case we find that qep is better than cep over the entire ( non - trivial ) range of @xmath4 . it is an open question whether this is generic behavior . we analyze the results of the transformation of each lattice in three regimes : near the critical density ( or densities ) ; near @xmath66 ; and between these two regimes . arguments near the critical densities typically rely simply on the fact that long - range entanglement is impossible on a lattice with density below the critical density . these results are the most insensitive to details of the definitions of entanglement and connectivity . near @xmath66 we compute high - density expansions for @xmath19 and related quantities ( see sec . [ sec : expansions ] . ) often the difference between the performance of cep and qep in this regime is marginal . it is important nonetheless to carry out the analysis if we hope to make general statements about transformations that hold for all @xmath4 . between the critical regime and high - density regime there are some techniques widely used in percolation theory that may be useful , such as russo s formula , which is used to prove inequalities in the rate of change of @xmath19 . however , we leave these techniques for future work and use monte carlo computations in the present paper . in our first example we compare cep on the kagom lattice to quantum entanglement percolation consisting of transformation of the kagom lattice to the square lattice via entanglement swapping at nodes specified in fig . [ fig : kagfive ] . although we do not treat them here , there are at least two more ways to transform the kagom lattice to the square lattice using the same kind of entanglement swapping . rigorous bounds have been obtained for @xmath69 @xcite ( as well as a high - precision monte carlo estimate @xcite ) while @xmath70 is known exactly , proving that the transformation gives an advantage for @xmath4 lying between @xmath70 and the lower bound for @xmath69 . as shown in table [ tab : series ] , the series for @xmath71 and @xmath72 are the same to the first non - trivial order in @xmath73 . but the next term shows that @xmath74 for @xmath4 close enough to @xmath38 . the mc data provides strong evidence that @xmath74 everywhere except near @xmath66 , where the statistical error is too large to distinguish the curves . but as is evident from the lower plot in fig . [ fig : kagfive ] , the terms in the expansions that we computed dominate @xmath75 already at values of @xmath4 for which the mc data is still accurate . thus we find that qep is advantageous over the entire range of @xmath4 . one of three transformations of the kagom lattice to the square lattice . pairs of qubits that are subjected to entanglement swapping are marked with loops . ] _ @xmath16 . the solid line is on the square lattice . the dashed line is on the kagom lattice . lower plot : normalized difference between @xmath71 and @xmath72 . the solid line is computed from mc data . the dashed line is computed from expansions from table [ tab : series ] . , title="fig : " ] _ v.s . _ @xmath16 . the solid line is on the square lattice . the dashed line is on the kagom lattice . lower plot : normalized difference between @xmath71 and @xmath72 . the solid line is computed from mc data . the dashed line is computed from expansions from table [ tab : series ] . , title="fig : " ] the hexagonal lattice with double bonds , each in the state specified by ( [ stateab ] ) ( see fig . [ fig : hontrtwo]a ) , was discussed in refs . and , in which it was shown that a transformation of the lattice to a triangular lattice offers an advantage over cep protocols for values of @xmath4 between @xmath76 and another critical value defined below . ( a ) double - bond hexagonal lattice . ( b ) transformed into the triangular lattice ( b ) . pairs of qubits that are subjected to entanglement swapping are marked with loops . ] here we extend the analysis to determine over the entire range of @xmath4 which of three protocols gives the highest probability of long - range entanglement . the first of the three protocols , which we call cep i , consists of performing an optimum singlet conversion to each bond separately so that the probability of getting at least one singlet connecting two nodes is @xmath77 . communication on the resulting lattice is then determined by bond percolation on the hexagonal lattice with bond density @xmath67 . in particular , the critical density ( see table [ tab : pcs ] ) is @xmath78 . in the second protocol , we perform a more efficient conversion , namely distillation , on all four qubits in a double bond . this protocol succeeds in producing a maximally entangled pair with probability @xmath79 ( see ( [ fourqubits ] ) and appendix [ sec : majorization ] . ) this results in bond percolation on the hexagonal lattice with density @xmath80\right\}.\ ] ] we refer to this method as cep ii , with critical density @xmath81 . in the third method , the entanglement swapping that maximizes scp is applied at every other node ( see fig . [ fig : hontrtwo ] ) , converting the double - bond hexagonal lattice to the triangular lattice with bond density @xmath4 , with @xmath82 . of @xmath83 _ v.s . _ @xmath16 for various strategies on the double - bond hexagonal lattice . ( a ) solid lines are monte carlo data . dashed lines are series expansions . ( b ) solid lines are mc . series expansions are indistinguishable from mc over the entire plot . ( c ) monte carlo data . [ fig : trihexone],title="fig : " ] of @xmath83 _ v.s . _ @xmath16 for various strategies on the double - bond hexagonal lattice . ( a ) solid lines are monte carlo data . dashed lines are series expansions . ( b ) solid lines are mc . series expansions are indistinguishable from mc over the entire plot . ( c ) monte carlo data . [ fig : trihexone],title="fig : " ] of @xmath83 _ v.s . _ @xmath16 for various strategies on the double - bond hexagonal lattice . ( a ) solid lines are monte carlo data . dashed lines are series expansions . ( b ) solid lines are mc . series expansions are indistinguishable from mc over the entire plot . ( c ) monte carlo data . [ fig : trihexone],title="fig : " ] the three protocols are compared in fig . [ fig : trihexone ] . we first observe that each method fails below its corresponding critical density , and these are known exactly . to compare the methods away from the critical points , we computed series expansions of @xmath19 about @xmath84 with results listed in table [ tab : series ] . defining @xmath85 we see that @xmath86 comparing ( [ thetacepi ] ) to @xmath87 from table [ tab : series ] we see that for @xmath88 the conversion to the triangular lattice is better than cep i. as is evident from figs . [ fig : trihexone]a and [ fig : trihexone]b , the two curves have the same leading behavior as @xmath89 , with the result that the the monte carlo can not distinguish them in this region . however , the series expansion lies well within the statistical error of the monte carlo points , even relatively far from @xmath66 , so we can be confident that @xmath90 even in the region where the mc can not distinguish the curves . finally , the mc clearly shows that @xmath90 for smaller @xmath4 , where the expansion fails . turning now to cep ii , we know that @xmath91 for @xmath92 and from ( [ distillp ] ) that @xmath93 for @xmath94 . it follows that there must be a crossover point . the mc data suggests that this occurs for @xmath95 . in summary , we see that qep gives an advantage over cep i for all @xmath4 , but that cep ii , which is more efficient in its use of double bonds , is better than qep at high densities . this transformation replaces the square lattice with two decoupled copies of the square lattice . to effect the transformation , at selected nodes the two horizontally opposing bonds are joined into one bond and likewise with the vertically opposing bonds . this procedure is applied at every other node , while staggering by one node when shifting rows as shown in fig . [ fig : doubsq ] . _ @xmath4 . upper points : @xmath96 _ v.s . _ @xmath4 . solid lines are series expansions . ] _ @xmath4 . lower points : @xmath97 _ v.s . _ @xmath4 . ] because the transformation splits the original lattice into two disjoint lattices , each taking half the surviving nodes , we can not compare connection between a single pair of nodes before and after the transformation . rather , we consider on the original lattice connection between , on the one hand , a pair of nodes @xmath98 separated by a small distance , and on the other hand , the same pair translated a distance much larger than the correlation length to a pair @xmath99 . we choose @xmath21 and @xmath100 so that each one goes to a different lattice in the transformation . after the transformation , there is the possibility of connection between @xmath21 and @xmath101 on one lattice and between @xmath100 and @xmath102 on the other . we choose the nodes as shown in fig . [ fig : doubsq ] . in addition to the complication of splitting , this example is subtle because the critical density is the same on all three lattices . in ref . it was shown that just above @xmath103 , the transformation gives an advantage over cep as measured by the choice of nodes @xmath98 . here we examine the connectivity for all @xmath104 . the probability that at least one of the pairs on the decoupled lattices is connected is @xmath105 we compare @xmath106 to the probability on the original lattice that at least one of @xmath21 and @xmath100 is connected to at least one of @xmath101 and @xmath102 . this probability is @xmath107 where @xmath108.\ ] ] ( in this paper ` or ' in ` @xmath109 or @xmath110 ' does _ not _ mean exclusive or . ) we prove that the doubling is advantageous in the limit @xmath88 using the power series about @xmath84 for @xmath71 and @xmath111 given in table [ tab : series ] . this follows from comparing @xmath112 and @xmath113 moreover , monte carlo evidence suggests that the transformation improves on cep for all @xmath114 in fig . [ fig : pi_pdoub_asymp ] one sees from the accuracy of the series expansion in the region of high - quality mc data that that the advantage is maintained even when the monte carlo becomes noisy near @xmath66 . the same mc data is not useful near @xmath70 for two reasons . firstly , both @xmath115 and @xmath107 vanish at the same critical point with infinite slope . secondly , the systematic error due to finite lattice size and large fluctuations in cluster statistics further complicate distinguishing the curves . we instead make use of an alternate expression for @xmath116 , that is @xmath117 where @xmath118 $ ] . the condition for advantage over cep then becomes @xmath119 the mc data for @xmath120 was generated by considering the largest cluster in the finite lattice to represent the infinite cluster @xmath22 even if it is not a spanning cluster . the mc data supports ( [ alternate ] ) , and furthermore , shows no evidence of non - analyticity at @xmath70 , as is evident in fig . [ fig : theta_omega_nolog ] . the foregoing analysis of the doubled square lattice was based on the choice of @xmath98 shown in fig . [ fig : doubsq ] . but this is not the only reasonable choice . in fact the question of whether the doubling transformation is better than cep depends somewhat on the details of how the nodes are chosen . although this is an extreme example , similar questions arise in analyzing other transformations . thus , the ambiguity in the measures comparing the various protocols must be addressed in future work . transformation of the bowtie lattice to decoupled square and triangular lattices . loops marking pairs of qubits represent swapping measurements . ] this is our second example that transforms a lattice into two decoupled lattices . figure [ fig : bowtie ] shows the measurements that decouple the bowtie lattice into the square lattice and the triangular lattice . for @xmath4 satisfying @xmath121 ( the exact critical values are listed in table [ tab : pcs ] ) , the transformation is obviously advantageous . in carrying out further analysis , the bowtie lattice presents a complication not present in other lattices studied in this paper : while the other lattices are regular in the sense that each vertex has the same environment up to rotations and reflections , the bowtie lattice has two kinds of vertices in this sense . notice first that all the surviving nodes on the square and triangular lattices shown in fig . [ fig : bowtie ] were generated from nodes of coordination number @xmath122 on the bowtie lattice , with all nodes of @xmath123 disappearing . each node of @xmath122 on the bowtie lattice has four nearest neighbors of @xmath122 , discounting nodes of @xmath123 ( for example node @xmath21 in fig . [ fig : bowtie ] . ) furthermore , two of these nearest neighbors are connected by a diagonal bond ( node @xmath124 ) and two are not ( node @xmath100 . ) in treating the example of doubling the square lattice in sec . [ sec : doubsquare ] , the fact that the lattice decouples forced us to consider connections between two widely separated pairs of nodes . we treat the present example in the same way , except that the more complicated local structure of the bowtie lattice forces us to consider a cluster of three nodes @xmath125 rather than a pair . in fig . [ fig : bowtie]b we see that @xmath126 are sent to the square lattice and @xmath21 is sent to the triangular lattice . we consider a distant cluster of nodes @xmath127 related to @xmath125 by translation , and examine the probability that a connection exists between at least one of @xmath125 and at least one of @xmath127 . on the bowtie lattice , this probability is @xmath128 where @xmath129.\ ] ] if instead we decouple the lattices via the lattice transformation , this probability is given by @xmath130 where @xmath131.\ ] ] a comparison of monte carlo estimates and high - density expansions of these quantities is shown in fig . [ fig : bt_sq_tr_log ] . monte carlo estimates of @xmath132 and @xmath133 . the inset is the same data on a linear scale . solid lines are high - density expansions . ] the mc data shows that for small densities , the lattice transformation gives an advantage , while for higher densities cep is the better protocol . the cross - over occurs at @xmath134 . as with the other examples , the curves representing the series expansions suggests that no cross - over occurs in the high - density region . .[tab : pcs ] @xmath6 for bond percolation on some lattices . @xmath135 is the unique root of @xmath136 . all critical densities are exact@xcite except for @xmath69@xcite . [ cols="<,<",options="header " , ] a ) the asymmetric triangular lattice . solid bonds represent the state @xmath15 ( density @xmath4 ) . dashed bonds represent the state @xmath137 ( density @xmath67 ) . b ) entanglement swapping is performed on pairs of solid bonds resulting in a triangular lattice with bond density @xmath4 . the remaining dashed bonds form a kagom lattice which is then transformed into a square lattice with bond density @xmath67 as in sec . [ sec : kag ] . ] with this example we demonstrate that qep can succeed when the initial bonds are not all in the same state , but fail if they are in the same state . consider the lattice in fig . [ fig : asymtr]a composed of two different kinds of bonds , each of the form given by ( [ stateab ] ) . the solid and dashed bonds represent the states @xmath15 ( with @xmath16 ) and @xmath137 ( with @xmath138 ) respectively . in general , we consider two entanglement distribution protocols : a ) classical entanglement percolation on the original lattice and b ) qep consisting of creating two decoupled lattices ( the square and triangular lattices ) via entanglement swapping , followed by cep on each of the resulting lattices . the transformation is described in fig . [ fig : asymtr ] . note that , for @xmath140 , the initial lattice is a triangular lattice with serial double bonds and therefore the critical point @xmath141 . on the other hand for @xmath142 and @xmath143 the initial lattice is the kagom lattice . we first examine the most robust measure of a protocol s effectiveness the binary measure that tells whether long - range entanglement is possible or not . the phase diagrams for this example before and after the transformation are shown in fig . [ fig : asymtr_phase ] . we see that there are regions in the phase space for which qep is better than cep , and vice versa . in the region in which both protocols allow long - range entanglement , a more detailed measure similar to those in previous sections is necessary . for example , for @xmath144 , we performed an analysis similar to the one used for doubling the square lattice . the two critical boundaries intersect at @xmath145 , but comparing connectivity just above the critical point shows that the transformation does not improve the probability of long - range entanglement . likewise for @xmath144 and @xmath4 near @xmath38 , series expansions showed that the transformation is not an improvement on cep . we have not yet determined whether qep in this scenario succeeds for some other @xmath144 , but this seems unlikely . phase diagram for the asymmetric triangular lattice . the heavy solid curve ( obtained by monte carlo ) separates the supercritical and subcritical regions on the original asymmetric triangular lattice ( cep ) . the dashed curve composed of two line segments separates the supercritical and subcritical regions after the transformation ( qep ) . after transformation , the supercritical region is defined to be the region for which at least one of the disjoint lattices is supercritical . in the light gray regions , long - range entanglement is possible with qep , but not with cep . in the dark gray region it is possible with cep , but not with qep . ] in this paper we have considered the problem of establishing long - distance entanglement in quantum pure - state networks on regular 2d lattices . we have discussed in detail several examples of quantum entanglement percolation strategies that are better than the corresponding classical strategies , i.e. those consisting of a direct attempt to convert bonds into singlets . our results illustrate nicely the interplay between quantum information theory and classical percolation theory . despite the fact that we do find certain rules and regularities governing qep strategies , many questions remain open . for instance , we can not say anything about optimality of our qep protocols and most probably they are not optimal . all of our protocols involve locc acting on pairs of qubits only ; can multipartite locc , which leads inevitably to creation of multipartite entanglement , help ? we know that for sufficiently large initial entanglement , perfect entanglement between remote nodes may be established with distance - independent probability @xmath7 . is it also possible for arbitrarily small initial entanglement ? we thank a. acn , j.i . cirac , and d. cavalcanti for discussions . we thank s. perseguers for kindly supplying figures . we acknowledge support of the eu ip programme `` scala '' , and spanish mec grants ( fis 2005 - 04627 , consolider ingenio 2010 `` qoit '' ) . j.w . thanks lluis torner and icfo for hospitality . he was partially supported by nsf grant dms 0623941 . m. l. thanks also the alexander von humboldt foundation for support . majorization theory has been applied to questions of transforming one bipartite pure state to another by locc . in particular , a theorem due to vidal@xcite gives the probability that such a transformation can be achieved via an optimal protocol ( without specifying that protocol . ) we state the result and apply it to the distillation procedure used in this paper . we must first introduce a certain partial order on vectors . consider two real , @xmath3-dimensional vectors @xmath146 and @xmath147 . we define the vector @xmath148 by reordering the elements of @xmath146 into non - decreasing order , and likewise with @xmath147 . we say that @xmath146 is submajorized by @xmath147 , denoted by @xmath149 , if @xmath150 for all @xmath151 . denoting the vector of schmidt coefficients of @xmath152 by @xmath153 , the theorem states that @xmath154 can be transformed into @xmath155 with probability @xmath4 , where @xmath4 is the largest number on @xmath156 $ ] such that @xmath157 . a simple , relevant application is the computation of the probability that optimal conversion of a state @xmath158 to a singlet will succeed . application of ( [ submaj ] ) gives @xmath159 . now we consider distillation , defined here as the optimal protocol for converting @xmath160 pure states @xmath161 to @xmath162 product states and one maximally entangled state in the schmidt bases . this operation is useful , for instance , in attempting to get a single , maximally - entangled bond from @xmath160 bonds connecting two nodes . explicitly , @xmath163 with @xmath164 and elements of the ordered set @xmath165 non - increasing . as usual , @xmath166 is shorthand for @xmath167 . for @xmath168 define the bijective numeration @xmath169 and @xmath170 . then we can write @xmath171 in order to apply the theorem , we need to show that this state can be written as a bipartite state with schmidt coefficients that can be chosen to satisfy our distillation condition . to this end collect all the first qubits of the bipartite states and all the second qubits , defining @xmath172 so that ( [ refactor ] ) becomes @xmath173 note that @xmath174 so that ( [ bigbipartite ] ) defines a state in @xmath175 already in schmidt form . the submajorization condition is @xmath176 for which the only nontrivial inequality is the penultimate one @xmath177 . thus the maximum distillation probability is @xmath178 note that this result agrees with the special case in ( [ fourqubits ] ) . we computed monte carlo estimates of @xmath19 and @xmath179 using the hoshen - kopelmann @xcite algorithm with modifications for efficiency @xcite and the mersenne twister @xcite random number generator . together with series expansions and exactly known critical densities , the quality of the data we obtained is more than sufficient to determine which transformations are advantageous . we estimated @xmath19 by computing the mean density of the largest cluster on an @xmath180 lattice . near the critical density we typically used lattices of size @xmath181@xmath182 . we took the value of @xmath183 at the inflection point of plots of @xmath19 _ v.s . _ @xmath183 at fixed @xmath4 as estimates of the correlation length @xmath25 . we typically found that @xmath184 for @xmath185 . we only expect significant systematic error in this region , but this poses no problem because the curves are never close to one another in these regions . ( we reformulated the problem when this is the case . ) we computed statistical error , but the error bars are at most barely visible on the plots , so we omitted them . exceptions are very near @xmath186 where fluctuations in the size of the largest cluster become large , and near @xmath66 , where finite clusters are rare , so collecting sufficient samples to distinguish curves is too expensive . we discuss the effects of these errors in the main body of the paper . we wrote a single computer code to study all the lattices . the code supports lattices with vertices that occupy points on @xmath33 with bonds connecting each pair of nearest neighbors as well as a diagonal bond from @xmath187 to @xmath188 . because the connectivity properties that we calculated depend only on the graph structure of the lattice , we embedded the graph of each lattice in the square lattice plus diagonals described above . depending on the lattice being modeled , some of the bonds in the underlying lattice are closed with probability one , and some are open with probability one , with the vertices identified in the embedded lattice . the graph structures of all lattices appearing in this paper were modeled in this way . we follow the ideas of the perimeter method @xcite to compute high - density series expansions of @xmath19 and @xmath179 listed in table [ tab : series ] . here we discuss the method for computing the series for @xmath19 , but our method for computing @xmath179 is similar . here , a cluster is any connected subgraph that contains at least one vertex . we denote by @xmath189 the collection of all finite clusters that include the vertex at the origin , and by @xmath190 the partition of @xmath189 induced by equivalence under translation , in other words the collection of free clusters . choosing an enumeration @xmath191 of clusters in @xmath190 , the probability that a randomly selected site is in the infinite open cluster is easily seen to be @xmath192 where @xmath193 is the number of sites and @xmath194 the number of bonds in cluster @xmath191 , and @xmath195 is the number of perimeter bonds ( bonds adjacent to cluster @xmath191 ) . clearly , we can find the series expansion in @xmath196 by enumerating the clusters in an order that is non - decreasing in @xmath195 . for instance , on the hexagonal lattice , only the cluster consisting of an isolated site has a perimeter @xmath197 of size less than or equal to @xmath198 ( see fig . [ fig : theta_hex_b ] ) , so that to lowest non - trivial order @xmath199 . percolation theory is a difficult subject precisely because the full enumeration is difficult . the series for @xmath200 ( for bond percolation ) was calculated by machine to high order in ref . . although tables of cluster numbers have been published , all the others that we are aware of are either for site percolation or for bond percolation with cluster size measured by the number of bonds rather than sites ( neither of which can be mapped to our problem . ) for our results we counted a few small clusters by hand , which is not so difficult . in practice however , we find that it is also essential to categorize the clusters by symmetry . as an example , the clusters contributing to @xmath23 on the hexagonal lattice to sixth order in @xmath196 are shown in fig . [ fig : theta_hex_b ] . 10 pt clusters on the hexagonal lattice contributing to @xmath23 to sixth order in @xmath196 . symmetries and multiplicities @xmath201 are : none@xmath202 , one axis@xmath203 , rotation@xmath204 , two axes@xmath198 , rotation and axis@xmath17 , all@xmath38 . the number of contributing clusters per figure is given by @xmath205 , the factor of @xmath206 accounting for the fact that each site only supports half the rotations . the contribution to @xmath23 is then @xmath207 . , title="fig : " ]
strong ovi absorptions observed around starforming galaxies at @xmath8 with impact parameter as high as 150 kpc reveal huge gaseous galactic haloes @xcite . even conservative estimates lead to amount of gas in them far exceeding the gas reservoir in galaxies themselves . such a conclusion is based on the assumption that gas in the haloes has solar metallicity . the arguments underlying this assumption stem from the standard estimate of oxygen mass from the observed column density @xcite [ mo ] m_o = 5r^2n__ovim__o f_hit ( 0.2 f__ovi ) = 1.210 ^ 7 ( 0.2f__ovi)m _ , where a typical column density @xmath9 @xmath3 , the halo radius @xmath10 kpc , the hit rate correction factor ( covering factor ) @xmath11 are assumed following by @xcite , @xmath12 is atomic mass of oxygen . the fractional abundance of ovi under thermal collisional ionization equilibrium never exceeds @xmath13 ( e.g. * ? ? ? * ; * ? ? ? * ) , such that within this assumption eq . ( [ mo ] ) provides a lower estimate of oxygen mass in the halo @xcite . it is readily seen from here that the mass @xmath14 , i.e. around 10 to 70% of oxygen mass in the ism , is indeed a conservative estimate . in principle , this conclusion for galactic haloes to bear such a large gas mass would might solve the problem of missing baryons and metals @xcite , though requiring enormously high oxygen production and mass ejection rates . moreover , the fraction @xmath13 under collisional ionization equilibrium is kept only in a very narrow temperature range : @xmath15 k ( e.g. * ? ? ? * ; * ? ? ? it is therefore totally unrealistic to assume that all observed haloes keep their temperatures within such a narrow range independent on distance from the host galaxy . in their estimates @xcite used the ionic fractions calculated under collisional and/or photoionization equilibrium conditions , i.e. independent of time @xcite . however , ionization state of gas situated in time - dependent ( nonequilibrium ) environment can differ qualitatively from the one settled on to equilibrium , particularly , for solar metallicity @xcite . this difference can be smaller for low - density gas exposed to a strong ionizing field . however , in the process of galactic evolution both the magnitude and the shape of the radiation spectrum can change . in addition the extragalactic spectrum , which can be important on the periphery of galactic haloes , does also evolve @xcite . under such conditions it is natural to expect the ionic composition to experience time variations . in this paper we therefore concentrate on the question of whether evolution of the ionizing radiation field can result in considerable changes of fractional ionization of oxygen such to make estimates of gas mass in galactic haloes more reliable . in next sections we will demonstrate that indeed under time - dependent conditions the observed column densities of ovi correspond to at least half order of magnitude less massive haloes , which can stay in a much wider range of physical parameters . the paper is organized as follows . in section 2 we describe the details of the model . in section 3 we present our results . section 4 summarizes the results . thermal and ionization state in our model is fully time - dependent : the model involves ionization and thermal evolution of gas located at radii @xmath16 kpc in the galactic halo , exposed to extragalactic and galactic time - dependent ionizing radiation field . in this paper we only briefly touch main features of calculation of the ionization and thermal evolution of gas immersed into external time - dependent ionizing radiation . the details can be found in @xcite . we study ionization and thermal evolution of a lagrangian gas element : a gas parcel is assumed to be optically thin to external ionizing radiation . in the calculations we include all ionization states of the elements h , he , c , n , o , ne , mg , si and fe . we take into account the following major processes : photoionization , multi - electron auger ionization process , collisional ionization , radiative and dielectronic recombination , collisional excitation as well as charge transfer in collisions with hydrogen and helium atoms and ions . the total cooling and heating rates are calculated using as a subroutine the photoionization code cloudy ( ver . 10.00 , more specifically , we input into cloudy code a given set of all ionic fractions @xmath17 calculated at temperature @xmath18 , gas density @xmath19 and external ionization flux @xmath20 and obtain the corresponding cooling and heating rates . the latter also includes compton heating from x - rays . for the solar metallicity we adopt the abundances reported by @xcite , except ne for which the enhanced abundance is adopted @xcite . in all calculations we assume the helium mass fraction @xmath21 , which corresponds to [ he / h]=0.081 , and is close to the observed one @xcite . we solve a set of 96 coupled equations ( 95 for ionization states and one for temperature ) using a variable - coefficient ordinary differential equation solver @xcite . we considered the two regimes of gas evolution : isochoric and isobaric . the isochoric regime suggests that gas density in a cloud is kept constant , while in isobaric model gas pressure is assumed constant . the two regimes correspond to two opposite limits of the ratio between the cooling and the sound crossing times : @xmath22 and @xmath23 , correspondingly . in the external heating radiation field isobaric models show essentially similar thermal and ionization evolution , though on longer time scales due to decreasing density coupled to increasing temperature . it results in an increase of the cross - section of gas clouds such that their covering factor increases as well . in the process of galaxy evolution the stellar content changes : massive stars produce enormous number of uv photons and form ultimately compact objects which emit hard ionizing photons . spectrum of the ionizing radiation escaping galactic interstellar medium ( ism ) and exposing further the halo depends on amount of metals in the ism disk absorbing ionizing photons , and thus on chemical evolution of the galaxy . in order to follow evolution of stellar mass , metallicity and galaxy spectrum , we use the spectro - photometric code pegase @xcite . we assume a schmidt - like power - law starformation rate ( sfr ) : @xmath24 , typical sfr for massive starforming galaxies , where @xmath25 is the normalized mass of gas in @xmath26 . in some regions of the galaxy sfr can be inhibited by gas outflows from the disk , however , when averaged over the whole disk the sfr remains sufficiently high over the whole period of active star formation . in our models we assumed a closed - box regime . in general , though this can not be applied to galaxies with active star formation . however , many parameters related to mass and energy exchange between galaxies and the intergalactic medium , such as the rates of mass ejection from and mass accretion from the ambient medium , gas metallicity and corresponding cooling rate , clustering of sne explosions , etc are highly uncertain and hard to be coherently described phenomenologically . gas in galactic haloes is exposed by a cumulative ionizing background consisting of the extragalactic and the galactic components . the extragalactic component is uniform on galactic halo scales , and is nearly constant on timescales @xmath27 myr , while changes significantly on longer times . for the extragalactic background we accept the spectrum described by @xcite . its evolution covers redshifts @xmath28 to @xmath29 divided by 49 equally spaced log bins . contribution from possible flux sources varies over cosmological time , and irregular changes in different bands of the spectrum with redshift can be met . for this reason simple linear approximation between neighbour redshift bins was used . the galactic component instead , may change on much shorter timescales . in general , the corresponding time scale is close to the lifetime of massive stars , i.e. @xmath30 myr . this galactic radiation component originates in the central star forming region , and is seen from outer parts of the halo at distances @xmath31 kpc as a nearly spherical domain of size of @xmath32 kpc . the ionizing radiation from stellar population is partly modified by absorption in the interstellar medium of underlying bulge and disk . in order to account this absorption we assume that ionizing photons pass through a layer of neutral gas in galactic disc with the optical depth @xmath33 , throughout the paper @xmath34 @xmath3 , @xmath35 @xmath3 are considered as fiducial . the corresponding optical depths at the hi and hei lyman limits are as high as @xmath36 and @xmath37 , respectively . as a result only photons with @xmath38 ev escape the galaxy and penetrate into halo ; the ionizing flux with photons of @xmath39 ev decreases as @xmath40 . in what follows we will discuss the dependence of our results on the hi column density . in our model the galactic uv spectrum is calculated with making use the pegase code @xcite , which gives spectral luminosity in the range from 91 to 160 @xmath41 m . in order to extend the spectrum to higher energies ( up to @xmath42 ev , responsible for ionization of highly charged ionic species ) we use the empirical relation between the x - ray luminosity and the star formation rate ` @xmath43 ' @xcite . this relation is well established for massive starforming galaxies considered here . overall the cumulative spectrum varies on time scales from several to hundreds of millions years . we consider gas in outer haloes of massive ( milky way type ) starforming galaxies with stellar mass of several @xmath44 . recent simulations of the milky way halo show that it extends up to the virial radius of the milky way ( i.e. , @xmath16 kpc ) with densities ranging within @xmath45 @xmath7 @xcite . observational estimates of the circumgalactic gas density around the milky way and other local group galaxies give similar numbers @xmath46 @xmath7 at @xmath47 kpc @xcite . in our calculations we follow these numbers and set @xmath48 @xmath7 in the circumgalactic volume . we consider both isochoric and isobaric regimes . we start the calculations at @xmath49 ( the lookback time is around 10 gyrs ) . this timescale is nearly the cooling time for hot gas with @xmath50 k and @xmath45 @xmath7 @xcite . the last major merging for milky way - type galaxies is thought to occur earlier than @xmath51 ( e.g. , * ? ? ? the initial ionic composition and the temperature are set equal to the ones corresponding to photoequilibrium in gas exposed to the extragalactic haardt & madau spectrum at @xmath49 . such radiation field is sufficient to force low density gas with @xmath19 in the accepted range to settle quickly onto photoequilibrium @xcite . calculations cover physical time scales much longer than the relaxation time scale of ionization and thermal processes in gas exposed to the time - dependent spectrum adopted here . gas metallicity in our models is assumed to range within @xmath52 to @xmath53 . higher metallicities are rare in dla qso and dla grb absorbers @xcite . the lower limit lies around ten times above the upper limit of the igm metallicity at @xmath54 ( e.g. , * ? ? ? * ; * ? ? ? ( solid black ) , and the stellar mass , @xmath55 ( solid grey ) ; the initial time corresponds to @xmath49 . the four models with the rate @xmath56 are shown : two models with fixed @xmath57 myr @xmath58 and the initial gaseous mass @xmath59 ( model a solid line ) , and @xmath60 ( model c dashed line ) . the other two models are with fixed @xmath60 and different @xmath61 myr @xmath58 ( model d dotted line ) , and @xmath62 myr @xmath58 ( model e dash - dotted line ) . , width=302 ] from the chemical evolution models calculated with the pegase code we obtain the time - dependent starformation rate , stellar mass , gas metallicity and spectral luminosity . we simulated several models corresponding to massive galaxies and chose two of them , whose parameters , namely the sfr and the stellar mass , are close to those of the starforming galaxies described by @xcite . with the accepted sfr law @xmath63 , where @xmath25 is the normalized mass of gas in @xmath26 and through the paper @xmath64 is assumed , the notations are as in @xcite . in total we consider four models . the first two suggest a fixed normalization factor @xmath57 myr @xmath58 , and two different values of the initial gaseous mass : @xmath59 model a , and @xmath65 model c. the other two models assume a fixed initial gaseous mass @xmath60 , and two different normalization constants @xmath66 : one with @xmath61 myr @xmath58 model d , and the other with @xmath62 myr @xmath58 model e. figure [ figssfr - time ] presents the star formation rate , @xmath67 , the specific starformation rate , @xmath68 , and the stellar mass , @xmath55 : model a ( solid line ) , model c ( dashed line ) , model d ( dotted line ) and model e ( dash - dotted line ) . two features are to note : first , the ssfr reveals differences between the models only after several hundreds of myrs , and second , models a and c show practically equal ssfr , and as a result gas in these models is equally converted into stars by @xmath69 gyr . instead , model d ( @xmath61 myr @xmath58 ) leaves the galaxies gas - rich only @xmath70 % of gaseous mass exhausts by this time . the characteristic time for the gas to exhaust is @xmath71 , which gives @xmath72 myr for all considered models , resulting in a nearly constant sfr ( upper panel in figure [ figssfr - time ] ) over first 200 myr with the sfr@xmath73 , and decreasing on later times as sfr@xmath74 . figure [ figsfr - mstar ] shows the dependence of specific starformation rate on the stellar mass for the models a ( pentagons ) , c ( circles ) , d ( up - triangles ) and e ( down - triangles ) . filled large symbols mark time moments shown nearby . data for the starforming and passive galaxies studied by @xcite are depicted by small squares and rhombs , correspondingly . the grey - scale map shows sdss+galex galaxies @xcite . it is readily seen that almost all points for the starforming galaxies are locked between tracks of the models considered here . one can expect therefore that spectral properties of starforming galaxies from @xcite are similar to those in the models a and c during the latest @xmath753 - 4 gyrs of their evolution . passive galaxies have an order of magnitude lower ssfr . . , width=302 ] and a distance from the galaxy @xmath76 kpc , which consists of the uv and x - ray galactic spectrum , attenuated by galactic neutral gas ( dash line ) , the extragalactic ionizing background ( dotted line ) . the galactic spectral luminosity is shown by dash - dotted line ( the right axis ) . , width=302 ] dash - doted line in figure [ figspec ] shows the galactic spectral luminosity at @xmath77 gyr , which is equal time elapsed from @xmath49 to 0.2 . a significant break is clearly seen at the wavelength 91 , which corresponds to the minimum wavelength in the spectrum reached in the pegase code . the break is due to an exponential decrease of the number of such hard photons emitted by stellar population ( see e.g. * ? ? ? * ) , as they are only produced by very massive stars , whose number is very small . as soon as we consider galactic evolution on timescales longer than 1 - 3 gyr x - ray binaries are expected to have already formed , and we extend the spectrum to higher energies assuming the empirical relation between x - ray luminosity and star - formation rate `` @xmath43 '' @xcite . by the thick grey line figure [ figspec ] shows also an example of the total spectral radiation flux exposing a given gas parcel located at distance @xmath76 kpc from the galaxy evolved till @xmath78 . the total spectrum consists of the galactic ( dash line ) and extragalactic ( dotted line ) ionizing photons . at low energies , @xmath79 ev , the extragalactic contribution dominates at large distances , @xmath80 kpc , while the stellar population turns into play in uv range at smaller distances . strong absorption of the galactic ionizing photons ( @xmath81 ev ) in the galactic disk with the fiducial @xmath82 and @xmath83 values results in the total dependence of the ionic composition of halo gas on the extragalactic background . the significance of absorption by the galactic disk can be understood from comparison of the galactic spectral luminosity shown by dash - dotted line and the cumulative flux ( thick grey line ) . a narrow bump at @xmath84 ev is due to the galactic photons survived against absorption in thick disk . its magnitude is obviously determined by our choice of the neutral column densities @xmath82 and @xmath83 in the disk . for the fiducial values of @xmath82 and @xmath83 the optical depth is about 3 for photons with @xmath85 ev . higher column densities can erase the bump and the extragalactic flux becomes dominating in the whole galactic halo . decrease of the galactic contribution is also seen at larger distances from the galaxy due to dilution @xmath40 . note that the ionization potential of ov , @xmath86 ev falls exactly in this the range @xmath84 ev . this means that the fraction of ovi ions can increase in galaxies with lower column densities ( @xmath82 and @xmath83 ) in their disks . ( dash lines ) and @xmath87 ( dash - dotted lines ) , where the superscipts g and e refer to galactic and extragalactic backgrounds . thick and thin lines correspond to redshifts 1.8 and 0.2 , respectively . , width=302 ] in general , the excess of galactic photons with energies higher than 113.9 ev contributes crucially into ionization kinetics of ovi ion . figure [ figflux - ratio ] presents the ratio of monochromatic galactic and extragalactic fluxes at 113.9 ev ( solid lines ) in model a for our fiducial column densities @xmath82 and @xmath83 . it is clearly seen that the galactic flux dominates within @xmath88 kpc at redshifts @xmath89 . at lower redshifts the region of predominance of galactic flux widens mostly due to a steep decrease of the extragalactic flux in these epochs . the monochromatic radiation at the hi lyman limit is fully absorbed in the disk for the fiducial @xmath82 and @xmath83 column densities . in order to estimate both the escape of hi ionizing photons and the efficiency of ov ionization we calculate radial dependences of the ratios of ionizing fluxes @xmath90 and @xmath87 , where the superscipts _g_ and _e_ refer to the galactic and extragalactic contributions . figure [ figflux - ratio ] shows these ratios at @xmath91 and 0.2 for model a. in the energy range @xmath92 ev the galactic contribution dominates up to distance @xmath93 kpc at @xmath91 , and extends even to @xmath88 kpc at @xmath78 due to strong decrease of the extragalactic background at low redshifts . it is obvious , that increase of ionizing flux enlarges radius of the zone of predominance of galactic ionizing photons : for instance , this zone increases from @xmath94 kpc in model a to @xmath95 kpc in model c at @xmath91 . decrease of absorption in the disk does also promote the zone of the galactic predominance to grow : it reaches @xmath96 kpc for @xmath97 @xmath3 even in model a. and @xmath83 . , width=302 ] figure [ figt - ovi - evol ] shows evolution of temperature ( lower panel ) and ovi fraction ( upper panel ) in gas located at different distances from the galactic center ; here we assume the sfr as in model a and our fiducial column densities @xmath82 and @xmath83 . in the beginning , @xmath51 , oxygen is mainly locked in the ovii state due to high ionizing flux . its fraction establishes around @xmath98 under the influence of the extragalactic ionizing background . after several hundreds million years the extragalactic background starts decreasing following the cosmic star formation rate . it should result in a quick transition from ovii to ovi . in a cooling plasma ovi recombines rapidly into lower ionization states and practically disappears shortly a well - known ovi `` fragility '' . however , in our case the excess of photons with @xmath99 ev emitted by starforming galaxies does not allow ovi to recombine . consequently , its fraction remains almost frozen at the level @xmath98 in the region @xmath100 kpc over the range @xmath101 till 0 . in our models @xmath102 reaches @xmath0 in a low - metallicity ( @xmath53 ) gas within @xmath103 kpc from @xmath104 to @xmath105 . this value is several times higher than the maximum ovi fraction , @xmath106 , reached in gas exposed only to the extragalactic background ( e.g. * ? ? ? * ; * ? ? ? * ) , and in nonequilibrium collisional gas evolved from @xmath107 k ( see e.g. * ? ? ? * ) . note that even at large distances @xmath108 kpc @xmath102 remains higher than @xmath109 . the temperature of gas with such high ovi fraction is within @xmath110 k ( see lower panel in figure [ figt - ovi - evol ] ) . the dependence of @xmath102 on metallicity is weak : for instance , ovi fraction in gas with @xmath111 is @xmath112 within distances @xmath113 kpc in model a. cooling of gas with @xmath114 exposed only to the extragalactic background is mainly due to hydrogen and helium , whereas metals ( oxygen and carbon ) play a minor role in radiation losses ( e.g. * ? ? ? * ; * ? ? ? * ) . a considerable increase of ovi fraction in the zones where galactic ionization dominates enhances the contribution of oxygen into cooling . as a result , gas temperature is lower in these regions as seen on the lower panel in figure [ figt - ovi - evol ] . deviation of column densities from the fiducial values accepted above may result in a considerable change of the overall picture due to changes of the interrelation between the fractions of galactic and extragalactic ionizing photons . in order to understand how sensitive is the oxygen ionization state to surface density of the underlying galactic gaseous discs we calculate several models with different @xmath115 and @xmath116 . for simplicity we assume that @xmath117 . figure [ fig - ovi - hi ] presents the galactic part of the spectrum at distance @xmath76 kpc and redshift @xmath118 for several values @xmath115 ( upper panel ) , and the evolution of ovi fraction in gas exposed to this radiation field ( lower panel ) . for column density @xmath119 @xmath3 the galactic spectrum in model a is fully absorbed in the range 13.6 - 80 ev . only for a hundredth of this column density the attenuation becomes less depressing . a lower column density @xmath120 @xmath3 allows a higher fraction of ionizing photons with the energy above the ionization potential of ov ( 113 ev ) to penetrate into the halo and support a higher fraction of ovi ( lower panel ) : the ovi fraction reaches @xmath121 and slightly increases for lower column densities . this value is considerably higher than in collisionally ionized gas and in gas photoionized only by the extragalactic background . the reason for such a high fraction of ovi is in the break at the energy slightly lower than the ovi ionization potential ( 136 ev ) . it is worth noting that the uncertainty of the spectral energy distribution in this energy range within stellar population codes is rather high . this partly stems from the stellar atmosphere models and the rarity of extremely massive stars @xcite . on one hand this is a shortcoming of our model , though on the other , the rarity of massive stars and their short lifetime allows us to think that stellar contribution into circumgalactic ionizing field falls steeply down at energies @xmath122 ev ( see the spectra of the hottest stars in * ? ? ? * ) , which is lower than the ovi ionization potential ( 136 ev ) . however it is important to note that that the predominance of galactic flux over the extragalactic background in the range around 110 - 130 ev results in ovi fraction as high as 0.6 - 0.8 . higher hi column density in underlying galactic discs , @xmath123 @xmath3 , heavily erases galactic flux in the range @xmath124 ev , which , for instance in model a becomes lower than the extragalactic background and the effect of an enhanced ovi fraction vanishes . kpc and redshift @xmath118 for several values @xmath115 ( upper panel ) , and the evolution of ovi fraction in gas exposed this spectrum ( lower panel ) . the thick black line on the upper panel shows the extragalactic part of the spectrum . redshift dependence of the fraction of ovi ions on the lower panel for different column densities on lower panel shows a catastrophic change at @xmath125 @xmath3 with a heavily depressed @xmath102 at lower @xmath126 . the ratio @xmath127 is assumed . , title="fig:",width=302 ] kpc and redshift @xmath118 for several values @xmath115 ( upper panel ) , and the evolution of ovi fraction in gas exposed this spectrum ( lower panel ) . the thick black line on the upper panel shows the extragalactic part of the spectrum . redshift dependence of the fraction of ovi ions on the lower panel for different column densities on lower panel shows a catastrophic change at @xmath125 @xmath3 with a heavily depressed @xmath102 at lower @xmath126 . the ratio @xmath127 is assumed . , title="fig:",width=302 ] k ( thick grey ) and @xmath129 k ( thin grey lines ) for surface gravity @xmath130 , @xmath131 and @xmath132 cm s@xmath133 ( dark grey , grey and light grey lines ) at solar metallicity taken from the stellar library calculated by @xcite are shown by grey lines ( the left axis ) . the galactic spectral luminosity accepted in model a is depicted by thick black line ( the right axis ) . several power - law functions @xmath134 are shown for comparison . _ middle panel : _ the spectral luminosity in models a ( black line ) and a+1 .. a+5 at @xmath78 , corresponding to different power - law index @xmath135 as shown by grey lines . _ bottom panel : _ the cumulative ionizing background flux at a distance from the galaxy @xmath76 kpc and redshift @xmath78 in models a , a+1 .. a+5 ( thick lines ) , the extragalactic ionizing background ( thin dotted line ) . models c+1 , c+2 , ... are defined as the model c with the spectra accepted for models a+1 , a+2 , ... , correspondingly . , title="fig:",width=291 ] k ( thick grey ) and @xmath129 k ( thin grey lines ) for surface gravity @xmath130 , @xmath131 and @xmath132 cm s@xmath133 ( dark grey , grey and light grey lines ) at solar metallicity taken from the stellar library calculated by @xcite are shown by grey lines ( the left axis ) . the galactic spectral luminosity accepted in model a is depicted by thick black line ( the right axis ) . several power - law functions @xmath134 are shown for comparison . _ middle panel : _ the spectral luminosity in models a ( black line ) and a+1 .. a+5 at @xmath78 , corresponding to different power - law index @xmath135 as shown by grey lines . _ bottom panel : _ the cumulative ionizing background flux at a distance from the galaxy @xmath76 kpc and redshift @xmath78 in models a , a+1 .. a+5 ( thick lines ) , the extragalactic ionizing background ( thin dotted line ) . models c+1 , c+2 , ... are defined as the model c with the spectra accepted for models a+1 , a+2 , ... , correspondingly . , title="fig:",width=283 ] k ( thick grey ) and @xmath129 k ( thin grey lines ) for surface gravity @xmath130 , @xmath131 and @xmath132 cm s@xmath133 ( dark grey , grey and light grey lines ) at solar metallicity taken from the stellar library calculated by @xcite are shown by grey lines ( the left axis ) . the galactic spectral luminosity accepted in model a is depicted by thick black line ( the right axis ) . several power - law functions @xmath134 are shown for comparison . _ middle panel : _ the spectral luminosity in models a ( black line ) and a+1 .. a+5 at @xmath78 , corresponding to different power - law index @xmath135 as shown by grey lines . _ bottom panel : _ the cumulative ionizing background flux at a distance from the galaxy @xmath76 kpc and redshift @xmath78 in models a , a+1 .. a+5 ( thick lines ) , the extragalactic ionizing background ( thin dotted line ) . models c+1 , c+2 , ... are defined as the model c with the spectra accepted for models a+1 , a+2 , ... , correspondingly . , title="fig:",width=283 ] kpc from the galaxy exposed to the spectra corresponding to models a , a+1 .. a+5 . , width=302 ] it is obviously clear from discussion in section 3.2 that the enhanced ovi fraction is mainly due to the excess of galactic photons with @xmath136 ev in the bump at @xmath84 ev . this bump maintains the ovi fraction at high level and keeps them against recombination to lower ionic states . the left border of this spectral feature , formed by to strong absorption of the galactic ionizing radiation in the disc , is irrelevant from the point of view of the maintainance of ovi even in the absence of any absorption it remains enhanced ( see figure [ fig - ovi - hi ] ) . the right edge of the bump is , however , artificial in origin : as a matter of fact , the cut at 136 ev ( 91 ) is a choice of the developers of stellar population codes ( e.g. pegase , starburst99 , galaxev ) . therefore , in the following we discuss convergence of our results against possible galactic spectral variations at energy higher than 136 ev . extreme ultraviolet and soft x - ray photons ( @xmath137 ev ) can be produced by post - agb stars before entering the white dwarf cooling phase ( e.g. * ? ? ? during this very bright evolution phase lasting @xmath138 yrs the effective temperature may reach more than @xmath139 k , while the surface gravity varies as @xmath140 to @xmath141 cm s@xmath133 . figure [ fig - spec - z02 ] ( upper panel ) presents the two spectra for @xmath142 k and @xmath129 k for solar metallicity taken from the stellar library @xcite . it is seen that the amount of photons with @xmath143 ev increases when the surface gravity @xmath144 decreases from @xmath131 to @xmath132 cm s@xmath133 for @xmath145 k. on the other hand even a small decrease of effective temperature results in a shortage of photons with @xmath146 ev independent on @xmath144 . the majority of observed post - agb stars is known to have masses below @xmath147 and temperatures @xmath139 to @xmath129 k ( e.g. * ? ? ? * ) , thus falling into the range shown here . for several spectra a break around @xmath148 ev is clearly seen . one can think that in order to conservatively account this break in stellar population codes the value 136 ev ( 91 ) was choosen as an upper limit of energy in their stellar library . it is clearly seen though that even the spectra with a break at @xmath148 ev continue up to @xmath149 ev and do not show such a strong decrease of flux at @xmath143 ev as used in our calculations . one can think that this gradually decreasing flux above @xmath148 ev produces qualitatively similar effects as does the flux with the break at @xmath143 ev . in order to check this tentative conclusion we performed calculations of ionic composition for several models of the ionizing flux above 136 ev we construct several spectra and test the appearence of enhanced ovi fraction . we smooth the spectra calculated in pegase code at energies @xmath150 ev and match them at @xmath151 ev to the spectrum obtained from the `` @xmath43 '' relation as described in section 2.3 ( see * ? ? ? we proceed in the manner that the resulting spectrum would be qualitatively close to the spectra depicted in the upper panel . figure [ fig - spec - z02 ] ( middle panel ) shows these spectra normalized on the spectral luminosity in model a. we calculate the ionization and thermal evolution of gas exposed to a cumulative spectrum which includes the extragalactic and the galactic radiation field . figure [ fig - spec - z02 ] ( bottom panel ) presents an example of the cumulative ionizing background flux at a distance @xmath76 kpc from the galaxy in models a , a+1 .. a+5 located at a redshift @xmath78 ( thick lines ) . figure [ fig - ovi - spec ] shows evolution of ovi fraction in gas exposed to the ionizing flux as in models a , a+1 .. a+5 . it is readily seen that in all a+ models the ovi fraction is greater than 0.1 at redshifts @xmath152 . the ovi fraction is several times lower than in model a , but this is still obviously higher than @xmath153 , which establishes in gas exposed only to the extragalactic radiation ( compare to the dot - dot line for log n(hi)=20.5 in lower panel of figure [ fig - ovi - hi ] ) . in models a+1 and a+2 oxygen at @xmath154 is mainly locked in ovii ions . starting from @xmath104 the ionization production rate of ovii ions decreases such that recombinations become efficient to replenish lower ionization states , and as a result the fraction of ovi increases at @xmath155 substantially : from @xmath156 at @xmath157 to @xmath158 in a+1 and @xmath159 in a+2 at @xmath152 . this balance between ovii and ovi is kept then quasi - steady down to @xmath160 . at large distances the contribution from galactic photons decreases resulting in a higher fraction of ovi in models a+1 and a+2 . note that contrary to models a+1 and a+2 the flux in model a+5 is low and oxygen accumulates mostly in oiv - ov states . higher galactic luminosities presented in a set of models c and c+3 .. c+5 , result in a higher ovi fraction : @xmath161 in gas located at @xmath76 kpc at @xmath152 . however , in models c+1 and c+2 the ionization flux at @xmath146 ev is excessive and the fraction of ovi decreases to 0.03 and 0.07 at @xmath78 , and reaches only 0.05 and 0.1 , at @xmath28 . at larger distances from a galaxy where the flux drops models c+1 and c+2 show an increase of ovi fraction : e.g. , in model c+1 it increases to @xmath162 ( 0.11 ) at 150 kpc and @xmath163 ( 0.17 ) at 200 kpc at @xmath78 ( @xmath28 ) . it is readily seen thus that in the majority of these models ovi fraction remains high and reaches @xmath164 at distance @xmath80 kpc from a galaxy located at @xmath152 . this is more than an order of magnitude higher than can be reached in gas exposed only to the extragalactic radiation at @xmath152 . in model c+1 the ovi fraction reaches 0.1 only for @xmath165 kpc ( @xmath152 ) , while at closer distances most of oxygen is confined into ovii . it is worth stressing though that the model c+1 can be considered as a model with an extreme possible contribution from post - agb stars , in the sense that models with harder spectrum are unlikely ( e.g. * ? ? ? * ) , thus one can expect that within conservative models of galactic x - ray spectrum the effect of enhanced ovi remains stable . in the previous section we have found that ovi fraction can reach high values of @xmath166 under the action of ionizing radiation from the underlying galactic stellar population along with the extragalactic radiation field . such ovi fraction is at least an order higher than the maximum @xmath163 reached in commonly used models with gas ionized collisionally and/or by the extragalactic radiation . assuming the spherical symmetry and using the time - dependent ovi radial distribution around a host galaxy one can find the ovi column densities along the line of sight crossing the galactic halo at impact parameter @xmath167 . we integrate from @xmath168 kpc to 300 kpc . @xmath7 , the middle and lower panels show the column density for the isobaric ( ib ) regime with the initial density @xmath169 @xmath7 and @xmath170 @xmath7 , respectively ; the metallicity is @xmath53 . open rhombs and squares depict the observed column densities in the passive and starforming galaxies , correspondingly @xcite ; integration is over @xmath168 kpc to 300 kpc , different symbols correspond to different redshifts . , width=302 ] , but for gas metallicity @xmath111 . , width=302 ] figure [ figcd_ovi ] ( upper panel ) presents the dependence of @xmath171 on impact parameter @xmath167 in a @xmath172 isochoric gas within the models a and c. the ovi column density ranges from @xmath173 to @xmath174 @xmath3 at the impact parameter @xmath175 kpc a factor of 3 - 10 higher than observed by @xcite . therefore circumgalactic gas of an order of magnitude lower metallicity @xmath176 would provide good agreement with observations as shown in figure [ figcd_ovi2 ] ( upper panel ) . it is worth noting though that spatial distribution of metals in circumgalactic environment is highly nonhomogeneous ( e.g. * ? ? ? * ; * ? ? ? * ) , and it is most likely for metals to be locked in clumps of smaller covering factor resulting in a proportional decrease of the column density . figure [ figcd_ovi ] ( middle and lower panels ) show the column density of ovi in isobaric gas with the initial density @xmath169 and @xmath170 @xmath7 , respectively . isobaric regime is supported by radiation losses and heating from ionizing radiation . photo - heating grows at large distances from the galaxy , @xmath177 kpc where contribution from ovi ions to cooling decreases . at lower distances ovi fraction increases and it becomes a dominant coolant resulting in a rather efficient cooling . consequently , column densities grow under isobaric compression : @xmath171 can reach @xmath178 @xmath3 for the initial density @xmath169 @xmath7 , and @xmath179 @xmath3 for @xmath170 @xmath7 . similarly to the isochoric case gas with lower metallicity ( 0.01 @xmath180 ) fits observational data better . such high column densities are reached for galactic spectra with a cut at 91 ( see footnote in section 3.1 ) . in reality though galactic spectra do not show such sharp breaks . in section 3.3 we considered galactic spectra approximated smoothly around @xmath181 ev to account possible variations of contribution from post - agb stars . we have found that in the low redshift range , @xmath152 , ovi fraction in gas exposed to such approximated spectra remains sufficiently high in the majority of models considered here . in all cases we have a considerable ( an order of magnitude ) excess of the column densities caluculated in models a , c and with respect to the observed ones as seen in figure [ figcd_ovi ] ) . @xcite have argued that the circumgalactic medium can be a significant reservoir of the ejected material from galaxies . in order to explain large column densities of ovi observed in the haloes they have had to assume that the circumgalactic gas has nearly solar metallicity . this assumption meets though difficulties because the minimum oxygen mass in the halo obtained within such an assumption reaches around 10 - 70% of the total oxygen mass in the ism @xcite , which requires in turn unprecedentedly high mass exchange between the galactic ism disk and a huge circumgalactic reservoir extending up to 150 kpc . higher ovi fraction obtained in our model reduces the estimates to a more reasonable level , and as a consequence weakens constrains on the sources of oxygen in underlying galaxies . contrary , an order of magnitude lower metallicity results in a more consistent scenario of metal production and mass exchange between galactic and circumgalactic gas . indeed , the total mass of the circumgalactic gas in ovi haloes within @xmath182 kpc is about @xmath183 independent on redshift as estimated from hi data @xcite . if such gas would be enriched to the metallicity as high as @xmath53 , then the oxygen mass in it equals @xmath184 . this amount can be produced during a 0.2 gyr period in a galaxy with sfr @xmath185 ( as in our model a ) , with 0.015 @xmath26 of oxygen returned to the ism per each 1 @xmath26 star formed ( e.g. , * ? ? ? * ) . this timescale is significantly shorter than the whole period of the evolution several gyrs , and comparable to the initial period of evolution in our models with the sfr kept nearly constant ( upper panel on figure [ figssfr - time ] ) . because of the high sfr during initial several hundreds myrs , gas can be ejected from the disk with velocities @xmath186 km / s sufficient to reach @xmath75100 kpc in approximately 1 gyr . we assumed throughout that metals are honogeneously distributed in a spherical layer between 50 and 300 kpc . this can be an overestimate , because in general the volume occupied by ejected metals depends on the galaxy mass and for dwarf galaxies can be smaller ( e.g. * ? ? ? moreover , metals are locked most likely into small - size dense fragments ( e.g. * ? ? ? * ) , where radiative cooling is efficient . it can result in a rapid cooling and compression and make such fragments to escape detection . dwarf satellites may add metals in the halo , though their contribution is apparently small because of low metallicity in their ism : @xmath187 ( e.g. , * ? ? ? * ; * ? ? ? * ) and massive dwarfs are rare ( e.g. , * ? ? ? * ) . in extended galactic haloes gas with a low density , @xmath188 @xmath7 , is exposed to rather a high ionizing background consisting of galactic and extragalactic components . under such conditions the ionic state may reach photoequilibrium . ionic composition under photoequilibrium is calculated with making use the cloudy . we assume the time - dependent ionizing background described in section 2.3 . the difference between our calculations and the ones performed by @xcite , is that besides the haardt - madau extragalactic radiation field used in @xcite we add the galactic contribution , which as we showed above may have crucial consequences : the excess of photons with @xmath189 ev from the underlying stellar population competes the extragalactic radiation , while x - ray photons in the excess with @xmath136 ev substantially enhance ovi fraction ( section 3.2 ) . for the fiducial column densities @xmath190 @xmath3 and @xmath191 @xmath3 ovi fraction in photoequilibrium is several magnitudes smaller than in our time - dependent models . in photoequilibrium oxygen is mainly locked in less ionized states , oiv - ov , and ovi column density drops to @xmath192 @xmath3 . when @xmath82 and @xmath83 decrease the fraction of ovi instead increases . even for @xmath193 @xmath3 ( here @xmath194 is assumed ) the ovi column density reaches @xmath195 @xmath3 at the impact distances @xmath196 kpc and @xmath118 . further decrease of @xmath82 results in ovi fraction to grow up to @xmath197 , and consequently in higher ovi column densities . figure [ figcd_ovi - eq ] presents the dependence of the ovi column density on the impact parameter for photoequilibrium model with @xmath198 @xmath3 and time - dependent model with @xmath199 @xmath3 . the clearly seen considerable difference between photoequilibrium and time - dependent models originates from a well - known fact that gas under nonequilibrium conditions is overionized with respect to what occurs under equilibrium ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? when @xmath82 decreases , the absorption of galactic ionizing radiation falls down and the peak in the spectrum around @xmath200 ev grows . at such conditions photoionization time shortens , @xmath201 , and the ionic composition approaches photoequilibrium . at low @xmath126 the extragalactic flux decreases : for example , for @xmath118 @xmath202 erg s@xmath203 @xmath3 hz@xmath203 at @xmath204 ev ( figure [ fig - ovi - hi ] ) . as a result the phoionization timescale @xmath205 of ovi ions increases up to several hundreds myr , here @xmath206 @xmath3 and @xmath207 are assumed ; @xmath208 ev is the ovi ionization potential and @xmath209 is the width of the peak between 113 - 136 ev . this is about several times shorter than the recombination time for gas with @xmath210 @xmath7 making nonequilibrium effects important . smaller absorption in the galactic disc provides higher galactic ionizing flux and shorter photoionization timescale such that ionic composition shifts to photoequilibrium . thus , nonequilibrium effects are important for ovi ionization kinetics in the circumgalactic gas with @xmath211 @xmath7 at @xmath5 . t is therefore can be concluded that the main factor which provides high ovi column densities in massive starforming galaxies at low and moderate metallicities is the excess of photons with energies 113 - 130 ev from the stellar population of a host galaxy . @xmath3 , whereas in time - dependent model @xmath212 @xmath3 is adopted ( @xmath213 is assumed ) . thin and thick lines correspond to nonequilibrium and photoequilibrium , respectively ; density and metallicity are @xmath210 @xmath7 and @xmath53 , the other notations are as in figure [ figcd_ovi ] . , width=302 ] in this paper we have presented the photoequilibrium and nonequilibrium ( time - dependent ) ionization and thermal state of circumgalactic gas located at distances up to @xmath16 kpc around starforming galaxies , and exposed to both extragalactic and galactic time - dependent ionizing background . for the extragalactic background we considered the spectra obtained by @xcite . using the pegase code @xcite we have calculated chemical and spectro - photometric evolution of galaxies , and have chosen the two models , whose specific star formation rate ( @xmath214 ) and stellar masses are close to the starforming galaxies with large ovi column densities observed in @xcite . we have found that * the maximum ovi fraction can reach @xmath0 under physical conditions ( gas density and metallicity , and the spectrum shape ) , which are typical in haloes of starforming galaxies ; such a high ovi fraction is due to galactic photons with @xmath1 ev ; the effect of enhanced ovi remains stable within conservative models of galactic x - ray spectrum fluctuations at @xmath215 ev ; * due to such high fraction of ovi its column density ranges in @xmath216 @xmath3 even for a low metallicity @xmath217 , and @xmath218 @xmath3 for @xmath219 at impact parameters up to @xmath220 kpc ; this results in several times more conservative estimate of the oxygen mass in haloes compared to @xmath221 @xcite . * we have shown therefore that the large ovi column densities observed in haloes of starforming galaxies @xcite can be found in circumgalactic conditions with nearly 0.01 - 0.1 of solar metallicity , and correspondingly the requirements to the sources of oxygen in the extended haloes become reasonably conservative . high ovi column densities in haloes of starforming galaxies can emerge under photoequilibrium as well as under nonequilibrium conditions . the main source is a high radiation flux of photons with @xmath1 ev from the stellar population of starforming galaxies . nonequilibrium effects for ovi ionization kinetics are important in the circumgalactic environment with @xmath211 @xmath7 at @xmath5 . very recently @xcite have reported about high ovi column densities in the circumgalactic medium at @xmath54 with @xmath171 reaching @xmath222 @xmath3 ; the observational sample includes absorbers of different type : from lyman limit to damped ly@xmath135 systems . from our point of view such a high ovi column density can be due to the excess of galactic ionizing photons with @xmath223 ev over the extragalactic background , while the metallicity might be rather low . we thank jason tumlinson for providing data and discussion , and the anonymous referee for valuable comments and pointing to a mistake . this work is supported by the rfbr through the grants 12 - 02 - 00365 , 12 - 02 - 00917 , 12 - 02 - 92704 , and by the russian federal task program `` research and operations on priority directions of development of the science and technology complex of russia for 2009 - 2013 '' ( state contracts 14.a18.21.1304 , 2.5641.2011 and 14.a18.21.1179 ) . ev is grateful for support from the `` dynasty '' foundation . asplund , m. , grevesse , n. , & sauval , a. j. 2005 , in asp conf . 336 , cosmic abundances as records of stellar evolution and nucleosynthesis , ed . t. g. barnes iii & f. n. bash ( san francisco : asp ) , 25
the qcd sum rule ( qcd - sr ) approach to qcd continues to be a highly active field . the influence of the field is reflected in over 1500 references to the seminal paper of shifman , vainshtein and zakharov ( svz ) @xcite , with over 380 of these references following the beginning of 1992 @xcite . while the approach has been applied to a variety of hadronic observables in both vacuum and finite density nuclear matter , a comprehensive review of the systematic errors associated with the approach has not been considered . one of the key assumptions of the approach is the use of the so called `` continuum model '' to remove excited state contaminations from the hadron correlator under investigation . recently , the continuum model for the nucleon was tested by using correlation functions calculated via lattice regularized qcd @xcite . the results suggest that the continuum model adequately removes the excited state contaminations and allows the isolation of the ground state . the success of the continuum model warrants a more careful investigation of other aspects of the qcd - sr approach so that uncertainties in the predictions may be reliably and quantitatively determined . the focus of this paper is to establish a rigorous procedure for extracting quantities of phenomenological interest from qcd sum rules . the main conclusion of this investigation is that qcd sum rules work and they are predictive when the analysis is done rigorously . the ground state masses of the @xmath0-meson and the nucleon are used to introduce the concepts . however , the procedures introduced here are more generally applicable to quantities of current experimental interest . the goal is to shed light on the stability of qcd sum rule analyses and thus the accuracy to which observables may be reliably determined . implementation of the procedure introduced here does not lead to slight adjustments in the extracted parameters , but rather leads to qualitatively different conclusions . it will become apparent that most existing qcd - sr investigations of nucleon properties are in fact unreliable . previous calculations are based on sum rules in which neither the operator product expansion ( ope ) nor the phenomenological description are under acceptable control . the qcd - sr method is presently the best fundamentally based approach for investigating the properties of hadrons in nuclear matter , as the lattice approach is challenged by a number of formidable obstacles @xcite . issues surrounding the finite width of the rho - meson make reliable lattice calculations of @xmath2 mixing very difficult . these are two of the many topics of current interest to nuclear physicists where the best fundamentally based approach is that of qcd - srs . moreover , the approach has minimal model dependence . hence it is imperative to probe the predictive ability of the approach and the ideas presented here are currently being applied to these topics mentioned above @xcite . the most significant source of uncertainty in the qcd sum rule approach is an imprecise knowledge of the vacuum condensates appearing in the operator product expansion ( ope ) . this is particularly problematic for the higher dimension operators , where one usually invokes the vacuum saturation hypothesis ( factorization ) to replace higher - dimension - operator vacuum - expectation values by products of lower - dimension operators . a list of key sources of uncertainty in the qcd - sr approach should include : * the unfactorized condensate values , * factorization of higher dimensional operators , * small but neglected @xmath3 corrections , * the truncation of the operator product expansion ( ope ) , * the selection of the regime for matching the qcd and phenomenological sides of the sum rules , * uncertainties associated with the summation of perturbation theory in large orders , and * the possibility of significant direct instanton contributions . with the exception of direct instanton contributions , the effects of these uncertainties will be estimated via a monte - carlo error analysis . gaussian distributions for the condensate values are generated via monte carlo . these distributions are selected to reflect the spread of values assumed in previously published qcd - sr analyses , and the uncertainties such as operator factorization listed above . these distributions provide a distribution for the ope and thus uncertainty estimates for the ope which will be used in a @xmath4 fit . in turn , the ope distribution provides distributions for the phenomenological fit parameters , thus establishing the predictive ability of the qcd - sr approach . in the field of qcd inspired models , the focus is often on whether or not the model is able to encompass the known experimental data , as opposed to assessing the predictive potential of the method . a review of the qcd - sr literature leaves no doubt as to whether or not the qcd - sr method is able to encompass the established data . the poorly determined condensate values and arbitrary aspect of the matching borel regime provides tremendous freedom in devising ways to achieve the desired result in fact , it is difficult to find qcd - sr - based predictions of hadronic observables which fail to agree with experiment . this is somewhat surprising , since the physics responsible for some observables may be poorly represented in a truncated ope commonly limited to distances of less than 0.3 fm . most qcd - sr calculations of nucleon properties in the literature fall short in the analysis stage where the phenomenological parameters are determined by matching the qcd and phenomenological sides of the sum rules . typical analysis short - comings include : * selecting a single value for the borel parameter which gives `` nice '' results . a borel _ regime _ should be selected in order to evaluate the stability and reliability of the results . * a selection of the borel regime without careful regard to ope convergence nor maintaining ground state dominance of the phenomenological side of the sum rules . * the fixing of search parameters ( such as the continuum threshold ) to preferred values . it will become apparent that this introduces a strong bias to the remaining fit parameters which may not reflect the properties of qcd . * claiming an accuracy for qcd - sr predictions without supporting calculations . occasionally a `` stability analysis '' @xcite is considered , in which fit parameters are monitored as a single condensate value is varied . however , such analyses explore a relatively small corner of the condensate parameter space . in general , former procedures adopted for matching the two sides of the sum rules lack the level of rigor which will be presented here . this general lack of rigor can be traced back to the original paper of svz @xcite . in their conclusions , they comment `` we prefer not to deepen into the debris of computer calculations , sophisticated fit programs , arranging error bars here and there@xmath5 '' while this may be appropriate for a seminal paper on the qcd - sr approach , continuation of this philosophy has lead to strong negative criticism of the qcd - sr field . it is not uncommon to hear remarks such as `` qcd sum rules are useful only if you know the answer . '' or `` you can get anything you want from qcd sum rules . '' and `` 10% here , 10% there pretty soon you re talking about real numbers ! '' these remarks are unfortunate , as a rigorous and proficuous analysis is possible . in this paper , a method for the quantitative determination of phenomenological quantities with uncertainties will be presented . since this paper is directed to both qcd - sum - rule practitioners and skeptics alike , we begin by briefly outlining the qcd - sr approach in section [ qcdsr ] . a more pedagogical review of the approach may be found in the recent review article of ref . this section also introduces the `` ioffe formula '' @xcite , argued by many to qualitatively encapsulate a description of the nucleon mass in terms of the chiral symmetry breaking order parameter @xmath1 . uncertainties associated with the evaluation of the ope will be addressed via a new monte - carlo based procedure . estimates for these uncertainties are presented in section [ uncert ] . section [ analysis ] details the monte - carlo uncertainty analysis and discusses the optimization algorithm and cautious convergence criteria used in the following . section [ rho ] presents an analysis of @xmath0-meson sum rules . here the criteria for the selection of the regime for matching the qcd and phenomenological sides of the sum rules is reviewed . reasonable alternatives are considered via monte carlo . these considerations lead to the selection of the optimal interpolating field for nucleon sum rules . new insights into unconventional nucleon interpolators obtained from the lattice qcd investigation of ref . @xcite play a significant role here . section [ nucleon11 ] reviews the selection of the optimal nucleon interpolating field for correlators obtained from spin-1/2 interpolating fields . the effects of choosing non - optimal interpolators is demonstrated . section [ nucleon13 ] explores additional sum rules obtained from the overlap of the generalized spin-1/2 and a spin-3/2 interpolator . in section [ correl ] we explore the correlations between condensate values , fit parameters and sum rule consistency . these correlations serve to identify the roles of the leading terms of the ope in hadronic physics . the validity of the `` ioffe formula '' is evaluated here , and these results may be of interest to those modeling the qcd vacuum . the contingency table analysis reveals that the intimate relationship between the quark condensate and the nucleon mass suggested in the `` ioffe formula '' is invalid . the necessity of direct instanton contributions to the qcd sum rules is determined in section [ instantons ] . we demonstrate that there is no evidence indicating that direct instanton contributions are required to maintain sum rule consistency . the implications of the results of this analysis on coordinate space correlators are discussed in section [ coordimpl ] . finally , section [ conclusions ] summarizes the findings of this in - depth investigation . the underlying principle of field theoretic approaches to hadron phenomenology is the ansatz of duality . that is , it is possible to simultaneously describe a hadron as quarks propagating in the qcd vacuum , and as a phenomenological field with the appropriate quantum numbers . hadron masses are extracted from an analysis of the two - point function @xmath6 the interpolating field @xmath7 is typically constructed from quark field operators combined to give the quantum numbers of the hadron under investigation . the @xmath0-meson interpolator related to the decay constant is @xmath8 to maintain maximal overlap with the ground state relative to excited states , only interpolators without derivatives are considered . for the nucleon we begin by considering the most general spin-1/2 nucleon interpolator @xmath9 where @xmath10 and @xmath11 @xmath12 is the interpolator typically used in lattice qcd analyses . @xmath13 vanishes in the nonrelativistic limit . however in a theory with light relativistic current quarks , there is no reason to exclude such an interpolating field _ a priori _ @xcite . with the use of the fierz relations , the combination of the above two interpolating fields for @xmath14 may be written @xmath15 [ chisr ] giving the proton interpolating field advocated by ioffe @xcite and often found in qcd - sr calculations . at the phenomenological level one proceeds by inserting a complete set of eigenstates with the quantum numbers of the interpolator . the ability of the interpolator to annihilate a positive parity baryon to the qcd vacuum is described by the parameter @xmath16 in @xmath17 for state @xmath18 of momentum @xmath19 and spin @xmath20 . @xmath21 is a dirac spinor . negative parity states require a @xmath22 on the right - hand side of ( [ lambdadef ] ) . for the nucleon , the two point function has the form @xmath23 where @xmath24 corresponds to positive / negative parity baryons . while there is some suppression of excited state contributions to the two - point function , there is little hope of isolating ground state properties from such a function . hence one uses the celebrated borel transform @xmath25 = \lim_{{\scriptstyle -p^2 \to \infty \atop \scriptstyle n \to \infty } \atop \scriptstyle -p^2/n = m^2 } { 1\over n ! } ( -p^2)^{n+1 } \left ( { d \over d p^2 } \right ) ^n f ( p^2 ) \ , , \ ] ] at each dirac-@xmath26 structure to obtain exponential suppression of excited states . the parameter @xmath27 is commonly referred to as the borel mass . the borel transform is easily applied to a dispersion relation for @xmath28 at each dirac-@xmath26 structure @xmath29 and the polynomial subtraction terms are eliminated by the borel transform . at the structure @xmath30 one has @xmath31 where we have introduced the spectral density @xmath32 , @xmath33[gp ] at the structure @xmath34 @xmath35 where @xmath36 here , @xmath37 and @xmath38 account for excited state contributions , and @xmath24 in ( [ rho1 ] ) corresponds to positive / negative parity states . for the vector meson one has @xmath39 @xmath40 [ rhov ] at the structure @xmath41 . at the quark level , one exploits the operator product expansion ( ope ) to describe the short distance behavior of the two - point function @xmath42 [ ope ] where @xmath43 is the normalization point at which the coefficient functions and the operators are defined . here we have explicitly included all operators up to dimension eight , to leading order in the quark mass @xmath44 , having the quantum numbers of the vacuum . the first digit of the subscript of the wilson coefficients @xmath45 gives the energy dimension of the operator . @xmath46 may take any of the 16 independent dirac-@xmath26 matrices . upon fourier transforming to momentum space , ( [ ope ] ) becomes an expansion in @xmath47 , valid for large momentum transfers . thus one may use perturbation theory to calculate the wilson coefficients . the contribution of the gluon operator term 4.1 in the expansion for the proton is suppressed due a factor of @xmath48 which appears in the denominator of the wilson coefficient arising from integration over the quark - loop momentum which introduces the factor @xmath49 . terms involving more than two gluon field strength tensors @xmath50 and products of @xmath44 and @xmath51 are estimated to be small and are typically neglected . the standard treatment of the ope proceeds via inserting the explicit forms of the interpolating fields into the two - point function of ( [ twopt ] ) and contracting out pairs of time - ordered quark - field operators which are the fully interacting quark propagators of qcd . wick s theorem for the time - ordered product of two fermion fields provides @xmath52 taylor expansion of the normal ordered piece leads to @xmath53 the well known trick is to select the coordinate gauge @xmath54 such that @xmath55 and the partial derivatives of ( [ taylor ] ) may be replaced by covariant derivatives @xmath56 generally , one works with ( [ aexp ] ) to leading order . the following relations allow an easy determination of the covariant derivatives @xmath57 @xmath58 = i g_c g^{\mu \nu } \ , , \ ] ] @xmath59 in summary , the quark correlators employed in calculating the ope under the usual assumption of vacuum saturation of the intermediate states of composite operators include @xmath60 \bigl ( g_c g_{\alpha \beta}^n \bigr ) \sigma^{\alpha \beta } { \lambda^n_{ab } \over 2 } \ , . \label{quarkcorr } \\ & & \nonumber \\ \lefteqn{\langle \omega \bigm | t \left \ { q^a ( x ) \ , g_c g_{\alpha \beta}^n \ , \overline q^b ( 0 ) \right \ } \bigm | \omega \rangle = } \hspace{36pt } \nonumber \\ & & - { 1 \over 2 ^ 6 \ , 3 } \langle \overline q g_c \sigma \cdot g q \rangle \sigma_{\alpha \beta } { \lambda^n_{ab } \over 2 } \nonumber \\ & & + { i \over 2 ^ 8 \ , 3 } m_q \langle \overline q g_c \sigma \cdot g q \rangle \bigl ( \gamma \cdot x \ , \sigma_{\alpha \beta } + \sigma_{\alpha \beta } \ , \gamma \cdot x \bigr ) { \lambda^n_{ab } \over 2 } \nonumber \\ & & + { x^2 \over 2^{10 } \ , 3 ^ 2 } \langle g_c^2 g^2 \rangle \langle \overline q q \rangle \sigma_{\alpha \beta } { \lambda^n_{ab } \over 2 } \ , . \label{quarkgluoncorr}\end{aligned}\ ] ] finally the momentum space correlator is borel transformed . here one encounters the factorial suppression of higher - dimension operators as indicated in @xmath61 = ( -1)^k \ , { 1 \over ( k-1 ) ! } \ , { 1 \over ( m^2)^{k-1 } } \ , .\ ] ] it is worth noting that the factorial suppression does not really set in until one reaches the term @xmath62 . for the @xmath0 meson this term corresponds to an operator of dimension eight , and for the nucleon the dimensions are twelve and thirteen for the two nucleon sum rules obtained from spin-1/2 interpolating fields . since the ope for the @xmath0-meson is traditionally truncated at dimension eight @xcite , the correlator should be more reliable than for the nucleon sum rules which are truncated at dimension nine . it is well known that the standard treatment of the ope encounters difficulties at large orders in the perturbative expansion of the wilson coefficients . perturbative expansions are divergent at large orders and are asymptotic at best @xcite . in the standard treatment , these divergences limit the accuracy to which the perturbative coefficients may be determined . renormalons @xcite , a particular set of perturbative graphs , provide a simple illustration of the factorial growth in the contributions of perturbative graphs at large orders @xcite . the factorial growth in large - order contributions associated with soft infrared ( ir ) virtual momenta is problematic , as the series is not borel summable . as such , the wilson coefficients , by themselves , are ill defined in the standard treatment . the uncertainty resulting from the restriction to finite orders of the asymptotic expansion is commonly referred to as the ir renormalon ambiguity . of course the use of perturbative propagators to describe soft virtual momenta is incorrect . instead , one must utilize fully dressed nonperturbative propagators . hence , it is not surprising one encounters difficulties in the ir regime , as one is simply encountering the landau pole of the perturbative coupling constant . a solution to the ir renormalon ambiguity problem was provided by the itep group @xcite through the introduction of the normalization / separation scale @xmath43 of ( [ opemu ] ) . the ope becomes a separation of scales , with virtual momenta lying above @xmath43 represented in the wilson coefficients , and momenta below @xmath43 in the vacuum expectation values ( vevs ) of higher - dimension operators beginning at dimension four . the introduction of the scale @xmath43 eliminates the ir renormalon ambiguity of the perturbative expansion . however , in principle , both the wilson coefficients and the vevs of operators contain perturbative as well as nonperturbative physics . in principle , the ope defined in ( [ opemu ] ) is free of ir renormalon ambiguities . however , in practice it is difficult to implement the formal procedure of truncating virtual momenta below @xmath43 when calculating higher order @xmath3 corrections to the wilson coefficients . while these contributions are small for leading order corrections , their inclusion is a source of error which can be accommodated in the monte - carlo uncertainty analysis . in the ultraviolet ( uv ) virtual momenta regime , the renormalon series is sign alternating and borel summable . summation of the uv series can give rise to a power correction proportional to @xmath63 which closely resembles the squared quark mass contribution to the ope @xcite . for the vector correlator , the uv renormalon gives rise to an effective mass of roughly @xmath64 , which is much larger than the usual current quark mass the order of 5 mev @xcite . hence there is some concern that important power corrections associated with uv renormalons has been overlooked in the standard qcd - sr treatment . moreover , it is the power corrections that provide the link to nonperturbative phenomena . to the extent that the power correction is scheme dependent and not reliably known , we will refer to these quark - mass like power corrections as uv renormalon uncertainties . additional evidence for @xmath65-type power corrections independent of renormalons has been obtained through a consideration of constraints imposed by asymptotic freedom and analyticity on the large - order behavior of perturbation theory @xcite . in this case the perturbative series is not borel summable . given the absence of a local gauge - invariant field - operator product of dimension two , some might argue that the coefficient of the singularity in the borel plane governing the large - order behavior of perturbative series must have a zero . however , perturbative estimates of the coefficient do not provide evidence of such a zero @xcite and the issue is unsettled . hence the issue of possible dimension - two power corrections is of current interest . the techniques presented in this manuscript might be used to provide some phenomenological insight into the debate . however , such considerations would take us too far afield and will be deferred to a subsequent analysis @xcite . on the other hand , it is interesting to look for discrepancies which might be rooted in such power corrections , and this is done in the following analysis . to summarize the points of this section , we present the traditional qcd sum rules for the nucleon @xcite obtained from the consideration of the interpolating field @xmath66 of ( [ chisr ] ) to operator dimension 8 . @xmath67 [ tradit ] the dots allow for excited state contributions . ) has had an elusive history . early calculations @xcite appear to have calculated a subset of the diagrams contributing to the dimension 7 operator @xcite . however a more persistent error lies in a factor of 3 correction in the term @xmath68 \left < \overline q q \right > \left < g_c^2 g^2 \right > \sigma_{\alpha \beta } \lambda_{ab}^n/2 $ ] of ( [ quarkgluoncorr ] ) from that used in previous analyses . ] the ioffe formula for the nucleon mass is obtained from ( [ tradit ] ) by keeping the leading terms of the opes and assuming dominance of the ground state on the right - hand side of the equations , @xmath69 [ bigassumption ] the ioffe formula follows from the ratio of these two equations , @xmath70 if one prefers to eliminate the borel mass by selecting @xmath71 , @xmath72 here the quark condensate is estimated from the partially conserved axial current ( pcac ) relation and pion decay to give @xmath73 which corresponds to @xmath74 gev , @xmath75 gev and the average light quark mass is 7 mev . the qualitative features of chiral symmetry breaking coupled with the quantitative success of these relations has lead to enormous faith being placed into the validity of the ioffe formula . most practitioners expect this relation to be valid , at least in a qualitative manner of proportionality . the operator product expansion provides knowledge of the two - point function ( [ twopt ] ) in the near perturbative regime . as such the correlator represents contributions from both the ground state hadron and excited states . to maintain some predictive ability in the qcd - sr approach , a `` continuum model '' for excited state contributions to the correlator is introduced guided by the principle of duality , the leading terms of the ope surviving in the limit @xmath76 are used to model the form of the spectral density accounting for excited state contributions . consider , as an example , the identity operator for the nucleon interpolator @xmath77 of ( [ chisr ] ) . explicitly including the excited state term of ( [ gp ] ) in equation ( [ ioffe1a ] ) leads to @xmath78 for large borel mass . the laplace transform readily provides @xmath79 the continuum model contribution is obtained by introducing a threshold , @xmath80 , at which strength in the excited states becomes significant . such a model is motivated well by the experimental cross section for @xmath81 hadrons . in addition , this approach accommodates the finite widths and the multi - particle continuum of qcd . hence the excited state contributions associated with the identity operator are @xmath82 continuum model contributions associated with @xmath83 and @xmath84 terms are calculated in a similar manner . usually the continuum contributions are placed on the ope side of the sum rules . hence , terms surviving in the limit @xmath76 have the following factors associated with them : @xmath85 \ , , \\ m^4 \ , : & & \left [ 1 - e^{-w^2/m^2 } \ , \left ( { w^2 \over m^2 } + 1 \right ) \right ] \ , , \\ m^2 \ , : & & \left [ 1 - e^{-w^2/m^2 } \right ] \ , .\end{aligned}\ ] ] [ continuum ] the contributions of this model relative to the ground state , whose properties one is really trying to determine , are not small . they are typically 10 to 50% @xcite . the validity of this somewhat crude model is relied upon to cleanly remove the excited state contaminations . in the lattice qcd investigation of the continuum model formulated in euclidean space @xcite , the short time regime of point - to - point lattice correlation functions is described well by the qcd - sum - rule - inspired continuum model . there , the laplace transform of the spectral density appears to be sufficient to render any structure in the spectral density insignificant in the short euclidean time regime of point - to - point correlators . similar conclusions are expected to hold for the borel transformed sum rules under investigation here . it should be noted that the spectral density of the physical world is different than that encountered on the lattice . for example , in the nucleon channel one expects strength in the correlator above the ground state to start at the @xmath86 threshold @xcite , whereas in the lattice calculations such contributions are altered by the heavier quark mass and the quenched approximation . while the lattice results are encouraging , one may find that in some cases the continuum model is not sufficiently detailed when the continuum model contributions are large . an example of this will be presented in the discussion surrounding the @xmath0-meson sum rules , where discrepancies may be due to an over simplified continuum model . the origin of the uncertainty estimate for the ope lies in the uncertainties assigned to the qcd parameters appearing in the truncated ope . in this section we assess the uncertainties on these qcd parameters . in assigning these uncertainties one would like to select values conservatively enough such that the qcd - sr approach can test the validity of our present understanding of qcd . qcd - sr predictions that fail to agree with experiment at the @xmath87 level should indicate a possibly interesting discrepancy . at the same time one would also like to select uncertainties in accord with the generally accepted values in the literature . in addition , the error assessments need to reflect uncertainties associated with the factorization of higher dimensional operators and small but never - the - less neglected @xmath3 corrections . these considerations have been balanced in the following . while some may argue that some values are known better than that indicated here , others will no doubt argue that the errors are underestimated . in any event , one will learn for the first time how the uncertainties in the qcd parameters are mapped into uncertainties in the phenomenological fit parameters . in the lattice qcd investigation of ref . @xcite it was noted that interplay between the pole position and continuum threshold lead to rather large uncertainties in the fit parameters . it will be interesting to see how reasonable estimates of the qcd uncertainties are revealed in the fit parameters . for the quark condensate , @xmath88 we note that there are two values commonly found in qcd - sr analyses . pcac and pion decay considerations coupled with quark masses of @xmath89 mev lead to a value of @xmath90 gev@xmath91 . qcd - sr considerations of @xmath92 and octet baryon magnetic moments @xcite prefer a smaller magnitude at @xmath93 gev@xmath91 . expecting the value to lie some where in this regime , we take the average of these values with an uncertainty of half the difference at an 80% confidence level . introducing the standard qcd - sr notation we define @xmath94 this value and uncertainty is also in accord with @xmath95 obtained from more recent quark mass estimates @xcite . early estimates of the gluon condensate @xmath96 from charmonium sum rules @xcite place the value at @xmath97 gev@xmath98 , which is generally referred to as the `` standard value '' @xmath99 however , a number of more recent investigations place the gluon condensate at much larger values @xcite . for example , a finite - energy sum rule ( fesr ) analysis of the @xmath0-meson channel places the gluon condensate at 2 to 5 times the standard value @xcite . in this analysis the spectral density is taken from a fit of the isospin - one channel of @xmath100 scattering data . first results from the aleph and cleo ii experiments @xcite also suggest a larger value at @xmath101 gev@xmath98 . a recent laplace based sum rule analyses @xcite suggests a value of @xmath102 gev@xmath98 , more in accord with the standard value . however , we are somewhat skeptical of the refinements of this result as the analysis is based on an approach which is criticized in the following . the most reliable sum rules of sec . [ nucleon13 ] also prefer a larger gluon condensate in order to maintain consistency between the sum rules . there it is found that the nucleon mass may be reproduced with a value of @xmath103 gev@xmath98 which agrees with experimental estimate . since the gluon condensate plays a critical role in the nucleon sum rules as part of a factorized dimension seven operator we adopt an uncertainty of 50% and take @xmath104 the mixed condensate is parameterized as @xmath105 . early analyses of baryons @xcite as well as more recent analyses of heavy - light quark systems @xcite place @xmath106 gev@xmath107 , while an analysis of @xmath108 qcd - sr sum rules demand @xmath109 to be smaller @xcite at @xmath110 gev@xmath107 . hence we take the average and half the difference as the uncertainty . @xmath111 since the parameter @xmath109 is selected via monte carlo , the value of the mixed condensate is obtained by multiplying @xmath109 by the central value of the quark condensate , as opposed to a randomly selected value of the quark condensate . this is in contrast to factorized operators where both condensates appearing in the operator are selected via monte carlo . relatively little is known about the magnitude of the dimension - six , four - quark operators . early arguments placed the values of these condensates within 10% of the factorized values @xcite . however other analyses claimed significant violation of factorization @xcite for these operators in both nucleon and vector meson sum rules . parameterizing the four quark - operators as @xmath112 , estimates of factorization violation place @xmath113 or more . for the nucleon we consider @xmath114 if one adopts the standard value for the gluon condensate , then @xmath115 is sufficient to reproduce the @xmath0-meson mass . however , selection of the more recent and reliable estimates of @xmath116 demands @xmath117 to reproduce the @xmath0-meson mass . this value is in accord with the observations of refs . @xcite , however the uncertainty on this parameter is large . hence we adopt as opposed to 2 , even a small suppression of the chirally even four - quark condensate at finite density can give rise to substantial suppression of the @xmath0-meson mass and width . further investigation of these issues is warranted . ] @xmath118 figure [ distrib ] displays histograms drawn from 1000 qcd parameter sets for four of the qcd parameters discussed to this point . = variation of the qcd scale parameter @xmath119 has little effect on the results . however , we restrict the values to the conventional range adopted in qcd - srs @xmath120 variation of @xmath119 above 200 mev is only sensible if the separation scale @xmath43 of the ope , taken to be 500 mev , is increased to maintain plausible convergence of the power corrections of the ope . a change in @xmath43 redefines the condensate values . a systematic determination of the new condensate values is beyond the scope of this investigation and we choose to simply restrict the value as above . it is perhaps interesting to note that if @xmath119 is very large , say near the upper limit of the current world average @xcite , the ope separation scale would necessarily become large . consequently , the borel mass would be restricted to larger values , and access to ground state properties may not be possible . the correlator may be dominated by strength from excited states , similar to the present situation for the pseudoscalar pion correlator . fortunately , the lower bound on the current world average for @xmath121 bodes well for present qcd sum rule analyses . however , if the current world average for @xmath119 persists , a comprehensive reformulation of the approach should be pursued . the @xmath0-meson sum rule requires knowledge of @xmath122 at the scale of 1 gev@xmath107 . since this scale is rather low for two - loop perturbation theory , we adopt a 10% uncertainty @xmath123 finally , the @xmath0-meson sum rule requires knowledge of the average light quark mass @xmath44 . since it appears multiplied by the quark condensate , we will use the relation of ( [ gor ] ) and equate @xmath124 monte - carlo techniques might also be used to probe ope truncation errors by randomly selecting the size of the next term to follow the truncated series . the problem lies in determining a sensible and realistic estimate without actually calculating the term . instead , we will emphasize the importance of ope convergence and select the borel analysis regime such that truncated terms are very unlikely to be important . gaussian distributions are generated using established algorithms for the conversion of uniform distributions on the interval ( 0,1 ) @xcite . a particularly fast algorithm used in this investigation is as follows . if @xmath125 and @xmath126 are uniform on ( 0,1 ) , then construct @xmath127 and @xmath128 . calculate @xmath129 . if @xmath130 start over . otherwise , calculate @xmath131 @xmath132 and @xmath133 are independent and normally distributed with mean 0 and variance 1 . a set of gaussianly - distributed randomly - selected condensate values is generated , from which an ope in borel space , @xmath134 , is constructed . by repeating this procedure , an uncertainty for the ope may be determined . selecting @xmath135 evenly distributed points in the borel parameter space @xmath136 , the standard deviation in the ope at the @xmath137th borel mass is given by @xmath138 ^ 2 \ , , \label{sigmaope}\ ] ] where @xmath18 denotes the @xmath18th set of @xmath139 qcd parameter sets and @xmath140 in practice , @xmath141 was used in estimating the ope uncertainty . however , @xmath142 was sufficient to get a stable uncertainty estimate . the uncertainties in the ope are not uniform throughout the borel regime . they are larger at the lower nonperturbative end of the borel region where uncertainties in the higher - dimensional vacuum condensates dominate . hence , it is crucial that the appropriate weight is used in the calculation of @xmath4 . having determined the ope uncertainty , a @xmath4 measure is easily constructed . for the ope obtained from the @xmath143th set of qcd parameters , the @xmath4 per degree of freedom is @xmath144 ^ 2 \over \sigma_{{\rm ope}}^2(m_j ) } \ , , \label{chisq}\ ] ] where @xmath145 is the number of phenomenological search parameters , and @xmath146 denotes the phenomenological spectral representation of the qcd - sr . in practice , @xmath147 points were used along the borel axis . @xmath4 is minimized by adjusting the phenomenological parameters including the pole residue @xmath148 , the hadron mass @xmath149 , and the continuum threshold @xmath150 . distributions for the phenomenological parameters are obtained by minimizing @xmath4 for many qcd parameter sets . provided the resulting distributions are gaussian , an estimate of the uncertainty in the phenomenological results such as the hadron mass is obtained from @xmath151 in the event that the resultant distribution is not gaussian , we will report the median and asymmetric standard deviations from the median . we note that ( [ sigmam ] ) is the standard deviation of the distribution and not the standard error obtained by further dividing by @xmath152 , the number of qcd parameter sets in the sample . the standard deviation is roughly independent of the number of qcd parameter sets used in the analysis , and directly reflects the input parameter uncertainties . in the following , we generally select @xmath153 . while 100 qcd parameter configurations are sufficient for obtaining reliable uncertainty estimates , we also wish to explore correlations among the qcd parameters and the phenomenological fit parameters . the extra sets aid in resolving more subtle correlations . some care must be taken in the interpretation of the @xmath154 . while we are free to choose any number of points @xmath135 along the borel axis for fitting the two sides of the sum rule , it is important to recognize that the actual number of degrees of freedom is determined by the number of terms in the ope . for the @xmath154 to have true statistical meaning , a correlated @xmath4 calculation is required . @xmath155 \times \nonumber \\ & & \quad c_{j\,j'}^{-1 } \left [ \pi^{\rm ope}_k(m_{j ' } ) - \pi^{\rm sr}(m_{j ' } ; \lambda_k , m_k , w_k ) \right ] \ , , \label{corrchisq}\end{aligned}\ ] ] where @xmath156 is the inverse covariance matrix . the covariance matrix may be estimated by @xmath157 \left [ \pi^{\rm ope}_i(m_{j ' } ) - \overline{\pi^{\rm ope}}(m_{j ' } ) \right ] \label{covar}\ ] ] unfortunately , for most of the fits considered here the covariance matrix is ill - conditioned . inverting the covariance matrix via singular value decomposition leads to well known pathological problems @xcite such as best fit lines which do not pass through the data points etc . while one could experiment with different approaches to stepping around this problem , we prefer to use the robust method of determining best fit parameters by minimizing the uncorrelated @xmath4 of ( [ chisq ] ) . even if some correlation among the data is suspected , it is still acceptable to use the uncorrelated @xmath4 fit provided the errors on the parameters are estimated using a monte - carlo based procedure like that discussed here rather than from the dependence of @xmath4 on the parameters @xcite . when simultaneously fitting multiple sum rules which may have some fit parameters in common , ( [ chisq ] ) is simply modified to @xmath158 ^ 2 \over \sigma_{{\rm ope}}^2{}_i(m_{ij } ) } \ , , \label{chisqmulti}\ ] ] where @xmath18 runs over the considered sum rules . ( in this example , @xmath149 is a common parameter to all @xmath159 sum rules . ) the subscript @xmath18 in @xmath160 indicates that each sum rule has a unique set of @xmath135 borel masses @xmath161 spanning the region of interest in borel space . the subscript @xmath143 in ( [ chisq ] ) , denoting a particular set of qcd parameters , has been suppressed . in the following discussion it will become apparent that it is advantageous to weight the contributions in the sum over the various sum rules by the size of their valid borel regimes . if @xmath162 is the size of the valid borel regime for the @xmath18th sum rule , then ( [ chisqmulti ] ) is modified to @xmath163 ^ 2 \over \sigma_{{\rm ope}}^2{}_i(m_{ij } ) } \ , . \label{chisqweight}\ ] ] the optimization method utilized in this investigation is an updated version of a direction - set routine by powell originally published in ref . this algorithm finds the minimum value of a function by iterative variation of the function parameters which need not be independent . modifications of the original routine improve the selection of conjugate directions in the search and avoid the possible generation of linearly dependent search directions @xcite . powell s method is among the best methods for finding minima in a multidimensional parameter space when derivatives are not readily available @xcite . part of the reason for the robust nature of powell s routine is the cautious criterion for ultimate convergence . once a minimum is found , the optimal vector , @xmath164 , in the parameter space is displaced by ten times the requested accuracy . the search proceeds from this new point until a minimum is found at vector @xmath165 . the minimum on the line joining @xmath164 and @xmath165 is found at @xmath166 . convergence is reached if the components of the vectors @xmath167 and @xmath168 are all less than 10% of the required accuracy . otherwise the direction @xmath167 is utilized in a new search . the displacement of the first minimum by ten times the requested accuracy plays a crucial role in removing the sensitivity of the ultimate minimum from the initial parameter estimates . while there is sufficient flexibility in the algorithm parameters to search in the locality of the initial parameter estimates , the opposite criteria has been utilized in this investigation . one of the main goals of this investigation is to eliminate the possibility of fine tuning the initial parameter estimates which otherwise can bias the results . instead , we are interested in the predictions of the qcd sum rules . the fundamental qcd - sr for the @xmath0 meson follows from the consideration of ( [ twopt ] ) , ( [ chirho ] ) , ( [ rhov ] ) and ( [ continuum ] ) . we consider the correlator , @xmath169 , where the continuum model is subtracted from the phenomenological side and placed on the ope side . @xmath170 + c_1 + { c_2\over m^2 } + { c_3\over 2!\ , m^4 } + \cdots \nonumber \\ & & + { c_m \over ( m-1 ) ! \ , ( m^2)^{m-1 } } + \cdots \ , . \label{rhosr}\end{aligned}\ ] ] the wilson coefficients are given by @xcite @xmath171 the phenomenology of the qcd - sr is described by the vector - meson pole of interest plus the continuum model accounting for the contributions of all excited states . by working in a region where the pole dominates the phenomenology , one can minimize sensitivity to the model and have assurance that it is the spectral parameters of the ground state of interest that are being determined by matching the sum rules . in practice , these considerations effectively set an upper limit in the borel parameter space , beyond which the model for excited states dominates the phenomenological side . at the same time , the truncated ope must be sufficiently convergent as to accurately describe the true ope . since the ope is an expansion in the inverse squared borel mass , this consideration sets a lower limit in borel parameter space , beyond which higher order terms not present in the truncated ope are significant and important . monitoring ope convergence is absolutely crucial to recovering nonperturbative phenomena in the sum - rule approach , as it is the lower end of the borel region where the nonperturbative information of the ope is most significant . as we shall see , this information must also be accurate . in short , one should not expect to extract information on the ground state spectral properties unless the ground state dominates the contributions on the phenomenological side and the ope is sufficiently convergent . in this investigation , we will analyze each individual sum rule with regard to the above criteria . a sum rule with an upper limit in borel space lower than the lower limit is considered invalid . as a measure of the relative reliability of various sum rules we consider the size of the regime in borel space where both sides of the sum rules are valid . this quantity is used as a weight in the calculation of the @xmath154 as indicated in ( [ chisqweight ] ) . in addition , the size of continuum contributions throughout the borel region can also serve as a measure of reliability , with small continuum model contributions being more reliable . the fiducial borel region is chosen @xcite such that the highest - dimensional operator(s ) ( hdo ) contribute no more than @xmath172 to the qcd side while the continuum contribution is less than @xmath173 of the total phenomenological side ( i.e. , the sum of the pole and the continuum contribution ) . the former sets a criterion for the convergence of the ope while the latter controls the continuum contribution . while the selection of @xmath174 is obvious for pole dominance , the selection of @xmath175 is a reasonably conservative criterion that has not failed in practice . and @xmath174 criteria are automatically explored in the monte - carlo error analysis , as the condensate values and the continuum threshold change in each sample . ] see ref . @xcite for further specific examples . figure [ sr1hdo ] displays the valid borel window for the @xmath0-meson sum rule of ( [ rhosr ] ) . the hdo contributions limited to @xmath175 of the ope and the continuum model contributions limited to 50% of the phenomenology are illustrated . the continuum contributions are the order of @xmath176 of the phenomenological contributions at the lower end of the borel regime . here , the pole of interest truly dominates the qcd - sr . figure [ sr1fit ] displays the corresponding fit obtained by minimizing @xmath4 in a three parameter search including @xmath177 , @xmath178 and @xmath179 . figure [ binrhomass ] displays the distribution of @xmath0-meson masses obtained from the 1000 sum rule fits . the distribution corresponds to @xmath180 the uncertainty is somewhat larger than the commonly assumed 10% . figure [ binrhof ] displays the distribution of the square of @xmath177 which corresponds to @xmath181 which agrees with the experimental value of @xmath182 gev . the continuum threshold is @xmath183 gev which compares favorably with @xmath184 gev ( the @xmath185 mass less the half width ) , given the approximate nature of the continuum model . = 11.7truecm = = 11.7truecm = = 11.7truecm = = 11.7truecm = in addition to the fundamental sum rule of ( [ rhosr ] ) one also finds the use of sum rules obtained by differentiating ( [ rhosr ] ) with respect to @xmath186 . for the @xmath187th derivative @xmath188 \ , \label{derivsr}\end{aligned}\ ] ] while these sum rules are obviously unnecessary for the extraction of @xmath0-meson properties , the hope is that these additional sum rules will provide new useful information that might be used to further constrain the phenomenological properties . it is possible to isolate the @xmath0-meson mass as a function of the borel mass , by taking a ratio of the first derivative sum rule with the fundamental sum rule . this approach has some aesthetic appeal , and has widely become the method of choice for analyzing qcd - srs . the continuum threshold is selected to make the ratio of the two sum rules as flat as possible as a function of borel mass . finally , the @xmath0-meson mass is selected from the point at which the ratio is most flat or stable . unfortunately , this method has some very unpleasant shortcomings . the ratio method does not check the validity of each individual sum rule . it is possible to have individual sum rules that are not valid while their ratio is flat as function of borel mass . in addition , the ratio method can not account for the fact that sum rules do not work equally well . the borel region where a sum rule is valid can vary from one sum rule to another . moreover , large uncertainties in one sum rule can spoil the predictions of another sum rule . finally , the continuum contributions to the sum rules are not monitored in the ratio method . if the continuum contribution is dominant in a sum rule , one should not expect to get reliable information about the lowest resonance . the fact that sum rules do not work equally well deserves further discussion . in ref . @xcite a detailed comparison of fundamental and derivative sum rules is carried out for the @xmath0 meson . these arguments are not broadly known and are generally applicable to any derivative borel - improved qcd - sr . hence the important observations are highlighted here . if one could carry out the ope to arbitrary accuracy and use a spectral density independent of models for excited state contributions , the predictions based on ( [ rhosr ] ) and those based on the derivative sum rules of ( [ derivsr ] ) should be the same . in practical calculations , however , one has to truncate the ope and use a simple phenomenological ansatz for the spectral density . thus it is unrealistic to expect the sum rules to work equally well . compare the @xmath187th derivative sum rule of ( [ derivsr ] ) with the direct sum rule of ( [ rhosr ] ) . 1 . the perturbative contribution in the derivative sum rule has an extra factor @xmath189 relative to the corresponding term in ( [ rhosr ] ) , implying that the perturbative contribution is more important in the derivative sum rules than in the fundamental sum rule , and becomes increasingly important as @xmath187 increases . since the perturbative term mainly contributes to the continuum of the spectral density , maintaining dominance of the lowest resonance pole in the sum rule will become increasingly difficult as @xmath187 increases . 2 . in ( [ rhosr ] ) , the term proportional to @xmath190 is suppressed by a factor of @xmath191 , while it is only suppressed by @xmath192 in the derivative sum rule ( @xmath193 ) . this implies that the convergence of the ope is much slower in the derivative sum rule than in the fundamental sum rule . consequently , the poorly known high - order power corrections are more important in the derivative sum rule than in eq . ( [ rhosr ] ) , and become more and more important as @xmath187 increases . if one would like to restrict the size of the last term of the ope to maintain some promise of ope convergence , the size of the borel region in which the sum rules are believed to be valid is restricted . 3 . the power corrections proportional to @xmath194 do _ not _ contribute to the @xmath195 derivative sum rule but do contribute to ( [ rhosr ] ) . if one truncates the ope , part or all of the nonperturbative information will be lost in the derivative sum rules . it is also worth noting that the leading power corrections are the most desirable terms to have . they do not give rise to a term in the continuum model and they are not the last term in the ope , whose relative contribution should be restricted to maintain ope convergence . in practice , the predictions based on the fundamental sum rule of ( [ rhosr ] ) are more reliable than those from the derivative sum rules , which become less and less reliable as @xmath187 increases . this can also be demonstrated by analyzing the sum rules numerically . figure [ sr12hdo ] displays the valid borel windows for a simultaneous fit of the @xmath0-meson sum rules of ( [ rhosr ] ) and the first derivative @xmath196 of ( [ derivsr ] ) . the reliability of the fundamental sum rule over the derivative sum rule is obvious . the valid borel regime for the derivative sum rule is nearly nonexistent . all three points of the discussion surrounding derivative sum rules are displayed in this figure . item 1 indicates the continuum model contributions will be much larger in the derivative sum rule for a given borel mass . indeed the continuum model contribution never drops below 45% in the valid regime . the relative contribution of the last term in the truncated ope is larger in the derivative sum rule for a given borel mass , in agreement with the expectations of items 2 and 3 . figure [ sr12fit ] displays the corresponding fit obtained by minimizing the weighted @xmath154 of ( [ chisqweight ] ) in a three parameter search including @xmath177 , @xmath178 and @xmath179 . this time the logarithms of the left- and right - hand sides of equations ( [ rhosr ] ) and ( [ derivsr ] ) are plotted as a function of the inverse squared borel mass . this method has the advantage of revealing the possible breakdown of the ope signified by deviations of the dashed curves from linearity . uncertainties are displayed only at the ends of the borel region for clarity . = 11.7truecm = = 11.7truecm = while the central values of the best fit parameters @xmath197 gev , @xmath198 gev@xmath107 , and @xmath199 gev , remain essentially unchanged , the reduction of the uncertainties comes from rather unreliable information as discussed above . these arguments are further supported by the failure of the two sides of ( [ derivsr ] ) for @xmath200 to agree as illustrated in figure [ sr12fit ] . to further probe the reasons for this failure , the analysis of these sum rules is repeated with a spectral representation that accounts for the finite width of the @xmath0 meson . to account for the finite width of the @xmath0-meson , the @xmath201 function in ( [ rhov ] ) is replaced by a normalized breit - wigner form @xmath202 and the integral over the @xmath0-meson ground state @xmath203 is handled numerically @xcite . using the experimental width of 151.5 mev , a three parameter fit of the fundamental and derivative sum rules with the breit - wigner form leads to essentially the same results reported in figure [ sr12fit ] . the mass , coupling , and continuum threshold all agree within uncertainties , and have similar uncertainties . hence , these spectral properties may be reliably estimated without precise information on the @xmath0 width . however , @xmath2 mixing provides an interesting example where finite mesonic widths can play a central role in arriving at the correct physics @xcite . the origin of the discrepancy for the derivative sum rule is more likely to lie in short - comings of continuum model . for this sum rule the continuum model contributions never drop below 45% , which may be too large for the simple model typically used in qcd - sr calculations . alternatively , poorly known higher order terms , not present in the truncated ope may become important in the derivative sum rule . in the derivative sum rule , the next term in the series is suppressed by merely @xmath204 . the importance of carefully monitoring the validity of qcd - srs is nicely displayed in the recent debate over the behavior of vector - meson masses in nuclear matter . there an analysis involving invalid sum rules leads to the conclusion that vector - meson masses increase in finite density @xcite , in contradiction to a previous analysis @xcite also based on the ratio method . the debate is resolved in ref . @xcite where valid sum rules indicate that vector - meson masses decrease for increasing nuclear matter density . having firmly established the analysis procedure in the well understood @xmath0-meson channel , we are ready to apply the same approach to the nucleon sum rules . we will find that many results contradict the conventional wisdom . here we focus on the borel improved qcd sum rules for the generalized interpolator of ( [ chio ] ) . the sum rules proportional to the correlator @xmath205 where continuum model contributions are subtracted from the phenomenological side and placed on the ope side are @xmath206 \nonumber \\ & & \quad + { 5 + 2 \beta + 5 \beta^2 \over 256 } \ , b \ , m^2 \ , l^{-4/9 } \left [ 1 - e^{-w_1 ^ 2/m^2 } \right ] \nonumber \\ & & \quad + { 7 - 2 \beta - 5 \beta^2 \over 24 } \ , a^2 \ , l^{4/9 } -{13 - 2 \beta - 11 \beta^2 \over 96 } \ , { m_0 ^ 2 \ , a^2 \over m^2 } \ , l^{-2/27 } \nonumber \\ & & \qquad = \widetilde \lambda_{\cal o}^2 \ , e^{-m_n^2/m^2 } \ , , \label{nucl11atp } \end{aligned}\ ] ] and @xmath207 \nonumber \\ & & \quad -{3 ( 1 - \beta^2 ) \over 16 } \ , m_0 ^ 2 \ , a \ , m^2 \ , l^{-14/27 } \left [ 1 - e^{-w_2 ^ 2/m^2 } \right ] \nonumber \\ & & \quad + { 19 + 10 \beta - 29 \beta^2 \over 2 ^ 7 \ , 3 ^ 2 } \ , a \ , b \nonumber \\ & & \qquad = \widetilde \lambda_{\cal o}^2 \ , m_n \ , e^{-m_n^2/m^2 } \ , , \label{nucl11at1}\]][nucl11 ] where the dirac-@xmath26 structures are indicated on the left , and @xmath208 . the factors of @xmath209 $ ] account for the anomalous scaling of the operators and interpolators in the leading logarithmic approximation @xcite . let us begin with the traditional or standard nucleon sum rule analysis where @xmath3 corrections are ignored . the more common interpolator of ( [ chisr ] ) is utilized by setting @xmath14 in ( [ nucl11 ] ) . in addition , the traditional condensate parameter set with @xmath210 is selected in order to evaluate the validity of previously published results . for a preliminary analysis of the validities of the sum rules , the nucleon mass is fixed to 1 gev . figure [ sr34k2hdo ] illustrates the valid borel regimes for these two sum rules found by implementing the procedure discussed for the @xmath0-meson . = 11.7truecm = a comparison of figure [ sr34k2hdo ] with the analogous figure [ sr12hdo ] for the @xmath0-meson suggests the nucleon sum rules will not be as reliable . in the nucleon sum rules , the continuum model contributes 20% or more to the phenomenological side of the sum rules . in particular , the continuum model contribution in the sum rule of ( [ nucl11atp ] ) exceeds that for the derivative @xmath0-meson sum rule which has already been shown to be incompatible with the fundamental sum rule . it should not be surprising to find a similar disaster for the nucleon case . as in the @xmath0-meson case , there is an analogous ratio method for obtaining the nucleon mass that has also become popular . here the nucleon mass is isolated as a function of the borel mass by taking the ratio of equations ( [ nucl11at1])/([nucl11atp ] ) . the mass is usually extracted from a regime in which the ratio is flat . again ope convergence and ground - state dominance criteria are not checked in this approach , and too often the mass is extracted from a regime in which the continuum model dominates . figure [ sr34k2fit ] illustrates the three parameter fit of the residue @xmath148 , and the two continuum model thresholds of @xmath211 and @xmath212 . the small borel regime and the relatively huge ope uncertainties of the sum rule of ( [ nucl11atp ] ) indicate that this sum rule plays a negligible role in determining the spectral properties of the nucleon . this is a perfect example of a case where large uncertainties in the sum rule of ( [ nucl11atp ] ) would spoil the predictive ability of ( [ nucl11at1 ] ) if the ratio method was used . the origin of the large uncertainties lies in the poorly known value of the four - quark condensate . the resultant fit parameters for this sum rule are also unphysical as the continuum threshold lies at zero below the nucleon pole . in other words , the present form of the sum rule at @xmath213 is unable to provide any information on the ground state that is consistent with the more reliable sum rule of ( [ nucl11at1 ] ) at the structure 1 . = 11.7truecm = unfortunately , the community has incorrectly identified ( [ nucl11atp ] ) as the more reliable sum rule . this belief is based on the simple fact that the leading term of the ope ( the identity operator ) has a well known vacuum expectation value . however , the @xmath3 corrections for the identity operator @xcite are uncomfortably large . the first order corrections are independent of @xmath214 and are @xmath215 where @xmath216 . this correction is @xmath217% and suggests perturbative corrections are completely out of control for this sum rule . since the magnitude of the identity operator contribution is closely tied to the size of continuum model contributions , it is not surprising to find the valid borel regime for this sum rule disappears when @xmath3 corrections are added . another factor in the traditional preference of ( [ nucl11atp ] ) over ( [ nucl11at1 ] ) is that the second term of ( [ nucl11at1 ] ) vanishes for ioffe s choice of nucleon interpolator @xmath218 . this is really nothing more than a good reason to exclude the interpolator @xmath77 of ( [ chisr ] ) as an optimal interpolator . when one considers the monstrous leading - order @xmath3 corrections , the possibility of a significant dimension - two power correction arising from a summation of the perturbative series @xcite , and the large uncertainties associated with the value of the four - quark condensate , one can not help but reach the conclusion that the qcd side of the sum rule of ( [ nucl11atp ] ) is really unknown . since the sum rule of ( [ nucl11atp ] ) has traditionally been the nucleon sum rule of choice , it is interesting to consider the predictions of this sum rule alone . applying the same techniques as used for the @xmath0-meson to ( [ nucl11atp ] ) , a three parameter fit of @xmath148 , @xmath211 and @xmath219 yields @xmath220 while setting @xmath221 slightly increases the nucleon mass to 0.47 gev , adding leading order @xmath3 corrections has a negligible effect on the fit parameters . consideration of the more recent condensate parameter set discussed in sec . [ uncert ] leads to similar results @xmath222 this may be the first qcd sum rule to predict spectral properties that significantly contradict experiment . as such , this may be the most interesting of all qcd sum rules , and we shall return to this sum rule after examining the selection of the most optimal interpolator for nucleon sum rule analyses . the predictions of ( [ nucl11atp ] ) summarized in ( [ failure ] ) are extremely different from that found in the literature . hence it is important to understand how these new results come about . a critical examination of the nucleon ground state results reported in ref . @xcite by the present author sheds considerable light on the importance of the rigorous qcd - sr analysis presented here . @xcite was the first analysis to give careful regard to the issues of ope convergence and ground - state dominance of the phenomenological side . moreover , all phenomenological parameters were included as fit parameters to be determined by the sum rules . however , there are three important differences between the analysis presented here and that of ref . @xcite . 1 . the standard logarithmic measure @xcite of sum rule agreement was used to fit the sum rules . this approach fails to apply an appropriate weight to sum rule discrepancies , unlike the @xmath4 measure of ( [ chisq ] ) . 2 . the algorithm used to fit the sum rules did not employ the cautious convergence criteria discussed in section [ algorithm ] . implementation of the new criteria aids in avoiding insignificant local minima . 3 . only the standard condensate parameter set was considered in determining the nucleon mass . here 1000 sets are considered . the optimal parameter set found in ref . @xcite corresponds to a local minimum which is sensitive to the measure of sum rule agreement , the initial parameter estimates and the condensate parameter set . other authors have managed to keep the nucleon mass closer to the experimental result by extracting the result from a regime in borel space that provides the correct answer or fixing the continuum threshold to values suggested by the particle data tables . as we shall see in section [ correl ] , fixing the continuum threshold to phenomenologically motivated values largely determines the mass . hence you _ can _ get anything you want provided you are clever enough to fix the threshold to the _ right _ value . and since the excited states are modeled there is tremendous freedom in the threshold parameter space . of course , such input may not be in agreement with the content of the ope . in any event , such results must not be regarded as predictions of qcd sum rules . having established the nucleon sum rule of ( [ nucl11at1 ] ) at the structure 1 to be the potentially more reliable sum rule , it is important to determine the optimal mixing of spin-1/2 interpolators for this sum rule . it has long been established that there are two independent interpolating fields with no derivatives having the quantum numbers of spin 1/2 and isospin 1/2 . both are expected to excite the ground state nucleon from the vacuum . as such , there is a history of argument over the optimal nucleon interpolating field to be used in analyses @xcite and this issue is recognized to be of paramount importance @xcite . while some advocate an interpolating field for which the leading terms of the ope are stationary with respect to the interpolating field mixing parameter @xcite , others argue that a balance between ope convergence and qcd continuum contributions must be maintained @xcite . ideally , one would like to simply calculate with alternate interpolating fields and confirm that the nucleon properties remain unchanged . however , previous implementations of the qcd - sr approach have hindered any firm conclusions . the rigorous analysis introduced here will shed considerable light on these issues . fortunately , the lattice approach to qcd is not plagued with the same limitations and the debate over the optimal interpolating field was resolved in ref . there is was found that , to a good approximation , @xmath223 of ( [ chin2 ] ) excites pure qcd continuum . since @xmath223 has negligible overlap with the ground state nucleon , it is tempting to simply conclude that the optimal interpolating field is @xmath224 . while this is certainly the case for lattice qcd investigations , it is not the case for qcd - sr analyses . by including @xmath223 components in an interpolating field , one can reduce the continuum contributions excited by @xmath224 and allow a broader borel analysis window in which better contact with the ground state of interest is obtained . one of the most difficult things to monitor in the qcd - sr approach is whether the ope is sufficiently convergent for a particular value of borel mass . the lattice results of ref . @xcite indicate the @xmath225 correlator has the fastest converging ope , as its overlap with the nucleon ground state is negligible . similarly , the combination @xmath226 produces an ope with the slowest convergence , as this correlation function is dominated by the ground state nucleon for small borel masses . hence , errors made in truncating the ope are dominated by errors in the @xmath226 component of the general correlator . the relative error in the ope truncation can be reduced by adding @xmath223 components to the correlator . however , the @xmath223 components in the ope are simply subtracted off again by the continuum model terms . hence the relevant error is the absolute error . for @xmath227 , this error is dominated by @xmath226 components of the correlator . as a result , ope truncation errors are approximately independent of @xmath214 . this crucial point has been neglected in previous arguments regarding the optimal nucleon interpolating field . since the errors made in truncating the ope are dominated by errors in the @xmath226 component of the general correlator the criteria of a 10% limit on the hdo contribution to the ope should be applied to sum rules when @xmath228 . once this lower limit is established , it may be used for all @xmath227 . we will refer to this approach for determining the valid borel regime as the `` refined '' method in the following . the optimal nucleon interpolator must involve @xmath224 , as this interpolator is required to maintain overlap with the ground state . the task is to determine the optimal mixing of @xmath223 . since @xmath223 has negligible overlap with the nucleon , the ground state contribution to a sum rule is also independent of @xmath214 . hence , the size of the continuum model contribution is the predominant factor in determining the optimal interpolator . figure [ optcon ] illustrates the contributions of the continuum model terms in ( [ nucl11at1 ] ) for @xmath229 gev and @xmath230 gev . the following discussion is not dependent on the precise values of these parameters . the first point to be made is that contributions from the continuum model are largest for @xmath231 . this selection of mixing is the worst possible choice for extracting information on the ground state nucleon . = 11.7truecm = figure [ optcon ] also indicates it is possible to have vanishing continuum model contributions at @xmath232 or @xmath233 . however , we are relying on the continuum model to account for strength in the correlator that does not have its origin in the ground state . without a continuum model , one would need to include additional poles on the right - hand - side of ( [ nucl11at1 ] ) to account for positive and negative parity excitation strength . for @xmath234 or @xmath235 the correlator is negative indicating the sum rule is saturated by a negative parity state . thus the optimal interpolator is @xmath236 or @xmath237 . to discriminate between these two regimes , we turn to the higher - dimension - operator contributions which do not contribute to the continuum model . it is these terms that provide crucial information on whether the strength in the correlator lies in the ground state or the excited states . if these terms are absent , the optimal fit of the correlator is obtained when @xmath238 and @xmath239 . in this case the continuum model and the ope become equivalent via the laplace transform of ( [ laplacetrans ] ) and the fit is perfect . hence the higher - dimension - operator contributions should be large in magnitude . a change in sign from the leading terms of the ope will also assist in distinguishing ground state strength from excited state strength as the change in the curvature of the correlator will be more prominent . the last term of ( [ nucl11at1 ] ) is independent of the continuum model , and its value is plotted as a function of beta in figure [ optcon ] . the contributions are larger for @xmath236 than for @xmath240 . in addition , the sign of the contribution is opposite that of the continuum model contributions . hence the preferred regime is @xmath236 . these arguments are supported by the numerical analysis described below . in summary , the lattice results indicate ope convergence , and ground state pole contributions are approximately independent of @xmath214 . considerations of the size of continuum model contributions and the sign and magnitude of operators independent of the continuum model leads to the preferred value of @xmath241 a more precise determination of @xmath214 will depend on the details of limits for continuum model and hdo contributions , condensate values , and other parameters of the sum rules . at @xmath242 , where the leading term of the ope is stationary with respect to @xmath214 @xcite , the continuum contributions are maximal . the positive value and small magnitude of the hdo contribution indicates the stability of the leading terms of the ope will not be realized as stability in the ground state mass , coupling , nor in the continuum threshold . another point of view argues that the optimal interpolating field is one that eliminates direct instanton contributions from the correlator @xcite . however , such arguments are based on the assumption that direct instantons necessarily contribute to the nucleon correlator in a significant manner . section [ instantons ] will demonstrate that direct instantons are not necessary to maintain sum rule self consistency for a wide range of interpolating fields . the @xmath214 dependence of the direct instanton contributions to ( [ nucl11at1 ] ) is proportional to @xmath243 . minimization of instanton contributions would place @xmath244 , very near @xmath245 where the hdo term of ( [ nucl11at1 ] ) vanishes and where continuum contributions are near their peak . isolation of ground state contributions is nearly impossible in this regime . therefore minimization of direct instanton contributions does not lead to an optimal interpolating field . while it is important to establish the optimal mixing of interpolating fields for qcd - sr analyses , one should not overlook the fact that there is a range of values for @xmath214 where the sum rules are expected to work . moreover , the ground state contribution to all these sum rules is equivalent to the 1% level @xcite . in other words , the right - hand side of ( [ nucl11at1 ] ) for a single pole plus continuum model is independent of @xmath214 . after the first sum rule is written down , additional sum rules may be introduced with merely one new fit parameter ( the continuum threshold ) per sum rule . since direct instanton contributions to the sum rules are not independent of @xmath214 @xcite , one has an excellent opportunity to see if direct instanton contributions really are necessary to maintain sum - rule consistency . this issue is addressed in section [ instantons ] below . having discovered that @xmath223 has negligible overlap with the nucleon ground state , it might be interesting to return to the `` ioffe formula '' this time generalized for arbitrary @xmath214 @xmath246 \ , { 4 \ , \left ( 2 \pi \right ) ^2 \ , \langle \overline q q \rangle \over m^2 } \ , .\ ] ] for @xmath228 , @xmath247 gev when @xmath248 gev , or @xmath249 gev when @xmath71 gev . the explicit dependence on the choice of interpolating field and the borel mass limits the significance of the `` ioffe formula '' to a qualitative role at best . here we focus on the more reliable sum rule of ( [ nucl11at1 ] ) at the optimal interpolator mixing of @xmath250 . fortunately , the @xmath3 corrections for the leading order operator of the ope are much smaller than for the identity operator at the order of 10% . the @xmath214-dependent coefficient of the quark condensate in ( [ nucl11at1 ] ) becomes @xcite @xmath251 these corrections are small relative to uncertainties in the other terms of the correlator and can be absorbed by a small shift in the condensate itself . since our qcd parameter set is extracted from analyses with neglected @xmath3 corrections , we shall do the same here . armed with knowledge of the optimal interpolator , and a refined understanding of ope convergence tests , we now present the qcd - sr predictions for the nucleon spectral properties from spin-1/2 interpolators . figure [ sr4hdo ] displays the valid borel regime for this sum rule . the hdo contributions are less than 10% of the ope contributions for this fit when @xmath228 , as indicated by the solid curve . as described above , contributions from correlator components known to be more convergent increase the hdo operator contributions making resolution of the pole from the continuum more probable . = 11.7truecm = = 11.7truecm = figure [ sr4fit ] displays the fit for this sum rule . the near perfect linearity of the ope@xmath252continuum side of @xmath205 supports the arguments presented above . despite 40% contributions of the hdo term to this sum rule , the ope@xmath252continuum side displays the anticipated behavior of reasonable ope convergence . the optimal fit parameters obtained from consideration of 1000 qcd parameter sets are @xmath253 the uncertainty of the nucleon mass is more than double the traditional expectation at 260 mev . figures [ sr4binm ] , [ sr4binw ] , and [ sr4binlambda ] display histograms for the nucleon mass , threshold and residue respectively . these histograms will aid in the discussion of the following section . = 11.7truecm = = 11.7truecm = table [ comparesr ] compares the prediction for @xmath254 with other estimates of @xmath255 . this residue corresponds to ioffe s interpolator of ( [ chisr ] ) and is expected to be independent of @xmath227 to the order of 1% @xcite . 0.0(0 ) .comparison of predictions for @xmath256 for various approaches to qcd . [ cols="<,<,^ " , ] what this indicates is that the intimate relationship between the quark condensate and the nucleon mass suggested in the `` ioffe formula '' is invalid . considering the assumptions made in deriving the `` ioffe formula '' one might expect the quark condensate value to be tied more tightly to the continuum model than the ground state pole , and this is borne out in table [ nucleoncorr ] . there the quark condensate plays the dominant role in determining the continuum model threshold . of course , all the condensate values themselves are correlated by the theory of qcd , and this correlation has not been included in this analysis . it may very well be that the nucleon mass is positively correlated with the quark condensate when the higher dimension condensates are also correlated with the quark condensate . the important point to recognize is that the correlation does not come about through some direct connection such as the `` ioffe formula '' . moreover , we are extracting nucleon properties in a borel regime in which these leading order condensates just happen to be the dominant ones . one could also extract the mass from much deeper in the non - perturbative regime , perhaps by a lattice qcd calculation . there , it is much higher dimension operators which account for the same nucleon mass @xcite . therefore , it is really impossible to identify a particular condensate for giving rise to ground state hadron properties . the only true statement that may be made about the relationship between condensates and spectral properties is what we have known all along : the condensates reflecting chiral symmetry breaking play a significant role in determining hadron properties . correlations among the phenomenological fit parameters themselves are qualitatively different from the correlations with the qcd parameters . figures [ rhowlambda ] , [ rhomw ] and [ rhomlambda ] display the close correspondence between all three of the fit parameters , @xmath179 , @xmath177 , and @xmath178 obtained from 1000 fits of ( [ rhosr ] ) . this close association between the @xmath0-meson spectral parameters suggests an underlying constraint on the fit parameters . indeed , local duality necessarily leads to correlations among the spectral parameters . to examine the correlations of local duality in an analytical but qualitative fashion , consider the finite energy sum rules ( fesr ) @xcite obtained from @xmath257 here , @xmath258 is given by ( [ rhovb ] ) and @xmath259 is the qcd evaluation of ( [ twopt ] ) at the structure @xmath260 . the weighting function @xmath261 for fesr is taken to be @xmath262 . these functions weight excited state contributions in favor of ground state contributions and , as such , fesr are inappropriate for a quantitative determination of ground state properties . however , the first moment and , to a lesser extent , the second moment of the fesrs can provide some general insight into the relationships among the spectral parameters . taking @xmath263 and @xmath264 yields the following two relationships @xmath265 [ fesrrho ] in a qualitative sense , the contributions from the ope on the right - hand side of ( [ fesrs ] ) are negligible . in this case ( [ fesrrho ] ) implies @xmath266 [ fesrrel ] these relationships , capturing the features of local duality in a qualitative sense , are independent of the precise values of the qcd parameters . the relationships remain prominent in the more quantitative borel sum rule analysis as indicated in figures [ rhowlambda ] , [ rhomw ] and [ rhomlambda ] . of course , the borel analysis is done independent of these qualitative constraints . the first relation , the most reliable fesr relation , is satisfied remarkably well . = 11.7truecm = = 11.7truecm = = 11.7truecm = = 11.7truecm = = 11.7truecm = an even more striking correlation is illustrated in figure [ sr4corrmw ] for the nucleon sum rule of ( [ nucl11at1 ] ) with the optimal @xmath250 . the association between the continuum threshold and the nucleon mass is nearly one - to - one despite the underlying scatter of qcd parameter values . the explanation for this phenomena is that there is a single term in ( [ nucl11at1 ] ) which is capable of resolving the pole from the continuum . the two leading terms of the ope are used in deriving the continuum model . hence if the third term is vanishingly small , the best fit is most certainly pure continuum @xmath239 as the sum rule is satisfied as a perfect laplace transform . the few points lying above the curve in figure [ sr4corrmw ] where the mass equals the threshold are cases where the third term in the sum rule has failed to resolve the pole from the continuum . figure [ sr4corrmlambda ] displays the corresponding association for the mass and the residue of the pole . in the @xmath0-meson case there are in fact two terms independent of the continuum model in ( [ rhosr ] ) . both terms act together to resolve the pole and continuum and hence give rise to some scatter in the mass threshold association . scatter in the residue of the pole reflects the fact that all terms affect the strength of the pole . the strong associations among nucleon spectral parameters reflect the underlying constraints of local duality . relationships analogous to the fesr for the @xmath0 meson may be derived for the nucleon spectral parameters @xcite . however , the larger continuum model contributions to the nucleon correlator make the use of fesr less desirable . the approximate nature of the continuum model may be too crude for a weighting function that fails to suppress excited state strength in the spectral density . the resultant fesr relations are relatively crude and do not capture the essence of local duality as well as in the @xmath0-meson case . a very different picture develops for the correlations among the nucleon fit parameters obtained from the analysis of ( [ nucl11at1 ] ) for the unfavorable value of @xmath267 . figure [ sr4at0.8corrmw ] displays the correlation between the mass and the continuum threshold , @xmath212 . the corresponding correlation between the mass and the residue is illustrated in figure [ sr4at0.8corrmlambda ] . = 11.7truecm = = 11.7truecm = the nature of the three regimes discussed earlier in section [ secnucl11at1.8 ] is resolved here . the near - pure continuum fits are revealed in the lower - left corner of the mass - threshold scatter plot of figure [ sr4at0.8corrmw ] . the collision of the continuum threshold and the nucleon mass gives rise to the intermediate regime where the gap between ground and excited states has not been resolved . the smaller residue compensating for strength from the continuum model in these fits is illustrated in figure [ sr4at0.8corrmlambda ] . the third pole - plus - continuum regime displays the same near one - to - one correspondence between the mass and the threshold as seen in figure [ sr4corrmw ] for the optimal @xmath250 . the correlations among the spectral parameters are interesting because they reveal the manner in which the sum rules work . correlations among the fit parameters from the simultaneous analysis of ( [ nucl13atgg ] ) and ( [ nucl13atgqq ] ) at @xmath228 are illustrated in figures [ sr467corrmw2 ] and [ sr467corrmlambda2 ] . these distributions are similar to those for the celebrated @xmath0-meson sum rule , and reflect the presence of two terms separating pole from continuum . = 11.7truecm = = 11.7truecm = arguments favoring the need for direct instanton contributions to the qcd side of qcd - srs @xcite are based on the assumption that the average instanton size @xcite , @xmath268 fm @xmath269 fm , the separation scale of the ope . even though @xmath270 , proponents of direct instanton contributions still argue that the effects of instantons are not sufficiently accounted for in qcd condensates alone . the analysis of direct instantons in nucleon sum rules are based on many additional assumptions . here we focus on the most recent investigation of ref . @xcite . there , @xmath3 corrections to the identity operator are neglected . as discussed in section [ nucleon11 ] these leading order corrections are 50% and suggest that @xmath3 corrections are completely out of control . moreover the possibility of a significant dimension - two power correction arising from a summation of the perturbative series , uncertainties surrounding the factorization of the four - quark operator , coupled with an absence of a valid borel regime for the sum rule of ( [ nucl11atp ] ) make any conclusions dubious . in addition , the ope is truncated at dimension six . as we saw in section [ nucleon11 ] the dimension six , seven , and eight operators are crucial to separating the ground state pole from the continuum . without the dimension seven term , the best fit of ( [ nucl11at1 ] ) is pure continuum . the fit is perfect when @xmath239 and @xmath271 . hence , the nucleon sum rule at the structure 1 in ref . @xcite ( equation 12 ) is unstable by design . any further discussion of stabilizing the sum rule by adding instanton contributions is uninteresting . finally , issues surrounding the problems of double counting were handled crudely by simply eliminating ope contributions beyond dimension five . hence one should not be prejudiced by existing analyses suggesting nucleon sum rules require direct instanton contributions . the issue is really not settled . for the spin-1/2 interpolator of ( [ chin1 ] ) , direct instanton contributions to the sum rule of ( [ nucl11at1 ] ) are governed by the factor @xcite @xmath272 the direct instanton contributions were argued to be of vital importance in ref . @xcite in the vicinity of @xmath273 . in the lattice qcd investigation of @xcite it was discovered that the overlap of the interpolator @xmath223 with the nucleon ground state is one hundred times smaller than that for @xmath224 . while the nucleon mass is certainly independent of the interpolating field mixing parameter @xmath214 , one also has the residue of the pole , @xmath148 , independent of @xmath214 at the 1% level . this anticipated independence of both ground state properties provides new opportunities for evaluating the necessity of direct instanton effects . as a first test of sum rule consistency , we utilize the sum rule of ( [ nucl11at1 ] ) to extract the mass as a function of the interpolating field mixing parameter @xmath214 . as @xmath274 the terms contributing to the continuum model become very small and additional poles must be added to the phenomenological side of the sum rule to account for strength in the correlator lying above the ground state mass . as a result we restrict @xmath275 . valid borel regimes were found over the range @xmath276 . the coefficient @xmath277 changes by 50% over this range . if direct instanton contributions are indeed important at @xmath278 , the absence of direct instanton contributions in this investigation might be revealed by a significant @xmath214 dependence in the nucleon mass . figure [ sr4massevol ] displays the ratio of the nucleon mass extracted at @xmath214 to the value obtained at the optimal @xmath279 . uncertainties are obtained from two uncorrelated samples of 200 qcd parameter sets . as such the central value at @xmath250 need not be 1 . the large uncertainties associated with this sum rule prevent a sensitive test of the necessity of direct instantons . it is clear that all ratios accommodate the ratio of 1 . = 11.7truecm = a more demanding test may be performed using the sum rules arising from the consideration of the overlap of spin-1/2 and spin-3/2 interpolators . the smaller uncertainties allow a more interesting examination of these issues . in addition , the continuum model terms are independent of @xmath214 and provide a wider range of @xmath214 in which a valid borel regime exists . figure [ massevol ] addresses the dependence of the nucleon mass obtained from the sum rules of ( [ nucl13atgg ] ) and ( [ nucl13atgqq ] ) on the interpolating field mixing parameter @xmath214 . here , ratios of the mass obtained at finite @xmath214 to @xmath228 are displayed . sum rule consistency demands that all ratios agree with one , and this is satisfied in figure [ massevol ] without direct instanton contributions . the uncertainty estimates are obtained from two uncorrelated samples of 200 qcd parameter sets . we note that a valid regime does not exist for @xmath280 . hence the use of the ioffe interpolator ( @xmath14 ) is a particularly unfavorable choice for these sum rules . the large error bar at @xmath281 reflects the small valid borel regime used in the fit . = 11.7truecm = the corresponding dependence of @xmath148 on @xmath214 is plagued by huge uncertainties . small changes in the nucleon mass which appears squared in the exponential of the phenomenological side of the sum rules leads to large fluctuations in @xmath148 . hence it is difficult to gain any information on the possible size of direct instanton contributions . however , they are certainly unnecessary as all ratios encompass 1 . in order to proceed , we consider fixing the nucleon mass to the value determined at @xmath228 in an attempt to reduce the uncertainty in @xmath148 . figure [ lambdaevolmfixed ] displays these ratios . the corresponding plot for the continuum threshold is given in figure [ contevol ] . = 11.7truecm = = 11.7truecm = for @xmath282 the continuum threshold tends to rise and the distribution of the residue remains largely independent of @xmath214 without direct instanton contributions . for @xmath283 the continuum threshold drops to account for strength that would otherwise be absorbed by the rising nucleon mass as illustrated in figure [ massevol ] . with the threshold near the pole position , strength which might otherwise rest in the pole has been suppressed as discussed in section [ phenomcorr ] . hence the drop in the residue is an artifact of our fixing the nucleon mass in an attempt to reduce the error bar . in summary , the independence of the mass and residue from @xmath227 is satisfied within uncertainties in the absence of direct instanton contributions . as the sum rules are improved in the future and qcd input parameters are better determined , it will be important to repeat this analysis to again evaluate the necessity of direct instanton contributions or other refinements of the qcd - sr approach . at present , direct instanton contributions are not required . this is good news indeed as the amount of modeling required for instanton estimates is relatively unpalatable when compared to that for the standard qcd - srs themselves . in fact a preferable approach would be to simply raise the ope separation scale further into the perturbative regime to ensure that direct instanton effects are more fully accounted . however , this does not yet appear to be necessary . ground state properties have been found to be independent of the interpolating field mixing parameter @xmath214 . on the other hand , the leading terms of the ope for ( [ nucl11 ] ) are very dependent on @xmath214 . this situation has interesting ramifications for previous analyses of coordinate space point - to - point correlation functions @xcite where there has been some confusion surrounding the physics represented in point - to - point correlation functions . in ref . @xcite some conclusions are drawn from an incorrect interpretation of the physics represented in two - point correlators . these authors suggest that the behavior of a ratio of the interacting to free correlator can reveal information on clustering in the scalar diquark channel . it should be emphasized that the phenomenological description of the two - point function involves the baryon mass , the continuum threshold describing the effective onset of excited states and the residue of the pole which indicates the ability of the particular interpolator to excite the baryon from the qcd vacuum . information on quark configurations in the nucleon wave function is absent in point - to - point correlation functions . the coordinate - space correlator may be obtained from ( [ rhop ] ) or ( [ rho1 ] ) via @xmath284 the important point here is that for large @xmath285 , @xmath286 is dominated by the contribution from the ground state pole . as such @xmath286 is independent of @xmath227 at the 1% level , or alternatively @xmath286 is independent of @xmath287 at the 10% level @xcite . the coordinate space analyses focused on a ratio of @xmath288 where @xmath289 is the free coordinate - space correlator . for the correlator at the structure @xmath290 studied in @xcite at @xmath14 , the generalized free correlator is given by @xmath291 since the denominator is @xmath214 dependent and the numerator is independent of @xmath214 for large @xmath285 , the ratio is dependent on @xmath214 . the rate at which the curve turns up as the ground state dominates and the free correlator continues to drop off for increasing @xmath285 is dependent on the interpolator mixing parameter @xmath214 . the rate at which the ratio increases above 1 describes the relative weightings of ground versus excited states generated by the interpolating field . it has nothing to do with the possibility of scalar diquark degrees of freedom being predominant in the nucleon wave function . moreover , a comparison with a @xmath292 correlation function for example has no meaning . a rigorous procedure for extracting quantities of phenomenological interest from qcd sum rules has been established . we have performed a comprehensive analysis of ground state @xmath0-meson and nucleon qcd sum rules . the inclusion of uncertainty estimates for the phenomenological parameters has allowed us to evaluate the predictive ability and self consistency of the qcd - srs . when the analysis is done rigorously , qcd sum rules are predictive . some have argued that fitting sum rules appears to be more of an art than a science . indeed , many of the findings here contradict the conventional wisdom of both practitioners and skeptics alike @xcite and suggests that past arguments have been based more on rhetoric than sound scientific arguments . * the nucleon sum rule of ( [ nucl11atp ] ) , traditionally taken to be the most reliable sum rule , fails to have a borel regime where both the truncated ope is reasonably convergent and the ground state dominates the phenomenology . * the nucleon sum rule of ( [ nucl11at1 ] ) , also obtained from spin-1/2 interpolating fields , resolves the nucleon mass to @xmath293 mev . this uncertainty is more than double the value commonly assumed by most practitioners . * the nucleon sum rules of ( [ nucl13atgg ] ) and ( [ nucl13atgqq ] ) obtained from the consideration of a spin-3/2 isospin-1/2 interpolator projected by the generalized spin-1/2 isospin-1/2 interpolator are much more reliable than the traditional sum rules . they have greater overlap with the ground state pole relative to the continuum model , have broader valid borel regimes and provide phenomenological estimates of nucleon spectral properties which are more stable than those obtained from the conventional sum rules . * the nucleon mass obtained from a simultaneous analysis of ( [ nucl13atgg ] ) and ( [ nucl13atgqq ] ) at the optimal @xmath228 is resolved to @xmath294 mev using qcd parameter estimates from the literature . * the dependence of the spectral parameters on the interpolating field mixing parameter is in accord with expectations without resort to direct instanton contributions . this is compelling evidence that instanton physics is adequately accounted for in the nonperturbative vacuum expectation values . the latter point is truly exciting for the qcd sum rule approach . the amount of modeling required for instanton estimates is relatively unpalatable when compared to that for the standard qcd - srs themselves . * the leading order @xmath3 corrections are the order of 50% . as such , perturbative corrections to this sum rule are very likely out of control . * summation of the perturbative series may give rise to an important dimension - two contribution @xcite . the size of such a correction is presently unknown . * contributions from the four - quark condensate are not well known and appear to be large . unlike the @xmath0-meson sum rule , the four - quark condensate is not suppressed by a factor of @xmath3 . * a regime in which the ope is reasonably convergent while the phenomenological description is dominated by the ground state pole is non - existent . if one proceeds with this sum rule , the results are very sensitive to the model for the continuum , or inaccuracies in the ope lead to incorrect results as seen here . the `` ioffe formula '' directly relating the nucleon mass to the local chiral condensate has been found to be misleading at best . while there is a strong connection between chiral symmetry breaking and hadron properties , such detailed or direct relationships are impossible to make and it is inappropriate to be more specific on the origin of the nucleon mass . the fundamental and derivative sum rules for the @xmath0 meson have also been examined . once again , the popular sum rule analysis has been found to be undesirable . the use of derivative sum rules for the determination of @xmath0-meson spectral properties is unfavorable . increased continuum model contributions , poorer ope convergence , and loss of vital information in the derivative sum rule results in discrepancies between the fundamental and derivative sum rules . the associations among the phenomenological fit parameters are particularly interesting as they reveal how the sum rules resolve the spectral properties . we have found that the separation of the ground state from the excited state continuum model is completely determined by the high dimension operators of the ope which are not used in formulating the continuum model . the practice of fixing the continuum threshold to a preferred value has been exposed as strongly determining the outcome of the sum rule analysis . qcd sum rule determinations of phenomenological spectral properties are necessarily dependent on the treatment of high dimension operators where a factorization assumption is usually invoked . moreover , careful attention to ope convergence is crucial to extracting meaningful results . these issues are somewhat unsettling , as we are working with expansions which are asymptotic . the emphasis here has been on exploring the qcd parameter space via monte carlo . further refinements of the parameter space might be made by adjusting the valid borel regime for each qcd parameter set such that the 10%50% criteria are satisfied . this was not done here in order to explore other criteria and reduce the sensitivity to a reasonable yet somewhat arbitrary criteria . in addition , the hlder inequalities might be applied as a further check of the validity and consistency of the sum rules @xcite . we are now in an excellent position to attack phenomena which is of current interest to the nuclear physics community . the techniques are currently being applied to an in - medium study of vector meson properties @xcite . there the debates over the behavior of the vector - meson masses and the sum rules to be used in extracting vector meson properties in nuclear matter are resolved . using standard in - medium condensate estimates , the analysis leaves no doubt that vector - meson masses decrease with increasing density . these techniques are also being applied to in - medium properties of the nucleon @xcite , by generalizing the vacuum sum rules analyzed here to finite nuclear matter density . the monte - carlo techniques are resolving the predictive ability of the in - medium sum rules . another investigation utilizing these monte - carlo techniques studies the qcd - sr determination of the @xmath295 meson mixing parameter @xmath16 @xcite . there , correlations between the ratio of @xmath296 and @xmath297 quark masses are particularly interesting . the relatively large uncertainty in the residue of the pole does not bode well for qcd sum rule analysis beyond two - point functions . for example , the residue of the pole appears as a pre - factor to any nucleon matrix element such as @xmath92 or magnetic form factors @xcite . it will not be surprising to find uncertainties in these observables as large as the 35% relative error in the squared residue itself . as the uncertainties in the additional vacuum matrix elements relevant to the three - point functions are folded in the uncertainties are likely to exceed 50% and may in fact approach 100% . analyses of these issues are currently underway @xcite . when compared to uncertainty estimates previously published in the literature , our uncertainty estimates are somewhat larger . this is not due to any shortcoming of the approach presented here . instead , previous authors have failed to provide a reliable method for determining realistic uncertainties . often , uncertainty estimates have been conjectured without any supporting quantitative analysis . this in - depth examination of qcd sum rule self consistency paints a favorable picture for further quantitative refinements of the approach . of course such progress will be very difficult , as there are many aspects of the approach which must be refined for any particular improvement to be relevant . the techniques introduced here , such as maintaining independence of ground state properties from the interpolating field mixing parameter , might be used to further narrow the uncertainties of the approach . it is extremely important to refine the qcd - sr approach such that uncertainties in the qcd input parameters may be reduced while maintaining @xmath298 . research in this direction is required before meaningful qcd - sr - based predictions for nucleon matrix elements and moments of structure functions are possible . i am indebted to kai hencken for his careful checking of the wilson coefficients and his discovery of new corrections in the dimension seven operators . his contribution is key to resolving the discrepancy between the sum rules presented here and modern estimates of the vacuum condensates . thanks also to xuemin jin for numerous beneficial discussions . dick furnstahl and xuemin jin have provided a critical examination of this analysis and i thank them for their insightful comments . thanks also to javed iqbal for his contributions to the evaluation of the finite width integrals for the @xmath0-meson correlator . many years ago , jimmy law provided the optimization routine used in this investigation , and i thank him for providing this useful program . this research was supported by the natural sciences and engineering research council of canada and the u.s . department of energy under grant de - fg06 - 88er40427 . d. b. leinweber , in _ lattice 93 , proceedings of the international symposium _ , edited by t. draper , s. gottlieb , a. soni , and d. toussaint , * 34 * , ( 1994 ) 407 , dallas , tx , 1993 , nucl . b ( proc . suppl . ) . f. x. lee , d. b. leinweber , and x. jin , new qcd sum rules for nucleon axial vector coupling constants , in _ panic 96 , proceedings of the xiv international conference on particles and nucleii _ , river edge , nj , ( 1996 ) , tjnaf , world scientific publishing co. , u. washington pp doe / er/40427 - 10-n96 .
insulating states of non - trivial topological order have attracted much attention both theoretically and experimentally . a topological insulator has a remarkable property of being metallic on the surface albeit insulating in the bulk . recently much focus is on a specific type of topological insulators,@xcite which are said to be `` @xmath0-nontrivial '' . @xcite the latter occurs as a consequence of interplay between a specific type of spin - orbit interaction and band structure . @xcite such systems are invariant under time reversal and show kramers degeneracy . from the viewpoint of an experimental realization , the original idea of kane and mele ( km ) @xcite was often criticized for being unrealistic , since it relies on a relatively weak spin - orbit coupling in graphene . in order to overcome such difficulty , bernevig , hugues and zhang ( bhz ) proposed an alternative system @xcite , which is also @xmath0-nontrivial but not based on graphene . bhz model was intended to describe low - energy electronic properties of a two - dimensional ( 2d ) layer of hgte / cdte quantum well . conductance measurement in a ribbon geometry @xcite showed that the system exhibits indeed a metallic surface state , which is also called _ helical _ edge modes . this paper highlights the spectrum and wave function of such helical edge modes in the bhz model . respecting appropriately the crystal structure of original hgte / cdte quantum well , one can safely implement it as a tight - binding model on a 2d square lattice.@xcite note , however , that in contrast to km model which can be represented as a purely lattice model , in bhz an internal spin-@xmath14 degree of freedom , stemming from the @xmath15-type @xmath16 and @xmath17-type @xmath18 orbitals , resides on each site of the square lattice in addition to the real electronic spin . we also mention that in the continuum limit with vanishing topological mass term , km model has two valleys ( @xmath19 and @xmath20 ) , whereas bhz has a single valley ( at @xmath21 ) . another idea which we can borrow from graphene study is the sensibility of edge spectrum on different types of edge structure , i.e. , either zigzag of armchair type . @xcite this applies also to the helical edge modes of bhz topological insulator in a ribbon geometry , since the edge spectrum is predominantly determined by how the 2d bulk band structure is `` projected '' onto the 1d edge axis . indeed , the structure of bhz helical edge modes has been extensively studied in ref . @xcite in the @xmath7-edge geometry and in the tight - binding implementation . however , in the practical experimental setup , @xcite this is certainly not the only one which is relevant to determine the transport characteristics at the edges . here , in this paper our main focus is on the other representative geometry , the @xmath11-edge . in the @xmath11-geometry , as a consequence of the specific way how `` hidden '' dirac cones ( or valleys ) in the 2d spectrum is projected onto the @xmath11-axis , edge modes show some unexpected behaviors . in order to motivate further the present study , let us first recall the importance of edge modes in the quantum hall state under magnetic field that exhibits a finite and quantized ( charge ) hall conductivity @xmath22 . in realistic samples with a boundary , quantization of hall conductivity is attributed to dissipationless transport due to a gapless chiral edge mode . in contrast to charge hall effect , a finite _ spin _ hall current does not need breaking of the time reversal symmetry . in the quantized spin hall ( qsh ) effect , the chern number in the bulk takes an integral value , and correspondingly there appears integral pairs of gapless edge modes . on the other hand , the @xmath0 topological insulator is characterized by an odd number of gapless modes per edge that are robust against weak perturbations preserving the time - reversal symmetry . the existence of such gapless edge states is generally guaranteed by a general theorem under the name of bulk / edge correspondence . @xcite the bhz model has a convenient feature that the location of the minimum energy gap can be controlled by changing the parameters in the model . in particular , the sign of spin hall conductivity changes discontinuously as the mass parameter of the model is varied . hence , the corresponding change of edge spectrum should provide useful information on the bulk / edge correspondence . furthermore , understanding of the nature of edge modes under specific edge geometries should serve as possible application of topologically protected phenomena in nano - architectonics . we take the representative cases of the @xmath7- and @xmath11-edges , which we call also the straight and zigzag edges , respectively . this paper is organized as follows : in sec . ii , we clarify our motivations to study the lattice version of a @xmath0 topological insulator ( bhz model ) , implemented as a nearest - neighbor ( nn ) tight - binding model . explicit form of the bhz lattice hamiltonian is introduced in sec . it is demonstrated that by considering a lattice model , one naturally takes into account _ hidden _ dirac cones , the latter lacking in the analyses based on the continuum dirac model . in secs . iv and v , we study the detailed structure of gapless edge modes under two different edge geometries : straight and zigzag edges . v is the highlight of this paper , demonstrating that the zigzag edge modes of bhz lattice model show unique features . we first point out that at @xmath23 , a pair of completely flat spectrum appears at @xmath24 ; besides the edge wave function can be trivially solved . we then show , in a half - empirical way , that this analytic exact solution at @xmath23 can be generalized to the case of an arbitrary @xmath25 $ ] ( this idea is schematically represented in fig . [ zig_concept ] ) . using the exact solution thus constructed , we highlight the nature of _ reentrant _ edge modes , another unique feature of the edge modes in the @xmath11-edge geometry . the reentrant edge modes possess two contrasting characters in real and momentum spaces : though well distinguished in real space , they live in an extremely small energy scale in the spectrum . vi is devoted to conclusions . the gapless edge modes of bhz topological insulator are also treated in the framework of continuum dirac model in appendix a. it has been well recognized that the quantized spin hall conductivity is determined by wave functions of bloch electrons over the entire brillouin zone . on the other hand , only the electronic states near the fermi level are relevant to the change of the hall conductivity when the fermi level is shifted . simplified effective models are useful for the latter case since various theoretical techniques can be employed in the low - energy range . in this paper , we work mainly with the lattice version of the bhz model , and make some comments in the low - momentum limit . why do we have to go back to a lattice model ? firstly , because we need to recover the _ correct _ absolute value of spin hall conductance @xmath26 . the latter is defined as the difference of hall conductance for up and down ( pseudo ) spins multiplied by @xmath27 : @xmath28 in quantized spin hall ( qsh ) systems , the spin hall conductance is quantized to be an integer in units of @xmath29 . in the language of charge conductance . ] this is completely in parallel with the quantization of _ charge _ hall conductance in units of @xmath30 in quantized hall system . in both cases , such integers are topological invariants and protected against weak perturbations . on the other hand , if one calculates , using kubo formula , the contribution of a single dirac fermion , e.g. , of the continuous dirac model at the @xmath21-point @xcite to spin hall conductance , then this gives _ half _ of the value expected from the topological quantization . @xcite in order to be consistent , it is naturally assumed that there must be _ even _ number of dirac cones . @xcite however , the low - energy effective theory with which we are starting contains obviously a single dirac cone . @xcite as we will see explicitly , such trivial discrepancy is naturally resolved by considering a lattice version of the bhz model . another aspect motivating us to employ the lattice version of bhz model is the fact the idea of an edge states is a real space concept , and we need _ a priori _ to go back to real space to give an unambiguous definition to it . in this paper , we highlight the detailed structure of gapless edge modes under a specific edge geometry . clearly , introduction of the latter needs a precise description in real space . recall also here that edge modes of graphene nano - ribbon exhibit contrasting behaviors in zigzag and armchair edge geometries : e.g. , the system becomes either metallic or semi - conducting in the armchair geometry , depending on @xmath31 , with @xmath32 being the number of rows , whereas a completely flat edge mode appears in the zigzag geometry . @xcite where does the difference comes from ? in momentum space , the question is how the bulk dirac cone structure look like when viewed from the edge . note that in the zigzag geometry the flat edge mode connects the two dirac points : @xmath19 and @xmath20 , whereas in the armchair geometry these two points are projected onto the same point at the edge . in a topological insulator , this bulk to edge projection is even a more subtle issue , since not all the dirac cones are explicit ( see table i ) . in a sense , zigzag edge in the square lattice bhz model ( see fig . [ geo_zig ] ) plays the following double role : it resembles the zigzag edge in graphene at @xmath33 , whereas it may rather resemble the armchair edge at @xmath34 and at @xmath35 . [ cols="<,<,^,^,^,^,<,^,^",options="header " , ] [ 4dpts ] spin hall conductance is a topological quantity determined by the _ global _ structure of entire 2d brillouin zone . the helical edge modes , encoding the same information , is , therefore , _ a priori _ not derived from a local description in the 2d brillouin zone . one exception to such a general idea is the study of ref . @xcite ( see also appendix ) , in which gapless edge modes are derived from the continuous model in a strip geometry by simply applying the condition that all the pseudo spin components of wave function vanish at the boundary . ( @xmath3)-sublattice component of the wave function vanishes at one ( the other ) boundary . for details , see , ref . @xcite . ] this implies that information about the helical edge modes is actually encoded in the original ( single ) dirac cone . this seems to be rather surprising , if one recalls that in the case of km model , the distinction between trivial and non - trivial phases is made by a relative sign of the mass gap at @xmath19- and @xmath20- points , which are , of course , macroscopically separated in momentum space . here , in the bhz model , the same distinction is made by relative sign between the mass ( @xmath36- ) term and the @xmath37- term added to the dirac hamiltonian . motivated by this observation , we investigate the structure of helical edge modes under different boundary conditions for the periodic bhz model , implemented as a square lattice and nearest - neighbor ( nn ) tight - binding model model . in parallel with the arm chair and zigzag edges for graphene , we consider ( a ) usually considered @xmath7- ( straight ) boundary @xcite , as well as ( b ) @xmath11- ( zigzag ) boundary for the tight - binding bhz model . the idea of focusing on dirac fermions in the description of quantized hall effect has appeared in the context of transitions between different plateaus . for describing the transitions , half - integer quantization is not a drawback , since the difference of hall conductance before and after the gap closing is quantized to be an integer in units of @xmath30 : @xmath38 or vice versa . a discrete jump in the hall conductance across the transition is indeed consistent with counting based on the emergence of massless dirac fermions at the transition @xcite . the absolute value of hall conductance is , on the other hand , a winding number , and determined by the vortices @xcite . here , each vortex gives , in contrast to a dirac cone , an integral contribution to the hall conductance ( in units of @xmath30 ) . an interesting question is whether the total hall conductance , often expressed as a topological invariant @xcite , can be also written as a sum of contributions from emergent dirac electrons in the spectrum . our empirical answer is yes,@xcite indicating that the number of dirac electrons emergent in the spectrum is always even , reminiscent of the no - go theorem of nielsen and ninomiya in 3 + 1 dimensions @xcite . it should be noted that here not only explicit dirac electrons ( gapless for a given set of parameters ) but also hidden dirac electrons ( massive for that set of parameters ) must be taken into account . such massive dirac electrons are called `` spectators '' in ref . @xcite , in the sense that they are inactive for the transition . spectators are indispensable to ensure the correct integer quantization of the hall conductance . ( vertical axis ) of a bhz lattice model , here , implemented as a nn tight - binding model on a square lattice : cf . ( [ dk_nn ] ) . the spectrum is shown over the entire brillouin zone : @xmath39 ( horizontal plane ) . parameters are chosen such that @xmath40 and @xmath41 , i.e. , the system is in the topologically non - trivial phase : @xmath42 . the spectrum shows a typical wine - bottle structure around the @xmath21-point.,width=302 ] let us first consider the bhz model in the long - wave - length limit . the low - energy effective hamiltonian , describing the vicinity of gap closing at @xmath43 , is the minimal model to capture the physics of a @xmath0-topological insulator . this effective hamiltonian is also contrasting to the prototypical km model , in that the former describes only a single dirac cone . the distinction between the @xmath0- trivial and non - trivial phases is , therefore , made by adding a @xmath37-term to the usual dirac hamiltonian . the explicit form of bhz hamiltonian is implemented by the following @xmath44 matrix : @xmath45 , \label{htot}\end{aligned}\ ] ] where @xmath46 is a 2d crystal momentum , here measured from the @xmath21-point . the lower - right block @xmath47 is deduced from @xmath48 by imposing time reversal symmetry . the bulk energy spectrum : @xmath49 is then given by solving the eigenvalue equation for the upper - left block @xmath48 of the @xmath44 bhz hamiltonian , i.e. , @xmath50 in order to represent @xmath48 in a compact form , we introduce a @xmath51-vector , @xmath52 , each component of which is either an even or an odd function of @xmath53 : @xmath54 , whose parity is determined by symmetry considerations . @xcite as far as the low - energy universal properties in the vicinity of @xmath21-point is concerned , we need to keep only the lowest order terms of @xmath55 , and in this long - wave - length limit , they read explicitly , @xmath56 other parameters which appear in ref . @xcite , i.e. , @xmath57 and @xmath58 are set to be zero , which , however , does not lose any generality . the bulk energy spectrum is thus determined by diagonalizing the following `` spin hamiltonian '' , @xmath59 , where @xmath60 , are pauli matrices . using the standard representation for @xmath61 , @xmath62 reads explicitly as , @xmath63 . \label{hk}\ ] ] each row and column of eq . ( [ hk ] ) represent an `` orbital spin '' associated with the @xmath15-type @xmath16 and the @xmath17-type @xmath18 orbitals of the original 3d band structure of hgte and cdte . @xcite then , by choosing the `` spin quantization axis '' in the direction of @xmath51-vector , one can immediately diagonalize the hamiltonian @xmath48 , i.e. , @xmath64 where the eigenvalue @xmath65 given by , @xmath66 this implies , @xmath67 where @xmath68 . when @xmath69 , the dispersion relation ( [ wine ] ) represents a wine - bottle structure ( fig . 2 ) , i.e. , @xmath70 shows a minimum at a finite value of @xmath71 . at the critical value @xmath72 , the density of states shows van hove singularity . in units of @xmath73 , obtained by numerically evaluating the @xmath53-integral in eq . ( [ tknn ] ) , plotted as a function of @xmath8.,width=8 ] lattice version of the bhz model is implemented as a tight - binding hamiltonian . to construct such a hamiltonian explicitly , we replace linear and quadratic dependences in @xmath48 on @xmath74 and @xmath75 as in eqs . ( [ dk ] ) , by a function which has the right periodicity of the square lattice . this can be implemented as , @xmath76 , \label{dk_nn}\end{aligned}\ ] ] where @xmath77 is the lattice constant . ( [ dk_nn ] ) corresponds to regularizing the effective dirac model on a square lattice with only nearest - neighbor ( nn ) hopping . in this setup , i.e. , eqs . ( [ htot ] ) , ( [ hk ] ) together with eqs . ( [ dk_nn ] ) , the lattice version of bhz model acquires four gap closing points shown in table i , if one allows the original mass parameter @xmath2 to vary beyond the vicinity of @xmath78 . the new gap closing occurs at different points in the brillouin zone from the original dirac cone ( @xmath21-point ) , namely at @xmath79 , @xmath80 and @xmath81 . the gap closing at @xmath82 occurs at @xmath83 , whereas the gap closing at @xmath84 and @xmath85 occurs simultaneously when @xmath86 . each time a gap closing occurs , one can re - expand the lattice model with respect to small deviations of @xmath53 measured from the gap closing . the new effective model in the vicinity of such hidden gap closing falls on the same dirac form as the original one at the @xmath21-point , up to the @xmath37-term . in order to quantify the emergence of such hidden dirac cones , one still needs the following two parameters : ( i ) the mass gap @xmath2 ( especially , its sign ) , and ( ii ) the chirality @xmath87 . the latter is associated with the homotopy in the mapping : @xmath88 . note that in the gap closing at @xmath84 and @xmath85 at @xmath86 , the role of @xmath89 is interchanged compared with the original dirac cone at the @xmath21-point . the former ( latter ) corresponds to @xmath90 ( @xmath91 ) . the missing dirac partner , in the sense of ref . @xcite , is found in this way . once the explicit form of effective dirac hamiltonian in the continuum limit is given , one can determine its contribution to @xmath92 . in systems with trs , i.e. , of the form ( [ htot ] ) , the contribution from @xmath48 to @xmath93 cancels with that of @xmath94 . on the other hand , their contribution to @xmath95 , remains finite , takes a half - integral value , @xmath96 in units of @xmath73 ( @xmath97 in conductance ) . contribution to @xmath98 from a dirac point with a mass gap @xmath2 and chirality @xmath87 is , @xmath99 provided that the fermi energy is in the gap . this can be verified explicitly by applying the kubo formula to the continuum model . as mentioned earlier , such counting based on the continuum dirac model , is known to describe correctly a discrete jump of @xmath100 in the quantum hall case . here , we apply the same logic to qsh case . imagine that one observes the evolution of @xmath98 , starting with the trivial insulator phase , where @xmath101 , and varying the mass parameter @xmath2 . the fermi energy is always kept in the gap unless there appears a dirac cone . each time such a gap closing occurs , @xmath98 shows a discrete change , which can be attributed to the above dirac fermion argument . in the present model , one can verify explicitly that this is indeed the case . if one evaluates the spin hall conductance from tknn formula , @xcite @xmath102 allows for the following representation , in terms of berry curvature integrated over the entire brillouin zone : @xmath103 where @xmath104 and @xmath105 is given , e.g. , by eq . ( [ dk_nn ] ) . for such an explicit choice of @xmath106 , eq . ( [ tknn ] ) is evaluated numerically , and plotted as a function of @xmath107 in fig . [ sigma_xy ] . when @xmath106 is given by eq . ( [ dk_nn ] ) , the plotted curve ( the solid curve shown in blue in fig . [ sigma_xy ] , which looks practically like steps ) is comparable with the column @xmath108 of table i. note that the absolute value of @xmath102 is susceptible of the concrete implementation of @xmath106 over the entire brillouin zone , whereas its parity ( whether it is even or odd ) in units of @xmath109 remains the same . as well known , the latter determines system s @xmath0-property . @xcite we have seen that @xmath102 takes a finite value @xmath110 when @xmath111 , i.e. , the system is in the topological ( inverted gap ) phase . this is also consistent with the gapless edge picture , in which the spin hall conductance of twice the unit of quantum conductance @xmath30 is attributed to two channels of edge modes , which form a pair of kramers partners . the apparent half - integer quantization at the @xmath21-point , in the sense of eq . ( [ 1/2 ] ) , is compensated by the contribution from missing dirac partner(s ) , and as a result , is indeed shifted by one - half , replaced by an expected integral quantization . edges ) are , here , chosen to be perpendicular to the @xmath112-direction . in this figure , the number of rows in the strip is @xmath113.,width=6 ] let us first review the behavior of gapless edge modes in the straight edge geometry , the latter commensurate with the square lattice , and can be chosen either normal to the @xmath7- or to the @xmath112-direction ( as in fig . [ geo_str ] ) . introducing an edge leads to breaking of the translational invariance in the direction perpendicular to the edge , inducing a coupling between dirac cones . in the straight edge geometry shown in fig . [ geo_str ] ) , electrons are confined inside a strip between the rows at @xmath114 and @xmath115 . the translational invariance along the @xmath116-axis is still maintained , allowing for constructing a 1d bloch state with a crystal momentum @xmath74 : @xmath117 where @xmath74 is measured in units of @xmath118 with @xmath77 being the lattice constant . @xmath119 is a one - body electronic state localized on site @xmath120 , and @xmath121 is an operator creating such an electron . it is also convenient to introduce @xmath122 , and express @xmath123 as @xmath124 . naturally , the two creation operators are related by fourier transformation similarly to eq . ( [ bloch ] ) , i.e. , @xmath125 . in order to introduce the edges , it is convenient to rewrite the bhz tight - binding hamiltonian in terms of the hopping between neighboring rows . let us first consider the bhz tight - binding hamiltonian in real space : @xmath126 , \label{ham_bhz}\end{aligned}\ ] ] where @xmath127 hopping matrices , @xmath128 and @xmath129 , are given explicitly as , @xmath130 in order to rewrite it in terms of the bloch state , eq . ( [ bloch ] ) , or equivalently , in terms of the corresponding creation and annihiration operators , @xmath122 ( @xmath131 ) , we perform fourier transformation in the ( @xmath116- ) direction along the edge . ( [ ham_bhz ] ) thus rewrites , @xmath132 \label{ham_str}\ ] ] where @xmath133 s are diagonal ( on - row ) components , which read explicitly , @xmath134 @xmath135 represents a hopping amplitude in the @xmath136-direction , i.e. , between neighboring rows . inside the strip , i.e. , for @xmath137 , these amplitudes take the same value as in the bulk , given in eqs . ( [ gamma_xy ] ) , i.e. , @xmath138 in the tight - binding implementation , a strip geometry can be introduced by switching off all the hopping amplitudes connecting sites on the edge of the sample to the exterior of the sample . in our straight edge geometry , such outermost rows are located at @xmath139 and @xmath140 . we turn off all the hopping amplitudes from @xmath139 to @xmath141 , and the ones from @xmath142 to @xmath143 , i.e. , @xmath144 this boundary condition , ( i ) breaks the translational invariance in the @xmath136-direction , and ( ii ) restricts the hamiltonian matrix into @xmath145 blocks . and @xmath146 plotted as a function @xmath8 in the limit @xmath147.,width=8 ] let us construct the eigenvector of the straight edge hamiltonian , eq . ( [ ham_str ] ) , with an eigenenergy @xmath148 . ( [ ham_str ] ) is already diagonal w.r.t . @xmath74 , we diagonalize eq . ( [ ham_str ] ) for a given @xmath74 , to find the energy spectrum @xmath149 . the corresponding eigenvector is thus specified by @xmath148 and @xmath74 , and takes generally the following form : @xmath150 where @xmath151 is a @xmath127 spinor specifying the amplitude and the pseudo spin state of eigenvector on row @xmath152 . one might rather regard , @xmath153 , \label{wf}\ ] ] as the wave function of the corresponding eigenstate . the eigenvalue equation , @xmath154 can be rewritten , in terms of the @xmath155 s , in the form of a recursive equation : @xmath156 all the information on the spectrum and the wave function of both the extended bulk states and the localized edge states is encoded in eq . ( [ rec_str ] ) and the boundary condition which we will specify later . since the recursive relation , eq . ( [ rec_str ] ) , is linear , its eigenmodes take the form of a geometric series : @xmath157 where @xmath158 is a solution of the characteristic equation which we will derive later . if @xmath159 , eq . ( [ ser_rho ] ) may represent an edge mode . since the recursive relation , eq . ( [ rec_str ] ) , is of second order , its characteristic equation becomes a quadratic equation , giving two solutions for @xmath158 , say , @xmath160 . on the other hand , our recursive equation has also a @xmath127 matrix form , we first have to solve a ( reduced ) eigenvalue equation for @xmath161 , assuming that @xmath158 is given . the reduced eigen value equation for @xmath161 reads , @xmath162 \psi_0 = e \psi_0.\ ] ] using eqs . ( [ gamma_xy ] ) this can be also rewritten as , ) corresponds to the choice , @xmath163 , or @xmath164 which is consistent with the bloch theorem . in the strip geometry with the periodic boundary condition , @xmath75 takes discrete values . with the open boundary condition relevant to the actual strip , @xmath75 is no longer a good quantum number . however , if the width of the strip is large , one may roughly interpret the 1d energy spectra in the strip geometry as composed of the many slices of bulk energy spectrum at different values of @xmath75 . in addition , a pair of edge modes appear as a characteristic feature of the nontrivial topological property . ] @xmath165 \psi_0 = e \psi_0 , \label{eeq_red}\ ] ] where @xmath166 is given in eq . ( [ dkx ] ) . this is a @xmath127 eigenvalue equation , and there are generally two solutions for @xmath148 and two corresponding eigenvectors for a given @xmath158 . recall here that for @xmath167 , our numerical data ( fig . [ spec_str ] ) show that the edge spectrum behaves as @xmath168 in the limit of @xmath147 . hereafter , we will focus only on such edge solutions . since in the same limit , @xmath169 , eq . ( [ eeq_red ] ) reduces to , @xmath170 \psi_0 = 0 . \label{eeq_0}\ ] ] note that we have multiplied both sides of eq . ( [ eeq_red ] ) by @xmath171 . it is clear from this expression that @xmath161 can be chosen to be an eigenstate of @xmath172 , i.e. , @xmath173 , where @xmath174,\ \ \left [ \begin{array}{r } 1\\ -1 \end{array } \right ] . \label{x+-}\ ] ] if one denotes the eigenvalue of @xmath172 by @xmath15 , as @xmath175 , then @xmath176 corresponds to @xmath177 , and @xmath178 to @xmath179 . namely , @xmath15 specifies the eigenspinors given in eqs . ( [ x+- ] ) . based on these eigenspinors , one can construct the total wave function . of course , one still needs to do determine the allowed values of @xmath158 . for an eigenstate specified by @xmath15 , @xmath158 must satisfy , @xmath180 for @xmath176 , @xmath181 in eq . ( [ quad ] ) , if @xmath158 is a solution of this quadratic equation for @xmath176 , then @xmath182 satisfies the same equation for @xmath178 . @xcite thus the general solution becomes a linear combination of the following four basic solutions : @xmath183 |x+ \rangle \nonumber \\ & + & % .\left [ c_{-1 } \left(1 \over \rho_1 ( 0 ) \right)^{j } + \left [ c_{-1 } \rho_1 ( 0 ) ^{-j } + c_{-2 } \rho_2 ( 0 ) ^{-j } \right ] |x- \rangle , \label{sol_0}\end{aligned}\ ] ] of course , at this point , this is just a solution at only one single @xmath184-point , @xmath185 . however , a solution in the form of eq . ( [ sol_0 ] ) , with @xmath186 given in eq . ( [ rho12_0 ] ) , can be easily generalized to satisfy eq . ( [ rec_str ] ) for an arbitrary , finite @xmath74 , after a simple replacement of parameters . as for the eigenmode of the form of eq . ( [ ser_rho ] ) , one has to solve a reduced eigenvalue equation for @xmath187 , and determine @xmath158 such that eq . ( [ eeq_red ] ) is satisfied . however , since @xmath133 has the structure given in eq . ( [ dkx ] ) , the eigenmodes for @xmath187 given as eq . ( [ x+- ] ) remain to be valid for an arbitrary @xmath74 . this might become clearer , if one decomposes eq . ( [ eeq_red ] ) into the following set of equations : @xmath188 \psi_0 & = & 0 , \label{eeq_2}\end{aligned}\ ] ] i.e. , eq . ( [ eeq_red ] ) is recovered by adding both sides of eqs . ( [ eeq_1 ] ) and ( [ eeq_2 ] ) . the first equation gives the energy dispersion , @xmath189 , if @xmath187 is chosen to be an eigenstate of @xmath172 , i.e. , @xmath190 , where @xmath191 note that this is an exact edge spectrum valid over the entire range of @xmath74 , as far as the edge solution is possible ( see fig . [ spec_str ] ) . on the other hand , eq . ( [ eeq_2 ] ) , analogous to eq . ( [ eeq_0 ] ) , justifies the previous conjecture : @xmath192 . while , in the characteristic equation for @xmath158 , one has to make the simple replacement : @xmath193 , i.e. , eqs . ( [ eeq_2 ] ) and ( [ eeq_0 ] ) are identical up to this replacement . for @xmath176 , the solution for @xmath158 reads , @xmath194 correspondingly , a general solution for @xmath195 can be constructed as , @xmath196 |x+ \rangle \nonumber \\ & + & \left [ c_{-1 } \rho_1 ( k_x ) ^{-j } + c_{-2 } \rho_2 ( k_x ) ^{-j } \right ] |x- \rangle , \label{sol_gen}\end{aligned}\ ] ] where the coefficients @xmath197 should be chosen to satisfy the boundary conditions . ( [ sol_gen ] ) is smoothly connected to eq . ( [ sol_0 ] ) in the limit : @xmath147 . three panels of fig . [ spec_str ] show the energy spectrum ( edge + bulk ) for different values of @xmath8 . as for the edge part of the spectrum , only a part of eq . ( [ spec_th_str ] ) is realized . in order to determine which part of the spectrum in eq . ( [ spec_th_str ] ) is indeed activated , we discuss below the case of semi - infinite geometry in some detail . [ spec_str ] also demonstrates one of another specific feature of straight edge mode that the main location of the mode moves from the zone center for @xmath198 , to the zone boundary for @xmath199 . thus , the group velocity intersecting with the fermi level reverses its sign , reflecting the sign change of @xmath200 in the bulk . this can be regarded as the concrete expression of bulk / edge correspondence in the present case . @xcite what kind of a boundary condition should we apply in eq . ( [ sol_gen ] ) ? suppose that here our system is _ semi - infinite _ , for simplicity , extended from @xmath201 to @xmath202 . such a boundary condition can be applied , by formally requiring that the wave function ( [ sol_gen ] ) vanishes at @xmath203 , i.e. , @xmath204.\ ] ] this means that the coefficients @xmath197 in eq . ( [ sol_gen ] ) should be chosen to satisfy , @xmath205 this turns out to be rather an important requirement for determining the range of validity of the solution given in eq . ( [ sol_gen ] ) , since the wave function @xmath195 must be normalizable . in eq . ( [ sol_gen ] ) , only the eigenmodes of the form of eq . ( [ ser_rho ] ) with @xmath159 should be kept in the solution ( to be precise , _ both _ @xmath206 and @xmath146 must be smaller than 1 ) . in a strip geometry , another solution , consisting of both @xmath207 , describes the edge mode localized at the opposite end of the system . and @xmath146 plotted as a function of @xmath74 at @xmath208 ( @xmath40 ) . reference lines are at @xmath209 , and at @xmath164 and @xmath210 as for the latter , cf . ( [ rho_flat ] ) . @xmath164 corresponds indeed to the point at which the edge modes merge with the bulk spectrum ( central panel , @xmath211 ) . lower panel : an enlarged image of the central panel , plotted also for a system of larger size : @xmath212.,title="fig:",width=8 ] and @xmath146 plotted as a function of @xmath74 at @xmath208 ( @xmath40 ) . reference lines are at @xmath209 , and at @xmath164 and @xmath210 as for the latter , cf . ( [ rho_flat ] ) . @xmath164 corresponds indeed to the point at which the edge modes merge with the bulk spectrum ( central panel , @xmath211 ) . lower panel : an enlarged image of the central panel , plotted also for a system of larger size : @xmath212.,title="fig:",width=8 ] and @xmath146 plotted as a function of @xmath74 at @xmath208 ( @xmath40 ) . reference lines are at @xmath209 , and at @xmath164 and @xmath210 as for the latter , cf . ( [ rho_flat ] ) . @xmath164 corresponds indeed to the point at which the edge modes merge with the bulk spectrum ( central panel , @xmath211 ) . lower panel : an enlarged image of the central panel , plotted also for a system of larger size : @xmath212.,title="fig:",width=8 ] in fig . [ sol_str1 ] , @xmath206 and @xmath146 are plotted as a function of @xmath8 in the limit @xmath147 . when @xmath213 , both @xmath206 and @xmath146 are _ smaller _ than 1 , namely both @xmath214 and @xmath215 are _ larger _ than 1 . this means that only the first two terms of eq . ( [ sol_gen ] ) , both corresponding to @xmath216 ( @xmath176 ) , should be kept in the solution , i.e. , @xmath217 outside this region , either @xmath206 or @xmath146 is larger than 1 . when @xmath207 , since this implies automatically @xmath218 , the eigenmode corresponding to the latter is still compatible with the boundary condition at @xmath219 . however , because of the boundary condition at @xmath203 , i.e. , eq . ( [ req_bc ] ) , when @xmath220 and @xmath221 , or vice versa , the only possible choice for the coefficients @xmath197 is @xmath222 namely , a solution of the type of eq . ( [ sol_gen ] ) , or an edge mode crossing at @xmath185 is _ inexistent_. this is consistent with the fact that an edge mode crossing at @xmath185 exists only in the region , @xmath223 $ ] in the straight edge geometry . ( when @xmath224 $ ] , the edge modes cross at @xmath225 , i.e. , at the zone boundary , which is also time - reversal symmetric . ) coming back to the regime in which edge modes are existent , i.e. , @xmath226 $ ] , one can clearly see in fig . [ sol_str1 ] that there are two different behaviors a flat region where @xmath206 and @xmath146 are degenerate , and the remaining part with two branches . this is due to the fact that the two solutions for @xmath158 could be either both real , or a pair of complex numbers conjugate to each other . in the latter case , the two solutions have the same absolute value , @xmath227 , whereas in the present case , one can verify that this degenerate value is independent of @xmath2 , i.e. , @xmath228 this explains the existence of a flat region in fig . [ sol_str1 ] . from eq.([rho12_0 ] ) we see that the square root becomes pure imaginary for all @xmath184 provided @xmath229 , where appear in the regime : @xmath230 , and also at the other end . in the approximation that becomes valid in the small wave number , this threshold value is given by @xmath231 . for @xmath40 , @xmath232 , whereas , @xmath233 is , of course , @xmath234 . if one expands eq . ( [ delta_pm ] ) in powers of @xmath235 , then at leading order @xmath236 coincides with @xmath233 . ] @xmath237 at @xmath164 , the edge solution is expected to merge with the bulk spectrum . this happens , when @xmath238 such a behavior becomes clearer by plotting @xmath239 s as a function of @xmath74 . [ sol_str2 ] illustrates this feature at @xmath41 for @xmath40 . one can indeed see that the edge spectra merge with the bulk at @xmath240 , satisfying eq . ( [ km ] ) . the latter reduces , at this value of @xmath2 , to @xmath241 , i.e. , @xmath242 . it is also instructive to investigate the nature of edge modes in _ real _ space , i.e. , the wave function , and compare it with the general solution ( [ sol_gen ] ) . in numerical experiments , one has to diagonalize the @xmath243 hamiltonian matrix , equivalent to eq . ( [ ham_str ] ) . an eigen wave function is , therefore , obtained as a @xmath244-component vector ; here , in the straight edge geometry , the latter can be chosen to be real . the edge wave function is easily identified if it exists , e.g. , by choosing the lowest - energy eigenmode @xmath161 in the upper band . by investigating the structure of such an edge wave function , one can explicitly verify that the eigenmodes are spanned by two eigenspinors given in eq . ( [ x+- ] ) . in repeating such numerical experiments for different @xmath74 and @xmath8 , one can naturally distinguish an edge state from a bulk state by focusing on the spatial distribution of the wave function . here , what deserves much attention is that one can recognize a one - to - one correspondence between localizability of the wave function @xmath161 and its spinor structure . let us turn to the case of a different edge geometry , the zigzag edge geometry , shown schematically in fig . [ geo_zig ] . as mentioned earlier , the zigzag edge geometry considered here is , in a sense , analogous to a more popular edge geometry of graphene ribbon , named in the same way , but defined on a hexagonal lattice . here , on a square lattice , a zigzag edge is introduced , either normal to @xmath11- or @xmath245-direction ( as in fig . [ geo_zig ] ) . electrons in the zigzag edge geometry are , therefore , confined to a strip diagonal in the cartesian coordinates , say , along the @xmath11-direction as in fig . [ geo_zig ] . -edge geometry ( here , chosen to be normal to the @xmath245-axis ) , defined in terms of the original square lattice , on which the new @xmath246- and @xmath247-axes are superposed in blue ( online ) . numbers on these axes are redefined indices @xmath248 and @xmath249 , i.e. , @xmath250 . the number of rows @xmath32 is here chosen to be , @xmath251 . when @xmath32 is even ( odd ) , the two edges are inversion asymmetric ( symmetric ) w.r.t . the center of strip.,width=6 ] intuitively , say , because the zigzag surface is literally , `` rough '' , one expects that this edge geometry might have a stronger tendency to trap electrons in the vicinity of the boundary . we show below , on one hand , that this intuition from the macroscopic world is still valid in the microscopic quantum mechanical world . but just as a result of this stronger tendency to keep the electrons in its vicinity , on the other hand , the zigzag edge shows various curious phenomena , e.g. , completely flat edge modes , and the reentrance of edge modes in @xmath184-space , etc . clearly , the translational invariance along the @xmath11-axis is maintained , on which @xmath246-axis is introduced , together with the conserved momentum @xmath252 in this direction . accordingly , @xmath247-axis is chosen to be in the @xmath245-direction . it may be also useful to _ redefine _ the indices @xmath248 and @xmath249 such that the lattice points in the original square lattice are located at @xmath250 , where @xmath248 is an even ( odd ) integer for @xmath249 : even ( odd ) . in the zigzag edge geometry , the spectrum @xmath253 and the wave function @xmath254 are determined by the following recursive equation : @xmath255 for @xmath195 , analogous to eq . ( [ rec_str ] ) in the straight edge geometry . in order to derive eq . ( [ rec_zig ] ) , we first rewrote the tight - binding hamiltonian ( 15 ) in the new labeling , and then considered a bloch state along the @xmath246-axis , analogous to eq . ( [ bloch ] ) , but with a crystal momentum @xmath184 conjugate to @xmath246 , i.e. , @xmath252 . as was the case in eq . ( [ rec_str ] ) , @xmath21 describes , here , in eq . ( [ rec_zig ] ) the hopping between adjacent rows , and reads explicitly as , @xmath256 note that here the crystal momentum @xmath184 is measured in units of @xmath257 so that the zone boundary is always given by @xmath258 . and @xmath259 . ] to find an edge solution , we first express the solution of eq . ( [ rec_zig ] ) , in the form of a geometric series , written formally in the same as eq . ( [ ser_rho ] ) . recall that @xmath158 is ( generally ) a ( complex ) number of , for an edge mode , amplitude smaller than unity ( with the understanding that the edge mode is localized in the vicinity of @xmath203 ) . @xmath187 is a two component eigenvector of the following reduced eigenvalue equation : @xmath260 \psi_{0 } = e \psi_0 . \label{reduced}\ ] ] from the analogy to the straight edge case , one may express @xmath21 as @xmath261 and rewrite eq . ( [ reduced ] ) into the following explicit form : @xmath262 \psi_{0 } = e \psi_0 . \label{}\end{aligned}\ ] ] where we @xmath263 and @xmath264 are short - hand notations for , respectively , @xmath265 and @xmath266 . comparing this form with the straight edge case , one can see that here one can not use the same recipe for solving the problem , i.e. , solving the problem at , say , @xmath267 and then extrapolate its solution to general @xmath184 . it seems not impossible to proceed in that direction and solve the problem analytically , but here , we choose to take another route , which is much simpler , to find still an exact solution , but in a half - empirical way . + some concrete examples of such energy spectrum are shown fig . [ spec_zig ] . a pair of gapless edges modes always appear iff @xmath42 . in contrast to the straight edge case , however , they appear always in the vicinity of @xmath267 , and intersects at @xmath267 . one of the last panels of fig . [ spec_zig ] shows a unique feature of edge modes in the zigzag edge geometry . at @xmath86 , the edge modes become completely flat , apart from a small finite - size gap around the zone boundary . we have already seen such flat edge modes in graphene in the case of zigzag edge geometry ( but on a hexagonal lattice ) . @xcite in graphene , such flat edge modes connect 1d projection of @xmath19 and @xmath20 points via the 1d bz boundary . this is a similar behavior to the present case , if one regards the former as the limit of vanishing intrinsic coupling ( or topological mass @xmath2 ) in the km model . one of the differences between the two cases is that here the flat edge modes cover the entire 1d brillouin zone . in order to elucidate the nature of flat edge mode at @xmath33 , first notice that at this value of @xmath2 the diagonal terms of ( the diagonal blocks of the hamiltonian matrix in ) eq . ( [ rec_zig ] ) vanish this implies , as in graphene nano - ribbon in the zigzag edge geometry , the existence of an eigenstate of the form : @xmath268 \label{psi_2j=0}\ ] ] i.e. , an eigenvector satisfying @xmath269 ( @xmath270 ) . here , a semi - infinite geometry is implicit ( for approximating a ribbon of sufficiently large width or @xmath32 ; our system extended from @xmath201 to @xmath271 ) , with a boundary condition of @xmath272 . under this setup , and with the condition of vanishing diagonal matrix elements , eq . ( [ rec_zig ] ) implies , @xmath273 simultaneously , it should also have a vanishing eigenenergy @xmath24 for consistency . clearly , eq . ( [ rec_d4b ] ) has a solution of the form of a geometric series , here , for @xmath274 ( @xmath275 ) : @xmath276 note that here @xmath277 plays , roughly , the role of @xmath278 , but their precise relation will become clearer when the entire problem is solved . in order to proceed , we recall that @xmath21 can be written explicitly as , @xmath279 . \label{gamma_zig_mat}\ ] ] then , by assuming a solution of the form of eq . ( [ ser_lambda ] ) , eq . ( [ rec_d4b ] ) can be reduced to the following eigenvalue problem for @xmath280 : @xmath281 with the eigenvalues , @xmath282 and the corresponding eigenvector , @xmath283 , i.e. , @xmath284 given explicitly as , @xmath285 . \label{u+-}\ ] ] the coefficient @xmath286 is a function of @xmath184 , which takes precisely the following form : @xmath287 notice that @xmath288 , and the two eigenspinors @xmath283 are orthogonal : @xmath289 . a general solution in the form of eq . ( [ psi_2j=0 ] ) can be thus constructed by applying @xmath290 , recursively , to @xmath291 and the result is , @xmath292 + c_- \lambda_-^j \left [ \begin{array}{c } \alpha_- ( 1- i)\\ 1 \end{array } \right ] . \label{sol_d4b}\ ] ] in this construction , the two eigenvectors @xmath283 always have the form of eq . ( [ u+- ] ) . this feature remains when @xmath2 is away from @xmath293 at which the edge modes are no longer completely flat , or rather even in the regime in which the edge spectrum is not flat at all . another remark , concerning the behavior of eq . ( [ sol_d4b ] ) is that under the choice of signs in eq . ( [ lambda+- ] ) , @xmath294 is always larger than 1 , whereas @xmath295 except at the zone boundary . this can be easily verified either numerically , i.e. , by plotting @xmath296 as a function of @xmath184 , or by showing @xmath297 using directly the expression for @xmath296 in eq . ( [ lambda+- ] ) . in numerical simulation for systems of a finite number of rows , both of these two solutions play a role giving rise to a pair of edge solutions . is odd , one solution , corresponding , say , to @xmath298 , is localized in the vicinity of @xmath201 , and indeed has the form of eq . ( [ psi_2j=0 ] ) with @xmath299 given as eq . ( [ sol_d4b ] ) and @xmath300 . the other solution , corresponding to @xmath301 , has the same structure of eq . ( [ psi_2j=0 ] ) but with @xmath299 increasing with practically an equal geometric ratio of @xmath301 , and naturally localized in the vicinity of the other edge : @xmath271 . on the other hand , when @xmath32 is even , the eigenmode of the system becomes a linear combination of the above two types of solutions . this even / odd feature occurs only at @xmath2 precisely equal to @xmath293 , since at this value of @xmath2 where the edge spectrum becomes completely flat , the two ( generally ) counter - propagating edge modes acquire the same ( zero ) group velocity , and get mixed . ] plotted as a function of @xmath184 at @xmath302 but for different values of @xmath3 , i.e. , @xmath303 and @xmath304 , corresponding , respectively , the colors : red , green , blue , cyan , magenta and orange ( upper panel ) . edge spectrum at @xmath86 and @xmath305 ( @xmath302 , @xmath212 , lower panel ) . two reference lines are at @xmath306 , the value of @xmath184 at which @xmath298 vanishes at @xmath305 . the axes are suppressed in the lower panel so as to highlight the completely flat edge modes.,title="fig:",width=8 ] plotted as a function of @xmath184 at @xmath302 but for different values of @xmath3 , i.e. , @xmath303 and @xmath304 , corresponding , respectively , the colors : red , green , blue , cyan , magenta and orange ( upper panel ) . edge spectrum at @xmath86 and @xmath305 ( @xmath302 , @xmath212 , lower panel ) . two reference lines are at @xmath306 , the value of @xmath184 at which @xmath298 vanishes at @xmath305 . the axes are suppressed in the lower panel so as to highlight the completely flat edge modes.,title="fig:",width=8 ] in fig . [ lambda ] , the upper panel shows @xmath298 plotted as a function of @xmath184 at @xmath302 but for different values of @xmath3 , naturally assuming @xmath86 . when @xmath3 is smaller than a critical value @xmath307 @xmath298 changes its sign ( has a zero ) at intermediate @xmath184 . the edge wave function is extremely localized , i.e. , onto a single row : @xmath201 or @xmath271 . ] this continues to be the case even in the limit @xmath3 vanishes . the lower panel shows the edge spectrum at @xmath86 , @xmath305 and @xmath302 . note that the zero of @xmath298 corresponds to the value of @xmath184 at which the bulk spectrum focuses onto a single point . as we will describe in detail in the next subsection , our `` recipe '' for constructing the _ exact _ edge wave function , and simultaneously its spectrum , lies in `` extrapolating '' the exact solution available at @xmath86 to a general value of @xmath308 $ ] ( recall also fig . [ zig_concept ] ) . to complete this program , we need to refer to some results of the numerical experiments performed for a system of finite number of rows . we have already seen the spectrum of such systems in fig . [ spec_zig ] ; here we focus on the behavior of wave function , i.e. , the behavior of @xmath195 as a function of @xmath152 . in the zigzag edge geometry , it is remarkable that ( not only ) the edge wave function ( but also the bulk wave function ! ) has the following particular form : @xmath309 . \label{evec_gen}\ ] ] when that eigenstate represents an edge mode , eq . ( [ evec ] ) further simplifies : @xmath310 \vspace{0.2 cm } \\ c_4 \left [ \begin{array}{c } \alpha ( 1-i ) \\ 1 \end{array } \right ] \vspace{0.2 cm } \\ c_6 \left [ \begin{array}{c } \alpha ( 1-i ) \\ 1 \end{array } \right ] \vspace{0.2 cm } \\ \vdots \\ \end{array } \right ] , \label{evec_edge}\ ] ] i.e. , for a given set of parameters @xmath311 , @xmath3 ( and @xmath2 ) as well as for a fixed @xmath184 , @xmath312 is a constant ( @xmath313 ) . the ratio , on the other hand , @xmath314 is a measure of , to what extent the edge mode is localized in the vicinity of a boundary , say , at @xmath201 . ( @xmath275 ) plotted ( red points with filling to the horizontal axis ) at @xmath315 for @xmath316 ( upper panel ) and @xmath317 ( lower panel ) . @xmath40 , @xmath318 . in the upper panel , the blue line corresponds to the `` theoretical '' value , @xmath319 , whereas in lower panel , the plots are fitted by a curve , @xmath320 , with the choice of parameters , @xmath321 , @xmath322,title="fig:",width=8 ] ( @xmath275 ) plotted ( red points with filling to the horizontal axis ) at @xmath315 for @xmath316 ( upper panel ) and @xmath317 ( lower panel ) . @xmath40 , @xmath318 . in the upper panel , the blue line corresponds to the `` theoretical '' value , @xmath319 , whereas in lower panel , the plots are fitted by a curve , @xmath320 , with the choice of parameters , @xmath321 , @xmath322,title="fig:",width=8 ] ( its magnitude , @xmath239 ) plotted as a function @xmath184 for @xmath323 ( @xmath40 ) . at @xmath324 there are two possible solutions for @xmath158 : @xmath325 and @xmath326 , the larger value of which determines large-@xmath152 behavior of @xmath327 . merger with bulk occurs when @xmath328 , i.e. , at @xmath329 close to the zone boundary ( @xmath330 ) , reentrance of edge solution occurs ( see fig . [ reent_k ] for details).,width=8 ] for @xmath331 ( @xmath40 ) . on the reference line at @xmath332 , @xmath333 ( @xmath334 , @xmath335 = \pm 0.133499 ... $]).,width=8 ] we have extensively studied such characteristic behaviors of the edge wave function in numerical experiments . in fig . [ rho_j ] , results of such analyses are shown . for the choice of parameters such that @xmath336 , @xmath40 and two different values of @xmath107 : @xmath337 and @xmath338 for comparison . at this value of @xmath336 , we first verified that the wave function @xmath195 takes indeed the form of eq . ( [ evec_edge ] ) , with @xmath286 approximately given by , @xmath339 as in the straight edge case , for @xmath184 corresponding to an edge mode , i.e. , for such a state that are localized in the vicinity of either of the two boundaries , this value of @xmath286 is common practically to all @xmath152 in the strip , as far as a finite amplitude exists . for bulk states which are well extended into the interior of the sample , the wave function takes no longer the form of eq . ( [ evec_edge ] ) , but keeps still a characteristic form as eq . ( [ evec_gen ] ) . in the two panels of fig . [ rho_j ] , @xmath340 is plotted as a function of @xmath152 for two different values of @xmath107 . at @xmath337 , @xmath340 saturates at rows away enough from the boundary at @xmath201 ( but not too close to the other edge , either ) , to a value close to @xmath319 ( blue line ) , a value which is later `` derived '' ( see fig . [ rho_th_d025 ] ) . let us assume , ) to a future publication . instead , we take it here as a plausible hypothesis fully justified by numerical experiments . ] as in the straight edge case , that the wave function in the zigzag edge geometry takes the following form : @xmath341 u_+ \nonumber \\ + \left [ c_{-1 } \rho_1^{-j } + c_{-2 } \rho_2^{-j } \right ] u_- , \label{sol_zig_gen}\ ] ] where the eigenspinors @xmath283 are always given by eq . ( [ u+- ] ) , the latter found analytically in the limit @xmath342 . we have extensively verified the validity of this hypothesis in numerical experiments . the coefficients @xmath197 are susceptible of system s geometry . here , in a strip geometry , they satisfy , @xmath343 for one edge mode , and @xmath344 for the other . under this hypothesis , such a behavior as seen in the upper panel of fig . [ rho_j ] ( @xmath337 ) is interpreted as a consequence of two `` real solutions '' for @xmath158 , which are also _ both _ smaller than unity ( cf . [ rho_th_d025 ] ) . on the other hand , at @xmath338 ( in the lower panel of fig . [ rho_j ] ) @xmath340 shows an oscillatory behavior . this implies , with the same hypothesis as above , i.e. , the wave function , @xmath195 , given as in eq . ( [ sol_zig_gen ] ) , with the choice of coefficients as eqs . ( [ cs_2 ] ) , a pair of complex solutions for @xmath158 . indeed , the plots at @xmath338 are nicely fitted by a curve , of the form , @xmath345,\ ] ] with the choice of parameters , @xmath334 and @xmath346 , which will be also ( a posteriori ) justified ( see fig . [ rho_th_d030 ] ) . ( @xmath347 ) calculated in a strip geometry ( here , @xmath318 ) is plotted against the theoretical curve for @xmath348 ( shown in black ) given in eq . ( [ a+- ] ) up to @xmath349 for the lowest - energy eigenmode at different values of @xmath350 ( @xmath40 fixed ) , corresponding , respectively , to the colors : red , green , blue , cyan and magenta . reference lines indicate the regime of @xmath184 in which the edge modes disappear at the given value of @xmath107.,width=8 ] comparing these two contrasting cases , notice that the coefficients @xmath286 , estimated to be such as eq . ( [ a_pm_k005 ] ) , are common to the two cases , i.e. , independent of @xmath107 . one can indeed verify ( by changing the parameter @xmath107 in numerical experiments ) that the eigenspinors @xmath283 remains always the same , as far as the state describes an edge mode ( see fig . [ ak ] ) ; only @xmath340 changes as a function of @xmath107 . in fig . [ ak ] , @xmath312 ( @xmath351 ) is plotted for the lowest - energy eigenmode @xmath161 ( in the upper band ) at different values of @xmath107 . one can see that in the range of @xmath184 at which @xmath161 is expected to represent an edge mode the plotted points fall roughly on the theoretical curve for @xmath348 cf . ( [ a+- ] ) apart from a small disagreement close to the zone boundary ( @xmath258 ) . this is indeed a key discovery allowing us to proceed to the next step , of extrapolating the earlier exact solution at @xmath86 to an arbitrary value of @xmath107 . let us reformulate the recipe for constructing the _ exact _ edge wave function and simultaneously its spectrum in the zigzag edge geometry , which has already been briefly outlined in the introduction ( recall also fig . [ zig_concept ] ) . 1 . we have seen in the previous subsection that the edge wave function @xmath254 in the zigzag edge geometry always takes , as far as it describes a _ localized edge mode _ , the form of eq . [ evec_edge ] with a parameter @xmath286 depending only on @xmath184 ( and @xmath311 , @xmath3 ) . all our numerical data agree with the hypothesis that @xmath286 , consequently the reduced two - component eigenvector @xmath283 , is independent of @xmath2 . 2 . on the other hand , we know that the problem can be solved exactly at @xmath352 . we have seen , in particular , that the wave function @xmath254 can be constructed from the same set of spinors @xmath283 with a choice of parameters @xmath286 given _ analytically _ as a function of @xmath184 in eq . ( [ a+- ] ) . taking also into account the fact that the edge modes , gapless at @xmath267 and characterizing the topological insulator , evolves continuously to the completely flat edge mode at @xmath352 , one can deduce , from these two observations , that the solution of the eigenvalue equation for @xmath187 , i.e. , eq . ( [ reduced ] ) for an _ arbitrary _ @xmath2 should be given , indeed , by @xmath283 , defined as in eq . ( [ u+- ] ) , with the parameter @xmath353 obtained analytically in the limit : @xmath33 ( recall fig . thus , for a general value of @xmath2 , only @xmath158 and @xmath148 are unknown ( recall that the edge spectrum is no longer flat for a general @xmath2 ) . but , clearly , they are solutions of @xmath260 \left [ \begin{array}{c } \alpha ( 1- i)\\ 1 \end{array } \right ] = e \left [ \begin{array}{c } \alpha ( 1- i)\\ 1 \end{array } \right ] , \label{mao - chan}\ ] ] where the @xmath127 matrix @xmath21 is given explicitly as eq . ( [ gamma_zig_mat ] ) . ) and ( [ gamma_zig_mat ] ) , notice that @xmath354 factors out . so all the coefficients become real . ] to find _ exactly _ the value of @xmath158 and @xmath148 , one has only to solve this set of equations , and at the end of the calculation , substitute the analytic expression for @xmath77 , i.e. , @xmath286 given in eq . ( [ a+- ] ) obtained in the limit of @xmath33 . clearly , eqs . ( [ mao - chan ] ) are a set of coupled equations , linear in @xmath148 and quadratic in @xmath158 . we expect , therefore , two sets of solutions for @xmath355 , which are given as a function of @xmath77 . to each of these two sets of solutions , one substitutes either @xmath356 or @xmath357 . there exist , therefore , four sets of solutions , in general . unfortunately , the analytic formula for these four sets of solutions are too lengthy to be shown here . instead , we plotted these four solutions in fig . [ 4sols ] , for different values of @xmath8 . plotted as a function of @xmath347 at @xmath40 and @xmath337 ( as in fig . [ rho_j ] , upper panel ) , but for @xmath358 ( upper panel ) and @xmath359 ( lower panel ) . in the upper panel , the blue line corresponds to the theoretical value , @xmath360 whereas , in the lower panel the plots ( in red ) are fitted by the curve , @xmath320 , with the choice of parameters , @xmath361 , @xmath362.,title="fig:",width=8 ] plotted as a function of @xmath347 at @xmath40 and @xmath337 ( as in fig . [ rho_j ] , upper panel ) , but for @xmath358 ( upper panel ) and @xmath359 ( lower panel ) . in the upper panel , the blue line corresponds to the theoretical value , @xmath360 whereas , in the lower panel the plots ( in red ) are fitted by the curve , @xmath320 , with the choice of parameters , @xmath361 , @xmath362.,title="fig:",width=8 ] plotted as a function @xmath184 at @xmath363 ( again , as in fig . [ rho_j ] , upper panel ) . enlarged picture for @xmath184 close to the zone boundary : @xmath364 the reentrance of edge solution occurs at @xmath365 , whereas two real solutions for @xmath158 is possible when @xmath366 two reference lines are at @xmath358 and at @xmath367 , on which @xmath340 was plotted in fig . [ reent_j ] . the value of @xmath158 on these lines are , @xmath368 and @xmath369 on @xmath370 ( case of real solutions ) , whereas @xmath371 on @xmath372.,width=8 ] ) . this is an enlarged image of the spectrum in the zigzag edge geometry as shown in fig . [ spec_zig ] . here , the parameters are chosen such that @xmath373 , @xmath40 , @xmath212 . two reference lines at @xmath374 and at @xmath375 introduce three different momentum regions : i ) @xmath376 $ ] , ii ) @xmath377 $ ] and iii ) @xmath378 $ ] , corresponding , respectively , to i ) the ordinary edge , ii ) the bulk and iii ) the reentrant regimes.,width=8 ] _ reentrance _ of the edge mode is another characteristic feature of the edge mode of zigzag geometry , and occurs close to the zone boundary , @xmath379 , when @xmath107 is not too large : @xmath380 . very remarkably , the spectrum looks completely `` innocent '' when this occurs , i.e. , the edge mode , say , the lowest energy ( @xmath381 ) mode in the upper band looks almost completely degenerate with the bottom of the ( bulk ) spectrum ( @xmath382 ) , in this regime of @xmath184 ( see fig . [ spec_zig_enl ] ) . existence of an edge mode of such specific character is , on the other hand , nothing exceptional in the zigzag edge geometry . at a value of @xmath2 , e.g. , @xmath383 or @xmath384 as in fig . [ rho_j ] , such reentrant edge modes are indeed existent . if one focuses on the wave function of , say , the lowest energy mode in the upper band , after touching the lower band at @xmath267 , it continues to be spatially localized when @xmath184 is small enough , but as the spectrum merges with the bulk continuum , the wave function also starts to penetrate into the bulk . however , close to the zone boundary , it starts to be localized again . this is what we call the reentrance of edge modes . figs . [ reent_j ] and [ reent_k ] highlight the behavior of such reentrant edge modes , naturally in a different regime of @xmath184 from , say , fig . [ rho_j ] . [ reent_j ] shows the behavior of @xmath340 at @xmath385 and @xmath367 for @xmath363 . at this value of @xmath184 , the wave function @xmath195 takes always the form of eq . ( [ evec_edge ] ) , but with a different set of parameters for @xmath286 from the case of fig . [ rho_j ] , upper panel , since it still depends on @xmath184 . the two plots for @xmath340 in fig . [ reent_j ] show two typical behaviors of the edge wave function , i.e. , one corresponding to real and the other to complex solutions for @xmath158 . as we have extensively studied in the case of ordinary edge modes ( appearing at @xmath386 ) , the two contrasting behaviors of @xmath340 as a function @xmath152 ( in fig . [ rho_j ] ) are naturally understood by referring to the theoretical curve of @xmath158 as a function of @xmath184 , e.g. , such as the one shown in fig . [ rho_th_d025 ] . what is rather remarkable here , in the case of fig . [ reent_j ] , is that this crossover between real and complex solutions occurs within a tiny change of @xmath184 , i.e. , from @xmath385 in the upper panel to @xmath367 in the lower panel . this drastic change is , however , quite reasonable from the viewpoint of fig . [ reent_k ] . the upper panel of fig . [ reent_j ] shows a monotonic decay , which converges asymptotically to a single exponential decay . this is consistent with the behavior of theoretical curve for @xmath158 as a function of @xmath184 in fig . [ reent_k ] . the latter implies two real solutions for @xmath158 at @xmath385 : @xmath368 and @xmath369 the latter coincides with the value of @xmath340 in fig . [ reent_j ] at which it saturates . at @xmath367 , on the other hand , the plots for @xmath340 are nicely fitted by the curve , @xmath320 , with the choice of parameters , @xmath387 and @xmath388 ( see the lower panel of fig . [ reent_j ] ) . this is a clear fingerprint that the reentrant edge mode at this value of @xmath184 corresponds to a pair of complex solutions for @xmath158 . is plotted as a function of @xmath184 at @xmath389 , @xmath40 . different curves correspond to different size ( width ) of the system : @xmath318 ( blue ) , 200 ( green ) and 300 ( red ) . the two reference lines are placed at @xmath390 and @xmath391 ( @xmath392 $ ] corresponds to the bulk regime ) . the plots reveal an extremely small but still a finite binding energy of the reentrant edge modes.,width=8 ] does the reentrant edge mode really have zero binding energy ? in order to address this question , we ( re)plotted the energy spectrum ( @xmath393 , to be precise ) but in an enlarged scale roughly by one thousand times in fig . [ en_bind ] . first , for an `` ordinary '' edge state , occurring at @xmath394 the value @xmath393 is much above the threshold at this scale . @xmath393 takes a value of order @xmath395 for such ordinary edge state . as for the reentrant edge mode , fig . [ en_bind ] reveals that it has indeed an extremely small but still a finite binding energy . notice different behaviors of @xmath393 as a function of @xmath32 in the bulk and reentrant regions of @xmath184 . the former ( the latter ) corresponds to @xmath396 $ ] ( @xmath378 $ ] ) . in the bulk region @xmath393 is expected to vanish in the thermodynamic limit . [ en_bind ] shows indeed that the binding energy of reentrant edge mode , @xmath397 , is thousand times smaller than that of the ordinary edge state . this implies the appearance of an extremely small energy scale which was not existing in the original hamiltonian ( cf . kondo effect ) . the reentrance of edge mode is indeed a unique feature , in its contrasting properties in real and momentum space , and in the appearance of an extremely small energy scale . we have highlighted in this paper various unique properties of helical edge modes in @xmath0 topological insulator . we have extensively investigated a lattice version of the bhz model , under different edge geometries . one of the specific characters of bhz model is that the spin hall conductance in the bulk changes its sign in the middle of topological phase ( at @xmath23 ) , i.e. , @xmath398 , respectively , for @xmath213 and for @xmath399 , though both represent a non - trivial value . from the viewpoint of bulk - edge correspondence , this information should be also encoded in the edge theory . we have seen that the change of @xmath98 manifests in a very different way in the @xmath7- ( straight ) and @xmath11- ( zigzag ) edge geometries . in the @xmath7-edge case , the edge spectrum changes its global structure in the two parameter regimes , i.e. , the main location of the mode moves from the zone center for @xmath9 , to the zone boundary for @xmath10 . as a result , the group velocity at the intersection with fermi level reverses its sign , leading to change of the sign in landauer conductance at the edge . in the @xmath11-edge case , on the other hand , the edge spectrum is symmetric w.r.t . @xmath23 , i.e. , neither change of the position of gap closing , nor the reversal of group velocity at @xmath23 . the change of @xmath98 is here encoded in the swapping of left- and right- going edge modes of the same spin . much of our focuses has been on the analysis of the zigzag or @xmath11-edge geometry , the latter showing , as a consequence of specific way in which the bulk topological structure is projected onto the 1d edge , a number of unique features , such as the completely flat edge spectrum at @xmath23 , and the reentrance of edge modes . we have also shown , here in a half - empirical way , that the exact edge wave function for zigzag edge geometry can be constructed , by extrapolating the solution at @xmath23 . the reentrant edge mode , though sharing much of its characteristics with the usual edge mode in real space , introduces a new extremely small energy scale which was absent in the original bhz model . ki , ay and ah have been much benefited from useful discussions with jun goryo on the bulk / edge correspondence . ki also acknowledges christoph brne , hartmut buhmann and laurence molenkamp for their detailed explanation of the experimental situations in stcm ( kyoto ) , ngss-14 ( sendai ) and qhsyst10 ( dresden ) conferences . ki and ay are supported by kakenhi ( ki : grant - in - aid for young scientists b-19740189 , ay : no . 08j56061 of mext , japan ) . let us first recall that the eigenvector @xmath400 , which has appeared in eq . ( [ dk_pm ] ) , is a standard su(2 ) spinor , here chosen to be single - valued . an eigenvector , corresponding to a positive energy eigenvalue @xmath401 , is @xmath402 with @xmath53 satisfying @xmath49 , and represented as , @xmath403 \\ \nonumber & = & { 1\over\sqrt{2d(d - d_z ) } } \left [ \begin{array}{c } d_x - i d_y \\ d - d_z \end{array } \right ] . \label{evec}\end{aligned}\ ] ] @xmath404 , @xmath405 are polar coordinates in @xmath51-space , satisfying the relations such as , @xmath406 in order to find an edge solution , we focus on a solution of the form , @xcite @xmath407 say , in a semi - infinite plane : @xmath408 . the spatial dependence in the @xmath136-direction can be taken into account by applying pierls substitution : @xmath409 to @xmath75 s in @xmath48 . the eigenenergy of such a solution is obtained by a simple replacement : @xmath410 in eq . ( [ wine ] ) , i.e. , @xmath411 this can be regarded as a quadratic equation w.r.t . its two solutions are , @xmath413 for a given set of @xmath74 and @xmath148 , there are two possible values for @xmath412 , or equivalently , four possible values for @xmath414 . of course , in a semi - infinite plane , say , @xmath408 the edge solution of the form , eq . ( [ evec_kappa ] ) should decay as @xmath415 , so only two of such solutions are relevant . we expect that the edge spectrum behaves as @xmath416 in the limit of@xmath147 . let us parametrize the two solutions in this limit as @xmath417 ^ 2\equiv p\pm q , \label{kappa0}\ ] ] where @xmath418 when @xmath419 , @xmath420 and @xmath421 is real . since @xmath422 as far as @xmath423 is real , eq . ( [ sol_kappa ] ) represents two positive solutions for @xmath412 , i.e. , the wave function represented by eq . ( [ evec_kappa ] ) shows simple exponential damping . on contrary , we expect a pure imaginary solution for @xmath414 for an extended state in the bulk . on the other hand , when @xmath424 , @xmath425 is always positive but @xmath421 becomes purely imaginary . thus two solutions for @xmath414 become complex numbers conjugate to each other . in this case , the wave function represented by eq . ( [ evec_kappa ] ) shows damped oscillation . @xmath427 . for a given value of @xmath74 and @xmath401 , we thus have identified two solutions characterized by different values of @xmath414 . in order to construct a general solution in the presence of a boundary , we need to take a linear combination of these two solutions , i.e. , @xmath428 e^{\kappa_+ y } + c_- \left [ \begin{array}{c } u_-\\ v_- \end{array } \right ] e^{\kappa_- y},\end{aligned}\ ] ] where @xmath283 , @xmath429 are short - hand notations for @xmath430 and @xmath431 . we now fix the boundary condition at @xmath432 , which we choose to be , @xmath433 \left [ \begin{array}{c } c_+\\ c_- \end{array } \right]= \left [ \begin{array}{c } 0\\ 0 \end{array } \right ] , \label{bc}\ ] ] which implies the following secular equation : @xmath434=0\ ] ] this leads to , @xmath435 let us check whether this solution contains the edge modes . we expect that the edge spectrum bahaves , as @xmath147 , @xmath416 . recall that in this limit , eq . ( [ sol_kappa ] ) reduces to eqs . ( [ kappa0 ] ) and ( [ ab ] ) . ( [ req_bc ] ) is also simplified in this limit , as @xmath436 focusing on the case , @xmath437 and @xmath438 , and using the parameterization in eqs . ( [ kappa0 ] ) and ( [ ab ] ) , one can readily verify , @xmath439 thus eq . ( [ req_bc0 ] ) is safely satisfied . how about the first order corrections ? i.e. , contributions of order @xmath440 to the energy spectrum , @xmath149 . first note that there is no @xmath440-correction to @xmath441 . one can , therefore , safely replace , at this order , @xmath441 s in eq . ( [ req_bc ] ) with their values at @xmath147 , @xmath416 , i.e. , @xmath442 . \label{req_bc1}\ ] ] we have already seen that the first two terms cancel , whereas @xmath443 ^ 2= 2a+2\sqrt{p^2-q^2}={a^2\over b^2}.\ ] ] thus , the edge spectrum in the continuum limit is determined to be , @xmath444 remarkably , the slope of the edge spectrum depends only on a single parameter , @xmath311 . an interesting question is to what extent this conclusion is general ? if one calculates the edge spectrum , using a tight - binding model , generally the results depend on the way edges of the sample are introduced with respect to the lattice . in the case of zigzag edge , in particular , apparently the edge spectrum does not converge to eq . ( [ edge_cont ] ) even in the long - wave - length limit : @xmath147 .
recent progress in creating artificial magnetic heterostructures @xcite gives an impetus to the theory of phase transitions in multilayers . while the related experimental and theoretical efforts have been so far mostly concentrated at microscopic effects , this paper addresses the problem of universal features associated with large scale ordering in superlattices . suppose the superlattice is built of atomically thick layers of two magnets , which taken separately both order through second order phase transitions at two respective bulk critical points @xmath5 . suppose also for simplicity that both components are _ ferro_-magnets and that the critical behavior at their bulk curie points is similar , i. e. both components belong to the same universality class . the question then is : at what temperature @xmath6 will long range ferromagnetic order appear in the superlattice and what critical behavior will be observed at that temperature ? one can reasonably expect that in the limit of thick layers the answers to these questions will be largely universal , depending only on the bulk critical properties of the two components and the geometry of the superlattice ; these universal aspects of the problem represent the subject of this paper . we will consider the simplest superlattice geometry ( fig . 1 ) constructed of two elementary building blocks : slabs , or layers , of two ferromagnets , 1 and 2 , of finite thickness @xmath7 and @xmath8 , respectively . the slabs are stacked periodically in the @xmath9-direction , so that @xmath10 is the period . the system is homogeneous in the remaining @xmath11 dimensions . while the dimensionality @xmath2 ( fig . 1a ) is naturally the most interesting one in view of experimental applications , the planar , @xmath12 , superlattice geometry can be quite conceivably realized by cutting a thin film out of a three - dimensional multilayered sample ( fig . 1b ) or by deposition of a magnetic film on a substrate , which is itself cut out of a superlattice . extension of our consideration to @xmath12 is important because of analytical tractability of two - dimensional models . being interested in the long wave length aspects of the problem , we will consider the limit of atomically thick slabs : @xmath13 , where @xmath14 are the thicknesses of elementary , molecular layers of the two components , respectively ; correspondingly , the interfaces separating two subsequents slabs do not have to be atomically smooth . naturally , universality extends applicability of the results obtained below beyond this particular magnetic model ; among other interesting applications are a superfluid in a periodic matrix and a superfluid film deposited on a periodic substrate . universal critical phenomena related to three different geometrical elements present in a superlattice have been extensively studied theoretically and , to a variable extent , experimentally . first , at a _ single _ interface separating _ half - infinite _ near - critical , say @xmath15 and disordered , @xmath16 , magnets , one expects _ ordinary surface critical _ phenomena to take place at the near - critical side of the interface . it has been established theoretically @xcite that the critical exponents and amplitudes characterizing surface values of various observables are quite different from those in the bulk , as well as from the @xmath17-dimensional ones . however , experimental confirmation of those theoretical findings has turned out to be difficult . note that near @xmath18 , where the other half of the single interface model goes through the criticality , the first one is already ordered and imposes a magnetic field acting at the surface of the second component . therefore , one expects _ normal surface critical _ behavior @xcite to take place at the interface near @xmath18 . second , near the bulk critical temperature @xmath19 of the @xmath20-th , @xmath21 , component a thick , @xmath22 , _ single _ slab of that component exhibits a dimensional crossover from the critical behavior characteristic of bulk , @xmath0-dimensional , samples to the @xmath11-dimensional critical behavior , as observed in thin films of the same material @xcite . we will refer to this crossover as _ bulk - to - film _ below ; it has been observed experimentally @xcite . finally , in a system of _ thin _ layers connected to each other by very interlayer bonds , near the critical temperature @xmath23 of a single layer one expects a crossover from @xmath17-dimensional to the bulk , @xmath0-dimensional behavior , which we will call _ film - to - bulk _ crossover below . this type of dimensional crossover has been well experimentally documented in a variety of magnetic @xcite and non - magnetic layered systems @xcite . one source of motivation for studying critical phenomena in multilayers is that the significantly increased surface - to - volume ratio may improve chances of observing features characteristic of the @xmath12-dimensional and , especially , the elusive @xmath2-surface critical behavior with respect to experiments involving single thin film or semi - bulk samples . however , finite separation between interfaces in a superlattice leads to correlations between fluctuations in different layers and at different interfaces . one therefore expects surface and two - dimensional scaling to be limited to a certain range of length scales , while eventually cooperative phenomena involving fluctuations on scales larger than the period of the superlattice will dominate the criticality . a crucial insight comes from the exact solution available for a ferromagnetic @xmath12-ising superlattice , as modelled by a planar ising model composed of alternating strips of two components @xcite . recent analysis @xcite demonstrated that the anomalous changes in the critical amplitudes and other interesting features of that model can be simply explained by _ reentrant dimensional crossover _ : as the temperature approaches the critical temperature of the superlattice the long wave length properties of the system are successively dominated by fluctuations characteristic of the the @xmath12-ising , then @xmath24-ising , and then again @xmath12-ising critical behavior . here we generalize and explore this picture by constructing and analyzing a renormalization group ( rg ) flow which provides a scaling description of criticality in experimentally interesting three dimensional superlattices as well as in the heizenberg ( isotropic ) and @xmath3 ( easy plane ) universality classes . in the proposed scenario as the scaled temperature @xmath25 , or @xmath26 , decreases , the critical behavior of the superlattice displays the features of the three simpler systems listed above ( fig . 2 ) : first , as @xmath27 is approached , the original superlattice ( ( i ) in fig . 2 ) can be approximated by sequence of thick layers of the first component weakly coupled through the 2-layers ( ( ii ) in fig . 2 ) . at larger values of @xmath28 this coupling can be neglected and the layers display regular bulk critical behavior with the ordinary surface critical behavior at the interfaces . then , as the correlation length in the first component becomes comparable to the thickness @xmath7 , each slab of the first component exhibits direct bulk - to - film , @xmath29-dimensional crossover . at still smaller values of @xmath28 , the thickness of these slabs is irrelevant and the system becomes equivalent to the third paradigm mentioned above : a weakly coupled layered system ( ( iii ) in fig . the reverse , film - to - bulk crossover finally takes the system back to the uniform , bulk @xmath0-dimensional behavior , characterized however by high degree of anisotropy ( ( iv ) in fig . 2 ) . this reentrant crossover behavior leads to dramatic changes in the critical amplitudes , as shown in fig . 3 for the @xmath2-ising class : the critical exponent characterizing divergence of a physical quantity , for which the specific heat per unit volume @xmath30 has been chosen in fig . 3 , changes two times , as @xmath28 goes to zero , between the values @xmath31 and @xmath32 characteristic of the @xmath2 and @xmath33-dimensional critical behavior , respectively . the relative widths @xmath34 ( subscript @xmath35 denotes _ direct _ , bulk - to - film crossover here ) and @xmath36 ( subscript @xmath37 stands for the _ reverse _ , film - to - bulk crossover ) of the temperature domains where the planar ising and the reentrant bulk critical behaviors are observed will be calculated below ; they provide convenient scaling combinations in terms of which a proper description of criticality in superlattices is achieved . two further points have to be added to this preview of the paper . first , a crucial element of our qualitative picture is that different layers of the first component remain effectively decoupled while the bulk - to - film crossover ( from ( ii ) to ( iii ) in fig . 2 ) happens _ within _ them . this turns out to be guarantied by the fact that the subsequent 1-slabs are coupled through the _ surface _ spins whose correlations are much weaker than of those in the bulk . thus the surface critical behavior , which is not directly seen in temperature dependencies of bulk quantities , like that shown in fig . 3 , appears crucial to the nature of criticality in the superlattice ; we will show that the values of surface critical exponents can thus be extracted from analysis of the dependence of the reentrant width @xmath36 on the thickness of the 2-slabs , @xmath8 . second , in the case of the @xmath2 ising superlattice the @xmath11-dimensional model , describing a single film of the first component , has a regular finite temperature phase transition . consequently , in the thick layer limit the critical temperature @xmath6 of the superlattice is only slightly depressed from @xmath27 and the critical fluctuations are essentially localized within the 1-layers . however , in all other cases , i.e. @xmath2 heizenberg and @xmath3-models and all models in @xmath12 , the corresponding @xmath17-dimensional universality classes lack true ferromagnetic long range order at finite temperatures . the interlayer coupling through the layers of the weaker , second component thus are crucial to existence of long range order . as a result , the critical temperature in these models is strongly shifted towards @xmath18 and the critical domain , in which large scale fluctuations are active , is extended to the temperature range of the order of @xmath38 . finally , a few remarks about the method of this paper . we focus here on the @xmath39-component vector spin systems with short range interactions . as usual , ising , @xmath3 , and heizenberg universality classes refer to the @xmath40 , or easy axis , @xmath41 , or easy plane , and @xmath42 , or isotropic , magnets respectively . the most important omission of this model , apart from restriction to the simplest type of the order parameter , is the absence of long - range dipolar forces ; it seems however natural to make the first step within the simpler realm of local models . the actual method employed here is to combine the well known linear rg flows at the fixed points representing the three paradigms discussed above ( fig . 2 ) , into the simplest _ global _ rg flow consistent with this linear behavior ( fig . the resulting scaling theory then passes the test of comparison to the exact results available for the planar ising model . the outline of the paper is the following : in the next section we start by taking the conceptually simplest case of the three - dimensional ising superlattice . the renormalization group flow will be proposed and scaling forms derived . in section iii we consider the planar ising case which requires extension of the previous consideration to the case of the zero - temperature critical point describing the @xmath43 dimensional ising model . the results will then be shown to agree with the exact forms available for this case @xcite . the extension of the method developed in section iii will be used in the following section , where the heizenberg and the @xmath3 universality classes are considered . the results are summarized in section v. we start now by focusing on the @xmath2-ising universality class . its most important feature is that a single thick slab of the first component is fully capable of ordering at a finite temperature @xmath44 , which continuously approaches the bulk transition temperature @xmath27 , as the thickness @xmath45 . we thus expect that the criticality in a three - dimensional ising superlattice happens close to @xmath27 and that the relevant fluctuations are localized within the 1-slabs . much of what will be said below will apply however to other systems , so we will keep using general @xmath0 and @xmath11 , as well as the general standard notation for the critical exponents in this section . the renormalization group procedure which we use here has been extensively discussed in @xcite for the half - infinite models bounded by one free surface , and applied to the superlattice geometry in @xcite . it is based on subsequent reduction , @xmath46 , of the upper cut - off @xmath47 imposed on the @xmath17-dimensional wave numbers @xmath48 characterizing the spatial variation of fluctuations in the directions parallel to the layers . since the superlattice geometry is uniform in this plane , expansion in plane waves @xmath49 can be used as a basis for a perturbative rg of the type discussed in @xcite . the details of the procedure will not be important to us here : all we need is the _ existence _ of an exact rg transformation ; then according to the general principles @xcite the linearized rg flow in the vicinity of the fixed points should be independent of the specific implementation . note that there is no need to do anything about the cut - off in the @xmath9-direction ( in fact one does not need such cut - off at all ) : elimination of fast modes in the directions parallel to the layers automatically _ induces _ coarse - graining in the @xmath9-direction @xcite . we will now follow the rg flow probing fluctuations on larger and larger length scales parallel to the layers . the two crucial length scales , to which the current ( running ) length scale has to be compared , are the _ bulk _ correlation lengths @xmath50 and @xmath51 , which can be ( at least in principle ) measured at the given temperature @xmath52 in independent bulk samples of the two components . near @xmath27 fluctuations in the second component are confined to scales smaller than the bulk correlation length @xmath53 , which remains finite while @xmath54 diverges . decreasing the upper momentum cutoff to @xmath55 essentially eliminates fluctuations in the 2-layers . on larger scales the 2-layers are adequately described by the ornstein - zernike spin density functional corresponding to the gaussian rg fixed point describing the low - temperature phase of the second component . as usual , gaussian degrees of freedom can be integrated out , the integration being equivalent to minimization of the ornstein - zernike functional . such minimization is performed in the appendix . it results in an effective interaction @xcite @xmath56 between the top surface spins @xmath57 of the @xmath39-th layer and the bottom surface spins @xmath58 of the @xmath59-th layer of the first component . at this point the thickness @xmath8 gets absorbed into the bare coupling constant @xmath60 . a major simplification has occurred after this initial crossover : the rg flow has taken the original superlattice system to the first fixed point , @xmath61 , equivalent to a sample of the first component containing a periodic sequence of defect ( hyper- ) planes , characterized by very weak vertical bonds @xmath62 ( ( ii ) in fig . 2 ) . note that all physics of the second component , as well as all details of the microscopic implementation of the i interface between the subsequent layers , have been absorbed into two parameters : the directly measurable correlation length @xmath63 and the nonuniversal amplitude @xmath64 . a necessary condition for the reentrant dimensional crossover , described in the introduction , is the smallness of the initial , bare value of the coupling in the temperature range around @xmath65 : @xmath66\ll k_bt . \label{condition1}\ ] ] under this condition , we can consider the coupling ( [ int ] ) as a weak perturbation with respect to the reference system consisting of _ noninteracting _ layers of the first component . further , for the values of the rg parameter @xmath67 such that @xmath68 , typical fluctuations are correlated only over lengths much shorter than @xmath7 , implying that the top and the bottom surfaces of each 1-layer are uncoupled . provided the correlation length @xmath54 of the first component has not been encountered yet , the system appears to be at the critical fixed point @xmath69 describing independent bulk critical behavior within the 1-layers coupled to the ordinary surface critical behavior at the noninteracting interfaces . note that this scenario implies another condition , @xmath70 both conditions , ( [ condition1],[condition2 ] ) , require the layers to be thick . in the case of two different materials , @xmath71 , so that @xmath72 and both inequalities are satisfied as soon as @xmath73 . the situation becomes less trivial in the case of a superlattice created by a weak periodic modulation of the properties of an originally uniform sample , then one has to be sure that the period of modulation is large enough to offset the smallness of the difference between the bulk curie temperatures implying a relatively large @xmath74 , where @xmath75 is the standard correlation length exponent in @xmath0 dimensions . the crossover is now described as an rg - flow between three critical fixed points ( fig . 4 ) : @xmath69 describing a sequence of uncorrelated _ thick _ layers of the first component ( ( ii ) in fig . 2 ) , with extensive properties dominated by the @xmath0-dimensional bulk critical behavior taking place inside the layers , while the associated ordinary surface scaling describes the observables localized at the surface . from @xmath76 the system flows to @xmath77 describing the @xmath78-dimensional criticality in a system of uncorrelated _ thin _ films of the first component ( ( iii ) in fig . 2 ) , and , finally , to @xmath79 describing uniform @xmath0-dimensional bulk behavior ( ( iv ) in fig . the flow between the fixed points is driven by two scaling fields ( cf . 4 ) , the inverse thickness of the 1-layers , @xmath80 , playing the role of a long wave length ( infrared ) cutoff in the @xmath9-direction ( cf . @xcite ) , and the interlayer coupling strength @xmath62 . yet another one , scaled temperature field @xmath26 controls the departure from the critical manifold , @xmath81 , containing the flow attracted by @xmath79 , towards the massive , high- and low - temperature fixed points . instead of the scaled temperature field @xmath82 characterizing the bulk criticality of the first component , one can use @xmath25 defined relative to the _ observed _ critical temperature @xmath83 of the superlattice . as we will see below , the scaling forms look simpler when expressed via @xmath28 , but lack any information about the _ shift _ in the critical temperature with respect to @xmath27 . in the rg approach the scaling fields become functions of the logarithmic length scale @xmath67 . while the explicit calculation of the crossover scaling functions requires application of non - perturbative methods such as the one used in @xcite , the leading singularities are determined by the rg flow in close vicinities of the fixed points . specifically , at @xmath69 the linearized rg flow equations are @xmath84 here @xmath75 is the standard correlation length exponent of the bulk @xmath0-dimensional universality class . the second equation ( [ ld ] ) reflects decrease in @xmath7 expressed in units of the running inverse length scale @xmath47 ; the equality of the rg eigenvalue in ( [ ld ] ) to one reflects asymptotic isotropy of the bulk critical behavior described by @xmath69 . the rg eigenvalue for @xmath62 is read from ( 1 ) . because the spins entering ( 1 ) are located at the surfaces , the scaling dimension of the spin density @xmath85 determining the rg eigenvalue of @xmath62 is that characteristic of the _ surface _ spin density . correspondingly @xmath86 is the exponent characterizing the susceptibility of the _ surface _ spins to a perturbation by a _ surface _ field @xcite . the crucial point now is that since the surface spins are correlated more weakly than the bulk ones , the surface susceptibility usually does not diverge ( see the estimates of surface critical exponents in @xcite ) : @xmath87 . consequently @xmath62 is irrelevant or marginal at the first encounter with @xmath76 . thus starting with a small coupling @xmath62 , we are guarantied that it remains small until the system arrives at @xmath77 . note that this picture implies a seemingly paradoxical prediction : a single layer of weak bonds cutting through a bulk sample effectively decouples the two halves , with two _ independent _ ordinary surface fluctuations developing at the two sides of it . the two other scaling fields , @xmath82 and @xmath80 are relevant at @xmath69 . as @xmath88 is positive definite , it is convenient to exclude @xmath67 from the equations ( [ td],[ld ] ) , rewriting them as @xmath89 the crossover to the @xmath17-dimensional fixed point @xmath77 occurs when @xmath88 decreases to the order of the microscopic length scale @xmath90 . as a result of this first crossover ( indicated by subscript @xmath35 , for `` direct , '' below ) the scaling field @xmath91 becomes irrelevant and disappears from the consideration , while the other two are renormalized to @xmath92 here the temperature rescaling factor @xmath93 conveniently characterizes the first crossover . we will reduce the number of nonuniversal parameters by defining @xmath34 not through @xmath90 , but rather through the critical amplitude @xmath94 characterizing the divergence of the bulk correlation length @xmath95 at the low - temperature side of the bulk critical point @xmath27 : @xmath96 in the rg formalism @xcite the correlation length @xmath54 characterizes the crossover from the critical to noncritical rg fixed points at nonzero values of @xmath82 . therefore @xmath34 determines the width of the temperature domain around @xmath27 in which the dimensional crossover @xmath97 actually occurs : for @xmath98 the @xmath82-field becomes large and drives the system away from the critical manifold before @xmath91 grows large ; as a result the system never reaches @xmath77 . if , on the other hand , @xmath99 the system arrives at @xmath77 with the values of the two relevant fields estimated by @xmath100 . in the absence of @xmath62 ( which is small at this stage ) @xmath101 completely determines the flow . the critical separatrix going into @xmath77 corresponds to a certain initial value @xmath102 . one expects @xmath103 , as finite thickness suppresses ordering . in fact , due to our definition of @xmath34 via the bulk correlation length @xmath54 , the parameter @xmath104 is universal : criticality in a _ free _ film of the first component of thickness @xmath7 happens at a temperature @xmath105 at which the ratio @xmath106 takes a universal value @xmath107 where the numerical estimate has been obtained by series expansion methods @xcite . expanding around the separatrix one obtains @xmath108 where the prime marks exponents related to the @xmath17-dimensional criticality . these equations are to be solved with the initial data given by @xmath109 . as a result of the first crossover the surface spins in ( [ int ] ) are strongly correlated with the rest of the corresponding layer , thus becoming @xmath17-dimensional in nature . correspondingly , the scaling dimension of the spin density @xmath85 in the rg flow equation for @xmath62 is changed from the _ ordinary surface _ value @xmath85 to the @xmath17-dimensional @xmath110 . correspondingly nonpositive @xmath86 is changed to positive @xmath111 , so that @xmath62 is a relevant perturbation at @xmath77 driving the second crossover back to @xmath76 . the following analysis of the reverse crossover ( as indicated by subscript @xmath37 below ) essentially repeats the one performed for @xmath97 above : we divide one of the two linear rg equations by another to obtain ( the coupling strength @xmath62 is positive definite ) @xmath112 a nonuniversal amplitude @xmath113 is defined so that at @xmath114 the @xmath17-dimensional hyperplanes become strongly coupled and the system crosses over to the uniform bulk behavior at @xmath79 . as this happens , @xmath82 is rescaled by the factor @xmath115 , i. e. @xmath116 where ( recall that @xmath87 ) @xmath117 just as previously , for @xmath118 , i. e. @xmath119 , the system flows away from the critical separatrix before the reentrant crossover to @xmath76 can take place . hence @xmath36 measures in units of @xmath120 the width of the temperature domain in which the reentrant scaling can be observed ; in the original units of @xmath82 the width is given then by the product @xmath121 . within the domain @xmath122 the system flows into the close vicinity of @xmath79 . the critical separatrix is again defined by a certain ( this time positive : interlayer bonds enhance ordering ) value of @xmath123 . in fact , as at this point we have no other scale to measure the interlayer coupling amplitude @xmath64 , we can fix the arbitrary constant @xmath124 by setting @xmath125 . having grown to the order of @xmath124 the interlayer coupling amplitude @xmath62 is absorbed into the rescaled value of the temperature field @xmath126 and becomes irrelevant . the reentrant @xmath79 is a regular critical fixed point with temperature ( in the absence of magnetic field ) being the only relevant perturbation measuring the deviation from the critical separatrix ( fig . the evolution of the latter is again described by the rg flow equation ( [ td ] ) which has to be solved with the initial condition @xmath127 having established the principle features of the rg flow we are in a position now to develop a scaling description of the observable quantities . the two temperature scales @xmath34 and @xmath36 defined by ( 2 ) and ( 3 ) , together with the original value of the scaled temperature of the first component , @xmath82 , conveniently parametrize the scaling functions . if the bulk properties of the two components are known , then the only new nonuniversal parameter appearing in our description of the superlattice is the dimensionless ratio @xmath128 , essentially characterizing the strength of coupling between the spins across the interface between the two components ( see the appendix for more precise definition ) . the standard two - scale factor universality @xcite of the bulk critical points is thus extended to what may be called @xmath129-factor universality : all scaling functions of a superlattice are universal apart from the two independent bulk critical amplitudes of the first component plus the amplitude @xmath128 characterizing the interface and the bulk correlation length of the second component @xmath130 . in fact , the two additional parameters enter the description via the initial value of the effective coupling @xmath131 $ ] . thus , if one were not interested in the ( singular ) dependence of the critical behavior of the superlattice on the thickness @xmath8 , only one extra amplitude , @xmath62 , would have to be added to complete the description of critical behavior in a superlattice . however , since for large @xmath8 the amplitude @xmath62 is going to be anomalously small , and since @xmath8 is easily measured , we prefer to split @xmath62 into the nonsingular amplitude @xmath64 and the singular exponential factor @xmath132 , adding the bulk correlation length @xmath130 to the list of empirical parameters . we start by analyzing the shift in the critical temperature . unfolding back the two renormalizations of the temperature field , one obtains from ( [ td],[t1c ] ) @xmath133=t'_c(l_1)+t_{c1}\tau_d\tau_r.\ ] ] the shift consists of a larger , @xmath134 , shift to lower temperatures , slightly corrected by a much smaller , @xmath135 $ ] , shift in the opposite direction . the last expression represents the latter as a shift with respect to @xmath136 $ ] , the critical temperature in a film of the first component of thickness @xmath7 with free boundaries . if this latter temperature is known , then the observed shift towards the higher temperatures , @xmath137 , can be used to estimate the unknown amplitude @xmath128 entering the definition of @xmath36 ( [ taur ] ) . let us consider now an extensive observable , say the specific heat per unit volume @xmath30 characterized by critical exponents @xmath31 and @xmath138 in dimensions @xmath0 and @xmath17 correspondingly . since the surface - to - volume ratio in a superlattice vanishes as @xmath139 , the observed signal at the first stage of the crossover is dominated by the interior of the 1-layers : as far as extensive properties are concerned @xmath69 is equivalent to @xmath79 and the rg flow is indeed reentrant . this flow can be represented by the scaling form @xmath140 where @xmath141 are given in ( [ t1d],[t1r ] ) , and @xmath142 is the bulk critical amplitude of the first component on the low - temperature side of the criticality . the factor @xmath143 represents the amplitude of the contribution to @xmath30 from the interior of the 1-layers at the first stage of the rg flow ( governed by @xmath69 ) below @xmath27 ; the choice of @xmath142 i instead of @xmath144 seems natural in view of @xmath145 . the scaling function @xmath146 is completely _ universal_. it has the following limits : it approaches 1 at @xmath147 and the universal amplitude ratio @xmath148 at @xmath149 ; at @xmath150 it develops a singularity @xmath151 which cancels the one generated by the @xmath152 factor in ( [ a1 ] ) . at @xmath153 one has @xmath154 . the asymptotic behavior of the new scaling function @xmath155 follows the same logic as the one employed above for @xmath146 : it takes finite limits at @xmath156 , develops a singularity , @xmath157 , when @xmath158 . as already mentioned above , because in the ising universality class the critical temperature shifts satisfy @xmath159 a simpler scaling form is achieved in terms of the scaled temperature @xmath160 shifted to the _ observed _ critical temperature @xmath161 . defining @xmath162 one can write @xmath163 where the universal scaling function @xmath164 takes the following limits : @xmath165 this form has a simple graphical interpretation given by the double logarithmic plot of fig . the continuity of @xmath166 requires that both universal amplitude ratios @xmath167 ( remember that the scaling function has been normalized by the bulk amplitude @xmath142 ) are numbers of the order of unity . the last asymptotic expression in ( [ a ] ) shows that while the ultimate divergence @xmath168 is characterized by the bulk exponent @xmath31 , the critical amplitudes @xmath169 are shifted on the logarithmic scale from the bulk amplitudes @xmath170 of the first component by @xmath171 ( see fig.3 ) , i.e. @xmath172+o(1 ) . \label{logcratio}\ ] ] more information about the critical state of a superlattice governed by @xmath79 can be obtained by studying the anisotropy of correlation functions . the basic ratio @xmath173 , of the critical amplitudes of correlation lengths across and along the layers , is given by expressions identical to ( [ cratio ] ) , ( [ logcratio ] ) , where one has to use anisotropic correlation length exponents : @xmath174 at @xmath175 , but @xmath176 at @xmath77 , since at @xmath77 the correlations grow only in the direction along the layers . the result is : @xmath177 similarly , at @xmath6 , the amplitude of the spin - spin correlation function , @xmath178 at @xmath179 is suppressed compared to that of @xmath180 by an amplitude ratio @xmath181 one thus arrives at the large - scale description of the critical state of the superlattice as a highly anisotropic realization of the bulk universality class of the first component ( ( iv ) in fig . 2 ) . in fact , following the standard ideas of scaling @xcite one can relate all anomalous changes in the critical amplitudes ( [ cratio ] ) , ( [ logcratio ] ) to the anisotropy @xmath182 of the basic length scales . despite the generality of the above consideration , of all the @xmath183 spin lattice models in three and two dimensions only the @xmath184-superlattice behaves strictly according to this scenario , as the ising model has regular finite - temperature critical fixed points without marginal operators in both @xmath2 and @xmath185 . the exactly solvable problem of a @xmath12-ising superlattice requires certain modifications because @xmath77 describing the one - dimensional ising model is a zero - temperature fixed point . this results @xcite in a _ large _ shift in the critical temperature , @xmath186 . more precisely , in the interesting case of a _ thick layer _ superlattice @xcite , criticality happens when the correlation length @xmath54 , characterizing the low - temperature phase of the first component , is much smaller than the width of the 1-layers , @xmath7 . this means that the rg flow in this case comes to @xmath77 not from @xmath69 , but from the low - temperature fixed point @xmath187 describing the ordered phases of the first component . nevertheless , many results can be formulated in terms of just one nonuniversal parameter characterizing @xmath187 : the linear energy @xmath188 of a domain wall separating spin - up and spin - down phases of the first component . let us start our consideration from a close vicinity of @xmath27 , where @xmath189 and the crossover @xmath97 proceeds directly , without encountering @xmath187 . after leaving @xmath69 the scaled temperature @xmath82 is renormalized to @xmath190 . since @xmath191 in the planar ising universality class , below but close to @xmath27 one can write @xmath192 ( note that the normalization @xmath193 , making @xmath194 has been adopted in @xcite ) . further , @xmath195 in this universality class @xcite , so that the coupling @xmath196 is not renormalized near @xmath69 to the linear order in @xmath62 . while one should expect logarithmic corrections to @xmath62 for this marginal case , those do not seem to affect the essential features of the exact solution . as discussed in @xcite , at @xmath197 @xmath198 the last identity , in which @xmath199 is the linear free energy of a domain wall between the two low - temperature phases of the first component , represents the dimensionless coarse - grained spin - spin coupling @xmath200 of the @xmath24-ising model onto which a strip of the first component is mapped after the direct , bulk - to - film crossover . at @xmath77 this spin - spin coupling transforms according to @xmath201 , or @xmath202 with the critical separatrix corresponding to the extreme value @xmath203 , as implied by @xmath204 in the @xmath43-dimensional ising class . note that the fugacity @xmath205 , taking the initial value @xmath206 , behaves at @xmath77 as a regular thermal scaling field with the rg eigenvalue equal to one . consequently , the criterion for @xmath77-dominated behavior is @xmath207 , or , equivalently , @xmath208 . meanwhile , magnetization scales trivially , @xmath209 , at @xmath77 . thus the eigenvalue of @xmath62 is @xmath210 and the corresponding rg equation at @xmath77 can be rewritten as @xmath211 . dividing one rg equation by another we obtain @xmath212 from which one can construct a renormalized temperature @xmath213 it is convenient to define @xmath124 here in such way that the criticality occurs at @xmath214 . then , in full agreement with @xcite , up to corrections vanishing as @xmath215 , the critical temperature @xmath6 of the superlattice can be obtained from the equation @xmath216 where the last identity defines what may be called the _ scaled thickness _ of the layers . if this equation is satisfied at @xmath217 , the case called the _ thin layer _ limit in @xcite , then the flow indeed never encounters @xmath187 . however , in this case the initial value of the coupling @xmath218 is not small , so that the criticality is reached before @xmath219 becomes large and negative so as to display any features of the @xmath43 critical behavior . the flow in fact never leaves @xmath76 . the reentrant crossover does happen in the opposite , _ thick layer _ limit @xcite , @xmath220 , however in this case the behavior of the system on length scales between @xmath54 and @xmath7 is governed by the low - temperature , non - critical fixed point @xmath187 . to describe criticality in this limit we start at scales larger than @xmath50 well below @xmath6 . if the rg is implemented with rescaling the block spin by the factor of the area of the block ( rather than the square root of the area leading to the gaussian fixed point described by the ornstein - zernike functional @xcite ) , then at these scales fluctuations in the amplitude of magnetization are completely suppressed . the only allowed fluctuations are domain walls separating domains of two different spin orientations . at these scales the walls look geometrically sharp and are characterized ( neglecting lattice effects ) by a single parameter : free energy per unit length @xmath188 . as discussed in @xcite , a single strip of the first component becomes identical to a @xmath24-ising model : spins are fully correlated across the strip ; a domain wall cutting across the strip has finite free energy @xmath221 playing a role of an effective ferromagnetic coupling between neighboring blocks of size @xmath222 . characterization of the effective @xmath24-model is completed by specifying the value of a block spin , @xmath223 , where @xmath224 is the bulk magnetization density in the first component . the crossover from @xmath77 to @xmath79 proceeds exactly as described in the beginning of this section , because the representation ( [ xi - sigma ] ) of @xmath120 via the linear surface energy @xmath188 is valid independently of whether the system has arrived at @xmath77 from @xmath69 or from @xmath187 . after the reverse crossover the thermal field is renormalized to @xmath225 leading to the universal relation for the critical temperature @xmath226 holding up to corrections of order @xmath227 . note that the relation @xmath228 used in ( [ xi - sigma ] ) is valid in the planar ising universality class up to corrections of order @xmath229 , which were neglected in the field - theoretical description of ref.@xcite . the expression ( [ tc2d ] ) however has a much wider range of validity : it requires only that the strips are thick , with no condition on the values of the correlation lengths in the two components . for an extensive observable like the specific heat per unit area we can write now @xmath230 here the exponent @xmath231 , equal to @xmath232 in the case of the specific heat , and the critical amplitude @xmath233 characterize singularity @xmath234 in the one - dimensional regime . the scaling function @xmath164 has the following limits : it approaches 1 at @xmath235 ; diverges as @xmath236 at @xmath237 with the amplitude ratios @xmath238 being universal ; an exponential singularity @xmath239 cancels the singular prefactor in ( [ 2dising ] ) at @xmath240 . the critical divergence of the specific heat , @xmath241 is characterized by the amplitude @xmath242 where we have taken into account that @xmath243 cancels the @xmath244 prefactor in ( [ c2dfull ] ) . in order to complete the estimate of @xmath245 we have to know the amplitude @xmath246 . the one - dimensional amplitudes of extensive observables ( per unit _ length _ of a layer ) can be generated from the well - known expression for the free energy density @xmath247 of a one - dimensional chain at temperature @xmath52 and magnetic field @xmath248 ( see @xcite and references there ) @xmath249 where we have taken into account that @xmath54 plays the role of the effective cutoff for the one - dimensional behavior along the layers , and both @xmath54 and @xmath250 are taken at @xmath251 . differentiating with respect to the temperature @xmath52 we obtain ( at @xmath251 and @xmath252 ) @xmath253 . combined with ( [ a2d-1 ] ) this gives @xmath254 this agrees fully with the result of @xcite . other amplitudes can be similarly estimated and agree with the exact solution . an interesting feature of the form ( [ 2dising ] ) is that below @xmath6 , at @xmath255 the exponential part of the scaling function @xmath164 compensating for the ( now spurious ) singular prefactor , leads to @xmath256 this can now be recognized as pointing to the dominance of the fluctuations in the _ second _ component as @xmath18 is approached : in fact , due to the self duality of the planar ising model , a completely similar construction can be developed starting from low - temperatures and focusing on the dual variables in the 2-layers . both scenarios lead to essentially the same results . the other @xmath183 models can be analyzed similarly ; only the most interesting results will be briefly discussed below . the case of the three - dimensional heizenberg ( @xmath42 ) model is very close to the planar ising one , as the @xmath257-class is believed to be characterized by a zero - temperature critical point with the effective coupling being marginally relevant @xcite . this effective coupling in a single heizenberg film of the first component is the dimensionless _ helicity modulus _ @xcite @xmath258 . in the low - temperature phase of the @xmath2-heizenberg model the helicity modulus @xmath259 has the dimension of inverse length , effectively defining the correlation length @xmath260 . in the thick layer limit the critical temperature @xmath6 of the superlattice is again shifted well below @xmath27 of the first component . thus on scales smaller than @xmath7 the first component is well ordered and the additive approximation @xmath261 is well justified @xcite . here @xmath259 can be ( at least in principle ) measured independently in a bulk sample of the first component . once the direct crossover to @xmath77 has occurred , one can form @xmath262 and use the well known @xmath12-heizenberg model rg equation @xcite @xmath263 while the coupling @xmath62 is trivially additive @xmath264 with @xmath185 here . combining these two equations we obtain , in complete analogy with ( [ tc2d ] ) , a universal expression satisfied at the critical point of a heizenberg superlattice : @xmath265 quite analogously to the consideration given above for the planar ising case one arrives at the picture of the critical temperature shift @xmath266 , large scale fluctuations between in the wide temperature range between @xmath27 and @xmath18 , and exponential dependence of critical amplitudes on the thickness of the layers . the case of a @xmath3 ( @xmath41 ) superlattice in @xmath2 is special because , on one hand , the @xmath185-class is characterized by a finite - temperature kosterlitz - thouless transition , so that we expect the shift in the critical temperature to vanish in the limit of ( atomically ) thick layers . on the other hand , the spontaneous magnetization is zero in @xmath185 at any finite temperature . thus the coupling @xmath62 through the 2-layers is crucial in supporting long - range order in a three - dimensional superlattice and we expect the observed spontaneous magnetization to be anomalously suppressed when @xmath196 is small , even well below @xmath267 . the suppression is easily estimated by combining the rg equation for the order parameter @xmath268 , where the scaling dimension @xcite @xmath269 with the additive growth of @xmath62 given by ( [ additive ] ) ; note that in the case of a superfluid the helicity modulus @xmath270 is proportional to the superfluid density @xmath271 . the result is @xmath272 note that @xmath273 essentially plays the role of @xmath54 in this universality class , as well . one can thus see that in the case of @xmath274 , spontaneous magnetization of a superlattice decreases exponentially as soon as @xmath54 exceeds @xmath63 , long before @xmath54 becomes comparable to the thickness @xmath7 . it would be interesting to see if this behavior could be observed at a superfluid transition of liquid helium-4 in some matrix with periodically modulated properties . similarly , a @xmath275 superlattice may be realized in a helium film deposited on a periodically modulated substrate ( the latter could be itself cut out of a _ solid _ three - dimensional superlattice ) . in this case the @xmath43 @xmath3-model orders only at @xmath276 . thus we expect @xmath6 of the superlattice to be shifted well below @xmath27 . at such temperatures the 1-layers are equivalent to a sequence of one - dimensional ( classical ) @xmath3-chains , described by the gaussian hamiltonians @xmath277 where @xmath278 is the phase of the @xmath3 order parameter , @xmath20 labels the layers . the effective one - dimensional helicity modulus @xmath279 is given by the product @xmath280 , as in all other cases considered in this section . each pair of neighboring layers is coupled through the 2-layers separating them via @xmath281 combining the flow of the one - dimensional gaussian coupling @xmath282 with the ubiquitous ( [ additive ] ) ( @xmath43 here ) we arrive at the condition of criticality @xmath283 , which with logarithmic accuracy can be rewritten as @xmath284 a little thinking shows that for @xmath274 the critical temperature of the superlattice is in fact shifted closer to the _ lower _ of the critical temperatures of the two components ; it stays however sufficiently above @xmath18 so that @xmath63 remains much smaller than @xmath8 and our focusing on fluctuations in the 1-layers remains well justified . in summary , this paper addresses the following , as yet hypothetical , experimental project : a sequence of superlattices is built of the same two ferromagnetic components , but with different basic thicknesses @xmath7 and @xmath8 ( fig . the bulk critical properties of the components are assumed to be well known , the thicknesses @xmath285 , i.e. large compared to any important microscopic length of the two materials , and no long - range dipolar interaction is present . given the well developed theory of critical phenomena in the ising , @xmath3 , and heizenberg universality classes , can one predict the variation of the critical properties of the superlattices with the changes in @xmath7 and @xmath8 ? naively one may expect that in the limit of @xmath286 all extensive properties of the superlattice can be expressed as a weighted average of the _ bulk _ densities of the components . indeed , if the limit of infinite thicknesses is taken at _ fixed temperature _ @xmath287 , surface corrections to the bulk density average vanish as @xmath288 , where the correlation lengths @xmath289 play the role of _ penetration depths _ for the perturbations induced by an interface in the two adjacent layers . however , experiment on a given sample studies the @xmath290 limit at _ fixed thicknesses _ @xmath291 , so that the correlation length(s ) are bound to exceed the period of the superlattice making the surface contributions crucial . the thick layer limit for the critical properties of a superlattice is _ singular _ , the thicknesses @xmath292 have to be combined with certain functions of the temperature to form proper scaling fields in terms of which a consistent scaling description is achieved . such scaling description has been constructed above based on the qualitative picture of reentrant dimensional crossover @xcite . in this picture 2 ) a superlattice , at its critical temperature @xmath6 , displays three different asymptotic types of critical behavior , when probed on different length scales . these three asymptotic models are , in the order of increasing wave length , ( ii ) a sequence of noninteracting semi - bulk samples , exhibiting _ ordinary _ surface critical behavior at the surfaces ; ( iii ) a sequence of noninteracting _ thin _ layers , exhibiting @xmath11-dimensional behavior ; ( iv ) a highly anisotropic , but uniform critical system of the same universality class as that of the components . away from @xmath6 the finite correlation length falls into one of the three regimes , so that instead of probing different length scales at @xmath6 one can study the crossover by observing the asymptotic behavior at different temperatures approaching @xmath6 . each of these limiting types of behavior corresponds to a previously studied renormalization group fixed point . combining those into a global rg flow ( fig . 4 ) we have constructed a rather complete scaling picture of critical phenomena in superlattices composed of thick layers of two components . the simplest case is that of a @xmath2-ising superlattice , characterized by the possibility of a finite - temperature phase transition in a single layer of the first component . as a result , in the limit of thick layers the critical temperature @xmath6 of the superlattice approaches that of the stronger of the two ferromagnets , @xmath27 . the two crossovers then happen in a ( relatively ) narrow vicinity of @xmath6 . they are conveniently characterized in terms of the reduced temperature scales @xmath293 given by ( [ taud ] ) , ( [ taur ] ) . these scales represent ( fig . 3 ) the relative widths of the temperature domains in which the superlattice behaves , respectively , as models ( ii ) and ( iii ) above . the simplest scaling description ( [ a ] ) is then achieved in terms of the reduced temperatures @xmath294 and @xmath295 , shifted to the observed critical temperature @xmath6 . as represented graphically in the double logarithmic plot of fig . 3 , the ultimate divergence is characterized by the critical exponents characteristic of the bulk , @xmath0-dimensional behavior . however the presence of the intermediate domain of the @xmath17-dimensional behavior ( model ( ii ) ) leads to anomalous changes in the critical amplitudes , as seen in ( fig . 3 ) and expressed analytically through ( [ taud],[taur ] ) in ( [ logcratio ] ) . the situation is somewhat more interesting in the other three- and two - dimensional spin - vector universality classes , where the weak bonds between subsequent 1-layers play a crucial role in maintaining the magnetic long - range order in the system . this leads to a large shift in the critical temperature of a superlattice , as implicitly given for various classes by the universal expressions ( [ tc2d],[tch3d ] , [ tcxy2d ] ) , as well as the anomalous , exponential suppression in the amplitude of spontaneous magnetization ( [ mxy3d ] ) in a three - dimensional @xmath3-superlattice . where applicable , the results obtained here by the method of this paper agree with the analysis of the exact solution for the two - dimensional ising superlattice @xcite given in @xcite . the three - peaked form of the temperature dependence of the specific heat , characteristic of that model generalizes to the three - dimensional heizenberg class . it has a transparent physical interpretation : most of the spin degrees of freedom in such a superlattice get ordered at one of the bulk critical temperatures of the two components , leading to finite , but large - amplitude peaks at both . however , a small , @xmath296 , fraction of the degrees of freedom participate in strong large - scale fluctuations at all @xmath297 , giving small - amplitude , but diverging contribution to the specific heat at @xmath6 . the abundance of interfaces in a three dimensional magnetic superlattice may greatly improve chances of experimental observation of surface and two - dimensional scaling . in particular , the elusive surface scaling exponents enter the definition of the parameter @xmath36 in ( [ taur ] ) and can be obtained , for instance , by analyzing the dependence of the critical amplitudes , ( [ logcratio ] ) and fig . 3 , on the thickness of 1-layers @xmath7 . finally , the main restriction of the present analysis seems to be the neglect of dipolar interactions . when present , these interactions will lead to demagnetization effects making heizenberg universality class asymptotically unstable . otherwise , most of the results presented above can be easily modified in the case when the interlayer interaction dependence on the thickness of the separating layers , @xmath298 , is a power - law rather than an exponential as was always assumed above . however , the assumption of local interactions enters our consideration in much deeper ways than just through the exponential form of @xmath298 , so that the required revision could be much more serious . _ anti_-ferromagnets , of course , are free from the complications arising from the dipolar interactions . however , pinning of the spin density waves by the interfaces between the layers in a superlattice gives rise to a different source of the thickness dependent effects beyond the scope of this paper . in any case , the above consideration seems to represent a reasonable first step towards a complete analysis of scaling in superlattices . note that recently critical phenomena in superlattices have been analyzed in the frames of the mean - field landau - ginzburg approximation @xcite . a direct comparison of our results to theirs is hardly feasible in view of the failure of the mean - filed approximation @xmath12 and 3 considered here . however a combination of the results of ref . @xcite with a renormalization group expansion in @xmath299 and the scaling framework developed above looks promising , although computationally very hard , direction for future work . comments by m. e. fisher have been highly appreciated . here we derive the effective spin - spin coupling ( [ int ] ) from the ornstein - zernike spin - density functional @xmath300=\frac{k_bt}{2\chi_2}\int_{0}^{l_2}dz\ d^{d ' } x ' \{{\bf s}^2+(\xi_2)^2(\nabla { \bf s})^2\ } \label{oz}\ ] ] describing a layer of the second component on the length scales exceeding the correlation length @xmath63 ; the other parameter entering ( [ oz ] ) is the bulk spin susceptibility @xmath301 . this functional corresponds to the gaussian rg fixed point describing the bulk paramagnetic phase of the second magnet . in a layer of finite thickness @xmath8 , the bulk part of the functional has to be complemented by the surface contributions , which at the paramagnetic fixed point take the form @xmath302 in the above equation @xmath303 , the phenomenological parameter @xmath304 accounts for the change in the number of nearest neighbors for the surface spins , while the surface fields @xmath305 are induced by the neighboring spins in the layers of the first component : @xmath306 where @xmath307 and @xmath308 , as defined after ( [ int ] ) , while @xmath309 is the effective coupling across the interface . the minimization of the total free energy @xmath310 proceeds via fourier expansion in the @xmath311-plane . for each fourier harmonic of the spin density one obtains an independent one - dimensional functional . minimizing the latter and summing all contributions one obtains in the limit @xmath312 an effective interaction between the surface spins of the first component @xmath313 ^ 2}.\end{aligned}\ ] ] since the fourier components of the spin density with wave numbers @xmath314 has been already eliminated by the rg transformation , the factors @xmath315 . using ( [ j12 ] ) one finally obtains the interaction of the form ( [ int ] ) with the amplitude @xmath316 ^ 2}j_{12}^2.\ ] ] c. f. majkrzak , _ et al _ , adv . phys . * 40 * , 99 ( 1991 ) ; refer to the japanese guy ? k. binder , in _ phase transitions and critical phenomena _ , edited by c. domb and j. l. lebowitz ( academic , new york , 1983 ) , vol . 8 , p. 51 ; and references there . h. w. diehl , in _ phase transitions and critical phenomena _ , edited by c. domb and j. l. lebowitz ( academic , new york , 1986 ) , vol . 10 , p. 75 ; and references there . l. v. mikheev and m. e. fisher , phys . lett . * 70 * , 186 ( 1993 ) ; phys . b * 49 * , 378 ( 1994 ) . t. w. burkhardt and h. w. diehl , phys . rev . b * 50 * , 3894 ( 1994 ) . h. au - yang and m. e. fisher , phys . b * 21 * , 3956 ( 1980 ) ; and references there . y. li and k. baberschke , phys . lett . * 68 * , 1208 ( 1992 ) ; and references there . l. j. de jongh , p. bloembergen , and j. h. p. kolpa , physica * 58 * , 305 ( 1972 ) ; see discussion in ref . @xcite below . see references in s. w. pierson , phys . 73 * , 2496 ( 1994 ) . a ) m. e. fisher , j. phys . * 26 * , 87 ( 1969 ) ; a. e. ferdinand and m. e. fisher ( unpublished ) ; b ) f. becker and h. hahn , phys . lett . , 9 ( 1972 ) ; i. decker and h. hahn , physica * 83a * , 143 ( 1976 ) ; physica * 89a * , 37 ( 1977 ) ; physica * 93a * , 215 ( 1978 ) . c ) j. zittartz , j. magn . magn . mat . * 31 - 34 * , 1105 ( 1983 ) , and references there . l. v. mikheev , j. stat . phys . * 78 * , 79 ( 1995 ) . k. g. wilson and j. kogut , phys . rep . * 12c * , 75 ( 1974 ) . m. e. fisher , in _ critical phenomena _ , edited by f. j. w. hahne , lecture notes in physics vol . 186 ( springer , new york , 1983 ) , p.1 . l. v. mikheev , unpublished ( cond - mat/9501018 ) . e. brezin , b. i. halperin , and s. leibler , j. physique * 44 * , 775 ( 1983 ) . d. oconnor and c. stevens , phys . lett . , 506 ( 1994 ) . n. nakanishi and m. e. fisher , j. chem . phys . , 3279 ( 1983 ) . v. privman , p. c. hohenberg , and a. aharony , in _ phase transitions and critical phenomena _ , edited by c. domb and j. l. lebowitz ( academic , new york , 1990 ) , vol . 14 , p. 1 . m. e. fisher in _ collective properties of physical systems _ , edited by b. lundqvist and s. lundqvist ( academic press , new york , 1973 ) . g. parisi , _ statistical field theory _ , ( addison - wessley , new york , 1988 ) . l. v. mikheev and m. e. fisher , j. stat . phys . * 66 * , 1225 ( 1992 ) . a. z. patashinskii and v. l. pokrovskii , _ fluctuation theory of phase transitions _ ( pergamon press , new york , 1979 ) . m. e. fisher and v. privman , phys . b * 32 * , 447 ( 1985 ) . x. hu and y. kawazoe , j. appl . phys . ( in press ) ; in preparation . fig . 2 : graphical representation of the reentrant dimensional crossover . the observed behavior of the system changes from a noncritical superlattice , ( i ) , to the sequence of _ thick _ layers of the first compnent , ( ii ) , connected by exponentially weak links mediated by noncritical fluctuations in the second component . from this limit , the direct crossover takes the system to a sequence of weakly coupled _ thin _ layers , ( iii ) , while the reentrant crossover takes it back to the bulk , although highly anisotropic , critical behavior , ( iv ) . fig . 4 : reentrant renormalization group flow in the _ critical manifold _ of the parameter space representing the @xmath2 ising superlattices . the three fixed points , defined in the text , are connected by critical separatrices ( bold lines ) . they correspond , in the order in which they are encountered by the flow , to the limits depicted by ( ii ) , ( iii ) , and ( iv ) in fig . 2 . the direct crossover , from @xmath69 to @xmath77 , and the reentrant one , from @xmath77 to @xmath79 , are driven by two scaling fields , measured by the inverse thickness @xmath80 and by the interlayer coupling strength @xmath62 , respectively . note that yet another relevant scaling field , @xmath28 , directed roughly perpendicular to the plane of the figure , drives the system away from this two - dimensional critical manifold to the high- and low - tmeperature fixed points .
el lanzamiento de los observatorios de rayos x _ chandra _ ( nasa ) y _ xmm - newton _ ( esa ) en 1999 ha revolucionado nuestra visin del universo , al proporcionar informacin astrofsica detallada sobre muchos tipos de fuentes . la alta eficiencia de _ xmm - newton _ lo convierte en el instrumento ideal para obtener espectroscopa de resolucin baja y moderada de fuentes dbiles y extensas . despus de 3 aos de operaciones , _ xmm - newton _ ha observado todo tipo de fuentes astronmicas y ha proporcionado resultados de la mayor importancia en muchas reas . en esta revisin , destacamos unos pocos aspectos donde la contribucin de _ xmm - newton _ ha incrementado de forma significativa nuestro conocimiento del universo energtico . x - ray astronomy is at the very heart of the enormous progress that our astrophysical knowledge of the universe has undergone over the last decades . x - rays are emitted in a wide range of physical situations in the universe , invariably linked to the presence of hot gas and strong gravitational fields . these phenomena , largely unsuspected in the early days of x - ray astronomy , are nowadays seen to be ubiquitous in a variety of astronomical objects . x - ray observations of previously known objects have often revealed phenomena ( e.g. , the presence of an accreting black hole ) just not seen at other wavelengths . in addition , new sources serendipitously discovered in x - ray surveys have uncovered new classes of objects , totally inconspicuous at optical wavelengths . one datum that illustrates the impact of x - ray observations in astronomy is that 20% of the papers published by the 3 main astrophysics journals in 2002 ( astrophysical journal , monthly notices of the royal astronomical society and astronomy & astrophysics ) contain the `` x - ray '' keyword in the abstract . the 2002 nobel prize in physics given to riccardo giacconi ( father and leader of x - ray astronomy ) , is also a sign of the good health reached by this discipline , as it is one of the very few granted to astronomers . x - ray astronomy began in 1962 , when a rocket flown to detect the moon in scattered x - rays from the sun , discovered instead the first extra - solar x - ray source ( sco x-1 ) and the cosmic x - ray background ( giacconi et al . 1962 ) . this already showed that the universe out there could be radically more energetic than what was known before , as sco x-1 had a much larger x - ray to optical flux ratio than the sun . more rockets followed this pioneering discovery , but there was a basic limitation in the amount of observing time available for observations in a single flight . _ uhuru _ ( which means _ freedom _ in swahili ) was the first orbiting x - ray observatory . it was launched from italy s san marco station in kenya on the 12th of december of 1970 , coinciding with the independence day of this country ( thence the name ) . the _ uhuru _ payload weighed less than 60 kg . it scanned the sky for over 2 years and produced the first all - sky catalogue of x - ray sources , containing a few hundred of them . subsequent work demonstrated that outside the galactic plane , most of the sources were active galactic nuclei ( agn ) and clusters of galaxies . in the galactic plane , _ uhuru _ found x - ray binaries , supernova remnants and other diffuse structures . this rough picture ( but with varying fractions of objects ) still describes to zero - th order the present knowledge of the x - ray universe . this early generation of x - ray orbiting observatories ( which also included _ heao-1 _ and more recently _ ginga _ , japanese word for milky way ) were equipped with a mechanical collimator as the only means to limit the field of view covered by the ( usually gas - filled proportional counter ) detectors , without any further optics . this provided a very rough angular resolution , of the order of several degrees , which ultimately limited the sensitivity of the observatory because of confusion of fainter sources . proper imaging x - ray optics was featured for soft x - rays ( below 4.5 kev ) with the _ einstein observatory _ first . with an angular resolution of a couple of arc minutes , _ einstein _ discovered many new x - ray sources ( including many at cosmological distances ) , showing that the word _ experiment _ was in the way of being replaced by _ observatory _ in x - ray missions . _ einstein _ made it possible to discover new classes of sources by their x - ray emission , resolved the spatial structure of the intracluster gas in galaxy clusters and produced large catalogues of x - ray sources that were subsequently used to conduct detailed astrophysical studies . it also showed , for the first time , that a significant fraction of stars are x - ray emitters . _ rosat _ , in some sense a successor to _ einstein _ , deepened our knowledge of the x - ray universe , by conducting first an all - sky survey ( this yielded the , so far , largest catalogue of x - ray sources , exceeding 50000 ) , and for almost a decade , targeted observations of a full range of astronomical objects , from comets to distant qsos . focusing x - ray photons is not an easy task . in a standard reflecting optical telescope , photons are directed almost normally to the reflector s surface and collected in the focal plane . if the same setup is used for x - rays , they end up either transmitted or absorbed , but never reflected . grazing incidence is needed to achieve total reflection for x - rays , the maximum angle with respect to the surface being of a few degrees for a photon of @xmath0 . reflection becomes more difficult at higher photon energies , as the angle for total reflection becomes smaller and the mirrors need to be placed almost parallel to the optical axis . it is easy to see that the effective area that a photon sees is very small with a mirror placed in grazing incidence , and therefore several ( or many ) mirrors are nested one inside each other to increase the collecting area . it is also clear that an x - ray focusing telescope of higher energy photons will have less effective area and longer focal length than a similar one focusing soft x - ray photons . the start of the nineties brought x - ray observatories able to produce images with photons of up to 10 kev , such as the japanese _ asca _ were launched . _ asca _ also opened the door to a more modern type of x - ray detector ( the _ charge coupled device _ or ccd ) that ended up burying the proportional counters . ccd - based detectors deliver an order of magnitude better spectral resolution for single photons than proportional counters . at the energy of the fe k@xmath1 line ( 6.4 - 6.7 kev ) , the resolution of @xmath2 achieved by these detectors enables a detailed study of the physical environment where it is produced . other x - ray observatories have been launched during the 1990 s , including _ bepposax _ ( which also featured higher energy detectors sensitive to many tens of kev ) and _ rossi x - ray timing explorer _ ( rxte ) which is tailored to timing analysis of bright sources . however , at the turn of the new millenium , both _ nasa _ and _ esa _ decided to launch their respective large x - ray observatories : _ chandra _ ( launched july 1999 ) and _ xmm - newton _ ( launched december 1999 ) . the coincidence of operations between both missions ( being just by chance , as _ chandra _ was over - delayed for several years ) has brought what can be called the _ golden age _ of x - ray astronomy . by virtue of their respective designs _ is mostly an imaging observatory , while _ xmm - newton _ has its major capabilities at spectroscopy . having both of them operating at the same time has opened a huge window to x - ray astrophysics , whose dimensions are just beginning to be realized after the first few years of operations . _ xmm - newton _ is an x - ray observatory launched and operated by the european space agency ( esa ) , with instruments contributed and funded by esa member states and nasa ( usa ) . _ xmm - newton _ was successfully launched by an ariane 5 on the 10th of december of 1999 . its orbit is highly eccentric with a period of @xmath3 days . although the original _ xmm - newton _ project was approved for 2 years of scientific operations , all systems and payloads are designed to last for @xmath4 years . specifically , fuel ( hydrazine ) is currently thought to last for more than that . in fact , an extension of the _ xmm - newton _ operations has already been approved to 2006 with a further 2 year period being revised by mid 2003 . for comparison with early missions , the full _ xmm - newton _ satellite weighs more than 3000 kg . currently , the _ xmm - newton _ spacecraft is operated by esoc at darmstadt ( germany ) , but the instruments and science operations in general are managed from vilspa ( villafranca del castillo , spain ) . [ xmm ] _ xmm - newton _ consists of 3 co - aligned grazing incidence x - ray telescopes ( jansen et al . 2001 ) , with angular resolution of 4 - 6 arcsec ( fwhm ) and 13 - 15 arcsec ( half energy width ) . each x - ray telescope nests 58 grazing incidence mirror pairs , with a total collecting area of @xmath5 ( see fig [ xmm ] for a schematic of the x - ray optics in the xmm - newton x - ray telescopes ) . one of these telescopes focuses all x - rays into a single spectroscopic imaging detector ( epic - pn , strder et al . 2001 ) , which has a pixel size of 4 arcsec and covers a field of view of about 30 arcmin in diameter . about half of the x - rays focused by the other two telescopes go undispersed to similar imaging spectrographs called epic - mos ( turner et al . the epic instruments can measure the energy of individual photons with a resolution of @xmath6 . epic is sensitive to photons from 0.2 to 12 kev and can detect x - ray sources as faint as @xmath7 in the 0.5 - 2 kev band in a few tens of kiloseconds ( ks ) . its main limiting sensitivity in soft x - rays is caused by confusion , due to the modest angular resolution of the x - ray telescopes . at higher photon energies the main limiting factor is photon counting , with an expected confusion limit in the 2 - 10 kev band reached in a several ms exposure only . the two x - ray telescopes that focus the x - rays into the epic - mos detectors are equipped with difraction gratings working by reflection . these disperse the remaining x - rays according to their wavelength and these are recorded in a further chain of ccd detectors . this wavelength - dispersive instrument is called the reflection grating spectrometer ( rgs ) and delivers a spectral resolution @xmath8 in first order dispersion ( i.e. , @xmath9 ) in the soft x - ray domain ( 5 - 35 , or 0.3 - 2.4 kev ) . the sensitivity of the rgs can be viewed in terms of a source with 0.5 - 2 kev flux of @xmath10 producing peak s / n@xmath4 per spectral resolution element in 100 ks . it is therefore an instrument designed to observe bright sources with relatively high spectral resolution . the moderate angular resolution of the xmm - newton mirrors makes it possible to obtain integrated moderate resolution of _ extended _ sources , a capability that is producing extremely interesting results ( see later ) . besides that , _ xmm - newton _ has a co - aligned optical / uv telescope ( the optical monitor - om ) which operates in the range from 1600 to 6600@xmath11 . at its focal plane the om has a photon counting device ( mason et al . 2001 ) , which delivers time - resolved information for the detected photons . the field of view of the om is 17 arcmin , with a point - spread - function of fwhm between 1.3 and 2.5 arcsec , with a limiting sensitivity of 23.5@xmath12 for 1000 sec integration . the om is equipped with a number of uv and optical filters , plus low resolution uv and optical grisms , whose calibration is being conducted at the time of writing this report . all three _ xmm - newton _ instruments ( epic , rgs and om ) are simultaneously operated . the ground segment operations of _ xmm - newton _ are conducted by the soc at villafranca del castillo ( spain ) with the assistance of an esa member - state funded consortium called _ survey science centre _ ( ssc ) . the ssc tasks include the development of science analysis software tasks in collaboration with the soc ; the pipeline processing of all _ xmm - newton _ data and the identification of the _ serendipitous _ sources discovered by _ xmm - newton_. this last task is specially demanding , as it is likely that _ xmm - newton _ will find about 50000 new x - ray sources per year . identifying and cataloguing these data will constitute a major legacy of _ xmm - newton _ ( see watson et al . 2001 ) . esa has already released the first version of the _ xmm - newton _ catalogue provided by the ssc , which contains over 30000 good - quality x - ray sources . some of these sources are identified thanks to the extensive archival search conducted by the ssc . at the time of writing this report the ssc has imaged in various optical filters a large fraction of fields that contain the catalogued sources . besides that , well over 500 of these x - ray sources have been spectroscopically identified . all this optical imaging and spectroscopic information will be included in future versions of the _ xmm - newton _ catalogue , delivering an extremely powerful tool to investigate the x - ray sky . one of the areas where the impact of _ xmm - newton _ has been strongest is the study of x - ray emission from `` normal '' stars . the possibility that stars without compact companions could be x - ray emitters had not been initially foreseen . though some normal stars had been detected by previous experiments ( e.g. , topka et al . 1978 ) , only after the observations by _ einstein _ was it fully realised that many solar - type stars emitted soft x - rays ( vaiana et al . equally surprising was the discovery that hot ob stars also appeared to be substantial soft x - ray sources . the x - ray emission from low - mass stars was quickly interpreted in terms of coronal activity , similar to that observed in the sun , but on a much larger scale . many m - type stars were found to display relatively strong x - ray emission , which correlated with other indicators of activity , such as emission in h@xmath1 or the caii doublet ( noyes et al . 1984 ; vilhu 1984 , fleming et al . 1989 ) . it was also found that there was a strong correlation between age and x - ray activity . in this sense , the all - sky survey by _ rosat _ greatly changed the previous view on the evolution of low - mass stars towards the main sequence . it was seen that many stars not displaying the typical characteristics of t tauri stars could be identified as very young objects because of their x - ray emission ( e.g. , neuhuser 1997 ) . this lead to the discovery of whole new associations of young stars . as previous missions lacked spectral resolution in the soft x - ray range , _ xmm - newton _ and _ chandra _ offer the possibility of obtaining , for the first time , information about the physical causes of this emission . coronal activity in chromospherically active low - mass stars is thought to result in x - ray emission through mechanisms similar to those observed in the sun . observations with _ xmm - newton _ provide spectra showing lines from a large variety of elements ( o , ne , mg , fe , n , etc ) , many of them in at least two different ionisation states . from them , coronal abundances , which are in many cases very different from the stellar atmospheric abundances , are derived . in the sun , elements with low first ionisation potential are overabundant with respect to their photospheric abundances , while elements with high first ionisation potential are not . in a survey of rs cvn binaries ( tidally locked binaries containing a highly active slightly evolved g - k star ) , audard et al . ( 2003 ) find that the most active stars display a behaviour opposite to that seen in the sun , suggesting that fractionation mechanisms should be revised . they indicate that this opposite effect is seen in all stars with high coronal activity , while stars with lower activity seem to present an effect similar to the sun . simultaneous observations of @xmath13 gem , an rs cvn binary with a k1iii primary , with _ xmm - newton _ and the vla have allowed the discovery of a correlation between the radio luminosity of a large flare and the time derivative of the x - ray luminosity ( gdel et al . this relationship is observed in solar flares , where it is known as the neupert effect , and supports flare mechanisms causing chromospheric evaporation . the presence of x - ray emission from young ( pre - main - sequence ) stars is now recognised as a widespread phenomenon ( feigelson & montmerle 1999 ) . it is assumed to be a consequence of the same magneto - centrifugal processes that drive the outflows and winds associated with classical t tauri stars ( e.g. , shu et al . 1994 ) . large - scale surveys with _ xmm - newton _ will allow an understanding of these phenomena . as a first example , favata et al . ( 2003 ) have observed the l1551 star forming complex . they find that the characteristics ( both temporal evolution and spectrum ) of classical t tauri stars and weak - lined t tauri stars are very different , suggesting that while weak - lined t tauri stars show an enhanced version of the coronal activity seen at older ages , classical t tauri stars have a different x - ray emission mechanism , related to their accretion disks . although emission from low mass stars can be understood in terms of coronal activity , the existence of relatively soft x - ray emission from massive stars came as a bit of a surprise ( harnden et al . models trying to explain it invoke hydrodynamic shocks resulting from intrinsic instabilities in their radiatively driven winds ( see feldmeier et al . _ xmm - newton _ rgs spectra have allowed the exploration of physical conditions in the regions where the emission is produced . observations of the o4ief star @xmath14 puppis have provided confirmation that the x - ray emission consists mostly of broad emission lines from h - like and he - like charge states of n , o , mg and si , as well as ne - like ions of fe ( kahn et al . 2001 ) . in this object with a very strong radiative wind , x - ray emission reaching us seems to be produced at large distances from the star . [ zpup ] conversely , the x - ray spectrum of the b0.2v mk standard @xmath15 scorpii is not very consistent with standard models . emission lines in its _ xmm - newton _ rgs spectrum are as narrow as the instrumental profile , suggesting that they are produced much closer to the star ( mewe et al . x - ray emission shows a hot component corresponding to a temperature in excess of @xmath16k . this is supported by a _ observation of the same star by cohen et al . ( 2003 ) , who suggest that the spectrum of @xmath15 scorpii is intermediate between those of o - type stars and coronal emitters . as there is rather strong evidence against the possibility of a cool companion for @xmath15 scorpii , a model in which dense clumps in the wind stall and decelerate , perhaps falling back onto the stellar surface , such as that presented by howk et al . ( 2000 ) , is preferred . the hard x - rays would then originate in bow shocks around those clumps . in some cases , emission from relatively hot stars seems to be due to late - type companions . simultaneous observations of castor with _ xmm - newton _ and _ chandra _ allow the resolution of its three components , all of which are close binaries . castor c is a pair of m1v stars , both of which appear to be moderately active ( stelzer et al . both castor a and castor b contain an a - type star and a low - mass companion , but their emission appears typical of coronal activity in the low mass stars ( stelzer & burwitz 2003 ) . another interesting result obtained with _ xmm - newton _ is the lack of x - ray emission from c - rich wolf - rayet stars . oskinova et al . ( 2003 ) failed to detect wr 114 with _ xmm - newton _ , implying a lower limit on its x - ray luminosity @xmath17 . until now , no single wc star has been detected in x - rays , while several wn star have . an example is wr 110 , detected with _ xmm - newton _ by skinner et al . surprisingly , the x - ray spectrum of this very hot star also shows a hard component which can not be explained with current models . in some cases , a relatively hard x - ray spectrum is produced when the winds of two massive stars forming a close binary collide . this is the case of hd 93403 ( o5.5i + o7v ) , where clear orbital modulation has been detected with _ xmm - newton _ ( rauw et al . 2002 ) . x - ray binaries ( xrbs ) are systems in which high energy radiation is emitted as a consequence of the accretion of material from a donor star ( in most cases , a hydrogen - burning `` normal '' star ) onto the surface of a compact object ( see lewin et al . 1995 , for extended reviews ) . binary systems in which the accreting compact object is a white dwarf are generally considered as a separate subclass , known as cataclysmic variables , though there are no strong physical reasons for this separation . xrbs emit most of the x - ray flux detected from normal galaxies , including the milky way . they have been discovered in large numbers by previous satellites and now a sufficiently large sample exists for population studies to be statistically significant ( e.g. , helfand & moran 2001 ) . _ xmm - newton _ offers several potentialities for their study . on the one hand , it opens up the opportunity of obtaining spectra of galactic sources with moderate resolution and high signal - to - noise ratio . on the other hand , because of its large collecting area and moderately high spatial resolution , it provides an excellent opportunity for studying the populations of nearby galaxies . last but not least , _ xmm - newton _ offers the possibility of obtaining x - ray spectra of faint galactic sources , opening the door to studies of low - luminosity accreting binaries . xrbs are generally divided into two main subclasses , high mass ( hmxbs ) and low mass ( lmxbs ) , depending on the nature of the donor star ( either an ob star or g@xmath18 m spectral type ) . additionally , depending on the temporal behaviour , they can be divided in persistent and transient sources . lmxbs containing an accreting neutron star can be either persistent or transient , but systems with black holes are generally transient . classical low mass x - ray binaries ( lmxbs ) consist of a neutron star with a moderately strong magnetic field ( @xmath19 g ) accreting from a low - mass star . the light - curves of these objects are rather complex ( displaying temporal features such as eclipses and dips ) . there is a large variety of behaviours among lmxbs , likely due to the complexity of their accretion geometries , which has prevented the creation of a unified model . as a matter of fact , while all lmxbs observed with _ asca _ could be well fit with a two component model , a blackbody representing the accretion disk and an extended comptonising region ( church & balucinska - church 2001 ) , this model failed to fit observations of several lmxbs with _ ( e.g. , oosterbroek et al . 2001 ) . many lmxbs also display bursts , very rapid increases in the x - ray flux followed by exponential declines ( typically lasting seconds to minutes ) . the bursts are interpreted as thermonuclear explosions on the stellar surface , after material has accumulated due to accretion ( lewin et al . 1993 ) . during this bursts , x - ray emission from the vicinity of the neutron star dominates that coming from the disk . _ xmm - newton _ has allowed the study of the accretion environments of low mass x - ray binaries , with the discovery of complex narrow absorption features in their x - ray spectra , corresponding to h - like and he - like ions of o and fe , and likely other elements ( cottam et al . 2001 ; sidoli et al . 2001 ; parmar et al . for example , comparing the number of absorbed photons at the ovii and oviii edges with the number of photons emitted in the oviii ly@xmath1 and ovii he - like complexes , cottam et al . ( 2001 ) argue that absorbing material must be aligned with the accretion disk and extend along our line - of - sight . jimenez - garate et al . ( 2002 ) are able to derive elemental abundance ratios from recombination emission lines seen in the spectrum of the lmxb her x-1 . they derive very high n / o ratios , which they interpret as proof of strong interactions during the formation history of this binary . lmxbs containing a black hole ( and a few transient sources containing neutron stars ) are generally termed soft x - ray transients ( sxts ) or x - ray novae . this is because they display very strong x - ray and optical outbursts ( generally also accompanied by radio emission ) separated by long periods of quiescence ( see tanaka & shibazaki 1996 ) . they are `` soft '' in the sense that their outburst spectra are dominated by a soft blackbody component at @xmath20kev . the x - ray spectra of sxts are very variable and characteristic low / hard and high / soft states have been identified . the geometry of the accretion flow and the source of soft photons during the outbursts is currently debated . in order to explain their long quiescent states optically thin advection - dominated accretion flows ( adafs ) have been invoked ( narayan et al . 1996 ; esin et al . some authors have argued that the very low x - ray fluxes during quiescence could be due simply to coronal activity from the donor . the large collecting area of _ xmm - newton _ allows the observation of very faint sources in quiescence . the sxt gu mus was observed in quiescence as a very faint source . the x - ray flux was characterised by a power law , ruling out coronal activity and apparently supporting adaf models ( sutaria et al . similar conclusions were drawn from an observation of the sxt gro j1655@xmath1840 ( hameury et al . however , observations of the black hole candidate and micro quasar grs 1758@xmath18258 found a very soft spectrum in the off state , against the predictions of simple adaf models ( miller et al . 2002 ) . another source , xte j1650@xmath18500 , was observed in the very high state by miller et al . ( 2003a ) . its spectrum displayed broad fe k@xmath1 lines , whose shape suggests that rotational energy is being extracted from a spinning black hole . cataclysmic variables ( cvs ) are systems containing a white dwarf accreting from a low - mass hydrogen - burning star ( e.g. , sion 1999 ) . as such , cvs may manifest themselves under the guise of classical novae , dwarf novae , recurrent novae , nova - likes and similar objects ( patterson 1984 ) . x - ray emission is generally detected from those cvs in which the white dwarf exhibits a strong magnetic field , the polars or am her stars , or a moderate magnetic field , the intermediate polars or dq her stars ( patterson 1994 ) . the interest of cvs stems from the fact that they allow a detailed study of accretion processes , with wide applications in several astrophysical contexts . as the x - ray spectra of cvs are not as hard as those of accreting neutron stars and their luminosities are also not very high , sensitive instruments are needed to detect their emission . _ rosat _ surveys resulted in the discovery of large numbers of new faint cvs and it is expected that _ xmm - newton _ will provide spectral information for a large number of sources , making statistical studies possible . such work is already in progress , with large numbers of cvs having been observed ( e.g. , ramsay & cropper 2002 ; ramsay & cropper 2003 ) , allowing determination of their basic parameters , such as white dwarf mass and mass transfer rate . of particular interest is the observation of the recent nova v2487 oph 1998 . this classical nova was detected with signatures of an accreting white dwarf less than three years after its nova outbursts ( hernanz & sala 2002 ) . this detection suggests that the system was back to its standard configuration after the phenomenal thermonuclear explosion which originated the nova outburst . discovered originally in the lmc , supersoft x - ray sources ( ssss ) display x - ray spectra peaking at energies much lower than traditional x - ray binaries ( @xmath21ev ) . because of this , they are heavily affected by interstellar absorption and unlikely to be detected in the milky way , unless they are very close . they are generally interpreted as white dwarfs accreting from a more massive hydrogen - burning donor . the mass transfer is therefore unstable and proceeds on the thermal timescale ( van den heuvel et al . because of their soft spectra , they have only been studied in any detail with _ rosat_. _ xmm - newton _ provides now the opportunity to observe their spectra . the first rgs spectrum of an sss , cal 83 in the lmc , proved that the x - ray emission originated from the photosphere of a very hot white dwarf by showing a rich spectrum of absorption complexes due to several elements ( paerels et al . even though this spectrum is qualitatively similar to expectations from white dwarf atmosphere models , rgs observations of the galactic sss mr vel reveal a wealth of emission lines displaying p - cygni profiles , indicative of a strong wind from the white dwarf s surface , which can not be reproduced with current models ( motch et al . also interesting is the likely detection of doppler shifts due to orbital motion in the x - ray emission lines of mr vel , probably the first detection of orbital motion at high energies ( motch et al . 2002 ) . in _ xmm - newton _ pointings , osborne et al . ( 2001 ) have found a transient sss close to the nucleus of m31 displaying an 865-s periodicity . king et al . ( 2002 ) interpret this as the spin period of a white dwarf and hence argue ( see also schenker et al . 2002 ) that cvs are descendants from ssss . supernova explosions have a very profound impact on the interstellar medium ( ism ) , both as sources of mechanical energy and heavy elements . supernova remnants ( snrs ) provide information about these issues and can be observed as high - energy sources . the interaction of high velocity ejecta with the ism generates very high temperatures , giving rise to line emission from heavy elements . by studying the spatial distribution of elements , we can gain insight into nucleosynthesis in the progenitor and the geometry of the explosions . snrs are also possible sites of cosmic ray acceleration . the characteristics of the rgs make it possible to obtain spectral information from relatively extended sources . rgs spectra of snrs allow the study of physical conditions all over the remnant ( e.g. , rasmussen et al . 2001 ) . by measuring the mass , temperature and bulk velocity of material , it is possible to obtain information about the supernova explosion . using _ xmm - newton _ observations of the young snr cas a , willingale et al . ( 2002 ) have derived kinematic information and abundance ratios for the supernova ejecta . from analysis of those data , they find evidence for beaming in the supernova explosion , suggesting that most of the material was ejected in two jets ( willingale et al . they also suggest that the progenitor was a very massive wolf - rayet star . the compact object born in the cas a supernova explosion has been identified with a point - like _ chandra _ source , cxo j232327.8 + 584842 . a long pointing with _ xmm - newton _ , however , has failed to detect the x - ray pulsations that would be expected from a neutron star ( mereghetti et al . the spectral properties of this source are also difficult to interpret in terms of a rapidly spinning neutron star . no pulsations have been detected either from the snr g21.5@xmath180.9 , which in many aspects resembles the crab snr , where a central pulsar is injecting high energy electrons ( la palombara & mereghetti 2002 ) . in this snr , the x - ray spectrum of the nebula softens as one moves out from the centre , reflecting the energy losses of the electrons due to synchrotron radiation , as they diffuse out ( warwick et al . similar conclusions are reached for the snrs g0.9 + 0.1 ( porquet et al . 2003 ) and 3c 58 ( bocchino et al . 2001 ) . while the central compact object of g0.9 + 0.1 has been detected with chandra ( cxou j174722.8@xmath18280915 ) , no obvious emission from a central object is seen in 3c 58 . the snr ic 443 is specially interesting , because it is interacting with a dense molecular cloud rather than with the low density ism . _ xmm - newton _ has resolved several discrete hard x - ray sources , which could be fragments of the snr interacting with the dense cloud ( bocchino & bykov 2003 ) . the previous generation of imaging x - ray satellites ( mainly _ rosat _ and _ asca _ ) have detected a large population of young neutron stars which appear as rotation powered x - ray pulsars . x - ray emission in these systems may arise from a variety of physical processes . in many of them , non - thermal emission from relativistic particles accelerated in the pulsar magnetic field shows a power law spectrum over a broad energy range . the large collecting area of _ xmm - newton _ has provided for the first time spectral information for many of these weak x - ray sources . its spatial resolution has also been necessary to separate them from their surrounding supernova remnants ( snrs ) . preliminary results have been advanced by becker & aschenbach ( 2002 ) , and the publication of many interesting results is expected in the near future . millisecond pulsars were first detected as radio sources ( see lorimer 2001 ) and only detected as x - ray pulsars with _ ( becker & trmper 1993 ) . it is generally believed that millisecond pulsars are `` recycled '' pulsars , which have been spun up to their present high rotational velocities by accretion in a low mass x - ray binary and therefore objects of the highest interest for our understanding of binary evolution ( bhattacharya & van den heuvel 1991 ) . because of this , the recent detections of millisecond pulsars in accreting binaries ( wijnands & van der klis 1998 ) has sparked enormous interest . _ xmm - newton _ observations of two of the currently four known accreting millisecond pulsars have been reported . the first accreting millisecond pulsar sax j1808.4@xmath18365 has been detected in quiescence ( campana et al . 2002 ) and at a slightly higher luminosity ( wijnands 2003 ) . in both cases , its spectrum was dominated by a power law , incompatible with thermal emission from the cooling surface of the neutron star ( heated by accretion ) . a much better spectrum was obtained for the 2.3-ms accreting millisecond pulsar xte j1751 - 305 ( miller et al . however , no spectral features were detected . the apparently isolated 4.86-ms pulsar psr j0030 + 0451 was detected by _ xmm - newton _ , which was able to measure its pulsed fraction . its spectrum could be fitted by a two component model ( either blackbody plus power law or two blackbodies ) or a broken power law ( becker & aschenbach 2002 ) . current models accounting for the spectra of millisecond pulsars predict that psr j0030 + 0451 should be detectable at optical wavelengths . however , a very deep search in its _ xmm - newton _ error circle conducted with the vlt has failed to detect any source down to b=27.3 , casting into doubt such models ( koptsevich et al . 2003 ) . observations of isolated neutron stars ( or neutron stars in quiescent x - ray binaries ) are extremely important in fundamental physics , as thermal emission from the surface of a neutron star carries signatures of its gravitational field , which may be used to infer its mass and radius . detection of absorption lines corresponding to elements on the neutron star atmosphere and measurement of their gravitational redshift would provide rather accurate data ( e.g. , zel & psaltis 2003).from the gravitational redshift at the surface of the neutron star , the ratio between its mass and radius may be measured , providing a very strong constraint on neutron star models . such models give physics an experimentally testable handle on properties of matter at ( supra-)nuclear densities . unfortunately , observations so far have not been very successful at detecting these atmospheric features.the first _ xmm - newton _ observation of an isolated neutron star ( rx j0720.4 - 3125 ) yielded no spectral features on top of the black body continuum in a 62.5 ks exposure ( paerels et al . also , the isolated neutron star rx j1856.5@xmath183754 was observed for 57 ks with _ xmm - newton _ without revealing any spectral feature . an extremely long ( 505 ks ) exposure was then performed with _ chandra _ , also failing to detect any feature ( burwitz et al . this result is against predictions by most current model atmospheres for neutron stars . the spectrum of rbs 1223 does show a very broad absorption line , but this is interpreted as a cyclotron feature caused by electrons moving in its very large magnetic field of @xmath22 ( haberl et al . no lines have been seen in the spectra of several other isolated neutron stars observed with _ chandra _ or the anomalous x - ray pulsar 1e1048.1@xmath185937 observed with _ xmm - newton _ ( tiengo et al . the only secure detection is that of two absorption features , discovered with _ , in the isolated neutron star 1e 1207@xmath185209 . _ xmm - newton _ observations have shown a phase dependence in at least one of these features ( mereghetti et al . 2002b ) , but their identification is still not clear . absorption features from the surface of a neutron star have likely been identified during a burst from the lmxb exo 0748@xmath18676 ( cottam et al . if the identification of the observed features with fexxvi and xxv and oviii lines is correct , they imply a @xmath23 gravitational redshift . as indicated above , x - ray emission from normal galaxies arises from their populations of x - ray binaries . by comparing the populations of different galaxies , we are able to obtain information about their star forming histories ( e.g. , grim et al . 2003 ) . in order to resolve the populations , imaging capabilities are necessary . because of this , thorough exploration of the most nearby galaxies , the magellanic clouds , started only with _ einstein _ and flourished with _ rosat _ ( e.g. , sasaki et al . 2000 ; haberl et al . _ rosat _ also allowed the study of other nearby galaxies such as m31 ( supper et al . 2001 ) or m33 ( haberl & pietsch 2001 ) . the potentialities of _ xmm - newton _ for this sort of study are enormous . deep pointings of the magellanic clouds result in the detection of a wealth of x - ray sources , many of which are accreting binaries . the sensitivity of _ xmm - newton _ permits the detection of effectively any active xrb . a pointing at the north of the lmc resulted in the detection of 150 discrete sources , among which a newly discovered hmxb , a new likely sss and several new snrs ( haberl et al . 2003 ) . a pointing to the smc resulted in the detection of the pulse periods of two previous candidates to hmxbs and the discovery of two new candidates ( sasaki et al . the wide field allows a large number of sources to be observed at the same time , considerably improving our knowledge of hmxb populations . [ haberl ] _ xmm - newton _ can also extend this sort of work to more distant galaxies . pointings to m 31 have allowed the detection of sources down to a luminosity of @xmath24erg s@xmath25 ( osborne et al . 2001 ) , resulting in the characterisation of the whole xrb population . as expected , the bulge population is dominated by bright lmxbs , while the disk population appears to be composed mostly of young hmxbs ( trudolyubov et al . . the temporal evolution of some bright transients has been studied with both _ xmm - newton _ and _ chandra _ ( trudolyubov et al . 2001 ) . among them , a bright lmxb has been detected in a globular cluster of the m 31 system ( trudolyubov et al . all this work suggests an overall galactic xrb population similar to that of the milky way . even further , _ xmm - newton _ has identified the first eclipsing x - ray binary outside the local group ( pietsch et al . this is rx j004717.4@xmath18251811 in the nearby starburst galaxy ngc 253 . one of the most interesting problems in extragalactic x - ray astrophysics is the existence of ultraluminous x - ray sources ( ulxs ) . these sources are variable on short timescales , but their luminosities are typically much higher than those observed in galactic and magellanic cloud xrbs , approaching in many cases @xmath26erg s@xmath25 . as this luminosity is much higher than the eddington limit for a neutron star or a low - mass black hole , it seems to require massive black holes with masses of the order of @xmath27 ( generally called intermediate - mass black holes , as their masses are between those of stellar - mass black holes and the supermassive objects at the heart of galaxies ) . severe problems are found when trying to understand the formation mechanism of such black holes , leading to the thought that perhaps ulxs can be explained as normal xrbs in which the x - ray emission is beamed towards the observer , requiring thus much lower fluxes ( king et al . 2001 ) . a very important result obtained with _ xmm - newton _ has been the discovery of quasi - periodic modulation in the flux from an ulx in m82 ( strohmayer & mushotzky 2003 ) . such modulation must arise in the immediate vicinity of the compact object and represents evidence against beaming . unfortunately , because of the relatively low spatial resolution , the identification of the ulx is not completely certain and a simultaneous observation with _ seems desirable . further evidence supporting intermediate - mass black holes has been collected from analysis of the spectra of two ulxs in the nearby spiral galaxy ngc 1313 . the blackbody temperatures inferred from the fits are a factor of ten lower than those of typical sxts , suggesting rather more massive black holes ( miller et al . one of the very first discoveries of x - ray astronomy was that clusters of galaxies are strong x - ray emitters ( see sarazin 1986 for an early review and mushotzky 2001 for a much more up - to - date compilation ) . their x - ray spectrum is well fitted by plasma emission at a temperature of @xmath28 , which includes thermal bremsstrahlung and line emission , most notably the fe k emission line at @xmath29 . the inferred fe abundance is about @xmath30 solar , with possible gradients across the cluster , but remarkable homogeneity across the cluster population . clusters ( and groups ) of galaxies are therefore filled with enriched gas ( likely deposited by the mass loss of the member galaxies ) , which appears to be trapped in the cluster potential well . the intracluster medium appears to be close to hydrostatic equilibrium except for the core ( with a few @xmath31 ) , cluster gas is isothermal ( or perhaps with a slowly decaying temperature ) out to the distances where x - ray emission can be detected ( less than the virial radius ) . relaxed clusters often exhibit a _ cooling flow _ phenomenon , whereby the gas in the central part of the cluster is significantly cooler and the density higher probably due to a highly subsonic inflow amounting typically to @xmath32 ( see fabian 1994 for an extensive review ) . one of the best studied clusters with _ xmm - newton _ is a1795 ( @xmath33 ) . this cluster shows a smooth gas density profile along the lines of the @xmath34 profile adopted for many clusters ( arnaud et al . the temperature of the gas is seen to be constant from 0.1 to 0.4 virial radii , but dropping significantly at smaller distances from the cluster centre . this is now seen as a common feature in many clusters , like a496 ( tamura et al . 2001 ) , a1413 ( pratt & arnaud 2002 ) . the possibility of mapping the intracluster gas structure with unprecedent detail is now also opening big questions on its physical state . in their detailed analysis of the @xmath35 a1413 cluster , pratt & arnaud ( 2002 ) argue that the gas departs from hydrostatic equilibrium when approaching the virial radius , as expected from the very long timescales involved . they also find evidence for a cuspy central density profile ( at variance of the usually assumed @xmath34-profile ) . the mass profile of the cluster , as derived under the assumption of hydrostatic equilibrium , is not far from the predictions of standard cold dark matter computations . another important contribution of _ xmm - newton _ to cluster science has been illustrated with the observations of the @xmath36 cluster cl0016 + 16 , a working horse for the use of the sunyaev - zeldovich effect to determine accurate values of the hubble constant . worrall & birkinshaw ( 2003 ) have been able to derive accurate values for the gas temperature ( to within 2.5% ) and of the emission measure ( electron density integrated along the line of sight ) to within a similar accuracy . the subsequent revised value of @xmath37 is now in agreement with the latest determinations using the _ hubble space telescope _ and the cosmic microwave background power spectrum obtained by wmap . for many years it has been thought that clusters build up from smaller structures , as predicted by popular cosmological scenarios . for instance , evidence for hierarchical merging in the coma cluster was provided by _ observations of x - ray emission of a merging sub - structure ( white , briel & henry 1993 ) . with its much improved sensitivity , _ xmm - newton _ is able to map intracluster gas down to very low surface brightness limits and therefore to strengthen or disprove the merging hypothesis . indeed , the very first epic images of the coma cluster ( briel et al . 2001 ) confirm the presence of the probably infalling lump from the sw ( around ngc 4839 ) already detected by _ rosat _ , but they also find a further lump probably ahead of this one on its infall into the coma cluster . n - body simulations do predict indeed that the infall into large clusters of galaxies proceeds along filamentary structures ! the structure of the gas around ngc 4839 does show clear traces of being into its first infall into coma ( neumann et al . the tail of the x - ray emission is very hot ( @xmath38 ) and indeed hotter than the temperature of the galaxy itself . together with a displacement of the x - ray emission with respect to the galaxy ngc 4839 ( well placed at the centre of the optical group ) this provides confirmation of the subgroup gas being ram pressure stripped in its infall into the coma cluster . evidence for filamentary structures infalling onto clusters is also seen in _ xmm - newton _ observations of the cluster a 85 ( durret et al 2003 ) . early _ rosat _ reports on a large @xmath39 filamentary structure towards the south of the central cd galaxy of a 85 are confirmed , together with a determination of the gas temperature of around @xmath40 . this suggests that the filamentary structure is made of a chain of several galaxy groups . further evidence in favour of clusters being built by merging blocks comes from observations of distant clusters . de filippis et al . ( 2003 ) observed the cluster cl 0939 + 4713 at @xmath41 and showed that it exhibits significant gas structure , with the gas in between the two dominant galaxies being the hottest . this implies a largely non - relaxed state with a likely collision between the two merging blocks within a few million years . hashimoto et al . ( 2002 ) observed the very distant cluster rx j1053.7 + 5735 at @xmath42 and also detected a double structure suggestive of a merger . therefore , _ xmm - newton _ has brought evidence for merging activity in clusters of galaxies , particularly at significant redshifts . assessing the incidence of mergers in the building up of galaxy clusters ( there are examples of dynamically relaxed clusters at high redshift , see arnaud et al . 2002 , for observations of rx j1120.1 + 4318 at @xmath43 ) has just started but will ultimately provide the most direct evidence on how large - scale structure forms in the universe . perhaps one of the most unexpected discoveries by _ xmm - newton _ has been the challenge to the standard picture of the cooling flow phenomenon . the cooling time in the centres of galaxy clusters is significantly smaller than the age of the universe , and therefore that gas will cool if no other phenomena prevent it ( see , e.g. , fabian 1994 for a comprehensive review ) . maintaining hydrostatic equilibrium with the hotter outer gas implies that gas should be steadily flowing inwards , increasing the density towards the cluster centre . mass deposition rates are of the order of @xmath44 , which integrated over a hubble time result in a substantial contribution to the mass of a large galaxy . cooling flows manifest themselves as a peaked surface emissivity at the centre of the cluster , where the x - ray spectrum shows cooler gas . they occur almost invariably in cd clusters ( i.e. , those dominated by a large central - dominant early type galaxy ) , but very rarely happen in more non - relaxed structures , possibly due to the disruption of cooling flows by mergers . [ m87 ] previous to the operation of _ chandra _ and _ xmm - newton _ there were doubts about the physics governing cooling flows . conduction need to be suppressed to prevent the gas from thermalising in the hotter cluster environment , but lower temperature gas was unambiguously seen . the most intriguing question was ( and still is ) where does all this gas go . cd galaxies present at the centres of cooling flow clusters are red and , although they do form some stars ( see the extensive work by cardiel 1999 on star formation in cooling flow clusters ) , they certainly do not form @xmath45 _ normal _ stars every year . a high pressure ambient as that present in cooling flows could favour the formation of low - mass stars . a major _ xmm - newton _ result has been to show that the gas in cooling flow clusters does not cool below a temperature which is 1/2 to 1/3 of the outer intracluster medium temperature ( see , e.g. , kaastra et al . 2001 , peterson et al . 2001 , tamura et al . 2001 , peterson et al . 2003 , sakelliou et al . cluster gas at temperatures @xmath46 kev should cool mostly through fe l lines , which should be very strong . instead they are almost absent of the _ xmm - newton _ rgs spectra . the amount of gas that cools is probably 1/10 of the gas expected from simple cooling flow models ( fabian & allen 2003 ) , suggesting a much modest role of cooling flows in galaxy formation ( see also molendi & pizzolatto 2001 ) . the reasons for that are still unclear . both conduction and heating by the central source ( usually a powerful agn ) have been revived . conduction should probably be decreased by the presence of magnetic fields , as a temperature gradient is indeed seen . heating is also non - trivial , as it should deposit heat uniformly over a large volume , but on the contrary _ chandra _ images show lots of structure in cluster cores . a further complicating point is that , despite the fact that the amount of gas seen at temperatures of millions of degrees is much smaller than predicted by the simplest cooling flow models , _ fuse _ observations reveal in some cooling flow clusters large ovi emission lines ( oegerle et al . 2001 , bregman 2003 , priv comm ) . the amount of gas cooling at a few @xmath47k is paradoxically in agreement with the _ cooling flow _ models , but the few million degree gas is missing . there is clearly much to learn in the coming years from this phenomenon . active galactic nuclei ( agn ) are the most populous class of x - ray sources in the universe , particularly at high galactic latitude . the large variety of agn manifestations results in distinctive x - ray emission properties . radio - loud agns were early claimed to have flatter x - ray spectra than radio - quiet ones ( wilkes & elvis 1987 ) , with flat spectrum radiosources having an even flatter x - ray spectrum ( canizares & white 1989 ) . this is now understood in terms of radio - loud agns being absorbed by cold gas ( sambruna et al . 1999 ) , plus relativistic beaming affecting the x - ray emission in core - dominated radio sources . radio - quiet type 1 agn ( seyfert 1 galaxies and qsos ) where optical broad emission lines are seen and type 2 agn ( seyfert 2s ) where these broad components are not seen in unpolarized light , should have different x - ray properties . indeed , the simplest version of the agn unified scheme ( antonucci 1993 ) , where the type 1/type 2 dichotomy results from different viewing directions of the same ( or very similar ) central engine , predicts that x - rays emerging from the nucleus would be directly seen in type 1 agns but absorbed / scattered in type 2 agns . the standard agn model , i.e. , that of a supermassive black hole fed by a geometrically thin accretion disk and collimated jets along the rotation axis , has found some of its best support in the x - ray observations . hard x - ray photons at a rate of @xmath48 are very difficult to produce by other means in a galactic centre . additionally , some spectral features , as a broad relativistic fe k@xmath1 emission discussed below , are distinctive tracers of accretion onto massive compact objects . x - ray emission from agn is characterised by an underlying power law with photon index @xmath49 ( on the steep side for radio - quiet agn ) , which probably rolls over at energies @xmath50 ( see the review by mushotzky et al . 1993 summarizing the situation pre-_xmm - newton _ ) . for radio - quiet objects and those which are radio - loud but the beaming towards the observer is not important , this power law is interpreted as the compton up - scattering of the uv photons produced in the accretion disk by the disk s relativistic electron atmosphere . an fe k@xmath1 emission line at 6.4 - 6.7 kev is the most prominent emission feature , along with a high - energy ` bump ' at @xmath51 ( pounds et al . 1990 ) , both of them probably arising from reflection of the x - rays in some thick material . for seyfert 1 galaxies the equivalent width of the fe line lies in the range 100 - 300 ev , while for seyfert 2 galaxies the emission line is much stronger ( up to 1 kev of equivalent width ) , in particular in compton - thick seyfert 2s . the most striking feature of the fe line is that it is very broad ( inferred doppler velocities up to @xmath52 ) . in the best studied case of the seyfert 1 galaxy mcg-6 - 30 - 15 it exhibits a relativistically broadened shape interpreted in terms of doppler shifts produced by the disk rotation , aberration and gravitational redshift in the red horn due to the proximity to the black hole s event horizon ( tanaka et al . similar shapes were observed in other seyfert 1 galaxies , suggesting that this is a relatively general feature of type 1 agns . in the best studied seyfert 2 galaxies ( e.g. ngc 1068 ) , _ asca _ showed a more complex emission line structure ( iwasawa et al . 1997 ) . agn often exhibit photoelectric absorption features in excess of those produced by cold gas in our galaxy . radio - loud agns at high redshifts , often exhibit large cold ( neutral ) absorbing gas columns ( cappi et al . 1997 ) but in radio - quiet agn the situation is more complex . in this case , type 1 agn appear to have little neutral absorbing gas ( nandra & pounds 1994 ) , but they often exhibit signs of ionised gas along the line of sight in the form of absorption edges , most notably ovii k at 0.74 kev and oviii k at 0.87 kev ( reynolds 1997 , george et al . 1998 ) . on the contrary seyfert 2 agn usually display large absorbing columns of neutral gas ( smith & done 1996 ) , which is attributed to the `` dusty torus '' that the unified agn scheme predicts to hide the broad line region . the average good correlation between optical spectral properties of radio quiet agn and the amount of neutral absorbing gas along the line of sight has been clearly shown in risaliti et al . ( 1999 ) on the basis of an [ oiii]-selected sample of agn ( the [ oiii ] emissivity is believed to originate further away than any obscuring material in agn , and therefore is regarded as an isotropic measure of the true intrinsic agn luminosity ) . _ chandra _ and to a larger extent _ xmm - newton _ have provided a qualitative leap forward in the study of fe k lines , due to the large increase ( in particular for _ xmm - newton _ ) of effective area at @xmath53 with respect to previous missions while preserving or improving the spectral resolution . the best - studied seyfert 1 galaxy mcg6 - 30 - 15 has indeed been the subject of several long _ xmm - newton _ observations . fabian et al . ( 2002 ) discuss a 300 ks observation , where they confirm the existence of a very broad fe line feature . a careful physical modeling of this feature shows that the line emissivity comes in two parts : the blue part of the line arises from the outer accretion disk ( radius @xmath54 , where @xmath55 is the gravitational radius of the black hole with mass @xmath56 ) , but to account for the very extended red tail displayed by the line ( extending down to 4 kev ) a very concentrated emission at much lower radii is required . this unambiguously confirms early claims based on _ asca _ that reflection features arise from radii as close as @xmath57 from the black hole , requiring a rapidly rotating ( kerr ) hole . in fact wilms ( 2001 ) caught the same source in the `` low '' state where the red wing of the fe line dominates , suggesting extraction of energy from the spinning black hole through magnetic fields . in any case it is clear that the fe line diagnostic of mcg6 - 30 - 15 can only be understood in the context of reflection in an accretion disk around a kerr black hole . observations of many other seyfert galaxies have shown that the profile of the fe line varies very much from source to source . in general , the reflection fe line arises either in the inner part of the accretion disk , where it is likely to be highly ionised and broad , and/or at much more distant regions ( perhaps the `` dusty torus '' ) where it betrays mostly neutral fe with a narrow line . prototypical cases showing both a narrow component at 6.4 kev and a broad component centered at 6.7 kev are mkn 205 ( reeves et al . 2001 ) and ngc 5506 ( matt et al . 2001 ) . [ mcg ] interestingly , one of the cases claimed to have a broad fe line with _ asca _ data , ngc 5548 , does only show a neutral narrow line in _ xmm - newton _ observations ( pounds et al . although the fe line appears to vary less than the continuum in some of the best - studied seyferts ( such as mcg6 - 30 - 15 ) , there are reports of varying fe lines . guainazzi ( 2003 ) studies non - simultaneous _ asca _ , _ xmm - newton _ and _ bepposax _ observations of eso198-g24 , and he reports a broad component appearing and then fading away along with a narrow component . it must also be stressed that the `` holy grail '' of using reverberation mapping ( i.e. , the response of reflected components , such as the fe line , to continuum variations ) to measure the mass of the black holes involved in agn , has turned out to be rather complex . the fe line intensity does not trivially follow the continuum variations and therefore much work is needed before this technique can be applied . a very lively debate has arisen since the launch of _ xmm - newton _ on whether or not there are weak , relativistically broadened emission lines corresponding to c , n and o at soft x - ray energies . this stems from the use of the grating spectrographs both in _ xmm - newton _ and _ chandra _ , which provide spectral resolutions of the order of 200 - 500 . branduardi - raymont et al . ( 2001 ) were the first to suggest the presence of these features in the rgs spectra of mcg-6 - 30 - 15 and mkn 766 , as an alternative to a standard warm absorber picture . using a _ chandra _ observation of mcg-6 - 30 - 15 , lee et al . ( 2001 ) identified a number of spectral features that were expected from a warm absorber , and even some of them only expected in fe atoms trapped in dust grains , supporting the dusty warm absorber picture and no evidence for broad emission lines . mason et al . ( 2003 ) have recently reported an analysis of rgs data on mkn 766 , where it is shown that broad relativistic emission lines of c , n and o give a better fit to the data than a dusty warm absorber . as of today , higher signal to noise _ xmm - newton _ data over a larger bandpass appear to favour the presence of relativistic soft x - ray emission lines , and higher spectral resolution _ chandra _ data appear to favour the dusty warm absorber interpretation . but the story continues @xmath58 the higher spectral resolution provided by rgs has enabled the possibility of studying the ionized absorbers in agn with much more detail . sako et al . ( 2001 ) conducted the first of such detailed analyses on the qso iras 13349 + 2438 , which was already suspected from the combined _ rosat _ , _ asca _ and optical data to have a complex absorption structure along the line of sight . the rgs spectrum shows a large number of absorption lines of c , n , o , ne with one and two electrons , as well as a number of fe lines in various ionisation states . most remarkably , an unresolved transition array ( uta ) of inner - shell lines of fe m is detected for the first time at x - ray wavelengths , which might be confused , at the lower resolution of non - dispersive ccd imaging spectrographs , with an absorption edge . absorption lines of fe are detected in fe vii - fe xii and in fe xvii - fe xx , but not in the intermediate ionisation states fe xiii - fe xvi . this suggests the existence of two distinct absorbers , a low ionisation one , which happens to be outflowing with a velocity @xmath59 and a high ionisation absorber basically at rest with the qso . the low ionization outflowing absorber implies ( for normal dust to gas ratio ) a reddening coincident with the optical one , but the high ionisation absorber , with a much larger column density , does not appear to have any dust . [ iras ] similar deep studies have been conducted on other targets ( e.g. , blustin et al . 2002 ) , revealing almost invariably the presence of utas , and a complex ( at least two - phase ) ionisation structure for the absorbers . it must be stressed that we are likely probing the gas in between the broad - line region and the `` obscuring torus '' in a manner that no other observational strategy could do . the existence of a cosmic x - ray background ( cxrb ) was among the very first discoveries of x - ray astronomy ( giacconi et al . its spectrum was well measured by the heao-1 mission ( marshall et al . 1980 , gruber 1992 ) and fits remarkably well a thermal bremsstrahlung model at a temperature of @xmath60 . at galactic latitudes @xmath61 the cxrb is remarkably isotropic claiming for a true cosmological origin ( see fabian & barcons 1992 for a review ) . most of the energy density in the cxrb resides at @xmath60 , an energy which is not accessible to sensitive enough instrumentation in the current era . below 1 kev the galaxy ( and probably a local hot bubble ) dominates the x - ray intensity , and the extragalactic component is shielded by the interstellar medium , making it difficult to measure . in spite of this , deep _ rosat _ observations resolved @xmath62 of the cxrb in the 1 - 2 kev ( hasinger et al . 1998 ) while _ chandra _ has resolved practically 90% of this and @xmath63 of the 2 - 7 kev cxrb ( mushotzky et al . 2000 ) . the lack of spectral distortions of the cosmic microwave background ruled out the presence of substantial amounts of intergalactic gas at the very high temperatures needed to produce the cxrb through thermal bremsstrahlung ( barcons et al . 1991 ) . finding the sources that produce the cxrb , their astrophysical nature , their cosmic evolution , spatial clustering and so on has been since then the main objective of cxrb studies . the most popular models for the cxrb make use of the unified model for agn , and assume a mixture of unabsorbed ( presumably type 1 ) and absorbed ( presumably type 2 ) objects as a function of redshift @xmath64 ( comastri et al . 1995 , gilli et al . 2001 ) . medium sensitivity ( mason et al . 2000 ) and deep ( lehmann et al . 2002 ) surveys carried out with _ rosat _ revealed that at soft x - ray energies the x - ray sky is dominated by agns , most of which are type 1 seyferts and qsos . a small fraction of the agn were type 2 and other narrow emission line galaxies ( e.g. , starburst galaxies ) . however , due to its bandpass limited to soft x - ray photons _ rosat _ missed most of the sources absorbed by h columns in excess to @xmath65 . the large field of view of the epic cameras ( @xmath66 in diameter ) makes _ xmm - newton _ a very powerful tool to conduct x - ray surveys . the survey science centre ( ssc ) is carrying out a series of serendipitous surveys in order to characterise the content of the x - ray sky at different depths . at high galactic latitudes , 3 samples of @xmath67 sources each , selected at 0.5 - 4.5 kev fluxes of @xmath68 , @xmath69 and @xmath70 will provide a major handle on this task . a number of surveys are being conducted besides the ssc ones , most of which are discussed in the proceedings of the workshop _ x - ray surveys , in the light of the new observatories _ held in santander during 4 - 6 september 2002 ( astronomische nachrichten , vol 324 , issues 1 and 2 ) . the axis programme is providing a substantial leap forward in the _ xmm - newton _ ssc medium sensitivity survey . this is a flux limited survey at @xmath71 , in the 0.5 - 4.5 kev band , with a source density of @xmath72 , which has the goal of covering @xmath73 both in the north and in the south . preliminary results were presented in barcons et al . ( 2002 ) and an update can be found in barcons et al . . [ lh ] the deepest x - ray survey conducted by _ xmm - newton _ so far is in the lockman hole area , with a total exposure approaching 1000 ks ( adding the payload verification , ao-1 and ao-2 data ) . the scope of this survey is not only to detect the faintest sources , particularly at hard x - ray energies where confusion is still unimportant and _ xmm - newton _ is 5 - 10 times more efficient than _ chandra _ , but more importantly to obtain x - ray spectra of the faintest sources . a preliminary account , based on 100 ks data , is presented in hasinger et al . ( 2001 ) . _ xmm - newton _ is also providing evidence that the one to one association of type 1 agns with unabsorbed sources and type 2 agns with absorbed sources fails clearly in some cases . for example barcons et al . ( 2003b ) have shown that the h1320 + 551 z=0.068 seyfert 1.9 galaxy ( and therefore with a broad line region putatively obscured ) does not show any signs of neutral or ionised gas along the line of sight . this is in direct contradiction to the unified agn model . on the opposite side , type 1 qsos are found which display large absorbing columns of neutral gas ( e.g. , georgantopoulos et al . 2003 ) , while none or very little was expected . a possible explanation to this is that x - ray absorption occurs very close to the central engine where dust just sublimates and therefore the optical spectrum is unaffected by the surrounding material . this might be a non - isolated situation , as explained by mateos et al . ( 2003 ) in an analysis of the x - ray spectra of xms sources , where a fraction of type 1 agns do require absorbing columns along the line of sight . if anything , the deep observation of hasinger et al . ( 2001 ) demonstrates that the basic picture for the origin of the cxrb , based on absorbed sources , is roughly correct . indeed , some @xmath74 of the faint sources detected show some hint of absorption in their x - ray colours , in agreement with the general expectations . quantitatively , however , the situation is still far from clear . one of the most striking results obtained from the optical identification of _ chandra _ deep surveys is that the redshift distribution of type 2 agn is peaked at significantly lower redshifts ( @xmath75 ) than that of type 1 agn ( @xmath76 ) ( hasinger 2002 ) . similarly , in the xms all type 2 agns are concentrated at low redshifts , while the type 1s reach @xmath77 . this probably means that the models for cxrb need some revision . following this , franceschini , braito & fadda ( 2002 ) have suggested that type 1 agn arise in the rare highest peaks of the density field in the universe , but that type 2 agn result from the evolution and merging of the very massive galaxies , a process that occurs later in the cosmic history . a still intriguing issue raised by the new observatories and studied in detail by _ xmm - newton _ is the existence of optically inactive but x - ray luminous galaxies . such objects were first highlighted in a _ _ deep survey by mushotzky et al . ( 2000 ) , as early type galaxies with a large x - ray to optical ratio . comastri et al . ( 2002 ) conducted a multi - wavelength study of one of these objects discovered in their hellas2xmm survey , which shows no optical or infrared emission lines , but does show a rather hard x - ray spectrum . they concluded that the counterpart galaxy at @xmath78 hosts a completely obscured active nucleus . severgnini et al . ( 2003 ) have recently reported on the study of 3 of these objects discovered in the ssc bright source survey . their x - ray properties are varied , but in all cases reflecting clearly the presence of an x - ray luminous agn . in that work it is shown that the emission lines related to the more or less obscured agn are diluted with the galaxy light and therefore very weak and inconspicuous in some cases . all this appears to support previous suspicions that most , if not all , galaxies host a black hole in their center , many of them active but obscured . obscured accretion is probably 3 times more frequent in the universe than unobscured accretion , if the cxrb models are roughly correct . a detailed accounting on the black hole density can then be inferred by assuming a radiative efficiency of the black holes ( typically 10% ) and compared to other independent estimates of the local black hole density . fabian ( 2003 ) concludes that the estimate based on the cxrb and with what is known from the redshift distribution of obscured sources , converges towards @xmath79 , a value which is consistent with local estimates . obscured accretion onto black holes is therefore important , although the `` hidden '' black hole mass is of the same order than that seen in unobscured agn . _ xmm - newton _ is making a big effort in observing gamma ray bursts ( grb ) following alerts provided by other higher - energy observatories . the absolute minimum reaction time reached is 6 - 8 hours after the alert . in all observations ( conducted as target of opportunity observations , and therefore immediately public ) the grb afterglow has been detected and much has been learnt on this phenomenon from the x - ray data . perhaps the most interesting case is that of grb011211 ( identified to occur in a galaxy of @xmath80 at @xmath81 ) observed by _ xmm - newton _ for 27 ks , 11 hours after its detection by _ bepposax_. during the first 10 ks of that observation , the epic x - ray spectrum showed emission lines which reeves et al . ( 2002 ) identified as mg xi , si xiv , sxvi , ar xviii and ca xx but at a significantly blueshifted velocity @xmath82 ( the detection confidence of these lines has been challenged by rutledge & sako 2003 though ) . if the reeves et al . ( 2002 ) interpretation is correct , then this would provide direct evidence for outflowing material , favouring the model of the collapse of a massive star for grbs . the work reported herein is based partly on observations obtained with _ xmm - newton _ , an esa science mission with instruments and contributions directly funded by esa member states and the usa ( nasa ) . we are grateful to fred jansen , christian motch , richard mushotzky and an anonymous referee for comments on an early version that resulted in substantial improvements . xb acknowledges partial financial support by the ministerio de ciencia y tecnologa ( mcyt ) , under grant aya2000 - 1690 . in is a researcher of the programme _ ramn y cajal _ , funded by the university of alicante and the mcyt . he acknowledges partial financial support by the mcyt under grants esp2001 - 4541-pe and esp2002 - 04124-c03 - 03 . antonucci , r.r.j . 1993 , unified models for active galactic nuclei and quasars , araa , 31 , 473 becker , w. , & aschenbach , b. 2003 , x - ray observations of neutron stars and pulsars : first results from _ xmm - newton _ , in becker , w. et al . eds . , proceedings of 270th we - heraeus seminar on neutron stars , pulsars and supernova remnants , mpe report 278 , p. 64 ( astro - ph/0208466 ) jimenez - garate , m.a . et al . 2002 , high - resolution x - ray spectroscopy of hercules x-1 with the _ xmm - newton _ reflection grating spectrometer : cno element abundance measurements and density diagnostics of a photoionized plasma , apj 578 , 391 peterson , j.r . 2003 , high resolution spectroscopy of 14 cooling - flow clusters of galaxies using the reflection grating spectrometers on _ xmm - newton _ , in _ new visions of the universe in the xmm - newton and chandra era _ , in the press ( astro - ph/0202108 )
in this paper , we study the behavior of different physical quantities at the resonances generated by the maximum of the potential of a semiclassical schrdinger operator . in particular , we show quantitatively to what extent the presence of these resonances drives the behavior of the scattering amplitude and of the schrdinger group . the resonances generated by the maximum point of the potential ( usually called barrier - top resonances ) have been studied by briet , combes and duclos @xcite and sjstrand @xcite . these authors have given a precise description of the resonances in any disc of size @xmath0 centered at the maximum of the potential . in particular , they have shown that the resonances lie at distance of order @xmath0 from the real axis , which is in very strong contrast to the case of shape resonances ( the well in the island case ) , with exponentially small imaginary part ( see helffer and sjstrand @xcite ) . the description of resonances in larger discs of size @xmath1 , @xmath20,1]$ ] has been obtained by kaidi and kerdelhu @xcite under a diophantine condition . for small discs of size one , this question has been treated in the one dimensional case by the third author @xcite with the complex wkb method . in the two dimensional case , the resonances in discs of size one have also been considered by hitrik , sjstrand and v ngc @xcite ( see also the references in this paper ) . here , we consider only the resonances at distance @xmath0 of the maximum of the potential and we recall their precise localization in section [ k5 ] . resonances can be defined as the poles of the meromorphic continuation of the cut - off resolvent ( see e.g. hunziker @xcite ) . the generalized spectral projection associated to a resonance is defined as the residue of the resolvent at this pole : @xmath3 as an operator from @xmath4 to @xmath5 . if @xmath6 were an isolated eigenvalue , this formula would give the usual spectral projection . many physical quantities can be expressed in terms of these generalized spectral projections . in the case of shape resonances , their semiclassical expansion has been computed by helffer and sjstrand in @xcite . in section [ a1 ] below , we obtain the semiclassical expansion of the generalized spectral projection for barrier - top resonances . since the resonances in the present case have a much larger imaginary part , our result is very different from that of the shape resonance case . resonances appear also in scattering theory ( they are called scattering poles in this context ) . in @xcite , lax and phillips have shown that they coincide with the poles of the meromorphic extension of the scattering amplitude . this result , proved for the wave equation in the exterior of a compact obstacle , was extended by grard and martinez @xcite to the long range case for the schrdinger equation ( see also the references in this paper for earlier works ) . for shape resonances , the residue of the scattering amplitude was calculated in the semiclassical limit by nakamura @xcite , lahmar - benbernou @xcite and lahmar - benbernou and martinez @xcite . more generally , upper bounds on the residues of the scattering amplitude have been obtained by stefanov @xcite ( in the compact support case ) and michel @xcite ( in the long range case ) for resonances very close to the real axis . in section [ k2 ] , we give the semiclassical expansion of the residues of the scattering amplitude for barrier - top resonances and we will see that these upper bounds do not hold in the present setting . it is commonly believed that resonances play also a crucial role in quantum dynamics . indeed , it is sometimes possible to describe the long time evolution of the cut - off propagator ( for example , the schrdinger or wave group ) in term of the resonances . typically , if the resonances are simple , the propagator @xmath7 truncated by @xmath8 satisfies @xmath9 here , @xmath10 is the generalized spectral projection defined previously . such a formula generalizes the poisson formula , valid for the operators with discrete spectrum . the resonance expansion of the wave group was first obtained by lax and phillips @xcite in the exterior of a star - shaped obstacle . this result has been generalized , using various techniques , to different non trapping situations ( see e.g. vanberg @xcite and the references of the second edition of the book @xcite ) . the trapping situations have been treated by tang and zworski @xcite and burq and zworski @xcite for very large times . on the other hand , the time evolution of the quasiresonant states ( sorts of quasimodes ) has been studied by grard and sigal @xcite . a specific study of the schrdinger group for the shape resonances created by a well in a island has been made by nakamura , stefanov and zworski @xcite . there are also some works concerning the situation of a hyperbolic trapped set . we refer to the work of christiansen and zworski @xcite for the wave equation on the modular surface and on the hyperbolic cylinder , and to the work of guillarmou and naud @xcite for the wave equation on convex co - compact hyperbolic manifolds . section [ k3 ] is devoted to the computation of the asymptotic behavior for large time of the schrdinger group localized in energies close to the maximum of the potential . for the proof of the different results of this paper , we use an estimate of the distorted resolvent around the resonances , polynomial with respect to @xmath11 . indeed , such a bound allows to apply the semiclassical microlocal calculus . this estimate is established in section [ k4 ] . to prove it , we proceed as in @xcite and use the method developed by martinez @xcite , sjstrand @xcite and tang and zworski @xcite . similar bounds around the resonances are already known in various situations ( see e.g. grard @xcite for two strictly convex obstacles , michel and the first author @xcite in the one dimensional case ) . note that , in our setting , a limiting absorption principle have been proved in @xcite . we consider the semiclassical schrdinger operator on @xmath12 , @xmath13 , @xmath14 where @xmath15 is a smooth real - valued function . we denote by @xmath16 the associated classical hamiltonian . the vector field @xmath17 is the hamiltonian vector field associated to @xmath18 . integral curves @xmath19 of @xmath20 are called classical trajectories or bicharacteristic curves , and @xmath18 is constant along such curves . the trapped set at energy @xmath21 for @xmath22 is defined as @xmath23 we shall suppose that @xmath15 satisfies the following assumptions [ a1 ] @xmath24 extends holomorphically in the sector @xmath25 for some @xmath26 . moreover @xmath27 as @xmath28 in @xmath29 . [ a2 ] @xmath15 has a non - degenerate maximum at @xmath30 and @xmath31 with @xmath32 and @xmath33 . [ a3 ] the trapped set at energy @xmath34 is @xmath35 . notice that [ a3 ] ensures that @xmath30 is the unique global maximum for @xmath15 . moreover , there exists a pointed neighborhood of @xmath36 in which all the energy levels are non trapping . in the following , @xmath37 denote the strictly increasing sequence of linear combinations over @xmath38 of the @xmath39 s . in particular , @xmath40 and @xmath41 . the linearization @xmath42 at @xmath43 of the hamilton vector field @xmath44 is given by @xmath45 and has eigenvalues @xmath46 . thus @xmath43 is a hyperbolic fixed point for @xmath44 and the stable / unstable manifold theorem gives the existence of a stable incoming lagrangian manifold @xmath47 and a stable outgoing lagrangian manifold @xmath48 characterized by @xmath49 moreover , there exist two smooth functions @xmath50 , defined in a vicinity of @xmath51 , satisfying @xmath52 and such that @xmath53 near @xmath43 . since @xmath22 is a schrdinger operator , we have @xmath54 . under the previous assumptions , the operator @xmath22 is self - adjoint with domain @xmath55 , and we define the set @xmath56 of resonances for @xmath22 as follows ( see @xcite ) . let @xmath57 be a large constant , and let @xmath58 be a smooth vector field , such that @xmath59 for @xmath60 and @xmath61 for @xmath62 . for @xmath63 small enough , we denote @xmath64 the unitary operator defined by @xmath65 for @xmath66 . then the operator @xmath67 is a differential operator with analytic coefficients with respect to @xmath68 , and can be analytically continued to small enough complex values of @xmath68 . for @xmath69 small enough , we denote @xmath70 the spectrum of @xmath71 is discrete in @xmath72 , and the resonances of @xmath22 are by definition the eigenvalues of @xmath71 in @xmath73 . we denote their set by @xmath56 . the multiplicity of a resonance is the rank of the spectral projection @xmath74 where @xmath75 is a small enough closed path around the resonance @xmath6 . the resonances , as well as their multiplicity , do not depend on @xmath76 and @xmath77 . as a matter of fact , the resonances are also the poles of the meromorphic extension from the upper complex half - plane of the resolvent @xmath78 ( see e.g. @xcite ) . in the present setting , sjstrand @xcite has given a precise description of the set of resonances in any disc @xmath79 of center @xmath36 and radius @xmath80 . this result has also been proved simultaneously by briet , combes and duclos @xcite under a slightly stronger hypothesis ( a virial assumption ) . _ [ e17 ] assume [ a1][a3 ] . let @xmath81 be different from @xmath82 for all @xmath83 . then , for @xmath84 small enough , there exists a bijection @xmath85 between the sets @xmath86 and @xmath87 , where @xmath88 such that @xmath89 . _ in particular , the number of resonances in any disk @xmath90 is uniformly bounded with respect to @xmath0 . for @xmath91 , we denote @xmath92 . _ we shall say that @xmath93 is simple if @xmath94 implies @xmath95 . _ _ [ k1 ] if @xmath96 is simple , the corresponding resonance @xmath97 is simple for @xmath0 small enough and proposition 0.3 of @xcite proves that @xmath97 has a complete asymptotic expansion in powers of @xmath0 . _ _ the analyticity of @xmath15 in a full neighborhood of @xmath12 is used only for the localization of the resonances . indeed , if the conclusions of theorem [ e17 ] and remark [ k1 ] hold for @xmath15 smooth and analytic outside of a compact set , then the results of this paper still apply under this weaker assumption . _ the semiclassical pseudodifferential calculus is a tool used throughout this paper , and we fix here some notations . we refer to @xcite for more details . for @xmath98 an order function and @xmath99 , we say that a function @xmath100 is a symbol of class @xmath101 when , for all @xmath102 , @xmath103 if @xmath104 , the semiclassical pseudodifferential operator @xmath105 with symbol @xmath106 is defined by @xmath107 for all @xmath108 . we denote by @xmath109 the space of operators @xmath110 . the rest of this paper is organized as follows . in section [ k4 ] , we prove a resolvent estimate in the complex plane that we use in all the rest of the paper . then , in section [ a1 ] , we compute the spectral projection associated to a resonance . in section [ k2 ] , we give the asymptotic expansion of the residue of the scattering amplitude at a simple resonance for long range potentials . section [ k3 ] is devoted to the computation of the asymptotic behavior for large @xmath111 of the schrdinger group @xmath112 , where the spectral projection appears naturally . at last , we have placed in appendix [ k6 ] some geometrical considerations about hamiltonian curves in a neighborhood of the hyperbolic fixed point , that we need in section [ a1 ] . in this section , we prove a polynomial estimate for the resolvent of the distorted operator @xmath71 around the resonances . this estimate is used throughout the paper to control remainder terms . more precisely , we prove the following result . _ [ aaa ] assume [ a1][a3 ] . there exists @xmath113 such that , for all @xmath81 and @xmath0 small enough , _ @xmath114 the operator @xmath22 has no resonances in @xmath115 + i [ -c h , 0 ] \setminus d ( e_{0 } , 2 c h ) .\ ] ] @xmath116 assume @xmath117 with @xmath118 . then , there exists @xmath119 such that @xmath120 for all @xmath121 + i [ -c h , c h]$ ] . in particular , the previous theorem states that all the resonances in @xmath122 + i [ -c h , 0]$ ] are those given by theorem [ e17 ] . the rest of this section is devoted to the proof of theorem [ aaa ] . we follow the approach of tang and zworski @xcite and we use the constructions of ( * ? ? ? * section 4 ) , where the propagation of singularities through a hyperbolic fixed point is studied , and of ( * ? ? ? * section 3 ) , where a sharp estimate for the weighted resolvent for real energies is given . @xmath124 the distorted operator @xmath71 defined in is a differential operator of order @xmath125 whose symbol @xmath126 satisfies @xmath127 with @xmath128 and @xmath129 we write the taylor expansion of @xmath130 with respect to @xmath76 as @xmath131 for some @xmath132 which vanishes in @xmath133 . notice that @xmath134 so that for @xmath135 small enough , there exists @xmath136 such that @xmath137 for all @xmath138)$ ] with @xmath139 . we want to gain as much ellipticity as we can near @xmath43 . as in ( * section 4 ) , we shall work with a weighted operator , and we start by defining the weights . let @xmath140 and @xmath141 . there exists a symplectic map @xmath142 defined near @xmath143 , with @xmath144 , such that , setting @xmath145 , @xmath146 here @xmath147 is a @xmath148 map from @xmath149 to the space @xmath150 of @xmath151 matrices with real entries such that @xmath152 let @xmath153 be a unitary fourier integral operator microlocally defined near @xmath154 and associated to the canonical transformation @xmath142 . then @xmath155 is a pseudodifferential operator in @xmath156 with a real ( modulo @xmath157 ) symbol @xmath158 , such that @xmath159 let @xmath160 . since the trapped set at energy @xmath34 for @xmath18 is @xmath161 , we recall from @xcite that , for the compact set @xmath162 ) \subset t^ { * } \r^{n}$ ] , there exist @xmath163 and a bounded function @xmath164 such that @xmath165 has compact support and @xmath166 as in @xcite , we set , for @xmath167 to be chosen later , @xmath168 where @xmath169 ) $ ] with @xmath170 on @xmath171 and @xmath172 ) $ ] with @xmath173 $ ] and @xmath174 in a neighborhood of @xmath175 $ ] . we also define functions on the @xmath176 side . we set @xmath177 here @xmath178 is a parameter that will be chosen later on . since we consider the semiclassical regime , we will assume that @xmath179 . moreover , @xmath180 , where @xmath181 is such that @xmath182 near @xmath183 and @xmath184 is such that @xmath185 near @xmath51 in @xmath186 . at last , we choose four cut - off functions @xmath187 such that , setting again @xmath188 , we have @xmath189 the notation @xmath190 means that @xmath191 near the support of @xmath192 . we define the operators @xmath193 for @xmath194 . notice that @xmath195 is acting on functions of @xmath196 , whereas the other operators are acting on functions of @xmath176 . the @xmath197 s are real constants that will be fixed below . then , @xmath198 for some @xmath199 . we define the operator @xmath200 splitting @xmath201 , we write @xmath202 and we compute the symbols of the operators @xmath203 separately . @xmath124 the goal of this part is to prove the following identity . _ [ e41 ] let @xmath123 be the operator defined in . then , @xmath204 _ _ [ e40 ] we will show in the proof of lemma [ e41 ] ( more precisely in ) that the operators @xmath205 and @xmath206 are invertible on @xmath207 and @xmath208 for @xmath209 and @xmath0 small enough . moreover , their inverses are polynomially bounded in @xmath11 . in particular , the resonances of @xmath22 are the poles of @xmath210 and to estimate @xmath211 , it is enough to estimate @xmath210 . _ the rest of this section is devoted to the proof of lemma [ e41 ] . in fact , is close to the equation ( 4.44 ) of @xcite and we will use some identities from @xcite when possible . @xmath212 first we consider @xmath213 . since we can assume that @xmath214 , we have @xmath215 with @xmath216 given , for any @xmath217 , by @xmath218 then again @xmath219 with @xmath220 given , for any @xmath221 , by @xmath222 the @xmath223-th term in is easily seen to be @xmath224 , so that choosing @xmath225 large enough , we conclude that @xmath226 . moreover @xmath227 modulo @xmath228 , and @xmath229 for some @xmath230 with @xmath231 modulo @xmath228 . by egorov s theorem , @xmath232 where @xmath233 verify @xmath234 modulo @xmath228 . moreover , from , we have @xmath235 with @xmath236 and a symbol @xmath237 satisfying @xmath238 modulo @xmath228 . since @xmath239 , we have @xmath240 and we get by pseudodifferential calculus @xmath241 then , using , , and , we obtain @xmath242 the first term in the right hand side of has already been computed in the equations ( 4.15)(4.41 ) of @xcite ( the reader should notice however that the symbol @xmath18 there has to be replaced by @xmath243 here ) . we have @xmath244 on the other hand , since @xmath245 , @xmath246 near the support of @xmath247 and @xmath248 . thus , @xmath249 and then , working as in , we obtain @xmath250 using and collecting and , the identity gives @xmath251 @xmath212 now we consider @xmath252 . as in , we have @xmath253 for some @xmath254 . moreover @xmath255 modulo @xmath228 and @xmath256 since @xmath257 , the pseudodifferential calculus gives @xmath258 . furthermore , using egorov s theorem , we obtain @xmath259 where @xmath260 and @xmath261 modulo @xmath262 . using @xmath263 , the supports of @xmath264 and @xmath265 are disjoint and @xmath266 @xmath212 it remains to study @xmath267 . working as in , we get @xmath268 with @xmath269 and @xmath270 modulo @xmath228 . as in , we have @xmath271 where @xmath272 . now and yield @xmath273 working as in the equation ( 4.43 ) of @xcite , we get @xmath274 combining with the last identity , we finally obtain @xmath275 @xmath212 the same way , one can prove @xmath276 and the same kind of estimate holds for the product the other way round . on the other hand , @xmath277 and @xmath278 . then the two operators @xmath205 and @xmath206 are invertible on @xmath279 for @xmath209 and @xmath0 small enough and they satisfy @xmath280 for some @xmath281 . the same thing can be done on @xmath208 since the operators we consider differ from @xmath282 by compactly supported pseudodifferential operators . this shows remark [ e40 ] . @xmath212 adding , and , we get lemma [ e41 ] @xmath124 let @xmath283 ) $ ] be such that @xmath284 near @xmath51 . we define @xmath285 for some large constant @xmath286 fixed in the following . _ [ e39 ] assume that @xmath26 , @xmath287 and @xmath117 with @xmath118 . denote @xmath288 . choose @xmath289 and fix @xmath290 large enough in this order . then , we have , for @xmath0 small enough , _ @xmath114 for @xmath291 + i [ - 2 c_{0 } h , 2 c_{0 } h ] $ ] and @xmath292 , the operator @xmath293 is invertible and @xmath294 @xmath116 for @xmath291 + i [ - 2 c_{0 } h , 2 c_{0 } h ] $ ] , the operator @xmath295 is invertible and @xmath296 this lemma is similar to proposition 4.1 of @xcite . we will only give the proof of part @xmath116 since the first part can be proved the same way ( using instead of ) . let @xmath297 ) $ ] be such that @xmath298 as usual , we denote @xmath299 . we now recall some ellipticity estimates proved in @xcite by means of grding s inequality and caldern vaillancourt s theorem . from the equations ( 4.50 ) , ( 4.51 ) , ( 4.54 ) , ( 4.55 ) and ( 4.64 ) of @xcite , we have @xmath300 for some @xmath301 which do not depend on @xmath0 , @xmath302 and the @xmath197 s . from and since @xmath117 , @xmath303 let @xmath304)$ ] be such that @xmath305 ) } \prec \omega_{6 } \prec \one_{b ( 0 , 2 r_{1 } ) \cap p^{-1 } ( [ e_{0 } - 3 \varepsilon , e_{0 } + 3 \varepsilon ] ) } .\ ] ] from the definition of @xmath306 , we have @xmath307 using grding s inequality , implies @xmath308 let @xmath309 ; [ 0,1])$ ] with @xmath310 near @xmath311 $ ] . using the functional calculus for pseudodifferential operators , we can write @xmath312 note that the operator @xmath313 is uniformly bounded on @xmath207 . grding s inequality together with give @xmath314 adding , , and and using grding s inequality , we obtain @xmath315 combining the formulas and and the estimates , and , we get @xmath316 now , assume that @xmath317 $ ] and @xmath318 is small . we choose the parameters , in this order , @xmath319 , @xmath320 , @xmath321 then @xmath322 and finally @xmath289 with @xmath323 . then , for @xmath0 small enough , grding s inequality implies @xmath324 on the other hand , from , we have @xmath325 then , @xmath326 for all @xmath0 small enough . summing and @xmath327 times the square of , we obtain @xmath328 for @xmath329 fixed large enough . then , using @xmath330 with @xmath331 , we finally obtain @xmath332 since we can obtain the same way the same estimate for the adjoint @xmath333 , we get the lemma . to prove the part @xmath114 of theorem [ aaa ] ( the resonance free zone ) , we will use in addition the following lemma . _ [ e37 ] assume @xmath334 . under the assumptions of lemma [ e39 ] , we have @xmath335 _ since @xmath336 , gives @xmath337 since the support of @xmath247 does not intersect the support of the symbol of @xmath338 , we obtain @xmath339 moreover , working as in , @xmath340 we now rescale the variables as in @xcite and in the equation ( 4.18 ) of @xcite . we define a unitary transformation @xmath15 on @xmath279 by @xmath341 if @xmath342 is a symbol , then @xmath343 if possible , we will identify in the following an operator with its conjugation by @xmath15 . as in ( * ? ? ? * ( 4.24 ) ) , we define the class of symbols @xmath344 , for an order function @xmath345 , by @xmath346 we refer to the appendix of @xcite for the pseudodifferential calculus in @xmath347 . from ( * ( 4.23 ) ) , we have that @xmath348 . since @xmath349 , we also have @xmath350 . then , the pseudodifferential calculus in @xmath347 implies @xmath351 the same way , ( * ? ? ? * equation ( 4.38 ) ) gives @xmath352 . so , @xmath353 working in @xmath354 , we get @xmath355 since @xmath356 , ( * ? ? ? * equation ( 4.48 ) ) yields that @xmath357 . using caldern vaillancourt s theorem for this operator , we finally obtain @xmath358 the lemma follows from , the choice of @xmath302 in lemma [ e39 ] and the estimates , , and . @xmath124 we first prove that holds for @xmath359 + i [ - c_{0 } h , c_{0 } h ] .\ ] ] here , @xmath360 is any fixed constant . we used a method due to tang and zworski @xcite . for @xmath361 + i [ - 2 c_{0 } h , 2 c_{0 } h]$ ] , the quantity @xmath302 can always be replaced by @xmath323 in lemma [ e39 ] ( see ) . then , @xmath362 is holomorphic in this set and @xmath363 . as usual ( see section 4 of @xcite for instance ) , we can find an operator @xmath364 such that @xmath365 , @xmath366 and such that holds with @xmath367 replaced by @xmath364 . furthermore , thanks to remark [ e40 ] , the resonances coincide with the poles of @xmath210 ( with the same multiplicity ) . mimicking the proof of proposition 4.2 of @xcite or lemma 6.5 of @xcite ( which are adaptations of lemma 1 of @xcite ) , the estimates and imply @xmath368 for some @xmath369 and any @xmath370 + i [ - c_{0 } h , c_{0 } h]$ ] . on the other hand , remark [ e40 ] gives @xmath371 for some @xmath372 . this proves for @xmath370 + i [ - c_{0 } h , c_{0 } h]$ ] . thanks to theorem [ e17 ] which describes all the resonances in any neighborhood of size @xmath0 of @xmath36 , it remains to prove that @xmath22 has no resonance in @xmath373 \setminus [ e_{0 } - a h , e_{0 } + a h ] \big ) + i [ - c_{0 } h , c_{0 } h ] , \ ] ] for one @xmath374 and that the resolvent satisfies in this region an upper bound polynomial with respect to @xmath11 . in particular , we can assume that @xmath334 . using lemma [ e39 ] , lemma [ e37 ] and @xmath375 , we get @xmath376 for some @xmath26 . then , summing the first identity with @xmath377 times the second one , we obtain @xmath378 since @xmath379 . if now we assume that @xmath380 , we get @xmath381 for @xmath382 large enough . thanks to remark [ e40 ] , this implies that @xmath22 has no resonance in the region given in and that holds in this set . the purpose of this part is to give the asymptotic expansion of the generalized spectral projection @xmath383 associated to an isolated resonance @xmath384 in some @xmath385 . we recall that @xmath383 is the operator from @xmath386 to @xmath387 defined by @xmath388 where @xmath75 is a simple loop in the complex plane , oriented counterclockwise , such that @xmath384 is the only resonance in the bounded domain delimited by @xmath75 . _ [ a2 ] assume [ a1][a3 ] . let @xmath83 be such that @xmath389 is simple . then , as operators from @xmath386 to @xmath387 , @xmath390 where @xmath391 and the function @xmath392 satisfies the following properties : _ @xmath114 it is locally uniformly in @xmath207 : for all @xmath393 , @xmath394 @xmath116 it satisfies the schrdinger equation : @xmath395 @xmath396 it is outgoing : there exists @xmath397 such that @xmath398 @xmath399 finally , locally near @xmath43 , we have @xmath400 where @xmath401 is a classical symbol satisfying @xmath402 we prove this result the following way . using @xcite , we compute @xmath403 for some well prepared wkb function @xmath404 and @xmath6 on a loop around the resonance @xmath384 . integrating with respect to @xmath6 , we get @xmath405 and thus the resonant state @xmath192 . to finish the proof , we obtain the constant @xmath406 computing @xmath407 by a stationary phase argument . _ @xmath114 since @xmath192 is not necessarily in @xmath408 , saying `` @xmath409 microlocally near @xmath410 '' means that there exists @xmath411 with @xmath412 such that , for every @xmath413 , @xmath414 in @xmath415 . _ @xmath116 the properties @xmath114@xmath399 of theorem [ a2 ] characterize uniquely the resonant state @xmath392 modulo @xmath416 . in particular , the usual propagation of singularities implies that this function is a classical lagrangian distribution of order @xmath51 with lagrangian manifold @xmath417 . for the punctual well in the island situation , the generalized spectral projection has been computed by helffer and sjstrand @xcite . in particular , they have proved that this operator is almost orthogonal . indeed , if the resonance @xmath6 is isolated and the cut - off @xmath418 is equal to @xmath419 near the well , then @xmath420 is exponentially close to the spectral projection associated to the dirichlet problem in the well and @xmath421 for some @xmath26 . the situation is very different in the present setting since , for @xmath422 , @xmath423 is of order @xmath424 . from the previous discussion , the polynomial upper bound on the resolvent proved in theorem [ aaa ] occurs effectively . more precisely , in every disc @xmath425 , with @xmath113 , the cut - off resolvent can not be bounded by anything smaller than @xmath426 for some @xmath427 . moreover , since @xmath428 , this constant can not be taken uniformly with respect to @xmath384 . one may perhaps prove theorem [ a2 ] with other methods than the one we use here . in the one dimensional case , the resolvent can be written in term of a basis of solutions of @xmath429 and of their wronskian . thus , it must be possible to use the results of @xcite in which the scattering amplitude , which can be expressed through the wronskians of the jost solutions , has been computed . in any dimension , another approach is perhaps also possible . one may first try to calculate the resonant state @xmath192 with various methods ( using , for example , the works of briet , combes and duclos @xcite , sjstrand @xcite or hassell , melrose and vasy @xcite ) . it then remains to calculate the constant @xmath406 . this question is equivalent to the calculation of the scalar product @xmath430 . if we neglect the problems of integration at infinity , this calculation is reduced to a problem of stationary phase at point @xmath51 . but , since @xmath431 vanishes to order @xmath432 , the knowledge of @xmath433 is not enough and we must explicitly know the @xmath434 first terms in the expansion of @xmath192 in powers of @xmath0 . in this computation , the situation becomes , in a sense , similar to that of the eigenvectors of the harmonic oscillator for which the `` good variable '' is @xmath435 . however , this is not the case in theorem [ a2 ] since the factor @xmath436 in @xmath192 has modulus @xmath419 . it may be possible to obtain some results when @xmath389 is not simple . in that case , various situations may occur : several resonances can be very close to each other , the resonances can have a non - trivial multiplicity and they can be multiple poles of the resolvent . we refer to ( * ? ? ? * section 4 ) where such phenomena are shown . in the remainder of this discussion , we consider the simplest case where a double resonance can appear . we assume that @xmath437 and that @xmath438 is a double resonance . then , near @xmath439 , the resolvent can be written @xmath440 where @xmath441 is holomorphic near @xmath439 . in that case , @xmath442 and @xmath443 . it seems possible to calculate @xmath444 with a proof similar to that of theorem [ a2 ] . using proposition [ d10 ] , we can construct two initial data @xmath445 such that the microsupport of @xmath446 and @xmath447 intersect along a hamiltonian curve which goes to @xmath51 along the @xmath448-th vector basis . then , computing the residue of @xmath449 , we obtain that @xmath450 is of the form @xmath451 modulo a constant . in the following , we can neglect @xmath452 as it gives lower order terms . since @xmath453 and @xmath454 can not be collinear , @xmath455 ( resp . @xmath456 ) forms a basis of @xmath457 ( resp . @xmath458 ) . to finish the computation of @xmath444 , it is sufficient to calculate @xmath459 . the scalar products @xmath460 can be calculated as in the proof of theorem [ a2 ] . but , according to the choice of the @xmath446 s and to the form of the @xmath461 s , @xmath459 appears to be smaller when @xmath462 . eventually , in the @xmath455 and @xmath456 bases , the operator @xmath444 seems to be a @xmath463-matrix whose diagonal coefficients are given by at the first order and whose off - diagonal coefficients are of lower order . one can probably also say something about @xmath464 . but , one may need to calculate several lower order terms in the semiclassical expansions ( for the resonance for example ) . this operator seems to have a smaller norm . @xmath124 [ a34 ] to prove the theorem , it is enough to show that @xmath465 for @xmath418 . let @xmath466 be the spectral projection of @xmath71 at the resonance @xmath384 . it is the operator on @xmath207 defined by @xmath467 we now assume that the distortion occurs outside of the support of @xmath468 . in particular , @xmath469 . let @xmath470 be the anti - linear operator on @xmath279 defined by @xmath471 since @xmath22 is a schrdinger operator with a real potential , @xmath472 and a direct calculation gives @xmath473 . thus , @xmath466 can be written @xmath474 with @xmath475 . the same way , @xmath476 for some @xmath477 . moreover , from @xcite , we can always assume that @xmath478 . in particular , @xmath479 . since @xmath389 is simple , for all @xmath480 ( where @xmath481 ) , @xmath482 with @xmath483 implies @xmath484 . then , from lemma [ d3 ] and proposition [ d10 ] , there exists a hamiltonian curve @xmath485 such that , for all @xmath480 , we have @xmath486 . we now construct the `` test functions '' , supported microlocally near the `` test curve '' @xmath487 , on which we will evaluate the spectral projection . let @xmath488 be a function defined in a vicinity of @xmath51 but not at @xmath51 . we assume that @xmath489 is a wkb solution of @xmath490 . more precisely , near the @xmath491-projection of @xmath492 , we have @xmath493 here @xmath494 is a @xmath148 function solving the eikonal equation @xmath495 . we assume that @xmath496 intersects transversely @xmath447 along @xmath487 . note that the construction of such a phase , whose associated lagrangian manifold projects nicely on the @xmath491-space in a vicinity of @xmath487 , can always be done thanks to ( * ? ? ? * proposition c.1 ) . the symbol @xmath497 is classical : for all @xmath498 , @xmath499 uniformly for @xmath500 . moreover , @xmath501 and the @xmath502 s are @xmath148 with respect to @xmath491 and analytic with respect to @xmath500 . finally , we assume that @xmath489 satisfies @xmath503 and @xmath504 near the @xmath491-projection of @xmath487 . for that , it is enough to solve the usual transport equations . finally , we suppose that @xmath505 outside a neighborhood of the spacial projection of @xmath487 . then , we set @xmath506 u , \ ] ] where @xmath507 with @xmath508 close to @xmath51 and @xmath509 near @xmath51 . we consider @xmath510 in all the proof of theorem [ a2 ] , we will work with @xmath6 in a ring @xmath511 such that @xmath389 is the unique element of @xmath512 in @xmath513 . note that theorem [ aaa ] implies that @xmath514 uniformly for @xmath515 , for some @xmath516 . @xmath124 we begin the proof by showing that @xmath517 is @xmath51 in the incoming region . more precisely , we have _ [ a4 ] let @xmath518 be such that @xmath519 and @xmath520- \infty , 0 ] h_{p } ) ( \rho ) $ ] does not meet the microsupport of @xmath404 . then , @xmath521 microlocally near @xmath522 , uniformly in @xmath515 . _ this lemma can be proved as theorem 2 of @xcite . first , assume @xmath523 . using the elliptic equation @xmath524 , the norm estimates @xmath525 and the condition @xmath526 , the standard pseudodifferential calculus implies that @xmath527 . more precisely , for all @xmath528 with @xmath529 near @xmath36 , we have @xmath530 uniformly in @xmath515 . assume now that @xmath531 . from the hypotheses , the half - curve @xmath532 , @xmath533 , does not meet @xmath534 and goes to @xmath535 as @xmath536 . then , one can find a symbol @xmath537 such that @xmath538 near @xmath522 , @xmath539 , @xmath520- \infty , 0 ] h_{p } ) ( \supp \omega ) $ ] does not meet @xmath540 and @xmath541 for some @xmath542 , @xmath543 and @xmath544 . here , @xmath545 . then , mimicking the proof of ( * ? ? ? * theorem 2 ) , we get @xmath546 , uniformly in @xmath515 . the unique difference with its proof is that the @xmath51 in the left hand side of ( * ? ? ? * ( 3.4 ) ) is replaced by @xmath416 ( here , we use that @xmath547 ) . we will now calculate @xmath517 on @xmath447 near @xmath51 . first , using @xmath548 , the previous lemma implies the following consequence . _ [ a7 ] we have @xmath549 microlocally near each point of @xmath550 . _ on the other hand , near @xmath487 , we have the following lemma . note that the results of this lemma and of remark [ a7 ] are uniform for @xmath515 . _ [ a6 ] let @xmath551 be a point close enough to @xmath51 . then , @xmath552 microlocally near @xmath522 . _ ( 0,0).,title="fig : " ] # 1#2#3#4#5 @font ( 12573,5760)(-1457,-5666 ) ( -1424,-2461)(0,0)[lb ] ( 1126,-2461)(0,0)[lb ] ( 5551,-886)(0,0)[lb ] ( 3151,-3586)(0,0)[lb ] ( 6376,-2161)(0,0)[lb ] ( 9676,-3136)(0,0)[lb ] ( 10126,-4411)(0,0)[lb ] ( 11101,-5611)(0,0)[lb ] we define @xmath553 u \ , d t , \ ] ] where @xmath554 is chosen such that @xmath555 - \infty , 0 ] h_{p } ) ( \rho ) = \emptyset$ ] ( see figure [ f1 ] ) . then , microlocally near each point of @xmath487 , we have @xmath556 u - e^{-i t ( p -z)/h } [ p , \tau ] u .\end{aligned}\ ] ] for the first equality , we have used that @xmath557 near the spacial projection of @xmath558 . thus , microlocally near @xmath487 , we have @xmath559 u .\ ] ] in particular , the choice of @xmath560 and the egorov theorem imply @xmath561 microlocally near @xmath520 - \infty , 0 ] h_{p } ) ( \rho ) $ ] . on the other hand , combining lemma [ a4 ] ( for @xmath517 ) and the egorov theorem ( for @xmath562 ) , we obtain @xmath563 microlocally near @xmath564 , for all @xmath565 large enough . using moreover that @xmath566 , the propagation of singularities implies that @xmath567 then , microlocally near @xmath522 , we have @xmath568 u \ , d t = \frac{i}{h } \int_{0}^{t } e^{-i t ( p - z ) /h } \big ( ( p - z ) \tau u - \tau ( p -z ) u \big ) d t \nonumber \\ & = \frac{i}{h } \int_{0}^{t } e^{-i t ( p - z ) /h } ( p - z ) \tau u \ , d t = - e^{- i t ( p - z ) /h } \tau u + \tau u = u , \end{aligned}\ ] ] which proves the lemma . in fact , one can prove more directly lemma [ a4 ] and lemma [ a6 ] by applying the proof of theorem 2 of @xcite to the function @xmath569 . @xmath124 we will use the variable @xmath570 , the notation @xmath571 and the set @xmath572 . note that @xmath573 and @xmath574 does not depend on @xmath0 and @xmath575 . since @xmath509 near @xmath51 , we have @xmath576 u = 0 , \ ] ] in a neighborhood of @xmath51 . on the other hand , let @xmath577 with @xmath578 small enough . from remark [ a7 ] and lemma [ a6 ] , we have @xmath579 microlocally near @xmath522 . moreover @xmath580 . then , we are in position to apply theorem 2.1 and theorem 2.5 of @xcite which give a representation of @xmath517 microlocally near @xmath43 . more precisely , theorem 5.1 of @xcite states that , microlocally near @xmath43 , @xmath581 concerning the symbol @xmath582 , we will only use that @xmath583 is a holomorphic function of @xmath584 which decays uniformly exponentially in @xmath111 ( see ( * ? ? ? * proposition 5.11 ) ) . the constant @xmath585 is defined by @xmath586 for all @xmath587 . the symbol @xmath588 , holomorphic for @xmath589 , is constructed the following way . there exists an expandible symbol @xmath590 of the form @xmath591 where the @xmath592 s satisfy @xmath593 we refer to helffer and sjstrand @xcite for the definition of expandible functions . here , @xmath565 is defined by @xmath594 the symbols @xmath595 are holomorphic for @xmath596 . moreover , as in ( * ? ? ? * ( 6.26 ) ) , @xmath597 does not depend on @xmath111 ( and @xmath598 ) and @xmath599 with @xmath600 , @xmath601 being the spatial projection . let @xmath602 be the expandible function @xmath603 constructed in ( * ? ? ? * section 5 ) . recall that @xmath604 satisfies the eikonal equation @xmath605 we consider the expandible symbol ( see ( 5.77 ) of @xcite ) @xmath606 for some @xmath607 fixed large enough , @xmath608 then , @xmath609 is a symbol , holomorphic with respect to @xmath589 , such that @xmath610 for some @xmath611 large enough . in the following , we will need some informations on the @xmath612 . let @xmath613 be such that @xmath614 or @xmath615 . since @xmath389 is simple , @xmath39 can not be written as a non - trivial combination of the @xmath616 s ( i.e. @xmath482 implies @xmath617 ) . therefore , calculating the term in @xmath618 of , we obtain @xmath619 working as in section 6.1 of @xcite ( see also ( 5.59 ) of @xcite for @xmath615 ) , one can prove that @xmath620 does not depend on @xmath111 ( i.e. @xmath621 ) , that @xmath622 and that @xmath623 since @xmath624 is collinear to the @xmath448-th vector of basis , we also denote this @xmath448-th component of the vector @xmath624 by @xmath624 . @xmath124 [ k9 ] let @xmath75 be a fixed simple loop in @xmath574 around @xmath51 oriented counterclockwise and @xmath625 . we integrate @xmath517 on the loop @xmath626 . first , since @xmath384 is a simple resonance for @xmath0 small and since @xmath404 is a holomorphic function with respect to @xmath627 , the equations and give @xmath628 on the other hand , we can also calculate this quantity microlocally near @xmath43 with the help of . since @xmath629 is holomorphic in @xmath630 , the second term in the right hand side of gives no contribution to this integral . moreover , for @xmath631 , the function @xmath632 is holomorphic for @xmath633 . this implies that only the terms of with @xmath634 give a non - zero contribution to the integral over @xmath598 . we now look for the terms with @xmath634 in . among these terms , the one which gives the higher possible power of @xmath11 , is given by @xmath635 and is equal to @xmath636 here , we have used the fact that @xmath389 is simple . note that , since @xmath597 and @xmath637 , with @xmath614 , does not depend on @xmath111 ( see the discussion before ) , this term does not depend on @xmath111 . then , @xmath638 satisfies , as @xmath639 , @xmath640 where the @xmath641 s are holomorphic with respect to @xmath589 and @xmath148 with respect to @xmath491 near @xmath51 . the function @xmath642 is holomorphic in @xmath643 . moreover , @xmath644 using the previous discussion , together with and , we obtain that @xmath645 microlocally near @xmath646 . moreover , @xmath647 to be more precise , in the @xmath148 case , theorem 2.1 of @xcite gives only uniqueness for @xmath6 outside of a set @xmath648 , which is finite uniformly with respect to @xmath0 . then , to prove , we integrate first on a loop @xmath649 of length of order @xmath0 and which may depend on @xmath0 in a non trivial way . but , since the function @xmath517 is holomorphic in @xmath650 , we can deform the contour @xmath651 to @xmath626 and thus justify . @xmath124 [ a46 ] we define the functions @xmath652 and @xmath653 by @xmath654 where , using the notation @xmath655 , @xmath656 as usual , we have @xmath657 if the distortion holds outside of the support of @xmath8 . from , @xmath652 ( resp . @xmath653 ) is in the image of @xmath383 ( resp . @xmath466 ) . moreover , using ( which gives that @xmath658 ) , , , , and , we have , microlocally near @xmath43 , @xmath659 where @xmath660 is a classical symbol satisfying @xmath661 in particular , @xmath652 is not identically zero . then , @xmath383 can be written as @xmath662 and @xmath652 satisfies @xmath399 of theorem [ a2 ] . furthermore , using lemma [ a4 ] , integrating over @xmath6 and coming back to the definition of @xmath652 ( see ) , we immediately obtain the point @xmath396 of theorem [ a2 ] . since @xmath652 is in the image of @xmath383 which is the spectral projection at a simple resonance , the point @xmath116 of theorem [ a2 ] is clear . combining @xmath396 , @xmath399 , , which gives a uniform bound outside of the energy level , together with `` the transport equation '' @xmath116 , we get the point @xmath114 by a standard argument of propagation of singularities . @xmath124 [ a53 ] here we calculate the scalar product between @xmath664 and @xmath665 . from and , the function @xmath404 is supported near @xmath666 and micro - supported near @xmath667 . then , if @xmath508 is close enough to @xmath51 , the previous section and imply that @xmath668 b e^{i \psi /h } , \overline{\widetilde{d } e^{i \varphi_{+ } /h } } \big ) + \co ( h^{\infty } ) .\ ] ] a direct calculus gives @xmath669 \big ( b e^{i \psi /h } \big ) = \widetilde{b } ( x , h ) e^{i \psi ( x ) /h},\ ] ] with @xmath670 then , using that @xmath671 , we get @xmath672 the critical points of the phase @xmath673 ( i.e. the points @xmath491 such that @xmath674 ) are the points in the spatial projection of @xmath675 . moreover , since this intersection is transversal , the phase function @xmath673 is non degenerate in the directions that are transverse to @xmath676 ( @xmath601 being the spatial projection ) . then , applying the method of the stationary phase in the orthogonal directions of @xmath676 ( written @xmath677 ) and parameterizing the curve @xmath676 by @xmath678 , gives @xmath679 with @xmath680 and @xmath681 from , we have @xmath682 for all @xmath683 . in particular , can be written @xmath684 with @xmath685 from and , we have @xmath686 using , we have the transport equation @xmath687 which gives @xmath688 on the other hand , since @xmath689 is @xmath148 and @xmath678 is expandible , the function @xmath690 is expandible . moreover , since @xmath39 can not be written as a non - trivial combination of the @xmath616 s , the taylor expansion of @xmath637 shows that the term in @xmath691 in the expansion of @xmath692 is @xmath693 . since gives another asymptotic expansion , the uniqueness of the development implies that @xmath694 then , combining with , we obtain @xmath695 note here that the curve @xmath487 has been chosen in section [ a34 ] such that @xmath696 . from the construction of @xmath489 in and since @xmath384 is a classical symbol ( see remark [ k1 ] ) with @xmath697 , the function @xmath698 satisfies the usual transport equation @xmath699 mimicking the proof of , we get @xmath700 therefore , gives @xmath701 from proposition c.1 of @xcite and since @xmath702 , we have @xmath703 using @xmath704 , we get @xmath705 and @xmath706 for @xmath111 large enough . finally , using the expansion of @xmath707 , we have @xmath708 combining the definitions of @xmath709 and of @xmath710 with the relations , , , and , the constant @xmath709 does not vanish if @xmath711 and the support of @xmath712 is sufficiently small near @xmath560 large enough . @xmath124 [ a57 ] from and , we have @xmath713 in particular , using and , we get @xmath714 with @xmath715 at this point , the function @xmath652 and the constant @xmath716 may depend on @xmath404 . nevertheless , since @xmath717 and @xmath718 ( the first term in the development of @xmath652 given in ) do not depend on @xmath404 , the constant @xmath719 also does not depend on @xmath404 . we choose a sequence of functions @xmath720 ( say @xmath721 ) , with @xmath722 , such that @xmath723 converges to the dirac mass @xmath724 for some fixed @xmath725 . then , from the definition of @xmath709 and of @xmath726 , we get @xmath727 combining , , , , and , we obtain @xmath728 then , letting @xmath111 going to @xmath729 and using that @xmath719 does not depend on @xmath111 , it follows @xmath730 we now consider a fixed @xmath404 as in the beginning of this subsection . with @xmath731 as in , gives that @xmath732 where @xmath733 now , we define @xmath734 . then , gives and the properties of @xmath192 given in theorem [ a2 ] follow from the properties of @xmath652 given in section [ a46 ] and @xmath735 . in this section , we give the semiclassical expansion of the residue of the scattering amplitude at an isolated resonance . to define the scattering matrix , we assume that the potential is long range : [ h7 ] for some @xmath736 , we have @xmath737 for all @xmath738 . using the constructions of isozaki and kitada ( see @xcite and @xcite ) , the assumption [ h7 ] allows to define the scattering matrix @xmath739 , @xmath740 0 , + \infty [ $ ] related to the pair @xmath741 and @xmath22 as a unitary operator @xmath742 in the short range case ( i.e. @xmath743 ) , this operator coincides with the usual scattering matrix . next , introduce the operator @xmath744 defined by @xmath745 its kernel @xmath746 is smooth away from the diagonal of @xmath747 ( see @xcite ) . here , @xmath748 ( resp . @xmath749 ) is called the outgoing ( resp . incoming ) direction . finally , the scattering amplitude is defined for @xmath750 by @xmath751 with @xmath752 in @xcite , grard and martinez have shown that for @xmath750 fixed , the scattering amplitude has a meromorphic continuation to a neighborhood of @xmath7530 , + \infty [ $ ] , whose poles are the resonances of @xmath22 . moreover , the multiplicity of each pole is less or equal to the multiplicity of the resonance . notice that , since the kernel of the residue of the scattering matrix is not singular at @xmath754 ( see theorem 1.1 ( iii ) of @xcite ) , we drop the assumption @xmath750 in the sequel . we will now make some hypotheses on the behavior of the classical curves . let @xmath755 be a hamiltonian curve in @xmath756 . under the hypotheses [ a1][h7 ] , there are only two possible behaviors for @xmath678 as @xmath757 : either it escapes to @xmath758 , or it goes to @xmath51 . from the long range assumption [ h7 ] , if @xmath678 escapes to @xmath535 , then @xmath759 has a limit in @xmath760 . moreover the set of points with asymptotic direction @xmath748 and @xmath749 , @xmath761 are lagrangian submanifolds of @xmath186 ( see @xcite ) . we suppose that [ h8 ] @xmath762 and @xmath47 ( resp . @xmath763 and @xmath48 ) intersect in a finite number @xmath764 ( resp @xmath765 ) of bicharacteristic curves , with each intersection transverse . we denote these curves , respectively , @xmath766 and @xmath767 note that , from proposition 2.5 of @xcite , the intersections @xmath768 and @xmath769 are never empty ( i.e. @xmath770 and @xmath771 ) . from @xcite , the curve @xmath772 with @xmath773 satisfies @xmath774 from lemma [ d3 ] , if @xmath39 satisfies @xmath775 , then @xmath776 . moreover , there always exists a @xmath777 such that @xmath778 . we define @xmath779 we know that @xmath780 is one of the @xmath39 s and that @xmath776 ( see ( * ? ? ? * ( 2.18 ) ) ) . we shall denote @xmath781 for some @xmath782 large enough which is equal to @xmath729 in the short range case @xmath783 . moreover , in the short range case @xmath783 , the bicharacteristic curves in @xmath784 , @xmath785 , are the bicharacteristic curves @xmath786 for which there exists a @xmath787 such that @xmath788 these trajectories are smooth with respect to @xmath789 . we denote by @xmath790 the impact parameter of the curve @xmath772 . let @xmath791 be the maslov determinants for @xmath772 which exist and satisfy @xmath792 ( see @xcite ) . we shall also denote by @xmath793 the maslov index of the curve @xmath772 . _ [ a47 ] assume [ a1][h8 ] . let @xmath83 be such that @xmath389 is simple . then , the residue of the scattering amplitude satisfies @xmath794 where @xmath795 moreover , @xmath796 if and only if @xmath797 . finally , in the short range case @xmath743 , we have @xmath798 _ in the last formula @xmath799 is a shorthand for @xmath800 where @xmath801 is identified with its @xmath448-th coordinate . to prove the theorem , we first obtain a representation formula for the scattering amplitude involving the resolvent . then we apply theorem [ a2 ] to express the residue of the scattering amplitude with the help of the resonant state @xmath192 . finally , the result follows from the computation of two scalar products which are done with the stationary phase method . _ stefanov @xcite ( in the compact support case ) and michel @xcite ( in the long range case ) have given a priori estimates for the residue of the scattering amplitude . for the resonances @xmath802 very close to the real axis ( more precisely @xmath803 ) and under a separation condition , they have proved that the residue satisfies @xmath804 in the present situation , these results do not apply since the resonances are `` too far '' from the real axis . furthermore , the previous estimate does not hold . indeed , the imaginary part of @xmath384 behaves like @xmath805 but the residue is typically of order @xmath806 . _ in the one dimensional case , theorem [ a47 ] can probably be deduced from the computation of the scattering amplitude obtained by the third author in @xcite . for a punctual well in the island case and under some geometrical assumptions , the asymptotic of the residue of the scattering amplitude has been computed by nakamura @xcite , lahmar - benbernou @xcite and lahmar - benbernou and martinez @xcite . it is possible to compare theorem [ a47 ] with the semiclassical expansion of the scattering amplitude for real energy obtained in @xcite . assume for simplicity that the @xmath39 s are non - resonant ( @xmath807-independent for example ) , @xmath808 , @xmath809 , @xmath810 and @xmath811 for all @xmath613 . in particular , we have @xmath812 . let @xmath813 be the first @xmath448 with @xmath814 ( thus , @xmath815 ) . in that case , theorem 2.6 ( a ) of @xcite gives @xmath816 for @xmath21 real with @xmath817 . here , @xmath818 and @xmath819 is an explicit function , analytic near @xmath36 . thus , the main term in , defined in @xcite for @xmath21 real , has a meromorphic extension in a fix neighborhood of @xmath36 . moreover , its poles are exactly the pseudo - resonances @xmath820 with @xmath821 and the corresponding residue coincides with that given in theorem [ a47 ] . in particular , this principal term does not contribute to the residue at the other ( pseudo)-resonances . the cases ( b ) and ( c ) in theorem 2.6 of @xcite only appear for resonant @xmath39 s and the corresponding main terms in the semiclassical expansion of the scattering amplitude have poles at some @xmath820 which are not simple . @xmath124 in this section , we recall a representation formula of the scattering amplitude for complex energies due to grard and martinez @xcite . their approach consists in extending the formula of isozaki and kitada @xcite to complex energies . for this purpose , they show that the phases and the symbols involved in that formula can be chosen to be analytic in a suitable complex neighborhood of @xmath822 . we only recall what will be useful in the following and refer to @xcite for the details . for @xmath823 large enough , @xmath824 , @xmath113 and @xmath8250 , 1[$ ] , we denote @xmath826 let @xmath113 , @xmath827 , @xmath828 and @xmath829 be sufficiently large . for @xmath830 , we denote @xmath831 . in @xcite , grard and martinez construct some phases @xmath832 and some symbols @xmath833 satisfying the general assumptions of isozaki and kitada @xcite and the following properties . the phases @xmath834 have a holomorphic extension to @xmath835 and satisfy @xmath836 uniformly in @xmath835 . moreover , @xmath837 . there exist two symbols @xmath838 supported inside @xmath839 , with @xmath840 such that @xmath841 moreover , for some @xmath578 with @xmath842 , we have @xmath843 for @xmath844 . finally , they extend holomorphically with respect to @xmath845 and @xmath846 for @xmath847 in @xmath848 and @xmath849 in a complex neighborhood of @xmath850 . furthermore , their extensions continue to satisfy estimates analogous to the previous ones . the symbols @xmath851 are then defined by @xmath852 and satisfy , for some @xmath853 , @xmath854 uniformly with respect to @xmath0 and @xmath855 . under the assumption [ h7 ] , grard and martinez @xcite have proved that the scattering amplitude can be written @xmath856 where @xmath857 has a holomorphic extension in a ( fixed ) neighborhood of @xmath36 , @xmath858 and @xmath859 by assumption , the resonance @xmath384 is simple for @xmath0 small enough . moreover , theorem [ aaa ] implies that @xmath860 for @xmath117 and some @xmath861 . then lemma 5.4 of @xcite ( see also proposition 5.1 of @xcite in the case of a well in the island ) states that @xmath862 where @xmath418 satisfies @xmath863 with @xmath864 . in particular , there is no distortion ( i.e. @xmath865 ) on the support of @xmath468 and theorem [ a2 ] implies @xmath866 where @xmath867 with @xmath731 given by . @xmath124 [ a68 ] we will calculate the scalar product @xmath868 by the stationary phase method . first , we will prove that this quantity has an asymptotic expansion in power of @xmath0 and then calculate the first term using a limit at the origin . we will use arguments close to the ones developed in section [ a53 ] or ( * ? ? ? * section 7 ) . ( 0,0 ) and @xmath869.,title="fig : " ] # 1#2#3#4#5 @font ( 10023,8033)(811,-7611 ) ( 9151,-286)(0,0)[lb ] ( 5851,-4636)(0,0)[lb ] ( 9376,-2011)(0,0)[lb ] ( 826,-361)(0,0)[rb ] ( 901,-1486)(0,0)[rb ] ( 901,-2761)(0,0)[rb ] @xmath870 ( 0,0 ) and @xmath869.,title="fig : " ] # 1#2#3#4#5 @font ( 10023,8033)(811,-7611 ) ( 9151,-286)(0,0)[lb ] ( 5851,-4636)(0,0)[lb ] ( 826,-361)(0,0)[rb ] ( 901,-1486)(0,0)[rb ] ( 901,-2761)(0,0)[rb ] ( 9376,-2011)(0,0)[lb ] denote @xmath871 and @xmath872 . from theorem [ a2 ] @xmath116 and , we have @xmath873 u \big ) = - \big ( f , [ p , \chi ] u \big ) .\ ] ] from and the choice of @xmath468 , @xmath874 microlocally near @xmath875 . moreover , since @xmath384 has an asymptotic expansion in power of @xmath0 , we can write , microlocally near @xmath875 , @xmath876 where @xmath877 has an asymptotic expansion in power of @xmath0 . note that @xmath878 . using maslov s theory , we can extend the function @xmath489 near @xmath879 , a small neighborhood of @xmath880 , such that @xmath489 is still a solution of @xmath881 microlocally in @xmath879 . let @xmath882 be such that @xmath883 out of @xmath879 ( see figure [ f2 ] ) . in particular , @xmath884 microlocally near the support of @xmath885 . so , we have @xmath886 u \big ) - \big ( ( p - z_{\alpha } ) f , \op ( \chi - \widetilde{\chi } ) u \big ) + \big(f , \op ( \chi - \widetilde{\chi } ) ( p - \overline{z_{\alpha } } ) u \big ) \nonumber \\ = & - \big ( f , [ p , \op ( \widetilde{\chi } ) ] u \big ) + \co ( h^{\infty } ) . \end{aligned}\ ] ] on the other hand , since @xmath887 , the microsupport of @xmath888 u$ ] satisfies @xmath889 u \big ) & \subset \lambda_{\varphi_{1 } ( \cdot , \sqrt{e_{0 } } \omega ) } \cap \supp \nabla \widetilde{\chi } \\ & \subset \big ( \lambda^{-}_{\omega } \cap \gamma^{- } ( r_{1 } , d , \sigma^{-}_{1 } /2 ) \big ) \cup \big ( \lambda^{+}_{\omega } \cap \omega \big ) .\end{aligned}\ ] ] moreover , theorem [ a2 ] gives @xmath890 . then , modulo @xmath891 , the non - zero contributions to @xmath892 u ) $ ] comes from the values of the functions @xmath192 and @xmath888 u$ ] microlocally on the set @xmath893 ( which constitute the intersection of the two microsupports ) . let @xmath894 be @xmath895 functions with support in a small enough neighborhood of @xmath896 such that @xmath897 in a similar neighborhood . then , becomes @xmath898 u \big ) + \co ( h^{\infty } ) .\ ] ] we now compute @xmath899 u$ ] . from proposition c.1 of @xcite , the lagrangian manifold @xmath900 has a nice projection with respect to @xmath491 in a neighborhood of any point of @xmath901 close to @xmath43 . then , maslov s theory implies that @xmath489 can be written as @xmath902 microlocally in such a neighborhood . from the construction of @xcite and @xcite , we see that @xmath903 for some @xmath904 large enough ( equal to @xmath729 in the short range case ) . the symbol @xmath905 has an asymptotic expansion @xmath906 with @xmath907 . moreover , in the short range case , equation ( 7.12 ) of @xcite gives @xmath908 where @xmath909 is the maslov index of the curve @xmath901 and @xmath910 is the maslov s determinant @xmath911 moreover , from section 6 of @xcite , we know that @xmath912 exists and satisfies @xmath913 . so , @xmath914 u = \widetilde{a}_{\ell}^{+ } ( x , h ) e^{i \psi_{\ell}^{+ } ( x ) /h } , \ ] ] with @xmath915 and @xmath916 since the support of @xmath917 is close enough to @xmath43 , theorem [ a2 ] @xmath399 and imply that @xmath918 we proceed now as in . in the support of the symbol @xmath919 , the critical points of the phase function @xmath920 ( i.e. the points @xmath491 such that @xmath921 ) are the points in the spacial projection of @xmath901 . since this intersection @xmath922 is transverse from the assumption [ h8 ] , the phase @xmath920 is non degenerate in the directions transverse to @xmath923 . therefore , performing the method of the stationary phase in the orthogonal directions of @xmath924 ( as in ) and parameterizing the curve @xmath924 by @xmath925 , gives @xmath926 with @xmath927 and @xmath928 since @xmath929 , @xmath930 and @xmath931 have the same derivative ( with respect to @xmath111 ) , and gives @xmath932 for all @xmath933 . then , combining with , we get @xmath934 with @xmath935 and @xmath936 from , @xmath937 does not depend on @xmath869 , modulo @xmath938 . in particular , changing @xmath869 in a neighborhood of a fixed curve @xmath901 , we obtain that each @xmath939 does not depend on @xmath869 . from proposition c.1 of @xcite , we have , up to a linear change of variables in @xmath12 , @xmath940 where @xmath941 is such that @xmath942 . since @xmath943 as expandible symbol ( see ( * ? ? ? * definition 5.2 ) ) , this implies @xmath944 as @xmath536 . on the other hand , gives @xmath945 we first consider the long range case ( @xmath946 ) . if @xmath947 , then and imply that @xmath948 . we will now prove that @xmath949 if @xmath950 . let @xmath951 be sufficiently large such that the quantities in , and do not vanish and holds at @xmath952 . then , if @xmath953 satisfies @xmath954 and has its support close enough to @xmath560 , the previous discussion , @xmath955 and imply that @xmath949 . let us now consider the short range case ( @xmath783 ) . assume that the support of @xmath956 is sufficiently negative . then , the formula and the estimates , , , , and give @xmath957 where the @xmath958 does not depend on @xmath869 . now , we take a sequence of functions @xmath869 such that the support of @xmath959 goes to @xmath960 and @xmath961 ( see figure [ f2 ] ) . since @xmath939 does not depend on @xmath869 , the previous expression gives @xmath962 @xmath124 following the approach of section [ a68 ] , one can prove that @xmath963 with @xmath964 and @xmath965 if and only if @xmath966 . moreover , in the short range case , we have @xmath967 then , combining the representation of the residue given in with the constants given in and , and the scalar products and , we obtain @xmath968 with @xmath969 and @xmath970 if and only if @xmath971 . moreover , in the short range case , and imply @xmath972 in this section , we prove a resonance expansion for the cut - off schrdinger propagator . the proof relies on the resolvent estimate in theorem [ aaa ] and on standard arguments . _ [ c1 ] assume [ a1][a3 ] . let @xmath973 be different from @xmath82 for all @xmath83 . let @xmath418 and @xmath974 ) $ ] for some @xmath135 small enough . then , there exists @xmath975 such that @xmath976 for all @xmath933 . in particular , if all the @xmath389 in @xmath977 are simple , we have @xmath978 for all @xmath933 . here , @xmath383 is the generalized spectral projection associated to @xmath384 and described in theorem [ a2 ] . _ _ note that the previous expansions make sense only for @xmath979 . _ one might think that the resonance expansion holds for shorter times . but , in fact , it is not possible to do much better . this follows from the paper of de bivre and robert @xcite which is stated with slightly different hypotheses . in the one dimensional case , they have proved that the coherent states propagate through a maximum of the potential for times of order @xmath980 and that they stay at @xmath43 before . on the other hand , the sum of the generalized spectral projections over the resonances appearing in theorem [ c1 ] can not be microlocalized only at @xmath43 thanks to theorem [ a2 ] . thus , if the resonance expansion with a small error holds at time @xmath933 , we have necessarily @xmath981 in the one dimensional case . if we only want to prove that @xmath982 as @xmath639 , we can more simply apply the standard propagation of singularities with an initial data microlocalized in @xmath983 . there is also a simplest way to justify this phenomena . let @xmath68 be such that @xmath984 . then , @xmath802 is the unique resonance in @xmath985 for @xmath0 small enough and @xmath986 is always simple . assume that , for some @xmath933 , we can write @xmath987 where @xmath988 is small . the left hand side is of order @xmath419 since the propagator is unitary . on the other hand , from theorem [ e17 ] and theorem [ a2 ] , the right hand side is of order @xmath989 . then , implies @xmath990 . remark that this critical time coincides with the one obtained by de bivre and robert in the one dimensional case . the situation is different for the well in the island case which was treated by nakamura , stefanov and zworski @xcite . in that setting , the cut - off schrdinger group is well approximated by the resonance expansion after a fix time . this is in adequacy with the geometrical interpretation since a fix time is enough to dispel the part of the initial data which is not localized in the well . nevertheless , grard and sigal @xcite have proved that the schrdinger equation with a quasiresonant state ( sorts of quasimodes ) as initial data is always well approximated by the resonance expansion for all time @xmath933 . _ when @xmath991 as @xmath639 , the sum over the resonances is negligible and theorem [ c1 ] simply yields @xmath992 . _ the remainder terms @xmath938 in theorem [ c1 ] come from the @xmath148 pseudodifferential calculus . thus , if the cut - off functions @xmath993 are in some gevrey class , it is perhaps possible to replace these remainder terms by @xmath994 for some @xmath578 . in that case , the sum over the resonances will dominate the remainders until @xmath111 is of order @xmath995 . burq and zworski @xcite ( see also tang and zworski @xcite ) have obtained a long time expansion of semiclassical propagators in terms of resonances close to the real axis . their result in the present situation gives @xmath992 for all @xmath996 for some @xmath997 . let @xmath998)$ ] be such that @xmath999 near @xmath1000 $ ] . then , from the pseudodifferential calculus , we get @xmath1001 where @xmath1002 , the spectral projection , is given by the stone formula @xmath1003 and @xmath1004 is analytic for @xmath1005 . then , @xmath1006 making a change of contour , we obtain @xmath1007 where @xmath1008 and @xmath1009 - \infty , e_{0 } - 2 \varepsilon ] $ ] , @xmath1010 $ ] , @xmath1011 - i \mu h$ ] , @xmath1012 $ ] and @xmath1013 ( see figure [ f3 ] ) . the theorem will follow from the estimates on the @xmath1014 s given below . ( 0,0).,title="fig : " ] # 1#2#3#4#5 @font ( 10844,3266)(504,-6405 ) ( 2401,-5311)(0,0)[lb ] ( 2101,-3661)(0,0)[lb ] ( 9001,-3661)(0,0)[lb ] ( 5701,-6061)(0,0)[lb ] ( 4501,-3286)(0,0)[lb ] ( 3901,-5461)(0,0)[lb ] ( 8701,-5461)(0,0)[lb ] ( 5776,-4561)(0,0)[lb ] @xmath212 estimations of @xmath1015 and @xmath1016 . using that @xmath1017 , there exists @xmath1018 such that @xmath1019 near @xmath1020 and @xmath1021 near @xmath1022 . then , by pseudodifferential calculus , @xmath1023 . therefore , yields @xmath1024 the same way , we get @xmath1025 . @xmath212 estimations of @xmath1026 . using theorem [ aaa ] for @xmath1027 , we obtain @xmath1028 @xmath212 estimations of @xmath1029 and @xmath1030 . let @xmath117 be as in theorem [ aaa ] and assume that the distortion occurs outside of the support of @xmath468 . then , @xmath1031 and @xmath1032 . in particular , we can write @xmath1033 let @xmath1034 be such that @xmath1035 near @xmath1036 and @xmath1037 near @xmath1022 ( see figure [ f4 ] ) . then , for @xmath1038 , @xmath1039 therefore becomes @xmath1040 where @xmath1041 ( 0,0 ) , @xmath223 and @xmath494.,title="fig : " ] # 1#2#3#4#5 @font ( 13469,3421)(-696,-4625 ) ( 376,-2461)(0,0)[lb ] ( 2926,-4561)(0,0)[lb ] ( 6526,-4561)(0,0)[lb ] ( 10576,-4561)(0,0)[lb ] ( -674,-4561)(0,0)[lb ] ( 8026,-2461)(0,0)[lb ] ( 5026,-2461)(0,0)[lb ] since @xmath223 and @xmath494 have disjoint support , the pseudodifferential calculus gives @xmath1042 . thus , theorem [ aaa ] implies @xmath1043 since @xmath1044 for @xmath1038 . on the other hand , @xmath1045 and @xmath1046 uniformly in @xmath1038 . moreover , @xmath1047 is a differential operator whose coefficients are supported outside of the support of @xmath468 . then , the microlocal analysis gives @xmath1048 uniformly in @xmath1038 . combining this estimate with theorem [ aaa ] , we get @xmath1049 using and to estimate , we conclude @xmath1050 . the same way , we have @xmath1051 . in this section , we construct hamiltonian curves in @xmath447 with a prescribed asymptotic expansion at infinity . they are used in section [ a1 ] , where test functions for the projection are built in a microlocal neighborhood of these curves . we will work on @xmath447 , but the same work can be done in @xmath417 . let @xmath1052 be a hamiltonian curve in @xmath447 . from ( * section 3 ) , the curve @xmath487 satisfies , in the sense of expandible functions , @xmath1053 the spectrum of @xmath42 is @xmath1054 . we denote by @xmath1055 the spectral projection on the eigenspace of @xmath42 associated to @xmath1056 . remark that @xmath1057 we have @xmath1062 . taking the taylor expansion of @xmath44 at @xmath51 , we get @xmath1063 where @xmath1064 is a polynomial of order @xmath223 and @xmath1065 . since @xmath39 can not be written as the sum of at least two terms @xmath1066 , the cross products @xmath1064 in the previous formula provide no term of the form @xmath618 . then , @xmath1067 which can be written @xmath1068 if @xmath1069 , the previous equation , together with , gives a contradiction thus , @xmath1060 and @xmath1061 from . we construct the coefficients @xmath1073 inductively . using the taylor expansion of @xmath44 at @xmath51 as in , one can see that it is enough to find @xmath1073 , @xmath1074 , such that @xmath1075 where the @xmath1076 depend only on the @xmath1077 for @xmath1078 . assume that the @xmath1077 have been chosen to satisfy for all @xmath1079 and for all @xmath1080 . if @xmath1081 , then it is enough to take @xmath1082 , @xmath1083 and , for @xmath1084 , @xmath1085 if @xmath1086 for some @xmath448 , then we take @xmath1087 and @xmath1088 here , @xmath1089 is the inverse of the map @xmath1090 . with these choices , and are always verified . let @xmath1093 where the @xmath1094 are given by lemma [ d6 ] and @xmath1095 will be fixed ulteriorly . since is verified for all @xmath1096 , we have @xmath1097 with @xmath1098 . we seek a solution of the form @xmath1099 . then , @xmath1100 must satisfies @xmath1101 let @xmath1102 be such that @xmath1103 and @xmath1104 for all @xmath1105 . we define inductively @xmath1106 by @xmath1107 define @xmath1111 we will prove the lemma inductively . first , @xmath710 satisfies . assume now that @xmath1112 exists on @xmath1109 and verifies . in particular , @xmath1113 for @xmath1105 . then , gives @xmath1114 for @xmath1105 . therefore , if @xmath1115 , @xmath1116 satisfies and the lemma follows . using standard arguments , lemma [ d14 ] provides us with a function @xmath1129 satisfying @xmath116 . then , part @xmath114 follows directly from lemma [ d11 ] . on the other hand , @xmath1130 as @xmath1131 . then , taking the limit @xmath1131 in , we obtain @xmath1132 thus , @xmath1124 and @xmath1127 satisfies @xmath1128 . to finish the proof of proposition [ d10 ] , we impose in addition that @xmath1133 . then , the function @xmath487 of lemma [ d16 ] is a hamiltonian curve in @xmath447 and , since @xmath1134 , lemma [ d6 ] assures that @xmath1135 for all @xmath1071 . p. briet , j .- combes , and p. duclos , _ on the location of resonances for schrdinger operators in the semiclassical limit . i. resonances free domains _ , j. math . * 126 * ( 1987 ) , no . 1 , 9099 . p. briet , j .- combes , and p. duclos , _ on the location of resonances for schrdinger operators in the semiclassical limit ii : barrier top resonances _ , comm . in partial differential equations * 2 * ( 1987 ) , no . 12 , 201222 . j. sjstrand , _ semiclassical resonances generated by nondegenerate critical points _ , pseudodifferential operators ( oberwolfach , 1986 ) , lecture notes in math . , vol . 1256 , springer , berlin , 1987 , pp . 402429 . j. sjstrand , _ a trace formula and review of some estimates for resonances _ , microlocal analysis and spectral theory ( lucca , 1996 ) , nato adv . 490 , kluwer acad . publ . , dordrecht , 1997 , pp .
one of the most celebrated successes of string theory is the ads / cft correspondence @xcite . this generates a powerful duality between asymptotically ads gravitational theories and conformal field theories on the ads boundary , which is the simplest and best - studied example of the more general notion of a ` gauge - gravity duality ' . as a strong - weak coupling duality , the correspondence allows for the translation of non - perturbative field theory calculations into more tractable , perturbative calculations in gravity and vice - versa . this has enabled the exploration of previously inaccessible regimes of theoretical physics . indeed , there are many examples of strongly coupled systems in condensed matter physics and it is hoped that gauge - gravity duality may allow for a better understanding of these . significant progress has already been made in this direction , leading to the development of the ads / cmt correspondence ( see @xcite and references therein ) . further recent progress has been to extend the correspondence to spacetimes which are not asymptotically - ads but rather exhibit hyperscaling violating and lifshitz ( hvlif ) behaviour @xcite , thus extending the dictionary between gravity and condensed matter systems living on the boundary . the central idea in gauge / gravity duality is that each state in the bulk has a corresponding state in the dual field theory . in particular , black objects are dual to thermal ensembles in the field theory with the same thermodynamic properties ( temperature , entropy , chemical potential , etc . ) as the bulk spacetime @xcite . a natural starting point for the correspondence is to look at charged ( reissner - nordstrm or ` rn ' ) extremal black holes and black branes in ads @xcite . however , like their asymptotically flat ` cousins ' they have a large non - zero entropy at zero temperature , thus violating the third law of thermodynamics , which states in its strictest version that the entropy of a system should vanish in the zero temperature limit @xcite . while a non - vanishing entropy for certain classes of extremal black holes is consistent with microstate counting for the corresponding d - brane configuration in string theory @xcite , this still begs the question of whether one can find other gravitational systems which have a zero entropy or entropy density at zero temperature . apart from being an interesting question about gravity , such systems are relevant for possible dualities between gravity and condensed matter systems . we remark that although ` nernst law ' is in the following used synonymously with ` third law of thermodynamics ' , nernst s original formulation only requires that the difference in entropy between two equilibrium states related through a change in external parameters goes to zero at zero temperature . this formulation is equivalent to the ` process version ' of the third law , which states that zero temperature can not be reached by any physical process in a finite number of steps . a process version of the third law of black hole mechanics was already established in @xcite . however , the nernst version or , equivalently , the process version of the third law does not imply by itself the slightly stronger version of the third law , due to planck , which states that the entropy itself goes to zero at zero temperature . this stricter version corresponds to systems with a unique ground state , and thus is the generic situation in condensed matter , although there is an extended debate about possible exceptions in specific systems , see for example @xcite . in the following we will be concerned with the explicit construction of families of gravitational solutions which have zero entropy ( or entropy density ) in the extremal limit . following conventions in the literature , we will refer to the third law in its stricter , planckian , version as the nernst law . extremal brane solutions with vanishing entropy density at zero temperature have recently been studied for a variety of bulk theories @xcite and could have important applications in extending the dictionary between condensed matter and gravity . they have been dubbed ` nernst branes ' in @xcite , and it is believed that the corresponding non - extremal solutions exist and satisfy the nernst law , that is , these non - extremal solutions have a finite entropy which goes to zero when the temperature goes to zero while external parameters are kept fixed . finding such non - extremal solutions is important , since extremal nernst branes are not completely regular solutions . while all curvature invariants remain finite at the horizon , tidal forces become infinite and scalar fields take infinite values , which suggests a breakdown of the underlying effective field theory @xcite . a first step in addressing this issue is to find non - extremal solutions , which can then be studied in the near extremal limit . in this context it is clearly desirable to have completely explicit , analytical solutions . however most results in the literature have to rely on a mixture of analytical and numerical methods . of course tidal forces may still get very large at the horizon when one approaches the extremal limit @xcite , but analytical solutions will enable one to identify the region in parameter space where the solution can be trusted and possibly be mapped to condensed matter systems . the second step in controlling the near horizon low temperature behaviour is to embed the theory under consideration into a uv - complete theory , for which string theory and its non - perturbative extension m - theory are arguably the best candidates . in the low - energy limit the relevant stringy gravitational backgrounds can be described in terms of supergravity . we will be working in a set - up which can be described by @xmath13 @xmath0 gauged supergravity with an arbitrary number of vector multiplets . theories with @xmath13 supersymmetry are natural generalisations of the einstein - maxwell - scalar theories underlying dilatonic black hole and black brane solutions which have been studied extensively as potential duals of strongly coupled electron systems @xcite . they have the advantage that one can often find exact , analytical answers , despite the fact that the couplings are not fixed by the matter content ( as is the case for @xmath14 supersymmetry ) , but depend on arbitrary functions of the scalar fields , which are subject to quantum and stringy corrections . while we do not discuss the string theory or m - theory embedding explicitly , note that such theories arise through heterotic flux compactifications on @xmath15 and type - ii flux compactifications on calabi - yau three - folds . we will not need to choose a specific model , and only assume that the vector multiplet couplings take the most general form that arises when working to leading order in the regge parameter @xmath16 , and within the validity of string perturbation theory . in other words , we only assume that the prepotential , which encodes the vector multiplet couplings , is of the so - called very special type reviewed below . by working in a gauged supergravity theory obtainable by flux compactification from string theory we will have the option to further address the issues related to singularities in the extremal limit at a later stage . for bps black holes with vanishing entropy it is known that the inclusion of stringy higher curvature corrections in supergravity @xcite leads to regular solutions with finite entropy @xcite , and the entropy function formalism demonstrates that this mechanism is robust and does not depend on supersymmetry and details of the higher curvature corrections @xcite . we refer to @xcite for a further discussion of the possible implications of quantum and string corrections to the zero temperature behaviour and the ` fate ' of the nernst law . within the framework of four - dimensional @xmath13 @xmath0 gauged supergravity coupled to vector multiplets , extremal nernst branes have previously been constructed in @xcite using a first - order rewriting of the equations of motion , and by considering a specific model : the so - called stu - model . however a similar rewriting for their non - extremal counterparts has so far proven elusive , and the only known examples @xcite have been constructed by deforming the metric of the corresponding five - dimensional extremal solution @xcite and imposing suitable consistency conditions . in this paper we are able to provide a systematic construction of non - extremal nernst branes by directly solving the second - order equations of motion . moreover , our results will not only apply to a particular model , but to all models where the prepotential is of the very special type . this gain in generality and systematics should help to expand the ads / cmt dictionary considerably in the future . we now present a brief overview of the results in this paper . we start with a theory of @xmath17 @xmath13 vector multiplets coupled to @xmath0 gauged supergravity , with prepotential @xmath18 where @xmath19 is homogeneous of degree three . if @xmath19 is a homogeneous _ polynomial _ of degree three ( which is not required for our methods to apply ) , then the corresponding theory can be obtained by dimensional reduction from five dimensions . moreover , such prepotentials capture perturbative string effects to leading order in @xmath16 if the model can be embedded into heterotic or type - ii string theory . in this case the supergravity lift to five dimensions becomes a lift from type - ii string theory to m - theory . within these models we restrict ourselves to static black brane solutions . apart from this we will impose that the scalar fields take purely imaginary values , as for such ` axion - free ' field configurations there is a systematic simplification of the equations of motion . since we impose stationarity in four dimensions , we can perform a time - like dimensional reduction to obtain an effective three - dimensional euclidean theory . the degrees of freedom in three dimensions can then be repackaged using the real formulation of special geometry developed in @xcite , which has been used to construct solutions to both gauged @xcite and ungauged @xcite theories of supergravity coupled to vector multiplets . since our ability to obtain explicit non - extremal solutions depends on using a specific formalism , let us briefly summarize the underlying principles without going into technical details . * instead of using the physical four - dimensional scalar fields @xmath20 , @xmath21 , we work on the ` big moduli space ' parametrized by scalar fields @xmath22 , the additional ( complex ) degree of freedom is compensated for by a local @xmath24 gauge symmetry . working on the big moduli space has the advantage that the number of scalar fields and gauge fields matches . * we use the real formulation of special khler geometry , which replaces the complex scalars @xmath22 by real scalars @xmath25 , @xmath26 and which replaces the holomorphic prepotential @xmath27 by a real hesse potential @xmath28 . this leads to a simpler , and fully covariant , behaviour of all relevant quantities under electric - magnetic duality . * upon dimensional reduction , the kaluza - klein scalar @xmath29 is absorbed into the real scalars @xmath25 , which results in the ` radial ' direction of the big moduli space becoming a physical ( rather than gauge ) degree of freedom . + we postpone fixing the remaining @xmath30 gauge symmetry to preserve electric - magnetic duality . the resulting three - dimensional theory depends on @xmath31 real scalars @xmath32 , subject to one local gauge symmetry , where @xmath33 are dual to the four - dimensional gauge fields and @xmath34 is dual to the kaluza - klein vector . while @xmath35 are vectors under electric - magnetic duality , @xmath34 is a scalar . * we impose an ansatz which corresponds , from the four - dimensional point of view , to a static solution with purely imaginary scalar fields . this determines @xmath34 and half of the fields @xmath35 in terms of the remaining fields , and also fixes the residual @xmath0 gauge symmetry . by abuse of notation , we denote the remaining independent fields by @xmath35 ( with a restricted range of @xmath36 , depending on the precise version of the ansatz ) . * when we now proceed to solve the time - reduced three - dimensional equations of motion , their particular structure allows us to obtain solutions in closed form . we remark that while some of the above ingredients are well known to people working on @xmath13 supergravity , it is critical that these elements are put together into a systematic structure . the key element that we use and preserve is electric - magnetic duality , which acts on the fields by symplectic transformations . our choice of variables , which all transform as symplectic tensors , leads to the simplifications and systematics that we exploit . we observe that this works despite the fact that the electric - magnetic duality group is broken to a discrete subgroup thereof by the presence of gauging ( a scalar potential ) , and despite the fact that our ansatz restricts us from the full symplectic group to a subgroup . solving the three - dimensional equations of motion directly results in an instanton solution depending on a number of integration constants , which are _ a priori _ undetermined . however , in order that this solution lifts to a _ regular _ black brane in four dimensions we have to impose suitable regularity conditions . in particular , we require that the four - dimensional solution has a finite entropy density , which happens to simultaneously ensure that the scalar fields take finite values on the horizon . for a given set of charges and fluxes , we are then left with a two - parameter family of black brane solutions parametrised by a temperature @xmath4 and chemical potential @xmath5 , which can both be freely varied . in the limit of zero temperature , we recover the extremal nernst branes of @xcite . therefore we interpret our solutions as non - extremal ( or ` hot ' ) nernst branes . indeed , it turns out that the entropy density goes to zero as @xmath37 for fixed charges / fluxes , in agreement with the nernst law . our solutions interpolate between hyperscaling violating lifshitz geometries with @xmath6 at the horizon and @xmath7 at infinity , where @xmath38 is the dynamical critical exponent and where @xmath39 is the hyperscaling violating exponent . in the zero temperature limit the near horizon geometry changes to @xmath8 . this paper is organised as follows . in section [ secrealform ] we review the real formulation of special geometry as applied to @xmath13 @xmath0 gauged supergravity with both electric and magnetic fluxes , relegating the more technical details to the appendices . we then reduce this theory over a time - like direction and determine the equations of motion of the three - dimensional theory for general static field configurations , before concentrating on the case of purely imaginary field configurations . in section [ stusec ] we solve the aforementioned equations of motion for the case where we have a single electric charge and some number of electric fluxes . having found a solution to the three - dimensional equations of motion we then lift it back to a four - dimensional solution and determine the conditions imposed on the various integration constants by regularity , before carrying out an analysis of the properties of the solution . in section [ secmagbrane ] we apply our method to the case where we instead switch on a single magnetic charge and a single magnetic flux , whilst keeping @xmath40 of the electric fluxes . section [ secdiscussion ] contains our conclusions . we also include a brief initial discussion of our results in the context of holography . we begin with the well - known bosonic lagrangian of @xmath13 fayet - iliopoulos @xmath41 gauged supergravity coupled to @xmath17 vector multiplets . our conventions follow those of @xcite @xmath42 where @xmath43 are four - dimensional spacetime indices , and @xmath44 label the four - dimensional gauge fields : @xmath17 from the vector multiplets and one , the graviphoton , from the gravity multiplet . for later convenience we use a formulation of the theory which contains @xmath45 complex scalar fields @xmath22 which are subject to local dilatations and @xmath0 transformations . the @xmath17 physical scalars remaining after gauge fixing can be parametrised as @xmath46 , where @xmath47 . while the physical scalars @xmath20 parametrise a projective special khler ( psk ) manifold , the @xmath22 parametrise a conic affine special khler ( cask ) manifold , which is a complex cone over the psk manifold . conversely , the psk manifold can be obtained as the khler quotient of the cask manifold with respect to the @xmath24-action generated by dilatations and @xmath0 transformations . in physical terms this quotient amounts to gauge fixing the local @xmath24 action , as discussed below . all terms in ( [ 4dlag ] ) except the scalar potential @xmath48 are completely determined by the holomorphic prepotential @xmath27 , which is homogeneous of degree 2 . prior to gauge fixing dilatations , the space - time ricci scalar , @xmath49 , is multiplied by the conformal compensator @xmath50 where derivatives of the prepotential are denoted @xmath51 , etc . the tensor @xmath52 is the horizontal lift of the physical ( psk ) scalar metric to the cask manifold . it has a two - dimensional kernel which reflects the fact that the @xmath22 only represent @xmath17 complex physical degrees of freedom . the vector couplings are given by @xmath53 where @xmath54 . we now turn to the @xmath24 gauge fixing . dilatations are fixed by imposing the d - gauge @xmath55 which in particular brings the einstein - hilbert term in ( [ 4dlag ] ) to the standard form @xmath56 . likewise @xmath0 transformations can be fixed by imposing any condition transverse to the @xmath0 action , such as @xmath57 . however , as discussed in more detail in @xcite , it is often advantageous to postpone @xmath0 gauge fixing until reducing the theory and starting to solve the resulting equations of motion . in particular , upon imposing the d - gauge one has @xmath58 where @xmath59 is the positive definite ( psk ) metric of the physical scalars @xmath20 . working with the scalars @xmath22 has the advantage that we retain covariance under symplectic transformations , and will result in a more convenient form of the equations of motion after reduction . note that while the d - gauge removes one real degree of freedom from the @xmath22 , the second unphysical degree of freedom is taken care of by the remaining local @xmath0 symmetry , see @xcite for details . geometrically , imposing the d - gauge while keeping the local @xmath0 symmetry corresponds to working on a @xmath0 principal bundle over the psk manifold . the four - dimensional lagrangian also includes a scalar potential @xmath60 , which as in @xcite is given as @xmath61 with a superpotential of the form @xmath62 where @xmath63 parametrize the @xmath0 gauging . since superpotentials of the form ( [ superpot ] ) arise in flux compactifications , we refer to them as magnetic and electric fluxes , respectively . note that we have included an explicit factor of @xmath64 in using dimensional analysis . we will use this later to rewrite the potential in terms of real variables . for reference , we note that the @xmath22 have mass dimension @xmath65 while the flux parameters have dimension @xmath66 , so that @xmath67 has dimension @xmath68 . since @xmath69 and , hence , its inverse @xmath70 are homogeneous of degree 0 , they have dimension 0 , and @xmath71 has dimension @xmath72 , as required . we also remark that for later convenience we have re - scaled the flux parameters by a factor of 2 relative to @xcite . moreover , we have not factorized the flux parameters into a dimensionful coupling and dimensionless parameters , but kept them dimensionful . imposing that the background is stationary , so that all of the fields are independent of time , we can reduce the four - dimensional action over a time - like direction in order to obtain an effective three - dimensional euclidean action . we decompose the four - dimensional metric as @xmath73 where @xmath29 and @xmath74 are the kaluza - klein scalar and vector respectively , and we have left the three - dimensional part of the metric undetermined for now . following the procedure for time - like dimensional reduction outlined in @xcite , and noting that the scalar potential remains unchanged throughout the reduction process , one obtains the three - dimensional lagrangian @xcite @xmath75 we have written all of the three - dimensional degrees of freedom using the conventions of the real formulation of special geometry developed in @xcite , and afterwards set @xmath76 for the remainder of the paper . while we give a brief summary here , more details can be found in appendix [ appendix : realspecial ] . the three - dimensional action contains @xmath31 scalar fields @xmath77 which are subject to one local @xmath0 symmetry and hence has @xmath78 independent scalar degrees of freedom . while the @xmath25 combine the four - dimensional scalars @xmath20 with the kaluza - klein scalar @xmath29 , the @xmath33 contain the degrees of freedom of the four - dimensional gauge fields , and @xmath34 is dual to the kaluza - klein vector . the constant tensor @xmath79 is the symplectic form of the cask manifold expressed in real variables @xmath25 . the tensor @xmath80 is given by @xmath81 where the hesse potential @xmath82 is related to the prepotential @xmath83 by the legendre transformation given in . as shown in the appendices , the scalar potential @xmath71 is given in terms of the real coordinates as @xmath84 \;,\ ] ] where the dual scalars @xmath85 are defined by @xmath86 . substituting this expression into the three - dimensional lagrangian becomes @xmath87 in the following we will restrict ourselves to static solutions , i.e. set @xmath88 in , for which the final term in vanishes @xcite . the equations of motion for @xmath33 are then given by @xmath89 whilst those for @xmath25 read @xmath90 finally , the three - dimensional einstein equations are @xmath91 we concentrate in this paper on purely imaginary ( pi ) field configurations , which we define to be those for which the complex scalars are rescaled versions of the scalars @xmath22 . see for the definition . ] @xmath92 are purely imaginary @xcite . moreover , we restrict ourselves to a class of prepotentials of the form @xmath93 where the function @xmath19 is homogeneous of degree three and real - valued when evaluated on real fields . for the case of ungauged @xmath13 supergravity , such models were considered in @xcite . note that those models with @xmath19 a cubic polynomial are precisely the ` very special ' prepotentials for which the solutions can be uplifted to five dimensions . since we choose to fix the @xmath0 gauge by taking @xmath94 , this is equivalent to setting @xmath95 to zero . models obtainable from five dimensions are invariant under constant shifts @xmath96 , and , hence , pi configurations will be referred to as ` axion - free ' . for the class of models the scalar fields @xmath25 take the form @xcite @xmath97 and hence we see that @xmath98 . following @xcite we extend the pi condition to the scalars @xmath33 by imposing @xmath99 which sets also @xmath100 . the quantities @xmath101 and @xmath102 encode the four - dimensional field strengths , see ( [ fieldstrengths ] ) . in the same way , we extend the pi condition to the fluxes @xmath103 by imposing @xmath104 which sets @xmath105 . we then find that the equations of motion and the three - dimensional einstein equations simplify to @xmath106 @xmath107 and @xmath108 it turns out to be useful to write the equations of motion in terms of the dual variables @xmath85 and @xmath109 defined in appendix [ appendix : realspecial ] . in terms of these , the equations become @xmath110 @xmath111 and @xmath112 in the next section we will look for solutions of which can be lifted to regular non - extremal black branes in four dimensions . our aim in this section is to construct a family of non - extremal black branes in the @xmath13 gauged supergravity theory with prepotential . restricting our attention to the pi configurations described in section [ ssecpi ] , it can be shown that the hesse potential takes the form@xcite @xmath113 for general functions @xmath19 , the form of the metric @xmath114 is fairly complicated @xcite . however , since the field @xmath115 decouples from the rest , we can compute @xmath116 and this will be sufficient to find solutions valid for any choice of @xmath19 . we remark here upon a slight abuse of notation which we will make throughout the remainder of this paper . specifically , we denote by @xmath117 with @xmath118 those scalar fields which are actually the @xmath119th components of the vector @xmath120 . the same is true of the components @xmath121 of the metric , which should properly be the @xmath122 components of @xmath80 . this notation is convenient since @xmath123 are the remaining non - trivial @xmath25-fields within our ansatz . for simplicity we will concentrate on solutions which are supported by a single electric charge @xmath124 and electric fluxes @xmath125 in this section however , as we will see in section [ secmagbrane ] , the methods introduced in the following can be easily extended to deal also with solutions with a single magnetic charge switched on and sourced by both electric and magnetic fluxes . the systematic investigation of dyonic black branes will be carried out in a future publication @xcite . we make a brane - like ansatz for the three - dimensional metric : @xmath126 where @xmath127 is some function to be determined . this form of the metric can always be obtained from the more commonly used form @xmath128 by a suitable redefinition @xmath129 . we also impose that all fields @xmath85 and @xmath109 depend only on @xmath130 . the coordinate @xmath130 has been chosen such that it is an affine parameter for geodesic curves on the scalar target space parametrized by @xmath25 and @xmath33 . equivalently , the @xmath130-dependent part of the three - dimensional laplace operator is given by @xmath131 . the non - zero components of the ricci tensor are given by @xmath132 where the dot denotes differentiation with respect to @xmath130 . with this choice the three - dimensional einstein equations become @xmath133 for @xmath134 and @xmath135 for @xmath136 , where we have used . equation is the hamiltonian constraint which needs to be imposed on solutions @xmath137 of the second order scalar field equations . we remark that since we have consistently reduced the full field equations , we do not need to impose this constraint by hand , but have retained it as a field equation following from an action principle . we now turn to the equations of motion for the fields @xmath85 and @xmath109 . we start with the @xmath109 equations of motion , which read simply @xmath138 and can be integrated once to find @xmath139 for some constants @xmath140 , which are proportional to the electric and magnetic charges of the solution , @xmath141 @xcite . the explicit relations between the @xmath109 and the field strengths can be found in appendix [ appendix : realspecial ] . for the case at hand we only have a single electric charge @xmath124 , and so the only non - zero component of @xmath142 is @xmath143 . we turn now to the @xmath85 equations of motion , which become @xmath144 for the models without magnetic flux , @xmath145 , on which we concentrate in this section , the @xmath115 equation of motion decouples from the others . indeed , using with @xmath146 the @xmath115 equation of motion becomes @xmath147 this takes precisely the same form as in the ungauged case @xcite and can be solved with @xmath148 for some constants @xmath149 and @xmath150 . since the solution is invariant under @xmath151 , we can take @xmath152 without loss of generality . it will turn out that @xmath149 acts as a non - extremality parameter for the full solution . furthermore , as we will see later explicitly , @xmath130 naturally takes values @xmath153 . thus in order that @xmath154 for @xmath155 we will have to require @xmath156 . the @xmath117 equations of motion , for @xmath118 , become @xmath157 multiplying by @xmath158 and summing over @xmath159 gives @xmath160 where we have made use of the homogeneity properties of the metric @xmath80 , viz . @xmath161 and @xmath162 . one can now compare this equation to , which for the model at hand becomes @xmath163 substituting from this into the last two terms of we obtain @xmath164 the left - hand side of this equation can be rewritten as a total derivative @xmath165 and so we can integrate to find @xmath166 for some integration constant @xmath167 , where we have chosen the factor for later convenience . now , using the identity @xmath168 @xcite one can show furthermore that @xmath169 substituting this expression into and further integrating gives @xmath170 where we have used the definition of @xmath171 given in , and have chosen the definition of the integration constant @xmath172 for later convenience . substituting the explicit expression for the hesse potential we therefore find @xmath173 let us now return to the hamiltonian constraint which , upon substituting the expression , becomes @xmath174 so far we have the following picture : the equations of motion for the @xmath117 are given by the set of coupled equations . the solutions @xmath175 of should then satisfy the two constraints and . we proceed by imposing that the @xmath117 are all proportional , which will in turn mean that all of the physical scalar fields @xmath20 are proportional to one another . specifically , we set @xmath176 for some constants @xmath177 . in terms of this ansatz , the constraints and ( the derivative of ) become @xmath178 we have made use here of the homogeneity properties of @xmath19 and the metric @xmath114 , as well as the identity @xmath179 @xcite which implies , for the models at hand , that @xmath180 the two equations can be combined into a second - order non - linear differential equation for @xmath181 : @xmath182 introducing the variable @xmath183 this becomes @xmath184 for @xmath185 and hence can be solved by @xmath186 where @xmath187 and @xmath188 are integration constants , and we have taken @xmath189 to be the positive root without loss of generality . note that the right hand side should be non - negative for all @xmath190 , and hence we should pick @xmath191 and @xmath192 . the solution now determines the function @xmath181 appearing in the metric ansatz in terms of some integration constants , which we will fix in section [ ssecphysicality ] . we can now use to find an expression for @xmath193 . indeed , differentiating with respect to @xmath130 and substituting into the second equation in we obtain @xmath194 this can be integrated up to find @xmath195 where @xmath196 is an integration constant . since we have set all of the @xmath117 proportional to each other , we can therefore write @xmath197 for some constants @xmath198 . substituting this into we find that @xmath199 , and that the @xmath117 equation of motion is satisfied provided the integration constants @xmath200 are related to the electric fluxes @xmath201 via @xmath202 returning to then determines the constant @xmath172 in terms of @xmath187 and the fluxes @xmath201 as @xmath203 finally , the kaluza - klein scalar @xmath29 appearing in the metric ansatz is determined in terms of the @xmath85 via the @xmath204-gauge condition and the explicit form of the hesse potential . to summarise , we find that the scalars @xmath85 are given by @xmath205 whilst the metric degrees of freedom are given by @xmath206 the @xmath207 signs in should be chosen such that the function @xmath208 is well - defined . in this section we want to look at the conditions on the various integration constants which give rise to regular black brane solutions in four dimensions . in particular , we impose that our solution has finite entropy density , which is the relevant regularity condition for solutions with non - compact horizon . let us recall the form of the four - dimensional metric in the @xmath130 coordinates : @xmath209 we will see below that for a suitable choice of integration constants @xmath210 is an event horizon , while @xmath211 is the asymptotic regime at infinite distance . the regularity of the solution within the bulk between horizon and infinity depends on the detailed properties of the function @xmath19 . in particular , when evaluating @xmath19 on the solution , we require that it has neither zeroes ( so that there are in particular no changes of sign of @xmath208 ) nor poles . given the experience with similar issues for black hole solutions and domain walls , one expects that such solutions exist for any prepotential arising in string theory upon suitable restriction of the integration constants @xcite . in any case , such questions can only be investigated explicitly on a case - by - case basis , while we restrict ourselves to questions that can be answered irrespective of the choice of @xmath19 . the position of the event horizon can be found by looking at the value of @xmath130 for which the norm of the killing vector field @xmath212 vanishes . since @xmath213 as @xmath214 , we can identify the horizon with the limiting value @xmath214 provided @xmath215 . if @xmath216 then the position of the horizon will change depending on the relative magnitudes of @xmath217 and @xmath149 , and so we will take @xmath215 in what follows . the area of the horizon is given by @xmath218 which is divergent since the @xmath219 and @xmath220 coordinates are non - compact . however , we can still define a finite entropy density @xmath221 provided the factor @xmath222 remains finite . from the expressions one can show that in this limit we have @xmath223 in order that this be finite and non - zero at the horizon we therefore require @xmath224 which turns out to be equivalent to fixing @xmath225 . note that in this case we likewise have @xmath226 . we still at this stage have four integration constants @xmath227 which are _ a priori _ yet to be determined . however , note that we can always absorb @xmath188 into a shift of @xmath130 and a redefinition of the constants @xmath187 and @xmath150 . indeed , it will be useful to set @xmath228 at this stage so that the asymptotic region of the solution is at @xmath229 . moreover , we see that in the extremal @xmath230 limit , the expression becomes @xmath231 . hence , we can scale @xmath130 to set @xmath232 , matching the conventions of the extremal nernst brane of @xcite . we are therefore left with a two - parameter family of solutions to the three - dimensional equations of motion , parametrised by @xmath149 and @xmath150 , which we will interpret in terms of thermodynamic quantities in section [ ssecproperties ] . before moving on to study properties of the solution , we summarise the results so far : the scalars @xmath85 and @xmath109 are given by @xmath233 whilst the metric degrees of freedom are given by @xmath234 the physical scalar fields @xmath92 can be determined from the expressions @xmath235 which were obtained in @xcite , see appendix [ appendix : realspecial ] . we find @xmath236 note that for @xmath237 , @xmath115 and @xmath117 all behave as @xmath238 when @xmath239 . we will show in the following section that this implies that the physical scalar fields take finite values on the horizon for @xmath237 . we now turn to an analysis of various properties of the solution obtained in section [ ssecphysicality ] , postponing a fuller discussion to section [ secdiscussion ] . it is convenient to introduce the radial coordinate @xmath240 via @xmath241 with this definition , the asymptotic region is situated at @xmath242 , while the horizon is at @xmath243 . in terms of @xmath240 , we find the expressions @xmath244 where we have introduced the function - model . ] @xmath245 .\ ] ] the physical scalar fields @xmath246 then take the form @xmath247 hence , for @xmath248 we find the asymptotic behaviour @xmath249 , whilst for @xmath250 we find @xmath251 . the four - dimensional line element becomes @xmath252 where we have found it convenient to define @xmath253 from this form of the metric , it is clear that the limit @xmath230 can be achieved simply by setting @xmath254 and @xmath255 in this case we reproduce the extremal nernst brane solutions of @xcite , albeit in different coordinates . this identifies @xmath149 as a parameter encoding the non - extremality of the solution . for @xmath250 , the harmonic function for both the extremal and non - extremal solutions becomes @xmath256 . the line element then becomes @xmath257 where we have defined @xmath258 with the sign chosen such that @xmath259 is positive . the corresponding extremal solution can be obtained by setting the ` blackening factor ' @xmath254 in . to investigate the near - horizon behaviour of the line element , we define @xmath260 and zoom in on the region @xmath261 . we then find that for @xmath237 the near - horizon metric looks like @xmath262 which is the product of a two - dimensional rindler spacetime with two - dimensional flat space . we also include , for comparison , the near - horizon behaviour of the extremal solution which , after putting @xmath263 , becomes @xmath264 .\ ] ] by wick rotating to euclidean time @xmath265 in and enforcing regularity of the @xmath266 circle we can read off the temperature @xmath267 we can also read off from the entropy density of the solution , which is given by @xmath268 note that from and we can eliminate the integration constant @xmath149 in terms of the thermodynamic quantities @xmath221 and @xmath269 via . we now turn to a consideration of the asymptotic @xmath242 properties of the line element , which for @xmath248 becomes @xmath271.\ ] ] note that this is the same for both the extremal and non - extremal solutions . making the coordinate change @xmath272 then brings this to the form @xmath273 , \ ] ] which is conformally @xmath274 with boundary at @xmath275 . for the case @xmath250 , the asymptotic limit corresponds simply to @xmath276 in , from which we find the asymptotic line element , after a suitable coordinate redefinition . the gauge field strength @xmath277 is determined from the scalar field @xmath278 via : @xmath279 substituting in the expression and integrating with respect to @xmath130 gives @xmath280,\ ] ] where we have chosen the integration constant such that @xmath281 , i.e. that the gauge fields vanish on the horizon . the chemical potential @xmath5 is then given by the asymptotic value of @xmath282 , @xmath283 , \ ] ] which diverges as @xmath284 . note that in the extremal limit @xmath230 with @xmath248 we get @xmath285 . we are now in a position to relate the integration constants @xmath149 and @xmath150 appearing in our solution to the thermodynamic quantities @xmath221 , @xmath269 and @xmath5 . in particular , we can rearrange to find @xmath286 where we have used . returning to we then find an equation determining the entropy density as a function of the electric charge @xmath124 , fluxes @xmath287 , temperature @xmath269 and chemical potential @xmath5 of the black brane : @xmath288 one consequence of is that , if we keep @xmath259 , @xmath124 and @xmath5 fixed and send @xmath289 , we see that @xmath290 , which is precisely the strict ( planckian ) formulation of the third law of thermodynamics @xcite . this identifies the solution constructed in section [ ssecphysicality ] as a non - extremal ( ` hot ' ) nernst brane . we can further analyse by looking at the dimensionless ratio @xmath291 . when @xmath291 is small , the second term in becomes negligible , and we find that the entropy density behaves as @xmath292 . on the other hand , when @xmath291 becomes large , the second term in dominates , and we find the behaviour @xmath293 . in figure [ stdiagram ] we plot equation for various values of @xmath5 , keeping @xmath124 and @xmath259 fixed . this shows a ) the nernst law behaviour @xmath290 as @xmath289 , and b ) the crossover from the behaviour @xmath292 to @xmath293 . varies with temperature @xmath269 for various values of the chemical potential @xmath5 , and with @xmath124 and @xmath259 fixed . ] we now turn our attention to a simple reformulation of the procedure in section [ stusec ] which for a certain class of prepotentials allows us to construct non - extremal black branes carrying magnetic charge . we will here simply present the supergravity solution , and leave a fuller discussion of the thermodynamics of magnetically - charged black branes for future work . in particular , we are interested in prepotentials for which one of the fields @xmath294 decouples from the others . without loss of generality , we can assume that @xmath295 decouples , and consider prepotentials of the form @xmath296 where the function @xmath297 is homogeneous of degree 2 . this class is particularly interesting from the perspective of embedding the model into string theory as it contains the tree - level heterotic prepotentials , which are linear in the heterotic dilaton @xmath298 . we consider black brane solutions which are supported by a single magnetic charge @xmath299 , a magnetic flux @xmath300 , and electric fluxes @xmath301 . in this case we see that the equations of motion can be solved in precisely the same way as in section [ stusec ] , with the field @xmath302 and magnetic charge @xmath299 playing the role of @xmath115 and @xmath124 in the preceding section . in particular , we have @xmath303 whilst @xmath115 and @xmath304 take the same form as after replacing @xmath305 with @xmath300 in the obvious place . moreover , the function @xmath306 remains unchanged and , since @xmath307 is symmetric in @xmath115 and @xmath302 , we find that the line element takes the same form as in section [ stusec ] . looking at the near - horizon behaviour we again find that regularity of the solution imposes the same relation between the integration constants , @xmath225 , as before . the entropy density is therefore @xmath308 whilst the temperature of the solution is given by @xmath309 in this paper we have provided a new technique for the construction of non - extremal black brane solutions to large classes of @xmath13 @xmath0 gauged supergravity models , utilising the techniques of time - like dimensional reduction followed by a rewriting of the effective three - dimensional degrees of freedom through the real formulation of special geometry . in section [ stusec ] we explicitly constructed a family of non - extremal black branes supported by a single electric charge and an arbitrary number of electric fluxes . this family of branes has an entropy density behaving as @xmath2 for @xmath37 , which therefore vanishes at @xmath310 , where we recover the extremal nernst brane solutions of @xcite . we anticipate that such non - extremal nernst branes will have interesting applications in the context of holography , where they could prove useful in describing dual field theory configurations at finite temperature and chemical potential which satisfy the nernst law . one issue with regards to a holographic interpretation is that our solutions do not fit naturally into the framework of ads / cmt , since they do not asymptote to @xmath274 , but rather conformal @xmath274 , as seen in . hence , the stress tensor of the dual field theory in the uv would not be scale invariant . however , in recent years much progress has been made in understanding the holographic description of such ` hyperscaling violating ' theories , as well as the more general class of hyperscaling violating lifshitz ( hvlif ) theories @xcite , which we now review . consider spacetime geometries of the form ( we use the conventions of @xcite ) @xmath311 where @xmath312 label the spatial directions on the boundary , @xmath38 is the ` dynamical critical ' ( lifshitz ) exponent , and @xmath39 is the ` hyperscaling violating ' exponent . note that for @xmath313 , @xmath314 one recovers the metric on @xmath315 . by looking at the near - horizon and boundary behaviour of our solutions , we see that the nernst brane interpolates between two hvlif geometries with @xmath316 . there are four cases of interest , corresponding to whether @xmath150 and @xmath149 are zero or non - zero : * @xmath250 , @xmath317 : the solution becomes globally hvlif with @xmath8 . it has zero temperature and infinite chemical potential . * @xmath250 , @xmath237 : the solution has finite temperature and infinite chemical potential , and interpolates between a near - horizon rindler geometry , with @xmath6 , and an asymptotic hvlif geometry with @xmath8 . * @xmath248 , @xmath317 : the solution has zero temperature and a finite chemical potential . it interpolates between a hvlif geometry with @xmath8 at the horizon , and the conformal @xmath274 geometry with @xmath7 at infinity . this is the nernst brane solution of @xcite . * @xmath248 , @xmath237 : the solution has finite temperature and chemical potential , and interpolates between a near - horizon rindler geometry with @xmath6 and the conformal @xmath274 geometry with @xmath7 at infinity . note that all of these values are consistent with the constraints imposed by the null energy condition @xcite . we have therefore found , _ analytically _ , a family of solutions which interpolate between two hvlif geometries . this family is parametrised by the two integration constants @xmath149 and @xmath150 , or equivalently by the temperature @xmath4 and chemical potential @xmath5 of the solution , both of which can be freely varied . both parameters have a distinct effect on the near horizon and asymptotic forms of the solution : while the extremal or zero temperature limit @xmath318 changes the near horizon solution from @xmath6 to @xmath8 , the infinite chemical potential limit @xmath284 changes the geometry at infinity from @xmath7 to @xmath8 . if both limits are performed we obtain a global hvlif solution with @xmath8 which we interpret as the ground state of the given charge sector . note that like any lifshitz solution different from ads it is not geodesically complete , and that the scalars are non - constant and run off to zero or infinity in the asymptotic regions . however , a similar behaviour can occur for domain wall solutions in gauged supergravity which , for lack of more symmetric solutions , are interpreted as ground states . sometimes this interpretation can be further justified by an embedding into string theory or m - theory , see for example @xcite . while we leave studying the string theory embedding of our solutions for future work , we remark that the interpretation is consistent with a limit where the temperature is zero and the chemical potential infinite . since so far solutions interpolating between hvlif geometries have only been found by relying on a mixture of analytical and numerical methods , we have made a significant step forward , and expect that the techniques used and described in this paper will be useful in making further progress . while we leave searching for a concrete holographic dual of the bulk geometries presented in this paper to future work , we can already make some interesting observations which shed some light on the properties which such a putative dual theory might possess . let us first consider the extremal ( @xmath317 ) solution with @xmath250 . since this is the gravitational ground state solution with @xmath8 , zero temperature and infinite chemical potential , we expect it to be dual to the ground state of a @xmath319-dimensional qft with hyperscaling exponent @xmath320 and lifshitz exponent @xmath321 . we remark that the specific value @xmath320 for a qft in @xmath316 space dimensions seems to be required for the description of states with hidden fermi surfaces , although a three - loop calculation gives @xmath322 rather than @xmath321 @xcite . now consider turning on some finite temperature @xmath323 on the field theory side . by a simple scaling argument , one can argue @xcite that the entropy density of the thermal state is related to the temperature as @xmath324 . we therefore expect that the non - extremal nernst brane with @xmath250 in provides us with the relevant gravity dual to the @xmath319-dimensional qft with @xmath320 and @xmath321 at finite temperature . indeed , taking @xmath325 in the relation we see that the entropy density of the brane solution is related to the temperature as @xmath2 which is the expected behaviour from the field theory arguments , and therefore consistent with our tentative interpretation . we now move on to consider what happens at finite chemical potential @xmath326 , which corresponds to @xmath248 . in this case , the extremal nernst brane interpolates between a hvlif geometry with @xmath8 at the horizon , and a hvlif with @xmath7 at infinity , which is conformal to @xmath274 . one possible interpretation is as an rg flow between two qfts : one with hyperscaling exponent @xmath327 in the uv ; and one with hyperscaling exponent @xmath320 and lifshitz exponent @xmath321 in the ir . as the gravity solution is smooth , and we do not seem to have a natural candidate for an order parameter identifying a phase transition , we think that the more likely interpretation is that the uv ` phase ' and the ir ` phase ' are related by smooth crossover . for the ir theory we expect that the entropy scales like @xmath328 , which agrees with the behaviour of the nernst brane solution for low temperature @xmath329 . adding temperature changes the near horizon geometry , but leaves the asymptotic geometry at infinity unchanged , which is consistent with interpreting these configurations as thermal states . we therefore expect that the ir behaviour is correctly described by the nernst brane solution , which in turn predicts a scaling @xmath330 of the entropy for high temperatures , @xmath331 . this however does not agree with the expected scaling of our tentative uv theory with @xmath7 , which predicts @xmath332 . we also note that the asymptotic uv geometry , while conformal to @xmath274 , can not be interpreted as an alternative ground state of our supergravity theory , because it is not , when taken as a global geometry , part of our family of solutions . moreover , the physical scalar fields @xmath333 run off to infinity in the uv region , which indicates strong coupling or decompactification . taken together this suggests that the description in terms of our four - dimensional gauged supergravity theory is incomplete in the uv , and that further degrees of freedom become relevant . if we accept that the uv geometry correctly captures the thermodynamic behaviour then the corresponding uv theory should have a scaling behaviour @xmath334 ( @xmath335 ) . the resulting tentative phase diagram is shown in figure [ phasediagram ] . the above mentioned analogy with domain walls together with the runaway behaviour of the scalars suggests to interpret the uv behaviour as a decompactification limit and to embed the four - dimensional supergravity theory into a higher dimensional theory . since the class of prepotentials that we have considered in this paper includes those ` very special ' prepotentials for which the theory can be uplifted to five dimensions , the most obvious embedding is into five - dimensional supergravity . there are grounds to believe @xcite that the dimensional reduction of theories admitting @xmath336 vacua would admit vacua with some nontrivial hvlif behaviour . therefore we expect that by lifting our solutions to five dimensions we will obtain new asymptotically @xmath337 finite temperature solutions in @xmath13 gauged supergravity which still satisfy the nernst law hot nernst solutions were constructed in @xcite , their solutions do not reduce to the finite temperature solutions presented here . ] . we will expand on this point in @xcite , and remark that an asymptotic @xmath337 leads to a scaling of the entropy @xmath338 , ( @xmath339 ) , which is consistent with our proposed uv theory . we should also point out that there are issues with the interpretation of our solutions if the temperature is strictly zero , since the nernst brane solution has infinite tidal forces and run - away behaviour of the scalars at the horizon in the extremal limit . this again indicates a breakdown of the effective description , and strictly speaking the supergravity solution should only be trusted at low but finite temperature . thus , as in the similar case of the holographic interpretation of hyperscaling violating solutions of einstein - maxwell - dilaton theories @xcite , the nernst brane solution is not a valid description of its ( tentative ) dual over the full range of the energy ( radial coordinate ) from the uv ( infinity ) to the ir ( horizon ) , but only over a finite interval outside the horizon . we leave it to future work to characterize the range of validity more quantitatively , and to identify the necessary completions in the uv and ir using a string theory embedding . one possible strategy to further investigate the zero temperature limit is to adapt formalisms that allow to include higher derivative terms . in @xmath1 supergravity a certain class of higher derivative terms ( those encoded in the so - called weyl multiplet ) , which are related to the topological string , lead to a generalization of the framework of special geometry , on which we relied in the article @xcite . one could also try to adapt the entropy function formalism @xcite , which employs universal properties of near horizon geometries and does not depend on supersymmetry . finally we comment on further possible future directions on the gravity side . here it would be interesting to find solutions where other and possibly more charges and fluxes have been turned on . we expect that our formalism is particularly suited to finding dyonic solutions , due to its built - in electric - magnetic covariance @xcite . for work in this direction it is encouraging that work on static bps solutions in @xmath0 gauged supergravity solutions with symmetric scalar target spaces has led to the construction of the general dyonic solution @xcite . we think that the systematic methods and explicit analytical solutions interpolating between hvlif geometries that we have presented in this paper will help to make progress towards a classification of solutions in gauged supergravity , and of the hvlif landscape , and to extend and deepen our understanding of the field theory / gravity dictionary . we would like to thank g. lopes cardoso for drawing our attention to the problem of constructing non - extremal versions of nernst branes in gauged supergravity , and for many further useful discussions . we also thank p. athanasopoulos , m. haack , s. gentle , p. rakow and o. vaughan for useful discussions . the work of pd is supported by stfc grant st / j50113x/1 and national research foundation of korea grants 2005 - 0093843 , 2010 - 220-c00003 and 2012k2a1a9055280 . the work of de is supported by stfc studentship st / k502145/1 . the work of tm is partially supported by the stfc consolidated grants st / g00062x/1 and st / l000431/1 . in this appendix we review the real formulation of special geometry introduced in @xcite , based on the work of @xcite , and extend it to include scalar potentials of the form ( [ scalpot ] ) , which result from a flux superpotential ( [ superpot ] ) . starting from the holomorphic formulation , where the complex scalars @xmath22 parametrise a conic affine special khler ( cask ) manifold , and where all vector multiplet couplings are encoded in a holomorphic prepotential @xmath27 , which is homogeneous of degree two , one introduces special real coordinates @xmath340 , where @xmath341 note that @xmath51 is homogeneous of degree one . in the real formulation all vector multiplet couplings are encoded in a hesse potential @xmath28 , which is homogeneous of degree two , and which is obtained from the imaginary part of the holomorphic prepotential by a legendre transformation , which replaces @xmath342 by @xmath343 as an independent variable : @xmath344 the special real coordinates @xmath25 are darboux coordinates , and the khler form on the cask manifold is simply @xmath345 it is useful to note that the first derivatives @xmath346 of the hesse potential are related to the imaginary parts of @xmath22 and @xmath347 by @xmath348 and provide an alternative , ` dual ' coordinate system on the cask manifold . to obtain the associated projective special khler ( psk ) manifold , one imposes the d - gauge @xmath349 , together with a condition which fixes a @xmath0 gauge . if one wants to preserve symplectic covariance , one postpones fixing a @xmath0 gauge and retains a local @xmath0 gauge invariance . geometrically this corresponds to working on the total space of a @xmath0 principal bundle over the psk manifold . in @xcite it was shown how to express all couplings appearing in the bosonic part of the vector multiplet lagrangian in terms of real coordinates . in particular the cask metric @xmath350 is replaced by the hessian metric @xmath351 for the purpose of this paper we need to rewrite the scalar potential @xmath48 of ( [ scalpot ] ) , and the associated flux superpotential @xmath352 of ( [ superpot ] ) , in terms of real coordinates . using that @xmath353 by homogeneity , and using the formulae given above , it is straightforward to obtain @xmath354 where we have defined @xmath355 . in order to obtain the potential @xmath71 as given in , we must compute the derivatives @xmath356 since this derivative involves the real coordinates @xmath357 rather than @xmath358 , we apply the chain rule to @xmath359 and compute @xmath360 after decomposing the second derivatives of the prepotential @xmath83 into real and imaginary parts ( including a conventional factor of 2 ) by @xmath361 , one can apply the chain rule to show that @xmath362 and read from @xcite that @xmath363 combining this , we find @xmath364 finally , we can put all of this together to obtain @xmath365 where @xmath366 , @xmath367 is the inverse hessian metric on the cask manifold ( see @xcite ) , and @xmath368 is the inverse of @xmath369 . using , we have that @xmath370 this can be substituted into , which after simplification becomes @xmath371 where we have used the identity @xmath372 @xcite . the final expression for the scalar potential given purely in terms of real coordinates then comes from using and as follows : @xmath373 , \end{aligned}\ ] ] where we have used homogeneity @xmath374 . lastly , we substitute the @xmath204-gauge condition @xmath375 into to obtain @xmath376.\ ] ] note that the expression within the square brackets is homogeneous of degree zero . this is useful in order to rewrite @xmath71 in terms of rescaled variables after dimensional reduction . we shall now define the various terms appearing in the three - dimensional lagrangian , which uses a modified version of the real formulation of special geometry that is adapted to dimensional reduction . we follow the conventions of @xcite , to which we refer the reader for further details . firstly , the complex scalar fields , @xmath22 , appearing in the four - dimensional lagrangian , are replaced by rescaled scalars @xmath377 where @xmath29 is the kaluza klein scalar . in the four - dimensional theory parametrised by the @xmath22 the radial direction of the cask manifold , which is generated by the vector field @xmath378 is a gauge degree of freedom . the above rescaling promotes it to a physical degree of freedom , which is equivalent to the kaluza - klein scalar . it turns out that this rescaling leads to a convenient parametrization of the reduced three - dimensional theory . rewriting the d - gauge condition ( [ dgaugex ] ) in terms of @xmath379 , we obtain @xmath380 which determines @xmath29 in terms of the scalar fields @xmath379 . due to the homogeneity properties of the prepotential and hesse potential , we can obtain a real parametrization which is based on the rescaled complex scalars @xmath379 . the associated real coordinates are defined by the decomposition @xmath381 as @xmath382 furthermore , after reducing to three dimensions it is possible to write the gauge degrees of freedom using scalar fields as well . in particular , we define @xmath383 where @xmath384 are the components of the four - dimensional gauge fields @xmath385 along the reduction direction , and @xmath386 are the hodge - duals of the three - dimensional vector parts . specifically , these scalars descend from the four - dimensional field strengths as follows : @xmath387 where @xmath388 are defined as @xmath389 we can make further use of hodge duality to encode the kaluza - klein vector degree of freedom using the scalar field @xmath34 @xcite , although we will not need this here since we deal only with static configurations . in terms of rescaled complex scalars @xmath379 and rescaled real variables @xmath25 , the relation between prepotential @xmath390 and hesse potential @xmath28 is @xmath391 we also note that the d - gauge , when expressed in terms of rescaled real scalars , reads @xmath392 in the lagrangian , we also use the tensor field @xmath393 this tensor can be interpreted as a metric on the cask manifold , which is related to @xmath394 by flipping the signature along the radial direction generated by the field @xmath395 , combined with a conformal transformation which changes the scale transformation @xmath396 , where @xmath397 , from being a homothety to being an isometry . this follows from the obvious fact that while @xmath398 is homogeneous of degree @xmath399 , @xmath400 is homogeneous of degree @xmath401 . note that the metric coefficients @xmath394 and @xmath80 are homogeneous of degrees 0 and @xmath66 , respectively . both tensors are related by @xmath402 it will be convenient for us to introduce a set of dual coordinates with respect to the metric @xmath80 defined by @xmath403 one can show that @xmath404 where we have used that @xmath405 is homogeneous of degree @xmath65 for the first identity and the chain rule for the second . finally , we re - express the scalar potential in terms of variables adapted to dimensional reduction . since , as we remarked , the expression in the square brackets of ( [ v ] ) is homogeneous of degree zero , it remains invariant if we rescale the real coordinates @xmath25 by @xmath408 . to express @xmath71 in terms of the tensor @xmath80 we use the relation to write @xmath409.\ ] ] finally , we use to re - write @xmath71 in terms of the dual coordinates @xmath85 , and take into account that upon dimensional reduction the term @xmath410 in the lagrangian gets multiplet by @xmath411 and obtain @xmath412.\ ] ] [ 2]#2 x. dong , s. harrison , s. kachru , g. torroba , and h. wang , _ aspects of holography for theories with hyperscaling violation _ , _ jhep _ * 1206 * ( 2012 ) 041 , [ http://xxx.lanl.gov/abs/1201.1905 [ arxiv:1201.1905 ] ] . a. chamblin , r. emparan , c. v. johnson , and r. c. myers , _ charged ads black holes and catastrophic holography _ , _ phys.rev . _ * d60 * ( 1999 ) 064018 , [ http://xxx.lanl.gov/abs/hep-th/9902170 [ hep - th/9902170 ] ] . g. lopes cardoso , b. de wit , and t. mohaupt , _ corrections to macroscopic supersymmetric black hole entropy _ , _ phys.lett . _ * b451 * ( 1999 ) 309316 , [ http://xxx.lanl.gov/abs/hep-th/9812082 [ hep - th/9812082 ] ] . g. lopes cardoso , b. de wit , j. kappeli , and t. mohaupt , _ stationary bps solutions in n = 2 supergravity with @xmath414 interactions _ , _ jhep _ * 12 * ( 2000 ) 019 , [ http://xxx.lanl.gov/abs/hep-th/0009234 [ hep - th/0009234 ] ] . s. barisch - dick , g. lopes cardoso , m. haack , and s. nampuri , _ extremal black brane solutions in five - dimensional gauged supergravity _ , _ jhep _ * 1302 * ( 2013 ) 103 , [ http://xxx.lanl.gov/abs/1211.0832 [ arxiv:1211.0832 ] ] . d. klemm and o. vaughan , _ nonextremal black holes in gauged supergravity and the real formulation of special geometry ii _ , _ class.quant.grav . _ * 30 * ( 2013 ) 065003 , [ http://xxx.lanl.gov/abs/1211.1618 [ arxiv:1211.1618 ] ] . g. l. cardoso , b. de wit , and s. mahapatra , _ non - holomorphic deformations of special geometry and their applications _ , _ springer proc.phys._ * 144 * ( 2013 ) 158 , [ http://xxx.lanl.gov/abs/1206.0577 [ arxiv:1206.0577 ] ] . p. bueno , w. chemissany , p. meessen , t. ortin , and c. shahbazi , _ lifshitz - like solutions with hyperscaling violation in ungauged supergravity _ , _ jhep _ * 1301 * ( 2013 ) 189 , [ http://xxx.lanl.gov/abs/1209.4047[arxiv:1209.4047 ] ] .
semi - supervised learning has attracted the attention of many researchers because of the importance of combining unlabeled data with labeled data . in many applications the number of unlabeled data points is so large that labeling training data is expensive and time - consuming . therefore , the problem of effectively utilizing a combination of unlabeled and labeled information is very important in machine learning research . this paper concerns the issue of how to address uncertainty quantification in such classification methods . in doing so we bring together a variety of themes from the mathematical sciences , including optimization , pdes , probability and statistics . we will show that a variety of different methods , arising in very distinct communities , can all be formulated around a common objective function @xmath0 for a real valued function @xmath1 on the nodes of a graph representing the data points . the matrix @xmath2 is proportional to a graph laplacian derived from the unlabeled data and the function @xmath3 involves the labelled data . the variable @xmath1 is used for classification . minimizing this objective function is one approach to such a classification . a probability distribution related to the objective function has density @xmath4 proportional to @xmath5 probability of the labelling variable @xmath1 is high where the objective function is small , and vice - versa . uncertainty quantification corresponds to using the probability distribution to compute expectations of test functions @xmath6 , defined on the nodes of the graph , which enable us to measure the variability of label variables , such as means and variances : @xmath7 in the settings of interest this will typically be a very high dimensional integral , with the dimension given by the number of unlabelled data points . carrying out this program requires computational algorithms to minimize @xmath8 or to draw samples , via monte carlo markov chain ( mcmc ) for example , from the probability distribution with density @xmath4 . these algorithms exploit the fact that @xmath9 is a graph analogue of the dirichlet energy and will leverage analogies with pde - based methodologies involving the classical euclidean dirichlet energy in order to derive effective computational methods . in this paper we will describe this confluence of ideas from different parts of the mathematical sciences , show how our approach builds on a broad range of advances in the field which we will review , and demonstrate the emergence of a problem area with many open challenges for the mathematical sciences . we emphasize that the variety of probablistic models considered in this paper arise from different assumptions concerning the structure of the data . our objective is not to assess the validity of these assumptions , which is a modelling question best addressed on a case - by - case basis , but rather we develop an overarching computational framework suitable for all the models arising from these different assumptions . an effective method for semi - supervised learning is to construct a similarity graph on both the unlabeled and labeled examples , and classify unknown labels by leveraging the graph structure . a central conceptual issue in the setting of this problem is that labels are discrete , whilst similarity information is often continuous . strategies to work with both of these settings simultaneously are at the heart of this subject . in @xcite , blum et al . posed the binary semi - supervised classification problem using a graph min - cut problem . this is equivalent to a maximum a posteriori ( map ) estimator with respect to a bayesian posterior distribution for a markov random field ( mrf ) over the discrete state space of binary labels @xcite ; the resulting optimization problem can be solved exactly in polynomial time . in general , inference for multi - label discrete mrfs is intractable @xcite . however , several approximate algorithms exist for the multi - label case @xcite , and have been applied to many imaging tasks @xcite . the probit classification method , using gaussian process priors , is described in @xcite ; however in that book the prior does not depend on the unlabelled data . gaussian priors which depend on the unlabelled data may be constructed by using the graph laplacian , an approach undertaken in @xcite . the model defined in @xcite is equivalent to a continuum relaxation of the discrete state space mrf in @xcite . the bayesian formulation which underpins our work in this paper was made explicit in @xcite where a variety of likelihood models are used to condition on the labelled data ; the probit approach , for example , could be used to accomplish this . probit utilizies the same prior as in @xcite but the data is assumed to take binary values , found from thresholding the underlying continuous variable , and thereby providing a link between the combinatorial and continuous state space approaches described in the previous paragraph . the probit methodology is often implemented via map optimization that is the posterior probability is maximized rather than sampled or an approximation to the posterior is computed , in the neighbourhood of the map estimator @xcite . for full posterior exploration , gibbs sampling is often used @xcite and this methodology has been applied recently in @xcite ; furthermore methods designed to break undesirable dependencies in the gibbs sampler are introduced in @xcite . in the context of map estimation , the graph - based terms act as a regularizer , in the form of the graph dirichlet energy @xmath10 . a formal framework for graph - based regularization can be found in @xcite . more recently , other forms of regularization have been considered such as the graph wavelet regularization @xcite . another link between discrete combinatorial optimization approaches and methods based on optimization over real - valued variables was made in the work of bertozzi et al . @xcite . the approach is based on the fact that the tv functional , when suitably generalized to weighted graphs , coincides with the graph cut energy . relaxation of the tv functional is well - understood in the context of partial differential equations ( pde ) and generalizing ideas applicable to the pde laplacian in the context of the graph laplacian leads to new optimization methods . based on this reasoning , in @xcite the graph ginzburg - landau functional was used as a relaxation of the graph tv functional for the task of binary classification . this was generalized to multiclass classification in @xcite . following this line of work , several new algorithms were developed for semi - supervised and unsupervised classification problems on weighted graphs @xcite . a further connection with pde based methods is the level - set approach to bayesian inversion , introduced recently in @xcite ; this is very closely related to our variant on the probit method , as we will demonstrate . there are a wide range of methodologies employed in the field of uncertainty quantification , and the reader may consult the books @xcite and the recent article @xcite for details and further references . underlying all of these methods is a bayesian methodology which is attractive both for its clarity with respect to modelling assumptions and its basis for application of a range of computational tools . nonetheless it is important to be aware of limitations in this approach , in particular with regard to its robustness with respect to the specification of the model , and in particular the prior distribution on the unknown of interest @xcite . whilst the book @xcite conducts a number of thorough uncertainty quantification studies for a variety of learning problems using gaussian process priors , most of the papers studying graph based learning referred to above primarily use the bayesian approach to learn hyperparameters in an optimization context , and do not consider uncertainty quantification . in this paper , we focus exclusively on the problem of binary semi - supervised classification ; however the methodology and conclusions will extend beyond this setting . our focus is on a presentation which puts uncertainty quantification at the heart of the problem formulation , and we make four primary contributions : * we define a number of different bayesian formulations of the graph - based semi - supervised learning problem and we connect them to one another , to binary classification methods and to a varierty of pde - inspired approaches to classification ; in so doing we provide a single framework for a variety of methods which have arisen in distinct communities and we open up a number of new avenues of study for the problem area ; * we highlight the pcn - mcmc method for posterior sampling which , based on analogies with its use for pde - based inverse problems @xcite , has the potential to sample the posterior distribution in a number of steps which is independent of the number of graph nodes ; * we introduce approximations exploiting the empirical properties of the spectrum of the graph laplacian , generalizing methods used in the optimization context in @xcite , allowing for computations at each mcmc step which scale well with respect to the number of graph nodes ; * we demonstrate , by means of numerical experiments on a range of problems , both the feasibility , and value , of bayesian uncertainty quantification in semi - supervised , graph - based , learning . the paper is organized as follows . in section [ sec:2 ] , we give some background material needed for problem specification . in section [ sec:3 ] we formulate the four bayesian models used for the classification tasks . section [ sec:4 ] introduces the mcmc and optimization algorithms that we use . in section [ sec:5 ] , we present and discuss results of numerical experiments to illustrate our findings ; these are based on four examples of increasing size : the house voting records from 1984 ( as used in @xcite ) , the tuneable two moons data set @xcite , the mnist digit data base @xcite and the hyperspectral gas plume imaging problem @xcite . we conclude in section [ sec:6 ] . to aid the reader , we give here an overview of notation used throughout the paper . * @xmath11 the set of nodes of the graph , with cardinality @xmath12 ; * @xmath13 the set of nodes where labels are observed , with cardinality @xmath14 ; * @xmath15 , feature vectors ; * @xmath16 latent variable characterizing nodes , with @xmath17 denoting evaluation of @xmath18 at node @xmath19 ; * @xmath20 the thresholding function ; * @xmath21 relaxation of @xmath22 using gradient flow in double - well potential @xmath23 ; * @xmath24 the label value at each node with @xmath25 * @xmath26 or @xmath27 , label data ; * @xmath28 with @xmath29 being a relaxation of the label variable @xmath30 ; * @xmath31 weight matrix of the graph , @xmath32 the resulting symmetric graph laplacian ; * @xmath2 the precision matrix and @xmath33 the covariance matrix , both found from @xmath32 ; * @xmath34 eigenpairs of @xmath32 ; * @xmath35 : orthogonal complement of the null space of the graph laplacian @xmath32 , given by @xmath36 ; * @xmath37 : orthogonal complement of the first @xmath38 eigenfunctions of the graph laplacian @xmath32 . * @xmath39 denotes the euclidean norm and @xmath40 the corresponding inner - product ; * @xmath41 : ginzburg - landau functional ; * @xmath42 , @xmath43 : prior probability measures ; * @xmath44 and @xmath45 : ( with suffices denoting different models ) ; * the measures denoted @xmath44 typically take argument @xmath18 and are real - valued ; the measures denoted @xmath45 take argument @xmath30 on label space , or argument @xmath29 on a real - valued relaxation of label space ; * @xmath46 denotes a gaussian random variable with mean @xmath47 and covariance @xmath48 ; * @xmath49 and @xmath50 denote the probability of an event , and the expectation of a random variable , respectively ; the underlying probability measure will be made explicit as a subscript when it is necessary to do so . in subsection [ ssec : ssl ] we formulate semi - supervised learning as a problem on a graph . subsection [ ssec : graph ] defines the relevant properties of the graph laplacian and in subsection [ ssec : gauss ] these properties are used to construct a gaussian probability distribution ; in section [ sec:3 ] this gaussian will be used to define our prior information about the classification problem . in subsection [ ssec : thr ] we discuss thresholding which provides a link between the real - valued prior information , and the label data provided for the semi - supervised learning task ; in section [ sec:3 ] this will be used to definine our likelihood . we are given a set of points denoted by @xmath51 , and a set of features @xmath52 associated with these points ; each feature vector @xmath53 is an element of @xmath54 , so that @xmath55 graph learning starts from the construction of an undirected graph @xmath56 with weights @xmath57 computed from the feature set @xmath58 . for graph semi - supervised learning , we are also given a partial set of ( possibly noisy ) labels @xmath59 , where @xmath60 has size @xmath14 . the task is to infer the labels for all nodes in @xmath11 , using the weighted graph @xmath56 and also the set of noisily observed labels @xmath61 . in the bayesian formulation which we adopt the feature set @xmath58 , and hence the graph @xmath56 , is viewed as prior information , describing correlations amongst the nodes of the graph , and we combine this with a likelihood based on the noisily observed labels @xmath61 , to obtain a posterior distribution on the labelling of all nodes . various bayesian formulations , which differ in the specification of the observation model and/or the prior , are described in section [ sec:3 ] . in the remainder of this section we give the background needed to understand all of these formulations , thereby touching on the graph laplacian itself , its link to gaussian probability distributions and , via thresholding , to non - gaussian probability distributions and to the ginzburg - landau functional . an important point to appreciate is that building our priors from gaussians confers considerable computational advantages for large graphs ; for this reason the non - gaussian priors will be built from gaussians via change of measure or push forward under a nonlinear map . the graph laplacian is central to many graph - learning algorithms . there are a number of variants used in the literature ; see for a discussion . we will work with the symmetric laplacian , defined from the weight matrix @xmath62 as follows . we define the diagonal matrix @xmath63 with entries @xmath64 if we assume that the graph @xmath56 is connected , then @xmath65 for all nodes @xmath66 . we can then define the symmetric graph laplacian that we use is that it is symmetric positive semi - definite . we could therefore use other graph laplacians , such as the unnormalized choice @xmath67 , in most of the paper . the only exception is the spectral approximation sampling algorithm introduced later ; that particular algorithm exploits empirical properties of the symmetrized graph laplacian . note , though , that the choice of which graph laplacian to use can make a significant difference see @xcite , and figure 2.1 therein . to make our exposition more concise we confine our presentation to the symmetric graph laplacian . ] as @xmath68 and the graph dirichlet energy as @xmath69 . then @xmath70 thus , similarly to the classical dirichlet energy , this quadratic form penalizes nodes from having different function values , with penalty being weighted with respect to the similarity weights from @xmath31 . furthermore the identity shows that @xmath32 is positive semi - definite . indeed the vector of ones @xmath71 is in the null - space of @xmath72 by construction , and hence @xmath32 has a zero eigenvalue with corresponding eigenvector @xmath73 . we let @xmath74 denote the eigenpairs of the matrix @xmath32 , ordered so that @xmath75 the upper bound of @xmath76 may be found in ( * ? ? ? * lemma 1.7 , chapter 1 ) . the eigenvector corresponding to @xmath77 is @xmath78 and @xmath79 , assuming a fully connected graph . then @xmath80 where @xmath81 has columns @xmath82 and @xmath83 is a diagonal matrix with entries @xmath84 . using these eigenpairs the graph dirichlet energy can be written as @xmath85 this is analogous to decomposing the classical dirichlet energy using fourier analysis . we now show how to build a gaussian distribution with negative log density proportional to @xmath86 such a prior prefers functions that have larger components on the first few eigenvectors of the graph laplacian , where the eigenvalues of @xmath32 are smaller . the corresponding eigenvectors carry rich geometric information about the weighted graph . for example , the second eigenvector of @xmath32 is the _ fiedler vector _ and solves a relaxed normalized min - cut problem @xcite . the gaussian distribution thereby connects geometric intuition embedded within the graph laplacian to a natural probabilistic picture . to make this connection concrete we define diagonal matrix @xmath48 with entries defined by the vector @xmath87 and define the positive semi - definite covariance matrix @xmath88 ; choice of the scaling @xmath89 will be discussed below . we let @xmath90 note that the covariance matrix is that of a gaussian with variance proportional to @xmath91 in direction @xmath92 thereby leading to structures which are more likely to favour the fiedler vector ( @xmath93 ) , and lower values of @xmath19 in general , than it does higher values . the fact that the first eigenvalue of @xmath33 is zero ensures that any draw from @xmath42 changes sign , because it will be orthogonal to @xmath94 to make this intuition explicit we recall the karhunen - loeve expansion which constructs a sample @xmath18 from the prior @xmath42 according to the random sum @xmath95 where the @xmath96 are i.i.d . @xmath97 since each @xmath92 with @xmath98 is orthogonal to @xmath99 it follows that @xmath18 is orthogonal to @xmath99 and the sign - change property is enforced because @xmath99 is of one sign . we choose the constant of proportionality @xmath89 as a rescaling which enforces the property @xmath100 for @xmath101 ; in words the per - node variance is @xmath102 . note that , using the orthogonality of the @xmath103 , @xmath104 thus the normalization implies that @xmath105 we reiterate that the support of the measure @xmath42 is the space @xmath106 and that , on this space , the probability density function is proportional to @xmath107 so that the _ precision matrix _ of the gaussian is @xmath108 in what follows the sign of @xmath18 will be related to the classification ; since all the entries of @xmath109 are positive , working on the space @xmath35 ensures a sign change in @xmath18 , and hence a non - trivial classification . for the models considered in this paper , the label space of the problem is discrete while the latent variable @xmath18 through which we will capture the correlations amongst nodes of the graph , encoded in the feature vectors , is real - valued . we describe thresholding , and a relaxation of thresholding , to address the need to connect these two differing sources of information about the problem . in what follows the latent variable @xmath110 is thresholded to obtain the label variable @xmath111 the variable @xmath112 is a real - valued relaxation of the label variable @xmath113 the variable @xmath18 will be endowed with a gaussian probability distribution . from this the variable @xmath30 ( which lives on a discrete space ) and @xmath29 ( which is real - valued , but concentrates near the discrete space supporting @xmath30 ) will be endowed with non - gaussian probability distributions . define the ( signum ) function @xmath114 by @xmath115 this will be used to connect the latent variable @xmath18 with the label variable @xmath30 . the function @xmath22 may be relaxed by defining @xmath116 where @xmath29 solves the gradient flow @xmath117 this will be used , indirectly , to connect the latent variable @xmath18 with the real - valued relaxation of the label variable , @xmath29 . note that @xmath118 , pointwise , as @xmath119 , on @xmath120 this reflects the fact that the gradient flow minimizes @xmath23 , asymptotically as @xmath121 , whenever started on @xmath120 we have introduced a gaussian measure @xmath42 on the latent variable @xmath18 which lies in @xmath122 ; we now want to introduce two ways of constructing non - gaussian measures on the label space @xmath123 or on real - valued relaxations of label space , building on the measure @xmath124 the first is to consider the push - forward of measure @xmath42 under the map @xmath22 : @xmath125 then @xmath126 thus @xmath127 is a measure on the label space @xmath128 the second approach is to work with a change of measure from the gaussian @xmath42 in such a way that the probability mass on @xmath122 concentrates close to the label space @xmath129 . we may achieve this by defining the measure @xmath43 via its radon - nykodim derivative @xmath130 we name @xmath43 the ginzburg - landau measure , since the negative log density function of @xmath43 is the graph ginzburg - landau functional @xmath131 the ginzburg - landau distribution defined by @xmath43 can be interpreted as a non - convex ground relaxation of the discrete mrf model @xcite , in contrast to the convex relaxation which is the gaussian field @xcite . since the double well has minima at the label values @xmath132 , the probability mass of @xmath43 is concentrated near the modes @xmath133 , and @xmath134 controls this concentration effect . [ sec : problem - spec ] in this section we formulate four different bayesian models for the semi - supervised learning problem . the four models all combine the ideas described in the previous section to define four distinct posterior distributions . it is important to realize that these different models will give different answers to the same questions about uncertainty quantification , just as different methods based around optimization will give different classifications . the choice of which bayesian model to use is related to the data itself , and making this choice is beyond the scope of this paper . currently the choice must be addressed on a case by case basis , as is done when choosing an optimization method for classification . nonetheless we will demonstrate that the shared structure of the four models mean that a common algorithmic framework can be adopted and we will make some conclusions about the relative costs of applying this framework to the four models . we denote the latent variable by @xmath17 , @xmath135 , the thresholded value of @xmath17 by @xmath136 which is interpreted as the label assignment at each node @xmath19 , and noisy observations of the binary labels by @xmath137 , @xmath138 . the variable @xmath139 will be used to denote the real - valued relaxation of @xmath140 used for the ginzburg - landau model . recall bayes formula which transforms a prior density @xmath141 on a random variable @xmath18 into a posterior density @xmath142 on the conditional random variable @xmath143 : @xmath144 we will now apply this formula to condition our graph latent variable @xmath18 , whose thresholded values correspond to labels , on the noisy label data @xmath61 given at @xmath13 . as prior on @xmath18 we will always use @xmath145 we willdescribe three different likelihoods . we will also apply the formula to condition relaxed label variable @xmath29 , on the same label data @xmath61 , via the formula @xmath146 we will use as prior the non - gaussian @xmath147 for the probit , level - set and atomic models , we now explicitly state the prior density @xmath141 , the likelihood function @xmath148 , and the posterior density @xmath142 ; in the ginzburg - landau case @xmath29 will replace @xmath18 and we will define the densities @xmath149 and @xmath150 . prior and posterior probability measures associated with letter @xmath44 are on the latent variable @xmath18 ; measures associated with letter @xmath45 are on the label space , or real - valued relaxation of the label space . the probit method is designed for classification and is described in @xcite ; in that context gaussian process priors are used and these do not depend on label data . a recent fully bayesian treatment of the methodology using unweighted graph laplacians may be found in the paper @xcite . in detail our model is as follows . * prior * we take as prior on @xmath18 the gaussian @xmath124 thus @xmath151 * likelihood * for any @xmath138 @xmath152 with the @xmath153 drawn i.i.d from @xmath154 we let @xmath155 and note that then @xmath156 similarly @xmath157 * posterior * bayes theorem gives posterior @xmath158 with probability density function ( pdf ) @xmath159 where @xmath160 we let @xmath161 denote the push - forward under @xmath22 of @xmath162 * map estimator * this is the minimizer of the negative of the log posterior . thus we minimize the following objective function over @xmath35 : @xmath163 this is a convex function , a fact which is well - known in related contexts , but which we state and prove in the supplementary materials for the sake of completeness . in view of the close relationship between this problem and the level - set formulation described next , for which there are no minimizers , we expect that minimization may not be entirely straightforward in the @xmath164 limit . this is manifested in the presence of near - flat regions in the probit log likelihood function when @xmath164 . our variant on the probit methodology differs from that in @xcite in several ways : ( i ) our prior gaussian is scaled to have per - node variance one , whilst in @xcite the per node variance is a hyper - parameter to be determined ; ( ii ) our prior is supported on @xmath165 whilst in @xcite the prior precision is found by shifting @xmath32 and taking a possibly fractional power of the resulting matrix , resulting in support on the whole of @xmath166 ; ( iii ) we allow for a scale parameter @xmath167 in the observational noise , whilst in @xcite the parameter @xmath168 this method is designed for problems considerably more general than classification on a graph @xcite . for the current application , this model is exactly the same as probit except for the order in which the noise @xmath153 and the thresholding function @xmath169 is applied in the definition of the data . * prior * we again take as prior for @xmath18 , the gaussian @xmath124 thus @xmath151 * likelihood * for any @xmath138 @xmath170 with the @xmath153 drawn i.i.d from @xmath154 then @xmath171 with pdf @xmath172 where latexmath:[\[{\phi_{\rm ls}}(u;y)=\sum_{j \in z ' } \bigl(\frac{1}{2\gamma^2 } denote the pushforward under @xmath22 of @xmath174 * map estimator functional * the negative of the log posterior is , in this case , given by @xmath175 however , unlike the probit model , the bayesian level - set method has no map estimator the infimum of @xmath176 is not attained and this may be seen by noting that , if the infumum was attained at any non - zero point @xmath177 then @xmath178 would reduce the objective function for any @xmath179 ; however the point @xmath180 does not attain the infimum . this proof is detailed in @xcite for a closely related pde based model , and the proof is easily adapted . this is a variant on the level - set method , but deals with the fact that categorical data , whilst maybe noisy , will often be discrete . the resulting data model could also be used within the ginzburg - landau formulation in the next subsection , but we do not describe this explicitly . the prior we employ is the gaussian @xmath42 , and in contrast to the level - set method , but like the probit method , the observation is assumed to take values in the set @xmath181 . in the atomic noise model the reporting of those values is accompanied by specified error rates @xcite , determined by the parameters @xmath182 . taking @xmath183 in the following corresponds to exact data with no error and is the same as level - set thresholding , or probit , in the limit @xmath184 * prior * we again take as prior for @xmath18 the gaussian @xmath124 thus @xmath151 * likelihood * we assume that the observation has the sign of @xmath18 with a specified probability . to be precise we assume that we have , @xmath185 and @xmath186 this defines a piecewise constant function of @xmath18 , with discontinuity at @xmath187 , for each value @xmath188 : we call this function @xmath189 in particular we have @xmath190 * posterior * bayes theorem gives posterior @xmath191 with pdf @xmath192 where @xmath193 we let @xmath194 denote the pushforward under @xmath22 of @xmath195 * map estimator functional * this is the minimizer of the negative of the log posterior . thus we minimize the objective function : @xmath196 we will not consider numerical methods for the atomic noise model because of space limitations . however it may be a natural choice for some measurement scenarios , hence its inclusion in this paper . its behaviour is very similar to the bayesian level set method because its likelihood is also piecewise constant with respect to latent variable @xmath18 ; and as with the bayesian level set there is no minimizer for the map estimation problem . for this model , we take as prior the ginzburg - landau measure @xmath43 defined by ( [ eq : gl - meas ] ) , and employ a gaussian likelihood for the observed labels . this construction gives the bayesian posterior whose map estimator is the objective function introduced and studied in @xcite . * prior * we define prior on @xmath29 to be the ginzburg - landau measure @xmath43 given by ( [ eq : gl - meas ] ) with density @xmath197 * likelihood * for any @xmath138 @xmath198 with the @xmath153 drawn i.i.d from @xmath154 then @xmath199 we see that bayes theorem gives posterior @xmath200 with pdf latexmath:[\[\begin{aligned } { \mathbb{p}_{\rm gl}}(v|y ) & \propto & \exp\bigl(-\frac12 \langle v , pv \rangle - { \phi_{\rm gl}}(v;y)\bigr),\\ { \phi_{\rm gl}}(v;y)&:=&\sum_{j \in z } w_\epsilon\bigl(v(j)\bigr)+\sum_{j \in z ' } \bigl(\frac{1}{2\gamma^2 } * map estimator * this is the minimizer of the negative of the log posterior . thus we minimize the following objective function over @xmath35 : @xmath202 this objective function was introduced in @xcite as a relaxation of the min - cut problem , penalized by data ; the relationship to min - cut was studied rigorously in @xcite . the minimization problem for @xmath203 is non - convex and has multiple minimizers , reflecting the combinatorial character of the min - cut problem of which it is a relaxation . in the small label noise limit @xmath204 the probit and level - set posteriors coincide with the atomic noise model in the limit where @xmath183 . furthermore all models then take the form of the gaussian prior @xmath42 conditioned to be positive on labelled nodes where @xmath205 and to be negative on labelled nodes where @xmath206 this can be linked with the original work of zhu et al @xcite which based classification on the measure @xmath42 conditioned to take the value exactly @xmath102 on labelled nodes where @xmath205 and conditioned to take the value exactly @xmath207 on labelled nodes where @xmath206 thus we see explicit connections between a variety of different bayesian formulations of graph - based semi - supervised learning . in figure [ fig : spectral - approx ] we plot the component of the negative log likelihood at a labelled node @xmath19 , as a function of the latent variable @xmath208 with data @xmath209 fixed , for the probit , bayesian level - set , and atomic noise models . the log likelihood for the ginzburg - landau formulation is not directly comparable as it is a function of the relaxed label variable @xmath139 , with respect to which it is quadratic with minimum at the data point @xmath210 . we set @xmath211 for probit and bayesian level - set , and @xmath212 , @xmath213 for atomic noise model . since @xmath214 for probit and bayesian level - set , we omit the plot for @xmath215.,width=264 ] the probit , bayesian level - set , and atomic noise models lead to posterior distributions @xmath44 ( with different subscripts ) in latent variable space , and pushforwards @xmath45 ( also with different subscripts ) in label space . the ginzburg - landau formulation leads to a measure @xmath200 in label space . uncertainty quantification in the widest sense is concerned with completely characterizing these posterior distributions . in practice this may be acheived by sampling using mcmc methods . in this paper we will study four measures of uncertainty : * we will study the empirical pdfs of the latent and label variables at certain nodes ; * we will study the posterior mean of the label variables at certain nodes ; * we will study the posterior variance of the label variables averaged over all nodes ; * we will use the posterior mean or variance to order nodes into those whose classificaions are most uncertain and those which are most certain . for the probit , level - set and atomic models , we interpret the thresholded variable @xmath216 as the binary label assignments corresponding to a real - valued configuration @xmath18 . the node - wise posterior mean of @xmath30 can be used as a useful confidence score of the class assignment of each node . the node - wise posterior mean @xmath217 is defined as @xmath218 with respect to any of the posterior measures ( pushed forward from latent variable space for probit , level - set and atomic models ) @xmath45 . note that @xmath219 $ ] and if @xmath220 then @xmath221 . for binary labels @xmath222 the mean also contains the variance information , and hence the formula captures posterior variance . specifically we have that @xmath223 later we will find it useful to consider the variance averaged over all nodes and hence define ] @xmath224 note that the maximum value obtained by var(l ) is @xmath102 . this maximum value is attained under all the prior distributions we use in this paper . the deviation from this maximum , under the posterior , is a measure of the information content of the labelled data . note , however , that the prior does contain information about classifications , in the form of correlations between vertices ; this is not captured in . from section [ sec:3 ] , we see that for all of the models considered , the posterior @xmath225 has the form @xmath226 for some function @xmath3 , different for each of the four models ( acknowledging that in the ginzburg - landau case the independent variable is @xmath227 , real - valued relaxation of label space , where as for the other models @xmath228 an underlying latent variable which may be thresholded by @xmath229 into label space . ) furthermore , the map estimator is the minimizer of @xmath230 note that @xmath3 is differentiable for the ginzburg - landau and probit models , but not for the level - set and atomic noise models . we introduce algorithms for both sampling ( mcmc ) and map estimation ( optimization ) that apply in this general framework . the sampler we employ does not use information about the gradient of @xmath3 ; the map estimation algorithm does , but is only employed on the ginzburg - landau and probit models . the samplers do use properties of the precision matrix @xmath2 , which is proportional to the graph laplcian @xmath32 ; in particular its spectral properties are relevant . figure [ fig : spectra - all ] demonstrates the spectral properties of @xmath32 for the four examples that we will apply our algorithms to in section [ sec:5 ] . we sample the posterior probability distribution using mcmc . to date probit models have typically been sampled by means of a gibbs methodology . however for three reasons we consider sampling algorithms which apply directly on all the nodes @xmath11 . these are : ( i ) we wish to highlight methods which apply to the ginzburg - landau , level - set and atomic noise models which precludes the explicit conditionally gaussian , or truncated gaussian , form of the methods described for probit ; ( ii ) all of our posterior distributions have a density with respect to the gaussian @xmath42 that is their densities are proportional to that of @xmath42 and as a result we may use mcmc methods which , in the case where the graph laplacian has a limit @xcite , have the potential for delivering samples from the posterior in a number of steps which is independent of the dimension @xmath12 of the state space , as overviewed in @xcite ; ( iii ) these mcmc methods are well - adapted to the use of approximation methods which exploit structure in the spectral properties of the graph laplacian . other classes of mcmc methods , such as the gibbs samplers in @xcite ; could be considered ; and other priors , relaxing the gaussian structure , could be considered ; but in taking these directions then the development of mcmc methods with @xmath231independent mixing rates which can also exploit approximations of the spectral properties of the graph laplacian is an open research direction . in order to induce scalability with respect to size of @xmath11 we use the pcn method described in @xcite and introduced in the context of diffusions by beskos et . al . in @xcite and by neal in the context of machine learning @xcite . the standard random walk metropolis ( rwm ) algorithm suffers from the fact that the optimal proposal variance or stepsize scales inverse proportionally to the dimension of the state space @xcite , which is the graph size @xmath12 in this case . the pcn method is designed so that the proposal variance required to obtain a given acceptance probability scales independently of the dimension of the state space ( here the number of graph nodes @xmath12 ) , hence in practice giving faster convergence of the mcmc when compared with rwm @xcite . we restate the pcn method as algorithm [ alg : pcnfull ] , and then follow with various variants on it in algorithms [ alg : pcneig ] and [ alg : pcneigfill ] . in all three algorithms @xmath232 $ ] is the key parameter which determines the efficiency of the mcmc method : small @xmath233 leads to high acceptance probability but small moves ; large @xmath233 leads to low acceptance probability and large moves . somewhere between these extremes is an optimal choice of @xmath233 which minimizes the asymptotic variance of the algorithm when applied to compute a given expectation . the value @xmath234 is a sample from the prior @xmath42 . if the eigenvalues and eigenvectors of @xmath32 are all known then the karhunen - loeve expansion gives @xmath235 where @xmath89 is given by , the @xmath236 are i.i.d centred unit gaussians and the equality is in law . for graphs with a large number of nodes @xmath12 , it is prohibitively costly to directly sample from the distribution @xmath42 , since doing so involves knowledge of a complete eigen - decomposition of @xmath32 , in order to employ . a method that is frequently used in classification tasks is to restrict the support of @xmath18 to the eigenspace spanned by the first @xmath38 eigenvectors with the smallest non - zero eigenvalues of @xmath32 ( hence largest precision ) and this idea may be used to approximate the pcn method ; this leads to a low rank approximation . in particular we approximate samples from @xmath42 by @xmath237 where @xmath238 is given by truncated after @xmath239 , the @xmath240 are i.i.d centred unit gaussians and the equality is in law . this is a sample from @xmath241 where @xmath242 and the diagonal entries of @xmath243 are set to zero for the entries after @xmath244 in practice , to implement this algorithm , it is only necessary to compute the first @xmath38 eigenvectors of the graph laplacian @xmath32 . this gives algorithm [ alg : pcneig ] . @xmath245 the accuracy of algorithm [ alg : pcneig ] as an approximation of algorithm [ alg : pcnfull ] depends to a large extent on the size of the eigenvalues of @xmath32 in the following sense : @xmath246 where the expectation is with respect to the centred unit gaussians @xmath247 note that the examples shown in figure [ fig : spectra - all ] gives an indiciation of the quality of this approximation ; the size and number of the smallest eigenvalues of @xmath32 play a very important role in determining whether the difference in is small . spectral projection often leads to good classification results , but may lead to reduced posterior variance and a posterior distribution that is overly smooth on the graph domain . we propose an improvement on the method that preserves the variability of the posterior distribution but still only involves calculating the first @xmath38 eigenvectors of @xmath248 this is based on the empirical observation that in many applications the spectrum of @xmath32 saturates and satisfies , for @xmath249 , @xmath250 for some @xmath251 . such behaviour may be observed in b ) , c ) and d ) of figure [ fig : spectra - all ] ; in particular note that in the hyperspectal case @xmath252 . we assume such behaviour in deriving the low rank approximation used in this subsection . ( see supplementary materials for a detailed discussion of the graph laplacian spectrum . ) we define @xmath253 by overwriting the diagonal entries from @xmath38 to @xmath254 with @xmath255 we then set @xmath256 , and generate approximate samples from @xmath42 by setting @xmath257 where @xmath258 is given by with @xmath259 replaced by @xmath251 for @xmath260 , the @xmath96 are centred unit gaussians , and the equality is in law . importantly samples according to can be computed very efficiently . in particular there is no need to compute @xmath92 for @xmath260 , and the quantity @xmath261 can be computed by first taking a sample @xmath262 , and then projecting @xmath263 onto @xmath264 . moreover , projection onto @xmath37 can be computed only using @xmath265 , since the vectors span the orthogonal complement of @xmath37 . concretely , we have @xmath266 where @xmath262 and equality is in law . hence the samples @xmath267 can be computed by @xmath268 the error induced by this approximation can be characterized through the formula @xmath269 under the stated empirical properties of the graph laplacian , we expect this to be a better approximation of the prior variance than the approximation leading to . the vector @xmath267 is a sample from @xmath270 and results in algorithm [ alg : pcneigfill ] . @xmath245 recall that the objective function for the map estimation has the form @xmath271 , where @xmath18 is supported on the space @xmath35 . for ginzburg - landau and probit , the function @xmath3 is smooth , and we can use a standard projected gradient method for the optimization . since @xmath32 is typically ill - conditioned , it is preferable to use a semi - implicit discretization as suggested in @xcite , as convergence to a stationary point can be shown under a graph independent learning rate . furthermore , the discretization can be performed in terms of the eigenbasis @xmath272 , which allows us to easily apply spectral projection when only a truncated set of eigenvectors is available . we state the algorithm in terms of the ( possibly truncated ) eigenbasis below . here @xmath273 is an approximation to @xmath2 found by setting @xmath274 where @xmath275 is the matrix with columns @xmath276 and @xmath277 for @xmath278 , @xmath279 thus @xmath280 , @xmath281 . in this section we conduct a series of numerical experiments on four different data sets that are representative of the field of graph semi - supervised learning . there are four main purposes for the experiments . first we perform uncertainty quantification , as explained in subsection [ ssec : uq ] . secondly , we study the spectral approximation and projection variants on pcn sampling as these scale well to massive graphs . finally we make some observations about the cost and practical implementation details of these methods , for the different bayesian models we adopt ; these will help guide the reader in making choices about which algorithm to use . we present the results for map estimations in the supplementary materials , alongside the proof of convexity of the probit map estimator . the quality of the graph constructed from the feature vectors is central to the performance of any graph learning algorithms . in the experiments below , we follow the graph construction procedures used in the previous papers @xcite ; those papers , together , applied graph partitioning to all of the datasets that we use here and so provide important guidance . moreover , we have verified that for all the reported experiments below , the graph parameters are in a range such that spectral clustering gives a reasonable performance . the methods we employ lead to refinements over spectral clustering ( improved classification ) and , of course , to uncertainty quantification ( which spectral clustering does not address ) . we introduce the data sets and describe the graph construction for each data set . in all cases we numerically construct the weight matrix @xmath31 , and then the graph laplacian @xmath32 . is symmetric in theory ; in practice we find that symmetrizing via the map @xmath282 is helpful . ] the two moons artificial data set is constructed to give noisy data which lies near a nonlinear low dimensional manifold embedded in a high dimensional space @xcite . the data set is constructed by sampling @xmath12 data points uniformly from two semi - circles centered at @xmath283 and @xmath284 with radius @xmath102 , embedding the data in @xmath285 , and adding gaussian noise with standard deviation @xmath286 we set @xmath287 and @xmath288 in this paper ; recall that then the graph size is @xmath12 and each feature vector has length @xmath289 . we will conduct a variety of experiments with different labelled data size @xmath290 , and in particular study variation with @xmath290 . the default value , when not varied , is @xmath290 at @xmath291 of @xmath12 , with the labelled points chosen at random . we take each data point as a node on the graph , and construct a fully connected graph using the self - tuning weights of zelnik - manor and perona @xcite , with @xmath292 = 10 . specifically we let @xmath293 , @xmath53 be the coordinates of the data points @xmath294 and @xmath19 . then weight @xmath295 from @xmath294 to @xmath19 is defined by @xmath296 where @xmath297 is the distance of the @xmath292-th closest point to the node @xmath19 . this dataset contains the voting records of 435 u.s . house of representatives ; for details see @xcite and the references therein . the votes were recorded in 1984 from the @xmath298 united states congress , @xmath299 session . the votes for each individual is vectorized by mapping a yes vote to @xmath102 , a no vote to @xmath207 , and an abstention / no - show to @xmath300 . the data set contains @xmath301 votes that are believed to be well - correlated with partisanship , and we use only these votes as feature vectors for constructing the graph . thus the graph size is @xmath302 , and feature vectors have length @xmath303 . the goal is to predict the party affiliation of each individual . we pick @xmath304 democrats and @xmath76 republicans at random to use as the observed class labels ; thus @xmath305 corresponding to less than @xmath306 of fidelity points . we construct a fully connected graph with weights given by with @xmath307 for all nodes @xmath308 the mnist database consists of @xmath309 images of size @xmath310 pixels containing the handwritten digits @xmath300 through @xmath311 ; see @xcite for details . since in this paper we focus on binary classification , we only consider pairs of digits . to speed up calculations , we subsample randomly @xmath312 images from each digit to form a graph with @xmath313 nodes ; we use this for all our experiments except in subsection [ ssec : cm ] where we use the full data set of size @xmath314 for digit pair @xmath315 to benchmark computational cost . the nodes of the graph are the images and as feature vectors we project the images onto the leading 50 principal components given by pca ; thus the feature vectors at each node have length @xmath316 we conduct a variety of experiments with different labelled data dimension @xmath290 , and in particular study variation with @xmath290 . the default value , when not varied , is @xmath290 at @xmath317 of @xmath12 , with the labelled points chosen at random . we construct a @xmath292-nearest neighbor graph with @xmath318 for each pair of digits considered . namely , the weights @xmath319 are non - zero if and only if one of @xmath294 or @xmath19 is in the @xmath292 nearest neighbors of the other . the non - zero weights are set using with @xmath320 . this is the only example we consider in this paper that does not have a fully connected graph . we choose the four pairs @xmath321 , @xmath322 , @xmath323 and @xmath315 . these four pairs exhibit increasing levels of difficulty for classification . this fact is demonstrated in figures [ fig : mnist1 ] - [ fig : mnist4 ] , where we visualize the datasets by projecting the dataset onto the second and third eigenvector of the graph laplacian . namely , each node @xmath294 is mapped to the point @xmath324 , where @xmath325 . the hyperspectral data set analysed for this project was provided by the applied physics laboratory at johns hopkins university ; see @xcite for details . it consists of a series of video sequences recording the release of chemical plumes taken at the dugway proving ground . each layer in the spectral dimension depicts a particular frequency starting at @xmath326 nm and ending with @xmath327 nm , with a channel spacing of @xmath328 nm , giving @xmath329 channels ; thus the feature vector has length @xmath330 the spatial dimension of each frame is @xmath331 pixels . we select @xmath332 frames from the video sequence as the input data , and consider each spatial pixel as a node on the graph . thus the graph size is @xmath333 . the classification problem is to classify pixels that represent the chemical plumes against pixels that are the background . we construct a fully connected graph with weights given by the cosine distance : @xmath334 this distance is small for vectors that point in the same direction , and is insensitive to their magnitude . we consider the symmetric laplacian defined in . because it is computationally prohibitive to compute eigenvectors of a laplacian of this size , we apply the nystrm extension @xcite to obtain an approximation to the true eigenvectors and eigenvalues ; see @xcite for details pertinent to the set - up here . we emphasize that each pixel in the @xmath332 frames is a node on the graph and that , in particular , pixels across the @xmath332 time - frames are also connected . since we have no ground truth labels for this dataset , we generate known labels by setting the segmentation results from spectral clustering as ground truth . the default value of @xmath290 is @xmath335 , and labels are chosen at random . this corresponds to labelling around @xmath336 of the points . we only plot results for the last @xmath337 frames of the video sequence since the first frame does not contain the chemical plume . in this subsection we demonstrate both the feasibility , and value , of uncertainty quantification in graph classification methods . we employ the probit and the bayesian level - set model for most of the experiments in this subsection ; we also employ the ginzburg - landau model but since this can be slow to converge , due to the presence of local minima , it is only demonstrated on the voting records dataset . the atomic noise model has a similar piecewise constant log likelihood as for the bayesian level - set method , and so we omit experiments on the atomic noise model . the pcn method is used for sampling on various datasets to demonstrate properties and interpretations of the posterior . in this subsection , we contrast the posterior distribution @xmath150 of the ginzburg - landau model with that of the probit and bayesian level - set ( bls ) models . the graph is constructed from the voting records data with the fidelity points chosen as described in subsection [ subsec : datasets ] . in figure [ fig : glvsprobitmcmc ] we plot the histograms of the empirical marginal posterior distribution on @xmath338 and @xmath339 for a selection of nodes on the graph . for the top row of figure [ fig : glvsprobitmcmc ] , we select @xmath337 nodes with low confidence " predictions , and plot the empirical marginal distribution of @xmath18 for probit and bls , and that of @xmath29 for the ginzburg - landau model . note that the same set of nodes is chosen for different models . the plots in this row demonstrate the multi - modal nature of the ginzburg - landau distribution in contrast to the uni - modal nature of the probit posterior ; this uni - modality is a consequence of proposition [ prop : convex - probit ] . for the bottom row , we plot the same empirical distributions for 6 nodes with high confidence " predictions . in contrast with the top row , the ginzburg - landau marginal for high confidence nodes is essentially uni - modal since most samples of @xmath29 evaluated on these nodes have a fixed sign . we also observe that the pcn algorithm converges far more quickly for probit than for ginzburg - landau , because of the presence of multiple modes in the latter ; this is manifest in the fact that the posteriors for ginzburg - landau are less well converged than for probit , for a similar amount of algorithmic time ; this issue is quantified in subsection [ ssec : cm ] . this undesirable feature of ginzburg - landau sampling can be ameliorated by choosing a larger value of @xmath134 ; however this then leads to a probability measure which is further from the label space which it relaxes . the bayesian level - set method behaves similarly to probit as is to be expected from the fact that , in the limit @xmath340 , they are formally identical as discussed in subsection [ ssec : sn ] . we construct the graph from the mnist @xmath315 dataset following subsection [ subsec : datasets ] . the noise variance @xmath167 is set to @xmath341 , and @xmath317 of fidelity points are chosen randomly from each class . the probit posterior is used to compute . in figure [ fig : mnistlowhigh ] we demonstrate that nodes with scores @xmath342 closer to the binary ground truth labels @xmath133 look visually more uniform than nodes with @xmath342 far from those labels . this shows that the posterior mean contains useful information which differentiates between outliers and inliers that align with human perception . in this set of experiments , we show that the posterior distribution of the label variable @xmath216 captures the uncertainty of the classification problem . we use the posterior variance of @xmath30 , averaged over all nodes , as a measure of the model variance ; specifically formula . we study the behaviour of this quantity as we vary the level of uncertainty within certain inputs to the problem . we demonstrate empirically that the posterior variance is approximately monotonic with respect to variations in the levels of uncertainty in the input data , as it should be ; and thus that the posterior variance contains useful information about the classification . we select quantities that reflect the separability of the classes in the feature space . figure [ fig : twomoonsfeature ] plots the posterior variance @xmath343 against the standard deviation @xmath344 of the noise appearing in the feature vectors for the two moons dataset ; thus points generated on the two semi - circles overlap more as @xmath344 increases . we employ a sequence of posterior computations , using probit and bayesian level - set , for @xmath345 . recall that @xmath346 and we choose @xmath291 of the nodes to have the ground truth labels as observed data . within both models , @xmath167 is fixed at @xmath341 . a total of @xmath347 samples are taken , and the proposal variance @xmath233 is set to @xmath348 . we see that the mean posterior variance increases with @xmath344 , as is intuitively reasonable . furthermore , because @xmath167 is small , probit and bayesian level - set are very similar models and this is reflected in the similar quantitative values for uncertainty . for the probit model and the bls model applied to the two moons dataset with @xmath287 . for each trial , a realization of the two moons dataset under the given parameter @xmath344 is generated , and @xmath291 of nodes are randomly chosen as fidelity . we run @xmath349 trials for each value of @xmath344 , and average the mean posterior variance across the @xmath349 trials in the figure . we set @xmath350 and @xmath351 for both models . , width=283 ] a similar experiment studies the posterior label variance @xmath343 as a function of the pair of digits classified within the mnist data set . we choose @xmath317 of the nodes as labelled data , and set @xmath350 . the number of samples employed is @xmath347 and the proposal variance @xmath233 is set to be @xmath348 . table [ tab : mnistpostvar ] shows the posterior label variance . recall that figures [ fig : mnist1 ] - [ fig : mnist4 ] suggest that the pairs @xmath352 are increasingly easy to separate , and this is reflected in the decrease of the posterior label variance shown in table [ tab : mnistpostvar ] . .mean posterior variance of different digit pairs for the probit model and the bls model applied to the mnist dataset . the pairs are organized from left to right according to the separability of the two classes as shown in fig.[fig : mnist1 ] - [ fig : mnist4 ] . for each trial , we randomly select @xmath317 of nodes as fidelity . we run @xmath353 trials for each pairs of digits and average the mean posterior variance across trials . we set @xmath350 and @xmath351 for both models . [ cols="<,<,<,<,<",options="header " , ] trials are run . we use spectral projection with number of eigenvectors @xmath354 . we plot the median accuracy along with error bars indicating the 25 and 75-th quantile of the classification accuracy of each method . we set @xmath350 for the probit model , and @xmath355 , @xmath356 for ginzburg - landau.,width=283 ] the non - convexity of the ginzburg - landau model can result in large variance in classification accuracy ; the extent of this depends on the percentage of observed labels . the existence of sub - optimal local extrema causes the large variance . if initialized without information about the classification , ginzburg - landau can perform very badly in comparison with probit . on the other hand we find that the best performance of the ginzburg - landau model , when initialized at the probit minimizer , is typically slightly better than the probit model . we note that the probit model is convex and theoretically should have results independent of the initialization . however , we see there are still small variations in the classification result from different initializations . this is due to slow convergence of gradient methods caused by the flat - bottomed well of the probit log - likelihood . as mentioned above this can be understood by noting that , for small gamma , probit and level - set are closely related and that the level - set map estimator does not exist minimizing sequences converge to zero , but the infimum is not attained at zero .
the gravitational distortion of light bundles from distant sources in the universe provides a unique means to investigate ( the statistical properties of ) the intervening mass distribution . being observable through the image distortion of the distant faint blue galaxy population , this cosmic shear effect offers the opportunity to study statistical properties of the large - scale structure . in contrast to almost all other methods for investigating the large scale structure ( lss ) with cmb being the only exception no assumption about the relation between dark and luminous mass is required . based on the assumption that the intrinsic orientation of the background galaxies is random , a net alignment in the observed galaxy images can be attributed to the tidal field ( shear ) . hence , the alignment pattern of the galaxy images directly reflects the properties of the mass distribution . for example , two - point statistical measures of the galaxy ellipticities ( understood here and in the following as two - component quantities , with an amplitude and an orientation ) can be expressed directly in terms of the power spectrum of the mass distribution , convolved with a filter function ( blandford et al . 1991 ; miralda - escud 1991 ; kaiser 1992 ; jain & seljak 1997 ; bernardeau et al . 1997 ; schneider et al . 1998 , hereafter svwjk ; kaiser 1998 ; van waerbeke et al . 1999 ; jain et al . 1999 , hereafter jsw ; and references therein ) . the power spectrum completely characterises a gaussian random density field , and so the two - point statistics , like the two - point correlation function of galaxy ellipticities or the rms shear in an aperture , suffices to extract the statistical information contained in the distorted galaxy images . whereas the earlier of the aforementioned papers concentrated mainly on predictions for the cosmic shear based on the linear evolution of the cosmic density field , it was pointed out by jain & seljak ( 1997 ) that even on scales as large as one degree , the non - linear evolution significantly affects the expected amplitude of the cosmic shear . the non - linear evolution transforms an initially gaussian field into a non - gaussian one , and thus the cosmic shear on small angular scales is expected to display significant non - gaussian features . as pointed out by bernardeau et al . ( 1997 ) and svwjk , the skewness of the resulting cosmic shear field is a sensitive measure of the density factor @xmath0 , since in quasi - linear perturbation theory the skewness is independent of the normalisation of the initial power spectrum . in order to define the skewness , mass reconstruction algorithms such as those developed for cluster reconstructions ( kaiser & squires 1993 ; seitz & schneider 1996 , seitz et al . 1998a ; lombardi & bertin 1998a , b ; and references therein ) can be employed to reconstruct the projected density field . this density field , appropriately spatially filtered , can then be used to calculate the skewness . in contrast to the two - point statistical measures mentioned above , which are defined directly in terms of the observable image ellipticities , this measurement of skewness is more indirect . this causes estimates of the statistical error from the data itself not to be straightforward . on the other hand , the aperture mass @xmath1 , introduced as a measure for cosmic shear in svwjk , is a scalar quantity directly defined in terms of the image ellipticities , and can thus be easily used for defining a skewness , as well as a second order statistics , the rms of @xmath2 , where @xmath3 is the angular scale of the circular aperture ( definitions are given in sect . 2 below ) . in particular , in contrast to the two - point ellipticity correlation function and the rms shear in an aperture , for which the filter with which the power spectrum of the projected density field is measured is broad , the corresponding filter function for the rms of @xmath2 is very narrow , and can be approximated very accurately by a delta `` function '' , so that @xmath4 directly measures the power of the projected density at wavelength @xmath5 ( bartelmann & schneider 1999 ) . the skewness defined in terms of @xmath1 has been considered in svwjk . as in the evolution of the three - dimensional density field , where highly non - linear structures like clusters of galaxies form , the projected mass density attains strongly non - gaussian features , e.g. , the projection of collapsed haloes . as a scalar quantity , the aperture mass is ideally suited to probe the full probability distribution of the projected mass density ; in particular , values of @xmath4 far out in the non - gaussian tail signal the presence of massive dark matter haloes . therefore , peaks in the distribution of @xmath1 can be used to search for such haloes , independent of their luminous properties ( schneider 1996 , hereafter s96 ) . indeed , a first application of @xmath1 to large scale structure simulations by reblinsky & bartelmann ( 1999 ) revealed that the detection of dark matter haloes through the aperture mass is more reliable in terms of completeness and spurious detections and suffers less from projection effects than optically selected cluster samples . assuming that highly significant peaks are caused by such haloes , one can predict their abundance ( i.e. , number of peaks above a certain threshold per unit solid angle ) by combining the spatial abundance as predicted by press & schechter ( 1974 ) theory with an assumed density profile , such as the universal dark matter profile found by navarro et al . ( 1996 , 1997 ; combined nfw ) . this idea has been put forward by kruse & schneider ( 1999a ; hereafter ks1 ) , who found that , depending on the cosmological model and the redshift distribution of background galaxies , of order 10 such haloes per square degree will be detectable in deep ground - based optical images , with a signal - to - noise ratio larger than 5 . given the rapid evolution of wide - field imaging , a mass - selected sample of dark matter haloes is now well within reach . indeed , a first example of a shear - detected mass concentration has recently been found by erben et al . ( 1999 ) . all of these predictions on cosmic shear are made using simplifying assumptions in order to make analytic progress . jsw tested some of these assumptions using ray - tracing simulation through a cosmic density field generated by very large n - body simulations ; similar tests have been carried out by van waerbeke et al . they found that the major approximations made in these analytic treatments , namely the so - called `` born - approximation '' ( which projects the density fields along `` straight lines '' ) , and the neglect of non - linear terms in the propagation equations , are very well satisfied . in particular , the twist of light bundles , which vanishes identically in the usual analytical treatments , is indeed very small . the predictions on the projected power spectrum using the approximation for the fully non - linear power spectrum of peacock & dodds ( 1996 ) agree very well with the numerical results , whereas the predictions concerning the skewness are less precise , indicating a breakdown of quasi - linear perturbation theory on small scales . in this paper , we extend the study of jsw to the particular application of the aperture mass statistics . we use the same simulations as jsw , resulting in a table of shear and projected mass density as a function of angular position . from the shear , we can simulate observations of the aperture mass as a function of position on the `` data '' field , and investigate its statistical properties . in particular , we calculate the probability distribution of @xmath4 , which we find to be highly non - gaussian , and from that we study the dispersion , skewness and kurtosis of the distribution . in agreement with jsw , we find that the dispersion is accurately predicted by analytic theory , whereas the skewness predictions can differ substantially from the numerical results . of particular interest is the kurtosis , since it enters the determination of the uncertainty of the dispersion measurement due to cosmic variance ( svwjk ) . we find that the kurtosis is a slowly decreasing function of angular scale @xmath3 , and attains values of @xmath6 even on scales as large as @xmath7 ; hence , the cosmic variance will be the major source of statistical error in the measurement of the power spectrum of the projected density field from @xmath1 . the analytic predictions on the abundance of significant peaks of @xmath1 , and thus presumably of dark matter haloes , turns out to be remarkably precise , given the strong assumptions made . we confirm the high number density of haloes detectable with this method . the shape of the probability distribution of @xmath1 in the highly non - gaussian tail , predicted by kruse & schneider ( 1999b ; hereafter ks2 ) , can also be confirmed to be well approximated by an exponential function . in this section , we briefly summarise the properties of the aperture mass , i.e. , its definition , its relation to the shear , and its signal - to - noise ratio . for more details , the reader is referred to s96 and svwjk . we define the spatially filtered mass inside a circular aperture of angular radius @xmath3 around the point @xmath8 by @xmath9 where the continuous weight function @xmath10 vanishes for @xmath11 . if @xmath10 is a compensated filter function , @xmath12 one can express @xmath1 in terms of the tangential shear @xmath13 at position @xmath14 relative to @xmath8 as @xmath15 ( fahlmann et al . 1994 ; s96 ) , where @xmath16 and @xmath17 is the polar angle of @xmath18 . the function @xmath19 is related to @xmath20 by @xmath21 we use the filter function for @xmath22 from the family given in svwjk : writing @xmath23 , and @xmath24 , we take @xmath25 and @xmath26 with @xmath27 for @xmath28 . an estimate of the shear field @xmath29 , and thus of the aperture mass @xmath30 through eq . ( [ mapshear ] ) , is provided by the distortions of images of faint background galaxies . the complex ellipticity of galaxy images is defined in terms of second moments of the surface - brightness tensor ( e.g. , tyson et al . 1990 ; kaiser & squires 1993 ) . specifically , we use here the ellipticity parameter @xmath31 ( schneider 1995 ; seitz & schneider 1997 ) , which is defined such that for sources with elliptical isophotes of axis ratio @xmath32 , the modulus of the source ellipticities is given as @xmath33 , and the phase of the @xmath34 is twice the position angle of the major axis . the complex image ellipticity @xmath31 can then be calculated in terms of the source ellipticity @xmath34 and the reduced shear @xmath35 by the transformation ( seitz & schneider 1997 ) @xmath36 this relation is valid only for noncritical clusters . for critical clusters , it has to be replaced by a different transformation . however , as we are mainly interested in the weak lensing regime , the above relation is sufficient here . it has been demonstrated ( schramm & kayser 1995 ; seitz & schneider 1997 ) that the ellipticity @xmath31 of a galaxy image is an unbiased estimate of the local reduced shear , provided that the intrinsic orientations of the sources are random . in the case of weak lensing , @xmath37 , one then has @xmath38 by averaging ( [ eps ] ) with the probability distribution of the source ellipticities . as for the tangential shear component @xmath39 occurring in ( [ tanshear ] ) , a similar quantity for the image ellipticities can be defined . consider a galaxy image @xmath40 at a position @xmath41 relative to the point @xmath8 with a complex image ellipticity @xmath42 . in analogy to ( [ tanshear ] ) the tangential ellipticity @xmath43 of this galaxy is then given by @xmath44 where @xmath45 is the polar angle of @xmath46 . we can now estimate the integral ( [ mapshear ] ) by a discrete sum over galaxy images , @xmath47 where @xmath48 is the number density of galaxy images . the discrete dispersion @xmath49 of the aperture mass @xmath50 is found by squaring ( [ map_1 ] ) and taking the expectation value in the absence of lensing , which leads to @xmath51 where @xmath52 . performing an ensemble average of eq . ( [ disp_d ] ) leads to the continuous dispersion @xmath53 @xmath54 finally , the _ signal - to - noise _ ratio @xmath55 at position @xmath8 is @xmath56^{1/2}}\;. \label{eq_sn}\ ] ] in the simulations , we draw the source ellipticities from a gaussian probability distribution , @xmath57}\ , \exp\left(-\frac{|\epsilon^{({\rm s})}|^2}{\sigma_\epsilon^2}\right)\;,\ ] ] where the width of the distribution is chosen as @xmath58 . throughout this paper , we assume that the number density of the background sources is @xmath59 . the aperture mass has been considered in the framework of blank field surveys in a variety of earlier publications . introduced as a convenient statistics for cosmic shear , svwjk have calculated the rms of @xmath1 as a function of angular scale , using the peacock & dodds ( 1996 ) approximation for the non - linear evolution of the power spectrum of density fluctuations . like other two - point statistics , the dispersion of @xmath1 is an integral over the power spectrum of the projected mass distribution , weighted by a filter function . the filter function corresponding to @xmath60 is very narrow and can be well approximated by a delta function ( bartelmann & schneider 1999 ) . hence , @xmath61 reproduces the shape of the projected power spectrum and , depending on the cosmological model and the redshift distribution of the sources , it reveals a broad peak at @xmath62 . one convenient property of the aperture mass is that the correlation function of @xmath1 of two apertures spatially separated by @xmath63 quickly decreases and already achieves values of @xmath64 for @xmath65 . this means that measurements of @xmath1 from a large consecutive area can be considered independent if the apertures are densely laid out on this data field ; this is in contrast to the rms shear in apertures which is strongly correlated , and thus must be obtained from widely separated regions on the sky . being a scalar quantity , @xmath1 can also be used for higher - order statistical measures of the cosmic shear . svwjk calculated the skewness of @xmath1 , using eulerian perturbation theory for the evolution of the three - dimensional density contrast @xmath66 . in agreement with bernardeau et 1997 ) they found that the skewness is a sensitive function of the cosmic density factor @xmath0 , and is in this approximation independent of the normalisation of the power spectrum . a measurement of the dispersion of @xmath1 is affected by two main sources of statistical error : the intrinsic ellipticity distribution of the source galaxies , and cosmic variance . to estimate the latter , one needs to know the kurtosis of @xmath1 which can not easily be determined analytically . values of @xmath1 much larger than its rms probe the highly non - gaussian regime of the projected density field . from its definition , one sees that large values of @xmath1 are expected if the aperture is centred on a density peak with a size comparable to the filter scale @xmath3 . therefore , the aperture map can be used to search for such density peaks , presumably collapsed dark matter haloes , in blank field imaging surveys . in this way it is possible to obtain a mass - selected sample of such haloes ( s96 ) . simple analytical arguments in s96 suggest that dark matter haloes with an approximately isothermal profile are detectable with a signal - to - noise ratio larger than 5 if their velocity dispersion exceeds @xmath67 km / s , assuming a number density of background sources of @xmath68 arcmin@xmath69 . indeed , this theoretical expectation was verified in the lensing investigation of the cluster ms1512 + 36 ( seitz et al.1998b ) . this cluster has a velocity dispersion of about @xmath70 km / s , as obtained from strong lensing modelling and from spectroscopy of cluster members , and is detected in the weak lensing analysis with very high statistical significance . assuming that the high signal - to - noise peaks of @xmath1 are due to collapsed dark matter haloes , one can attempt to estimate the abundance of such peaks using analytic theory . ks1 have calculated the number density of haloes with aperture mass larger than @xmath1 , @xmath71 , assuming ( 1 ) that dark matter haloes are distributed according to press & schechter ( 1974 ) theory which yields the number density of collapsed haloes as a function of halo mass and redshift , and ( 2 ) that the azimuthally - averaged projected density profiles of these haloes can be described by the projection of the universal halo density profile found in numerical simulations by nfw . depending on the cosmological model and on the redshift distribution of the faint galaxies , the number density of peaks of @xmath1 with a signal - to - noise ratio larger than 5 was estimated to be @xmath72 per square degree , and the redshift distribution of these haloes is strongly dependent on the behaviour of the linear growth factor for density perturbations , and thus on @xmath0 . this abundance is encouraging , since it allows one to obtain samples of haloes selected by their mass properties alone ( for a first example , see erben et al.1999 ) . using the same model , ks2 have calculated the probability distribution of @xmath1 for values of @xmath1 much larger than its rms , assuming that this non - gaussian tail of the probability distribution is dominated by dark matter haloes . they found that the distribution is very well described by an exponential ; i.e. , the tail is much broader than for a gaussian . all these analytic predictions are based on a number of approximations and simplifying assumptions . in sect . 4 below we shall compare these analytic results with those found in ray - tracing simulations through a cosmological mass distribution obtained from very large n - body calculations , as described in the next section . simulated shear maps due to weak lensing by large - scale structure are made by performing ray tracing simulations through the dark matter distribution produced by n - body simulations ( jsw ) . the n - body simulations used are a set of adaptive particle - particle / particle - mesh ( ap@xmath73 m ) simulations . the long - range component of the gravitational force is computed by solving poisson s equation on a grid . the grid calculation is supplemented with a short range correction computed either by a direct sum over neighbouring particles , or , in highly clustered regions , by combining a calculation on a localised refinement mesh with a direct sum over a smaller number of much closer neighbours . the parameters used by the n - body simulations are given in table [ cosmo ] . the simulations were run with a parallel adaptive ap@xmath73 m code ( couchman et al . 1995 ; pearce & couchman 1997 ) kindly made available by the virgo supercomputing consortium ( e.g. jenkins et al . 1998 ) . they followed @xmath74 particles using a force law with softening length @xmath75kpc at @xmath76 ( the force is @xmath77 its @xmath78 value at one softening length and is almost exactly newtonian beyond two softening lengths ) . @xmath79 was kept constant in physical coordinates over the redshift range of interest to us here . the simulations were carried out using 128 or 256 processors on cray t3d machines at the edinburgh parallel computer centre and at the garching computer centre of the max - planck society . these simulations have previously been used for studies of strong lensing by bartelmann et al . ( 1998 ) , for studies of dark matter clustering by jenkins et al . ( 1998 ) , and for studies of the relation between galaxy formation and galaxy clustering by kauffmann et al . ( 1999a , b ) , and diaferio et al . ( 1999 ) . the ray tracing simulations of weak lensing from which we use the convergence and shear maps were computed by jsw . they used a multiple lens - plane calculation that implements the discrete recursion relations for the position of a given photon and for the jacobian matrix of the lens mapping at this position ( schneider & weiss 1988 ; schneider et al . 1992 ; see seitz et al . 1994 for a thorough justification for this approach ) . aside from the distance factors , the main input into the recursion relations is the shear matrix at each lens plane . the ray tracing algorithm consists of three parts : constructing the dark matter lens planes , computing the shear matrix on each plane , and using these to evolve the photon trajectory from the observer to the source . the details involved at each step are as follows : \1 . the dark matter distribution between source and observer is projected onto @xmath80 equally spaced ( in comoving distance ) lens planes . the particle positions on each plane are interpolated onto a grid of size @xmath81 . since the three - dimensional mass distribution is taken from a single realisation of the evolution of the lss , the projected mass distributions of consecutive lens planes are correlated . in order to decorrelate them , the projection is carried out along a randomly chosen one of the three coordinate axes ; in addition , the origin of the coordinate system in each lens plane is translated by a random vector and the lens plane is rotated by a random angle . in this way , the projected mass distributions of consecutive lens planes are as independent as possible , given the restriction of only a single realisation of the 3-d matter distribution . \2 . on each plane , the shear matrix is computed on a grid by fourier transforming the projected density and using its fourier space relation to the shear . the inverse fourier transform is then used to return to real space . the photons are started on a regular grid on the first lens plane . perturbations along the line of sight distort this grid and are computed using the relation between deflection angle and projected density . once we have the photon positions , we interpolate the shear matrix onto them and solve the recursion relations for the jacobian of the mapping from the @xmath48-th lens plane to the first plane . \4 . solving the recursion relations up to the source plane yields the jacobian matrix at these positions . note that the ray tracing is done backwards from the observer to the source , thus ensuring that all the photons reach the observer . the first lens plane is the image plane and has the unperturbed photon positions . all sources are assumed to be at a redshift of @xmath82 . there are two kinds of resolution limitations in the ray - tracing simulations . the first reflects the finite size and resolution of our n - body simulations , the second the use of finite grids when computing deflection angles and shear tensors on the lens planes . at the peak redshift of the lensing contribution , both effects give a small scale resolution of order @xmath83 . however , since the lens efficiency is not very sharply peaked , effects at other redshifts also enter . thus depending on the statistical measure being used , the small scale resolution lies in the range @xmath84 . on large scales the finite box - size of the n - body simulations sets the upper limit on the angular scales available . the angular size of our simulation box at @xmath85 is about 3@xmath86 . thus on scales comparable to @xmath87 , only a few sample regions are available , leading to large fluctuations across different realisations . we therefore restrict our considerations to apertures with radius @xmath88 using one realisation for every cosmological model . for the @xmath89cdm model , we use ten different realisations of the ray tracing simulations ( i.e. , they differ in the direction of projections , the translation and rotation of the projected matter distribution in the individual lens planes ) to estimate the cosmic variance . @rllll@ + & & & & + [ 10pt ] + @xmath90 & @xmath74 & @xmath74 & @xmath74 & @xmath74 + @xmath91 kpc ] & 36 & 36 & 30 & 30 + @xmath92 & 0.5 & 0.21 & 0.21 & 0.21 + @xmath93 mpc ] & 85 & 85 & 141 & 141 + @xmath0 & 1.0 & 1.0 & 0.3 & 0.3 + @xmath94 & 0.0 & 0.0 & 0.7 & 0.0 + @xmath95 [ km / s / mpc ] & 50 & 50 & 70 & 70 + @xmath96 & 0.6 & 0.6 & 0.9 & 0.85 + @xmath97 & 1.0 & 1.0 & 1.4 & 1.4 + field size [ @xmath98 & 2.7&2.7&3.4&3.9 + + for each of the shear maps generated as described in the last section , we create a 2-dim . `` @xmath1 map '' by simulating `` observations '' of @xmath1 as a function of position on the 2-dim . shear maps . the probability distribution function of @xmath1 ( pdf ) and some of its moments are then calculated for every @xmath1 map and compared to the analytical model . it is most instructive to consider two different sets of simulated maps : in the first , we neglect noise from the intrinsic ellipticity distribution of the background sources and compute @xmath1 directly from the shear values on the grid according to eq . ( [ mapshear ] ) . we do this either in the limit of weak lensing , i.e. we use ( [ mapshear ] ) directly , or we replace @xmath39 in ( [ mapshear ] ) by the reduced shear @xmath99 , which is the quantity estimated from the observable galaxy ellipticities . in the second set of simulations we introduce ellipticities of background galaxies according to the distribution function ( [ distrib ] ) . the ellipticities add noise to @xmath1 . the noise - free results are the ones best compared to the analytic results , whereas the ones accounting for intrinsic ellipticities yield a more realistic description of the observational situation . in the following the term `` without noise '' will refer to the first set of @xmath1 simulations , while the term `` with noise '' will be used for the second one . as an illustrative example , the 2-dimensional distribution of @xmath100 for a standard cdm ( scdm ) and an open model ( ocdm ) is shown in fig . [ s_and_ocdm ] . in both cases high peaks in these maps correspond to haloes in the intervening matter distribution . it is possible to construct a shear - limited sample of haloes from these maps and to determine their abundance . comparing the two model universes , we see that the @xmath1 maps reflect the different growth of structure in different cosmologies . the @xmath1 map of the ocdm model is dominated by many isolated peaks which correspond to already collapsed dark matter haloes . the level of background noise coming from matter not yet collapsed is considerably smaller than for the scdm model in which the structure forms later . the peaks in the scdm model are less pronounced and isolated than in the open model . once we have computed the 2-dimensional distribution of @xmath1 , it is straightforward to determine the one - point probability distribution function ( pdf ) of @xmath1 and its moments . the pdf contains the cosmological information . the lower order moments like rms value and skewness can be derived analytically under simplifying assumptions , but the pdf itself can not be calculated . therefore , ray tracing simulations provide the only tool for testing the precision of the analytical calculations . the qualitative features of the pdf for different filter scales @xmath3 and for the four different cosmologies ( table [ cosmo ] ) can be studied in fig . [ histo ] . the first point to note is that the non - gaussian features , namely the tail of the pdf at high @xmath1 values , are less pronounced for larger filter scales . this is due to the fact that the smaller filter scales are more sensitive to the already collapsed , non - linear objects . the second feature to note is the exponential decrease of the tail of @xmath1 which was already obtained semi - analytically in ks2 . we shall discuss this feature in more detail later in this section . we now turn to the rms value @xmath101 of @xmath1 . fig . [ dispersion ] compares the analytical rms value of @xmath1 calculated using the nonlinear power spectrum of peacock & dodds ( 1996 ) to the rms values computed from the pdfs without noise for @xmath39 ( left panel ) and @xmath99 ( right panel ) . the comparison of the latter shows that the difference between shear and reduced shear is negligible even on filter scales as small as @xmath102 arcmin corresponding to the highly nonlinear regime of the mass distribution . in the left panel of fig . [ dispersion ] , there is an excellent agreement between the analytic predictions and the rms values computed from simulations for the scdm model . there is also good agreement for the @xmath103cdm and ocdm models , especially for the larger apertures . the notable exception is the @xmath89cdm model , for which the simulations for small filters deviate by a larger factor from the theoretical predictions . when interpreting this difference between analytical calculation and simulation in the @xmath89cdm model , one has to keep in mind that the numerical results of fig . [ dispersion ] are based on a single realisation . as the cosmic variance is relatively large , it is possible that the large deviation is due to the special choice of the realisation . this interpretation is supported by the fact that the mean for the 10 realisations is considerable lower than for the single realisation plotted . furthermore , the field sizes of the simulated fields used are too small to represent a characteristic region of the universe . the next higher moment of the pdf is the skewness , which is defined as @xmath104 for which we can perform a similar analysis as for the rms value of @xmath1 . as pointed out by bernardeau et al . ( 1997 ) , van waerbeke et al . ( 1999 ) , and jsw , the skewness defined in analogy to ( [ skew_def ] ) using a top - hat filter is a very sensitive probe of the cosmic density parameter @xmath0 . the dependence of the skewness on filter scale @xmath3 is displayed in fig . again , we compare the skewness computed from the pdf obtained from the ray tracing simulations without noise , both using @xmath39 and @xmath99 , to the skewness of @xmath1 obtained using quasi - linear theory ( svwjk ) . the error bars on the skewness for the @xmath89cdm model for 2 , 5 , and 10 arcmin are derived from the 10 different realisations and are centred on their arithmetic mean . again , the differences between the skewness obtained from simulations with @xmath39 and @xmath99 are small , though slightly larger than for the dispersion , owing to the larger contribution from high-@xmath105 regions to the skewness . this difference , which is of order a few percent at most , has been predicted to be small in the appendix of svwjk . when comparing the skewness as determined from second - order perturbation theory for the density evolution to that obtained from simulations ( either computed with @xmath39 or @xmath99 ) we see that the former underpredicts the skewness by factors of up to 2 . this failure of quasi - linear theory for the prediction of higher - order moments has been demonstrated previously ( jain & seljak 1997 ; gaztanaga & bernardeau 1998 ) . as we only determine the skewness on scales below 10 arcmin , we are in a regime where the density contrast is non - linear already . the skewness as calculated by hui ( 1999 ) using the so - called hyper - extended perturbation theory ( scoccimarro & frieman 1999 ) may provide a more accurate analytical prediction of @xmath106 than that from second - order perturbation theory . another point to note is the increase of the skewness towards smaller filter scales . generally speaking , such a behaviour is expected , as the non - linear structure growth becomes more and more important for small filter scales . this increase is described insufficiently by quasi - linear theory : for the two eds universes and even for the @xmath103 model on large filter scales above 5 arcmin , this increase ( not the absolute value ! ) is predicted satisfactorily , but the slope for the open model is larger than analytic values on all scales displayed . this discrepancy between fully non - linear simulations and quasi - linear theory can be attributed to the fact that the open model is much more dominated by already collapsed , non - linear objects than all other models . the highest moment we consider explicitly is the kurtosis @xmath107 @xmath108 the kurtosis is not only important by itself , but also for the determination of the error of the rms value of @xmath1 , as will be discussed . as for the skewness , the kurtosis for the noise - free simulations for both @xmath39 and @xmath99 is plotted , and the scatter for @xmath89cdm is determined from the 10 realisations . no analytic result for @xmath107 is available ; however , using third - order perturbation theory , bernardeau ( 1998 ) has calculated the kurtosis for a top - hat filter . we clearly see that the difference between @xmath39 and @xmath99 becomes important for the kurtosis , at least for the smaller filter scales , since it is even more dominated by the non - gaussian tail of the pdf than the skewness . the large error bars on the kurtosis are mainly due to large cosmic variance in combination with the small fields used ; thus , the current simulations are unable to provide an accurate determination of @xmath107 . we now turn to the error bars on the rms values of @xmath1 in fig . [ dispersion ] . in the right panel , they were estimated as the standard deviation from 10 different realisations for the @xmath89cdm model . the error bars in the left panel were calculated as follows : as shown in svwjk , an unbiased estimator of @xmath60 from a single aperture is given by @xmath109 where @xmath110 is the number of galaxies in the aperture , and @xmath111 is the value of the weight function @xmath19 for the @xmath40-th galaxy . the dispersion of this estimator is @xmath112 where the two terms in parenthesis correspond to the noise from the intrinsic ellipticity distribution , and the gaussian cosmic variance , respectively , whereas the term involving @xmath107 is the excess cosmic variance due to non - gaussianity . for a collection of @xmath113 independent apertures , all containing the same number of galaxy images , an unbiased estimator for @xmath60 is the mean @xmath114 of @xmath115 over these apertures , and the dispersion is @xmath116 note that this result does not assume that the density field is gaussian . if one had a collection of @xmath117 fields widely separated on the sky , they would be statistically independent , so that @xmath118 . in the opposite situation where a consecutive area on the sky is available , one can lay down apertures on that field , but they will not be statistically independent . however , as was shown in svwjk , the @xmath1 values of two apertures which touch each other ( i.e. , with separation twice their radii ) , are almost uncorrelated . whereas the fact that the two aperture masses in these two apertures are uncorrelated does _ not _ imply that they are _ independent _ ( which would mean that the joint probability distribution for the values of @xmath1 would factorize ) as would be the case for gaussian fields we assume the statistical independence for estimating the effective number of fields @xmath117 entering ( [ sigma_map ] ) . thus , the error bars in the left panel of fig . [ dispersion ] are obtained from ( [ sigma_map ] ) , assuming that the number of independent apertures is @xmath119 ^ 2 $ ] , where @xmath120 is the side length of the simulated shear field . in contrast , the error bars plotted in the right panel of fig . [ dispersion ] for the @xmath89cdm model at the three different filter scales @xmath121 arcmin are based on 10 different realisations of the ray - tracing simulations and allow one to obtain a rough estimate for the error from cosmic variance . notice that the error bars are centred on the arithmetic mean of the 10 realisations and _ not _ on the plotted results from a single realisation . comparing the size of the error bars in both panels , we see that both methods give errors of the same order of magnitude even though the errors estimated from the 10 realisations are smaller than the errors from the estimator of @xmath122 . there are two possible reasons for this : first , the effective number of independent apertures is probably larger than our estimate given above , so that the error bars on the left panel in fig . [ dispersion ] most likely overestimate the true error . second , in the calculation of the error bars in the right panel , it was assumed that the 10 realisations are independent ; but as argued in sect . [ ray_tracing ] it is possible that the realisations are not completely independent . this would lead to an underestimation of the cosmic variance . from fig . [ dispersion ] these two competing effects can not be quantified . it should be noted that at least on the largest scale plotted , the contribution of the intrinsic ellipticity distribution to the error ( [ sigma_map ] ) is completely negligible compared to the cosmic variance . as already indicated in sect . [ map_anal ] , high signal - to - noise peaks of @xmath1 can be identified with dark matter haloes , rendering the construction of a mass - limited ( more correctly : shear - limited ) sample feasible . analytically , the halo abundances can be modelled using the press & schechter ( 1974 ) prediction for the mass- and redshift - dependent halo number density , and the universal density profile of nfw , while in the simulated @xmath1 map all connected regions above the corresponding threshold are counted as haloes . we shall consider haloes with signal - to - noise ratio @xmath55 larger than 5 , i.e. , a peak in the @xmath1 map is counted as a halo if @xmath123 . we consider two differently constructed halo abundances in the following : the first sample is simply @xmath124 , the number density of haloes with an aperture mass larger than @xmath125 for a given filter size @xmath3 . the second sample selects peaks with @xmath126 within a connected , cross sectional area of @xmath127 , where @xmath128 is the corresponding cross section radius ; the number density of such peaks is denoted @xmath129 . hence , the size of the peaks in the second sample exceeds the threshold @xmath128 ; these peaks are expected to be more robust with respect to noise coming , e.g. , from the intrinsic ellipticity distribution and measurement errors . we use a fixed value of @xmath130arc minutes . in fig . [ no_noise ] the number density of the two halo samples as determined from the simulations without noise are compared to the results from the analytic calculation in ks1 over a range of filter scales @xmath131 . the four panels in fig . [ no_noise ] refer to the four cosmological models considered . the error bars for the @xmath89cdm model at 2,5 , and 10 arcmin are again obtained from the 10 different realisations centred on the arithmetic mean of the realisations . in general , the number counts determined from simulations agree astonishingly well with the analytical results , considering the simplifying assumptions entering the latter . the deviations between simulations and analytical calculation for three of the four cosmologies , namely scdm , ocdm , and @xmath103cdm , and especially for the filter scales above 5 arcmin , are less than 10 % . the largest deviation found for these three models is a factor of 2 , for the @xmath103cdm model at smallest filter scale . the only notable exception is the @xmath89cdm model where the deviation remains above 10 % even for the largest filter scales ( @xmath132 arcmin ) . this relatively bad agreement has already been noticed for the rms value of @xmath1 and is probably due to the fact that the realisation plotted is not characteristic for the mean properties of that model , as also indicated by the fact that the halo abundance lies above the mean of all realisations as indicated by the error bars . the good agreement between analytic estimates and numerical results for the halo number density are surprising , given that ( a ) press - schechter theory does not exactly reproduce the spatial number density of haloes when compared to n - body simulations , and ( b ) the universal density profile found by nfw has been obtained by spherical averaging , and therefore can not account for the non - axisymmetry of their projected density . furthermore , ( c ) the haloes found from the simulated @xmath1 are expected to be affected by projection effects ( reblinsky & bartelmann 1999 ) which are completely neglected in the analytic estimates . despite these effects which one might suspect to yield significant discrepancies , we find that the analytic estimates are very accurate . we also investigate the halo abundance in an observationally more realistic situation in fig . [ with_noise ] , including the noise from the intrinsic ellipticity distribution of the background sources . the plot displays the same quantities as fig . [ no_noise ] except for the fact that the halo abundances of the two different samples have been determined using the tangential ellipticities in the case of the simulations . the analytic estimates are obtained as in ks1 . for all four cosmologies , we determined error bars using 7 different realisations of the ellipticity distribution of the background sources ( [ distrib ] ) . the error bars from the 10 realisations shown for @xmath89cdm are slightly sub - poissonian , as in fig . [ no_noise ] . as expected from the large value of the kurtosis the error coming from the intrinsic ellipticity distribution is much smaller than the error coming from the cosmic variance . on the whole , the number of detected haloes is increased in all cosmologies because , due to the steepness of the press schechter mass function for massive objects , there are more objects just below the threshold than above it . so on average more objects will be lifted above the threshold by noise than brought down below it . in fig . [ tail ] we compare the pdf for @xmath133 as obtained from analytic calculations ( ks2 ) with that derived from the simulations without noise . the pdf is shown for four filter scales @xmath134 arcmin in the range @xmath135 , for which the analytic results predict a nearly exponential behaviour . indeed , the numerical pdf in the non - gaussian tail also seems to follow an exponential rather closely , with a slope very similar to the analytic result . in order to see how much the pdf varies between different realisations , we have plotted in fig . [ tail1 ] the pdf for @xmath135 obtained from the 10 realisations in the @xmath89cdm model , for 3 filter radii , together with their mean and the corresponding analytic prediction . we find that for the smallest filter scale @xmath136 , all 10 realisations are clearly below the analytic result , whereas for the larger filters , the realisation mean of the pdf agrees very well with the analytic prediction . remembering that the analytic predictions were made by assuming that all high values of @xmath1 are coming from regions close to collapsed haloes , in addition to the assumptions used for estimating the number density of @xmath1 peaks ( press - schechter halo abundance and nfw density profile ) , this good agreement is somewhat surprising . we used ray - tracing simulations through n - body - generated cosmic density distributions to study the statistical properties of the aperture mass @xmath1 as a statistics for cosmic shear measurements and for finding dark matter haloes from their shear properties . in particular , we have compared results from these simulations with the available analytic results and found in most cases a very good agreement , except for the skewness which is the least accurate of these predictions . whereas all other predictions tested here are based on manifestly non - linear results ( like the press - schechter halo abundance and the peacock & dodds power spectrum ) , the skewness was estimated analytically by using second - order eulerian perturbation theory which , on the scales considered , is not very accurate . comparing the results from our simulations with analytic studies , we obtain the following main results : ( 1 ) the rms of @xmath1 is accurately described by analytic results if the fully non - linear prescription of the power spectrum of density fluctuations is used . ( 2 ) the statistical error of this rms is dominated by cosmic variance , which in turn depends on the kurtosis of @xmath100 . this kurtosis turns out to be relatively large even on angular scales of @xmath137 , implying the need for many more measurements of @xmath1 than expected for a gaussian field , for a given accuracy of the estimated projected power spectrum . ( 3 ) the skewness is only approximately described by analytic considerations based on second - order perturbation theory . ( 4 ) the predicted abundance of dark matter haloes detectable at given statistical significance is very well approximated by the semi - analytic theory which combines the press - schechter number density of haloes with the universal density profile of navarro , frenk & white . ( 5 ) similarly , the functional form of the probability distribution of @xmath1 for values much higher than the rms ( i.e. , in the non - gaussian tail ) is found to closely follow an exponential form , of similar slope and amplitude as predicted by analytic theory which needs to assume that such high values originate due to collapsed haloes . thus , on the whole , we find that the analytical estimates for the statistical properties of @xmath1 are surprisingly accurate . we also find that our simulations are not sufficiently large for an accurate estimate of the higher - order statistical measures , owing to the finite size of the simulation box in combination with the large effect of cosmic variance . as discussed in svwjk , ks1 , ks2 , van waerbeke et al . ( 1999 ) and bartelmann & schneider ( 1999 ) , the aperture mass is a useful cosmic shear measure which will eventually allow one to constrain cosmological parameters , completely independent of any assumption on the relation between mass and light . for this purpose , the predictions from cosmology must be known precisely , and our results here indicate that analytic estimates are relatively accurate . unfortunately , we found a large cosmic variance ; e.g. , in the estimate of the variance of the rms value of @xmath1 , the kurtosis enters and it decreases only rather slowly with increasing filter scale . it can be expected that the first successful application of the aperture mass will be the definition of a sample of haloes defined in terms of their lensing properties only , with a first example given by erben et al . the combination of cosmic shear information and cmb measurements can be extremely useful , as shown by hu & tegmark ( 1999 ) , increasing the precision of the determination of cosmological parameters substantially over each of the two individual methods . their study was based solely on the dispersion of cosmic shear , i.e. , on second - order statistics . it is to be expected that a similar combination of cmb results with the pdf of @xmath1 will yield even more precise parameter estimates . a detailed study of this combination is expected to be very valuable , but requires a larger grid of cosmological n - body simulations . we thank m. bartelmann for his many valuable suggestions and a careful reading of the manuscript , as well as an anonymous referee for his constructive comments . this work was supported by the sonderforschungsbereich 375 - 95 fr astro - teilchenphysik " der deutschen forschungsgemeinschaft . bartelmann m. , schneider p. , 1999 , a&a 345 , 17 bartelmann m. , huss a. , colberg j.m . , jenkins a. , pearce f.r . , 1998 , a&a 330 , 1 bernardeau f. , 1998 , a&a 338 , 375 bernardeau f. , van waerbeke l. , mellier y. , 1997 , a&a 322,1 blandford r.d . , saust a.b . , brainerd t.g . , villumsen j.v . , 1991 , mnras 251 , 600 couchman h.m.p . , thomas p. a. , pearce f. r. , 1995 , apj 452 , 797 diaferio a. , kauffmann g. a. , colberg j.m . , white s.d.m , 1999 , mnras 307 , 537 erben t. , van waerbeke l. , mellier y. et al . , 1999 , a&a , submitted fahlman g. , kaiser n. , squires g. , woods d. , 1994 , apj 437 , 56 gaztanaga e. , bernardeau f. , 1998 , a&a 331 , 829 hu w. , tegmark m. , 1999 , apj 514 , l65 hui l. , 1999 , apj 519l , 9 jain b. , seljak u. , 1997 , apj 484 , 560 jain b. , seljak u. , white s.d.m . , 1999 , submitted to apj , preprint astro - ph/9901191 ( jsw ) jenkins a.r . , frenk c.s . , pearce f.r . et al . , 1998 , apj 499 , 20 kaiser n. , 1992 , apj 388 , 272 kaiser n. , 1998 , apj 498 , 26 kaiser n. , squires , g. , 1993 , apj 404 , 441 kauffmann g. a. , colberg j.m . , diaferio a. , white s.d.m . , 1999a , mnras 303 , 188 kauffmann g. a. , colberg j.m . , diaferio a. , white s.d.m . , 1999b , mnras 307 , 529 kruse g. , schneider p. , 1999a , mnras 302 , 821 ( ks1 ) kruse g. , schneider p. , 1999b , submitted to mnras , preprint astro - ph/9904192 ( ks2 ) lombardi m. , bertin g. , 1998a , a&a 330 , 791 lombardi m. , bertin g. , 1998b , a&a 335 , 1 miralda - escude j.,1991 , apj 380 , 1 navarro j.f , frenk c.s . , white s.d.m . , 1996 , apj 462 563 ( nfw ) navarro j.f , frenk c.s . , white s.d.m . , 1997 , apj 490 , 493 ( nfw ) peacock j.a . , dodds s.j . , 1996 , mnras 280 , l19 pearce f.r . , couchman h.m.p . , 1997 , newa 2 , 411 press w.h . , schechter p. 1974 , apj 187 , 425 reblinsky k. , bartelmann m. , 1999 , a&a 345,1 schneider p. , 1995 , a&a 302 , 639 schneider p. 1996 , mnras 283 , 837 ( s96 ) schneider p. , weiss a. , 1988 , apj 330 , 1 schneider . p. , ehlers j. , falco e. e. , 1992 , _ gravitational lenses _ , springer verlag , berlin schneider p. , van waerbeke l. , jain b. , kruse g. 1998 , mnras 296 , 873 ( svwjk ) schramm t. , kayser r.,1995 , a&a 299 , 1 scoccimarro r. , frieman j. , 1999 , apj 520 , 35 seitz c. , schneider p. , 1997 , a&a 318 , 687 seitz s. , schneider p. , 1996 , a&a 305 , 383 seitz s. , schneider p. , ehlers j. , 1994 , class . quant . grav . 11 , 2345 seitz s. , schneider p. , bartelmann m. , 1998a , a&a 337 , 325 seitz s. , saglia r.p . , bender r. , hopp u. , belloni p. , ziegler b. , 1998b , mnras 298 , 945 tyson j.a . , valdes f. , wenk r.a.,1990 , apj 349 , l1 van waerbeke l. , bernardeau f. , mellier y. , 1999 , a&a 342,15
in the past two decades , cosmology has acquired a new standing due to a wide range of observations of high precision data @xcite . these observations seem to reveal that the two dominant components of the universe are a non - baryonic form of matter , the so - called dark matter @xcite , and an exotic form of energy , dark energy @xcite . the former leads to structure formation while the latter is responsible for the present accelerated expansion of the universe . the search for an explanation for the existence and properties of these dark components of the universe has prompted a strong debate about their origin and nature . the behaviour of the galaxy rotation curves and the dynamical mass in clusters of galaxies suggest the existence of an exotic form of matter , dark matter , at galactic and extra galactic scales , respectively , where the presence of this sort of matter is revealed through its gravitational effects . notwithstanding , recent results demonstrated that dark matter may interact with ordinary matter , with the production of patterns of annihilations or decays of dark matter particles in the fluxes of cosmic rays @xcite . evidence for the existence of dark matter lies in the velocity dispersion of clusters that , combined with its morphological features , leads one to conclude that the overall mass should be far greater than the visible mass . in fact , the dark matter component of these clusters is well described by dark matter models , where the interaction of dark matter with baryonic matter determines the density profile of galaxy clusters @xcite . an alternative approach to the above mentioned problems is to consider modified theories of gravity . amongst modified gravity theories , those that introduce higher order curvature terms in the einstein hilbert action and , in particular , the @xmath0 theories of gravity @xcite offer alternative explanations for these problems , along with other cosmological , astrophysical , and high - energy physics motivations @xcite . recently , it has been shown that , in the context of power - law theories , @xmath1 , the rotation curves can be explained as a curvature effect @xcite ; a similar result @xcite can be obtained in another extension of gravity that relies not only on a non - trivial curvature term , but also on a non - minimal coupling between matter and geometry @xcite . clusters are the largest astronomical structures whose masses can be measured in a reliable form , since they constitute the largest configuration of objects that passed through gravitational relaxation and entered into virial equilibrium @xcite . the measuring methods , x - rays and gravitational lensing for visible and dark matter respectively , reveal that the masses of clusters are approximately seven to ten times larger than the total combined mass of stars and hot gas in the intracluster medium ( icm ) . this missing mass is predominant in the inner regions of clusters , but it is also extended beyond the core radius , as defined by the gas density distribution . in the context of modified newtonian dynamics @xcite , another popular ( albeit incomplete @xcite ) alternative to general relativity ( gr ) , it is difficult to explain cluster observations without the assumption of `` real '' dark matter , because the mass of the dark component is otherwise significantly reduced but not completely removed @xcite . the problem of clusters has recently been discussed in @xmath0 theories , by considering a generalized version of the virial theorem . previously discussed in the context of a possible unification and interaction of dark matter and dark energy @xcite . assuming a steady state , the virial theorem can be used to deduce the mean density of astrophysical objects , and therefore the corresponding total mass , such as galaxies and clusters , by observing the velocities of test particles moving around them . the overall mass of a sample of clusters can also be estimated through corrections in the gravitational potential that emerge in the weak field limit of these theories @xcite . along with these works , there were also conducted the first cluster abundance constraints on a modified gravity model , specifically the modified action @xmath0 model , which predict deviations in the abundance of massive dark matter halos @xcite . in the context of the non - minimal coupling between the ricci scalar curvature and matter @xcite , it has been shown in several works that both dark energy and dark matter can be effectively described within this framework , along with other cosmological and astrophysical implications : these include the perturbed hydrostatic equilibrium of the sun @xcite , the acceleration of the universe @xcite , the galaxy rotation curves @xcite , the reheating scenario after inflation @xcite , the mimicking of a cosmological constant @xcite , and the change in the gravitational potential @xcite . in order to extend this non - minimally coupled model one tries to obtain the large mass difference of clusters through a dark matter mimicking , an extra matter component that emerges due to the presence of the non - minimal coupling , and a subsequent modification of gravity . one first explores here the scenario of the abell cluster a586 , which is a nearby massive strong - lensing cluster , with a high x - ray luminosity , related with the mass of the intracluster gas . its morphology is well suited to the assumption of spherical symmetry , making it an ideal test bed for modifications of gravity at extra - galactic scales . one then extends this analysis to a larger set of galaxy clusters with the same features . this work is organized as follows . first , one shortly reviews the formalism leading to the total cluster mass and the field equations that result in the presence of the non - minimal coupling and stablishes the strategy to address the problem of mimicking the dark matter component of a cluster . one then studies the particular case of the abell cluster a586 , discusses the coupling function that leads to the dark matter mimicking , and extends the model to a larger sample of galaxy clusters . finally , the numerical results are presented and discussed . galaxy clusters are usually considered as closed gravitational systems with spherical symmetry and in hydrostatic equilibrium when virial equilibrium is attained despite the fact that recent observations reveal that clusters have more evolved structures with strong interactions and dynamical activity , particularly in their inner regions @xcite . under these assumptions , the structure equation can be derived from the collisionless boltzmann equation for an isotropic system @xcite , ( _ g _ r^2)=-_g ddr , [ boltzmann_equation ] where @xmath2 is the newtonian gravitational potential of the cluster , @xmath3 the mass - weighted velocity dispersion in the radial direction , and @xmath4 the gas - mass density . the pressure profile is related with these quantities through , @xmath5 , and inserting in eq . ( [ boltzmann_equation ] ) leads to @xmath6 , where the prime indicates differentiation with respect to the radial coordinate . the newtonian gravitational potential @xmath2 can be read from the newtonian limit of the modified einstein field equations ( as addressed later ) , _ i.e. _ a suitably modified poisson equation . however , this does not prevent us from formally writing the latter as the effect of the non - minimal coupling ascribed to the `` mimicked '' dark matter component , so that ^2 = 4 g ( + _ dm ) . [ poissoneq ] this approach shall be discussed in more detail in the following section . for a gas sphere , with temperature profile @xmath7 , the velocity dispersion becomes @xmath8 , where @xmath9 is the boltzmann constant , @xmath10 the mean mass particle , and @xmath11 the proton mass . introducing this expression into eq . ( [ boltzmann_equation ] ) gives ( kt m_p_g ) & = & -_g ddr , or , equivalently , -ddr & = & kt m_pr . [ eq : boltzmannpotential ] therefore , from the models for the gas density and temperature profiles , the total mass enclosed in a sphere with a given radius @xmath12 can be obtained from , m ( r ) & = & 4(_g + _ dm ) r^2 dr = r^2 g ddr , [ dynamical_mass_possion ] leading to the usual expression for the mass profile of a spherical mass distribution in hydrostatic equilibrium m ( r ) & = & -kt rgm_p(d_g dr+dt dr ) . [ dynamical_mass ] as already stated , this is the total mass that one should observe in a cluster and that , in general , agrees well with observations from gravitational lensing @xcite . however , the baryonic component of the cluster mass , galaxies mass in the visible spectrum and gas in x - ray spectrum , where the latter dominates , is much less than the total observed mass leading one to conclude the existence of a dark matter component that in the present work results from the presence of the non - minimal coupling . hence , the equilibrium equation , eq . ( [ dynamical_mass ] ) , which leads to the total mass within a cluster , can be used to derive the amount of dark matter present in a cluster of galaxies , and its spatial distribution , by the mass difference between this and the gas - mass estimates as provided by x - ray observations . in the following , one tries to reproduce the total mass profiles of a sample of galaxy clusters , where the additional density component emerges not from an exotic form of `` dark '' matter but from a change in geometry due to the presence of the non - minimal coupling in the action . clusters of galaxies present a velocity dispersion with a relative constant value along the radius , which reveals an implicit matter density profile of the form @xmath13 . the singular isothermal sphere is the simplest model whose density profile , @xmath14 , leads to a constant and isotropic velocity dispersion @xmath15 along the clusters profile . however , numerical fits to the observational data show that the density profile of clusters are flatter than the isothermal one at small radii and steeper at larger radii . a general form replicating this behaviour is given by the cusped density profile , _ m&=_0 & ( r r_s ) ^-p(1+r r_s)^p - q , [ cusped_profile ] where the parameters @xmath16 and @xmath17 determine the respective slopes of the power - law density profile in the inner and outer regions , and @xmath18 is the so - called core radius , signaling the steepening of the profile . these parameters usually present values within the ranges @xmath19 and @xmath20 , for models such as the navarro - frenk - white ( nfw ) profile ( @xmath21 and @xmath22 ) , the moore profile ( @xmath23 and @xmath22 ) , and the rasia profile ( @xmath21 and @xmath24 ) . in fact , one of the important results of hierarchical cdm models is that the density profile of cold dark matter halos is described by the nfw model @xcite . the more relevant baryonic component is the icm in the form of hot gas , which emits observable x - ray radiation . in general , the observed surface - brightness profiles of clusters that reveal the underlying visible matter profile of this intracluster gas are well described by the so - called @xmath25 model : _ g(r)=_g0(1+r^2 r^2_c)^-3/2 , [ visible_profile ] where @xmath26 is the core radius , which has a specific value for each cluster @xcite . in this cluster , the emission measured profile is well described by the @xmath25 model , eq . ( [ visible_profile ] ) , and , from a least - squares fit of two different sets of observational data , the obtained values for the free parameters are @xmath27 and @xmath28 @xcite . the icm particle number density profile is given directly by the analytic fit for the projected emission measure profile , eq . ( [ visible_profile ] ) , and is easily converted into the gas density . the deep potential well of the cluster compresses the icm gas to x - ray emitting temperatures , such that the gas temperature can be inferred from the derived spectrum . for a polytropic temperature model @xcite , t(r)&=&t_0 ( 1+r^2 r_c^2)^-3(-1)/2 , [ temperature_profile ] where @xmath26 and @xmath25 are parameters obtained in the fit for the visible mass density , and a least - squares fit leads to the values @xmath29 , @xmath30 @xcite . naturally , setting @xmath31 leads to the isothermal temperature profiles . the total mass follows from the hydrostatic equilibrium equation , eq . ( [ dynamical_mass ] ) , using the previous expressions for the visible density and temperature profiles m(r)&= & m_0 ( r r_c)^3(1+r^2 r^2_c ) ^-1 - 32(-1 ) , [ mass_profile ] with m_0 = 3 k t_0 r_c g m_p . from the relation for a spherical mass distribution , = 4r^2 ( r ) , one may approximate the dark matter density by the total density profile using eq . ( [ mass_profile ] ) in the last expression and solving for @xmath32 , [ dm_profile ] & & _ dm(r ) _ dm0(1+)(1+r^2r_c^2)^-2 . where _ dm0&=&3 k t_0 4 g m_p r_c^2 = m_0 4r_c^3 [ defdm0 ] , and one defines @xmath33\be$ ] , for convenience . finally , one introduces the virial radius @xmath34 , which signals the distance where the density of the cluster approaches a multiple @xmath35 of the critical density of the universe @xmath36 , where @xmath37 is the hubble expansion rate at the present . thus , from @xmath38 , one may estimate the virial radius from the relation eq . ( [ dm_profile ] ) , & & ( _ dm0_c ) ^1 /(2-+ ) , which , for @xmath39 , with @xmath40 , leads to a virial radius of @xmath41 . recently , a generalization of the @xmath0 modified theories of gravity was considered that further extends the presence of curvature invariant terms in the einstein hilbert action by non - minimally coupling the scalar curvature with matter @xcite . the action is obtained by adding an additional @xmath42 term , dependent on the scalar curvature @xmath43 and coupled with the matter lagrangian @xmath44 , to the usual metric theories of gravity , leading to @xcite s = d^4 x , [ model ] where @xmath45 ( with @xmath46 ) are arbitrary functions of the scalar curvature and @xmath47 is the metric determinant . setting @xmath48 one obtains the usual @xmath0 theories of gravity , and gr is recovered if one takes the linear function @xmath49 , where @xmath50 . variation with respect to the metric , @xmath51 , leads to the field equations , & & ( f_1 + 2 f_2_m)r_-1 2f_1g_= + & & = ( _ -g_)(f_1 + 2 f_2_m)+(1+f_2)t _ , [ field_equations ] where @xmath52 is the energy - momentum tensor . the respective trace equation is given by & & ( f_1 + 2 f_2_m)r-2f_1= + & & = -3(f_1 + 2 f_2_m)+(1+f_2)t . [ field_equations_trace ] from the covariant derivative of the field equations , eqs . ( [ field_equations ] ) , along with the bianchi identities , @xmath53 , one encounters the covariant non - conservation of the energy - momentum tensor @xcite , _ t^&= & ( g^_m - t^ ) _ ( 1+f_2 ) , [ non - cons ] which can be regarded as an energy - momentum exchange between matter and geometry . the usual covariant conservation for the energy - momentum tensor is naturally obtained when the non - minimal coupling vanishes , @xmath48 . this non - conservation implies that a massive particle in the absence of forces will not describe a geodesic curve . strong variations on the extra term in the last equation can imply a violation of the equivalence principle , leading to a possible way of testing and setting bounds on the coupling functions , @xmath42 . contrary to the usual jordan brans dicke theories , it is impossible to perform a conformal transformation to the einstein frame such that the coupling disappears and the covariant conservation of the energy - momentum tensor is recovered for all matter forms @xcite . indeed ( following ref . @xcite ) , using the conformal transformation g__= f_2 g _ , one obtains the covariant conservation law _ ^= 0 , only if @xmath54 and @xmath55 . from the energy - momentum tensor of a perfect fluid , @xmath56 , where @xmath57 is the energy density , @xmath16 the pressure and @xmath58 the four - velocity ( with @xmath59 ) , the respective trace is @xmath60 , such that one may write , 2= 3p - p=(2+)/3 . however , one obtains different equations of state depending on the chosen lagrangian density @xmath61 . for instance , if @xmath62 @xcite , one has the state equation @xmath63 , a perfect fluid with negative pressure , even though not a cosmological constant for which @xmath64 ; for @xmath65 , then @xmath66 , which is the equation of state for ultra - stiff matter . therefore , this non - conservation is , in fact , a fundamental property of the model , eq . ( [ model ] ) , meaning that even under a suitable conformal transformation the energy momentum tensor is not conserved for all different types of matter . assuming that matter is described by a perfect fluid , the non - conservation , eq . ( [ non - cons ] ) , results in an extra force of the form f^&=&1 + p h^ , [ extra - force ] which has units of an acceleration , and where the projection operator is given by @xmath67 , such that the extra force is orthogonal to the four - velocity , @xmath68 . in order to test the effect of the non - minimal coupling , one focuses on possible deviations from gr due to the presence of the coupling by considering the linear function @xmath69 ; following ref . @xcite , one adopts a power - law non - minimal coupling f_2(r)= ( r r_n ) ^n , [ powerlawcoupling ] where @xmath70 is a characteristic curvature . considering a pressureless perfect fluid , @xmath71 , and the respective trace equation , @xmath72 , along with the lagrangian density , @xmath73 ( a choice discussed in depth in ref . @xcite ) , the field eqs . ( [ field_equations ] ) become & & r_- 1 2rg_= + & & = n ( g_- _ ) + & & + 1 2u_u _ , [ first_equation ] and the respective trace , eq . ( [ field_equations_trace ] ) , reads r=1 2 - 3n . [ second_equation ] if one considers a strong coupling @xmath74 , the above has an implicit solution given by the vanishing of the derivative term in the _ r.h.s . _ , = _ dm _ n ^1/(1-n ) , [ implicit ] where one introduces the characteristic density , @xmath75 . one dubs this a `` static '' solution , not because it is time - independent ( which it is , if one assumes a time independent visible matter density @xmath32 ) , but in opposition to a more evolved , `` dynamical '' solution that would arise if this second term did not vanish . indeed , it was shown that ( see ref . @xcite ) the most general solution of the differential equation corresponds to a dominance of the gradient terms on the _ r.h.s . _ of this equation : however , it was found that this `` dynamical '' solution exhibits very small oscillations around the `` static '' one ( with a very short - wavelength providing the dominance of the gradient term ) . thus , the scaling law @xmath76 remains valid even if the gradient term dominates the dynamics . furthermore , it was shown that , although the above analytical treatment remains valid , this implies that the non - minimal coupling is actually perturbative , _ i.e. _ @xmath77 . this will be used in the following section to avoid the destabilizing effect of the extra force on closed orbits . as already seen in a previous work on the mimicking of the dark matter component in galaxies @xcite , considering the tensor character of the field eqs . ( [ field_equations ] ) leads to the following relation between the energy density and pressure of the mimicked dark matter component for a power - law coupling function @xcite : [ dm_density ] _ dm & = & 1-n1 - 4n2r , + p_dm & = & n 1 - 4n2r , which , combined with eq . ( [ implicit ] ) , leads to a relation between the visible and dark matter density profiles , = 1-n 1 - 4n^1/(1-n ) . [ dm_visible2 ] the above clearly shows that one may obtain a `` dark '' component from the visible matter density profile , with a change of slope given by the scaling exponent @xmath78 , _ i.e. _ , @xmath79 . conversely , if @xmath80 and @xmath81 , one has = m^mn=1-mm^ , [ n - estimative ] which leads to an estimate for the exponent @xmath82 of the power - law coupling function , eq . ( [ powerlawcoupling ] ) based on the knowledge of the visible and dark matter density profiles . for convenience , eq . ( [ second_equation ] ) can be rewritten in a dimensionless form , by defining the rescaled quantities @xcite , y & & r_n r , [ var1 ] + & & ( _ n)^1/(1-n ) , [ var2 ] + & & 2r _ n=1 r r_n , [ var3 ] where one defines the length scale , @xmath83 . the resulting dimensionless form of the trace eq . ( [ second_equation ] ) is & = & ^-n+(1 - 2n)^n - 6n |(^n-1 ) , [ rhovar ] where @xmath84 . the `` static '' equivalent solution , , is thus , @xmath85 , as may easily be checked . hence , by considering a known visible matter density profile @xmath57 ( which in a cluster is approximately equal to the icm gas density , @xmath86 ) , one may solve the differential equation eq . ( [ rhovar ] ) and thus read the resulting mimicked dark matter profile through eqs . ( [ var3 ] ) and ( [ dm_density ] ) . this component can account for the large mass difference in the cluster total density profile , as will be discussed in sec . [ results ] while the perturbative nature of the non - minimal coupling ensures that no deviation from the geodesic motion occurs . before further exploring the mimicking mechanism offered by the scaling law @xmath76 , one addresses an issue mentioned in sec . [ hydro ] , namely , that the ensued modification of the poisson equation due to the presence of a non - minimal coupling would imply that the derivation of is flawed . one begins by recalling that the poisson equation is obtained from the newtonian limit of the metric , ds^2 = -[1 + 2(r)]dt^2 + _ ijdx^idx^j .[metric ] inserting the trace of the einstein field equation ( for non - relativistic particles with @xmath87 ) into the @xmath88 component of the latter , yields the well - known result r = -t 2= 2 r_00 ^2 = 1 4 . since the above metric yields @xmath89 , one may instead resort directly to the scalar curvature , to obtain the same result : r = 2r_00 2 ^2 = 1 2 . since this study focuses particularly on the trace , the latter form is more suitable but both approaches are completely equivalent , as they depend only on the assumed metric , not the underlying gravity model . as before , may be written as @xmath90 , with _ dm = ( 1 - 2n)(r r_n)^n-6n . thus , the above shows that one trivially obtains the same form for the poisson equation , . one can argue that this rewriting obfuscates the effect of the non - minimal coupling , as its _ r.h.s . _ also involves the scalar curvature @xmath91 . this , however , misses the main point of this study : one is not aiming at fully determining the newtonian gravitational potential @xmath2 and the interconnected equilibrium density profile @xmath57 something that would require the derivation of a modified hydrostatic equilibrium equation , as followed in ref . @xcite . instead , the focus of this study ( cf . the previous work addressing galactic dark matter @xcite ) is to establish a relation between the observed gas density profile and what is usually interpreted as dark matter , not to explore why the former adopts a particular configuration : the mimicking mechanism proposed here acts as a `` translator '' between visible and dark matter . before dwelling into the analytical and numerical details of the scenario under scrutiny , one must first consider a possible issue : the presence of an extra - force , as given by eq . ( [ extra - force ] ) . considering a pressureless perfect fluid along with the lagrangian , @xmath92 , the same equation reduces to f^&=&-_(1+f_2 ) h^ . for spherical symmetry , the radial component is f^r&=&-_r ( 1+f_2 ) = - f_2(r)1+f_2(r ) r^ , where the prime denotes a derivative with respect to the radial coordinate . considering a power - law coupling of the form of , one has f^r & = & -n ( r / r_n)^n1+(r / r_n)^n r^r . [ f_r ] if the curvature @xmath43 decreases along the radius , a negative exponent @xmath82 will produce an inward binding force , and a positive one an outward force , which could break the stability of an orbit . as already seen in a previous work @xcite , different solutions for the differential equation , eq . ( [ rhovar ] ) , lead to different behaviours of this extra - force , depending on the relevance of the coupling function . if it dominates , @xmath93 , then the force becomes f_r - nr^r . on the contrary , a perturbative regime leads to a suppression of the expression above by a factor @xmath94 , f_r - n(r r_n)^nr^r . the `` static '' solution given by is derived from the dominant condition @xmath95 ; taking only the outer slope behaviour of cusped density profiles for the dark matter component , @xmath96 , and a positive exponent @xmath82 ( as will be used in the following section ) , one concludes that the extra force has an outward direction , f_r - nr^r ~-n_dm _ dm = nm r . for @xmath82 and @xmath97 of order unity , one obtains a force much greater than the newtonian counterpart , @xmath98 , thus destabilizing the orbital motion of test particles . in the inner region , @xmath99 , one has an approximately constant dark matter density , @xmath100 , yielding a negligible extra force . however , as discussed in the previous section , one finds that the `` dynamical '' solution to must be considered , as the gradient term on the _ r.h.s . _ of the latter dominates : this solution is essentially proportional to ( disregarding small , short - wavelength oscillations ) . thus , for a fixed @xmath70 , it is then clear that @xmath101 increases for lower values of a positive exponent @xmath82 ; thus , one expects that there is a lower bound on @xmath82 , such that the condition @xmath102 fails due to the insufficient suppression of the extra force @xmath103 . in order to assess the value of this lower bound , one first uses eqs . ( [ mass_profile ] ) and ( [ dm_profile ] ) to write the following approximations , valid in the large radius region @xmath104 : [ radial_scaling_0 ] m(r ) ~m_0 ( r r_c)^1 + - , _ dm~_dm0(r r_c)^-2 + - . using the above , one may write f_r & & -n(r r_n)^nr^r ~-n(_dm _ n)^n_dm _ dm ~ + & & ( 2-+)nr_c ( _ dm0 _ n ) ^n ( r r_c)^(-2 + -)n-1 , so that , writing @xmath105 , the perturbative condition @xmath106 reads ( ) ^n ( r r_c)^(-)n-12-n 1 . one now assumes that @xmath107 ( as is the case with the abell cluster a586 , discussed in the following section ) . thus , the relevance of the extra force decreases with the distance from the cluster s center , and one may saturate the _ l.h.s . _ of the above equation by setting @xmath108 , yielding b(r_n , n ) ( ) ^n r_0 r_c 1 . [ lower_bound ] finally , notice that , for a positive exponent @xmath82 , the perturbative condition @xmath109 translates into ( r r_n)^n 1 _ dm0 2r_n r_0 r_n . thus , a lower exponent @xmath82 will increase the _ _ , as expected . in the following section , this lower bound for @xmath82 will be obtained for the abell cluster a586 . in this section , one tries to describe the mass difference of the abell cluster a586 . this particular cluster is a nearby massive strong - lensing cluster and has a high x - ray luminosity , directly proportional to the mass of the intracluster x - ray emitting gas . these features imply a stringent bound for the two measuring methods for its density profile : strong lensing and hydrostatic equilibrium estimatives @xcite . the assumption of spherical symmetry is an accurate hypothesis for its morphology @xcite . the abell cluster a586 is also a relaxed cluster , as pointed out in ref . . recalling eqs . ( [ visible_profile ] ) and ( [ radial_scaling_0 ] ) , one writes the approximations valid in the large radius regime @xmath104 , _ g~(rr_c)^-3 , _ dm~(rr_c)^-2 - 3(-1 ) . [ radial_scaling ] therefore , from the scaling given in eq . ( [ dm_visible2 ] ) , one obtains the relation @xmath110 , allowing for an estimate , through eq . ( [ n - estimative ] ) , for the exponent @xmath82 : [ expected ] n=2 + 3(-2)2 + 3(-1)0.279 , using @xmath111 and @xmath112 @xcite . thus , for the isothermal case ( @xmath31 ) n=1 - 32(2-)0.223 . hence , the results arising from the non - minimal coupling are somewhat similar whether one considers an isothermal ( @xmath113 ) or a polytropic temperature profile . in order to mimic the dark matter component of this cluster , one numerically solves the differential eq . ( [ rhovar ] ) , for an assumed gas density profile , as given by eq . ( [ visible_profile ] ) , and varies the parameters @xmath70 and @xmath82 in order to obtain a best fit between the derived mimicked dark matter density @xmath114 and the observed curve . the considered range for the numerical integration lies between @xmath115 , the scharwzschild radius of the cluster , and a sufficiently large radius , larger than the observed size of the cluster . as a measure of the quality one tries to minimize the quantity @xmath116 , given by _ ^2=_i=1^n ( |_i-_i)^2 ( _ i)^2 , [ sigma2 ] where the sum is performed over a large number of equal spaced radii @xmath117 , the density @xmath118 corresponds to the dark matter profile obtained from the hydrostatic equilibrium estimatives for the radius @xmath117 , and @xmath119 is the value for the obtained mimicked profile in the same radius . one also presents the deviation of the final total mass of the dark matter component from the quantity = |m_mim - m|m , [ mass_deviation ] where @xmath120 is the amount of dark matter . the results of the numerical integration are shown in fig . ( [ densities ] ) and the resulting integrated mass radial profile , @xmath121 , in fig . ( [ masses ] ) , for the values of @xmath122 , @xmath123 , @xmath124 , @xmath125 , and @xmath126 , with a best fit attained for @xmath127 ; @xmath128 is kept constant , as its value does not change the mimicked dark matter density profile ( as discussed below ) . the best fit is given by the minimization of the relative difference between the obtained mimicked total mass and the expected one from the hydrostatic estimative , eqs . ( [ dynamical_mass ] ) and ( [ mass_profile ] ) , with values shown in table ( [ mass_difference ] ) . the relative mass difference is evaluated at the virial radius of the abell cluster a586 , @xmath129 @xcite . as one can see , these results show that , in the outer regions of the cluster , the density decreases with the value of @xmath82 along with a decrease in the total mass for the cluster . the best fit @xmath127 is the one that yields a mimicked profile , which minimizes the mass difference to only @xmath130 at @xmath131 , the virial radius for the density profile derived from eqs . ( [ visible_profile ] ) and ( [ mass_profile ] ) . in fact , in fig . ( [ masses ] ) , one can see that the values in the range , @xmath132 , lead to a dynamical mass larger than the upper limit set by the model , @xmath133 . notice that this best fit @xmath134 deviates from the expected value @xmath135 , computed in eq . ( [ expected ] ) . as the curves for @xmath136 and @xmath137 show , this stems from the fact that the mimicked dark matter density arising from the expected value is unable to follow the observed profile . indeed , while @xmath135 yields the same slope as the observed total mass density , it leads to a right - shifted curve , giving rise to a much greater total mass for the cluster . , where the dashing length increases with @xmath82 . the best fit is obtained for @xmath127 ( dark dashed ) . the horizontal dotted line corresponds to @xmath138 . ] ( polytropic case ) @xcite . ] .virial radius and respective mass for @xmath39 . mass differences for different values of the @xmath82 evaluated at @xmath139 . [ cols="^,^,^,^,^",options="header " , ] , after the integration of the differential equation , for different values of @xmath82 . ] in fig . ( [ extra_forces ] ) one presents the plot of the additional radial velocity @xmath140 arising from non - geodesic motion as a function of distance to the cluster s center , for the considered range of values for @xmath82 and a fixed @xmath141 . notice that , despite the fact that the observable quantity is the velocity dispersion @xmath142 , estimated from the velocities of individual clusters , these curves enable one to see the relevance of the extra force that appears due to the presence of the non - minimal coupling . since the scalar curvature ( _ i.e. _ the total density ) is decreasing , one concludes that the related force counteracts the gravitational attraction , as discussed in sec . [ subsectionnon ] . thus , stability requires that this additional velocity be much smaller than the newtonian radial velocity profile : for a fixed value of @xmath143 , this is attained for higher values of @xmath82 . in particular , one concludes that a power - law coupling with @xmath127 can effectively mimic the dark matter component of the abell cluster a586 , with a negligible effect on the velocity profile @xmath144 . as already discussed in sec . [ nongeodesicalmotion ] , one expects a decrease of the effect of the extra force for increasing exponent @xmath82 ( for fixed @xmath143 ) . indeed , as fig . ( [ couplings ] ) depicts , the considered range for @xmath82 always leads to a perturbative @xmath42 , for a fixed @xmath145 . however , values of the exponent in the range @xmath146 , which imply @xmath147 , do not enable a sufficiently strong suppression of the extra force : although the additional ( outward ) radial velocity is indeed smaller than @xmath148 , it may still be above the newtonian counterpart the converse would require one to compute the condition for @xmath137 , in order to obtain a bound on @xmath149 . + + using and the relevant values obtained from ref . @xcite , one has @xmath150 and @xmath151 . inserting this into , together with @xmath152 and @xmath153 , one finds that the extra force is smaller than the newtonian force when @xmath154 , which is equivalent to the constraint , @xmath155 , and should be compared with the aforementioned lower bound @xmath156 , obtained from fig . [ extra_forces ] . the lack of sensitivity of the mimicked dark matter density profile on the value of @xmath157 may be used to establish a more stringent upper bound on this parameter . to do so , one fixes the best fit value @xmath134 and searches for the value of @xmath143 that fulfills the condition @xmath158 , which is equivalent to the constraint @xmath159 . one now extends the dark matter mimicking mechanism to a large sample of galaxy clusters , aiming to obtain an encompassing description of their `` dark matter '' content . for this purpose , one uses a sample of nearby relaxed clusters derived from the high quality data of _ chandra _ @xcite . as already seen , in the case of abell cluster a586 , one resorts to the scaling @xmath160 to estimate the exponent @xmath82 of the power - law non - minimal coupling . however , since the density profiles are now more evolved , alternatively one considers the behaviour of the inner and outer regions , that is one resorts to two different exponents , @xmath161 and @xmath162 , respectively . indeed , the dark matter density profile follows a nfw profile given by _ t~(r r_s ) ^-1(1+r r_s ) ^-2 , such that for small radii , @xmath163 , and for outer regions , @xmath164 ; the gas density profile follows a generalized nfw model @xcite , ^2_g~n_0 ^ 2(r / r_c)^-(1+r^2/r_c^2)^3-/2 , [ generalized_beta ] for which , @xmath165 , in inner regions , and @xmath166 for outer regions . thus , for small radii one has , from eq . ( [ n - estimative ] ) , = 1/2n_in=1 - 2 , [ n1 ] and , for large radii , = 33n_out=1- . [ n2 ] the results of eqs . ( [ n1 ] ) and ( [ n2 ] ) for each cluster are shown in table [ clusters_values ] using the reported values for the @xmath167 and @xmath25 parameters @xcite . since @xmath82 is a parameter of the coupling function eq . ( [ powerlawcoupling ] ) , it should be universal : thus , one performs a simultaneous fit of the `` dark matter '' component of all clusters , with a fixed @xmath82 . the characteristic length scale @xmath143 is varied individually for each cluster , allowing for the analysis of the perturbative regime of the non - minimal coupling eq . ( [ powerlawcoupling ] ) and the extra force eq . ( [ extra - force ] ) , discussed below following the approach detailed for the abell a586 cluster in the previous section . numerically , one finds that the best fit occurs for @xmath168 ; the initial conditions ( _ i.e. _ the initial value for @xmath169 and its derivative ) are varied within one order of magnitude , reflecting the uncertainty on the dark matter distributions reported in ref . the resulting dark matter profiles yield the dashed curves presented in figs . ( [ mimic1 ] ) and ( [ mimic2 ] ) : these minimize the difference between the observed and mimicked dark matter components , leading to small values of the estimator @xmath170 , eq . ( [ sigma2 ] ) , and mass difference @xmath171 , eq . ( [ mass_deviation ] ) , as can be seen in table [ clusters_values ] . as in the case of abell cluster a586 , one finds that the mimicked dark matter profile is almost independent of the characteristic length scale @xmath172 , as long as the non - minimal coupling eq . ( [ powerlawcoupling ] ) is perturbative , @xmath173 . this leads to the bound @xmath174 , which is approximately independent of the cluster . as before , a more stringent bound , @xmath175 , is obtained from the requirement that the extra force eq . ( [ extra - force ] ) is smaller than the newtonian force . the independence of the obtained mimicked dark matter distributions on the value of @xmath143 again emerges because a different value for the latter leads to a change in the boundary conditions of eq . ( [ rhovar ] ) , which implies a change of the dimensionless solution @xmath169 ; this is compensated by the presence of @xmath143 in the relation between the latter and the mimicked dark matter density , eq . ( [ var1 ] ) . + + + notice that the fitting of the abell a586 cluster dark matter profile using the universal exponent @xmath176 ( final graph of fig . [ mimic2 ] ) leads to the values @xmath177 , and @xmath178 naturally higher than those obtained from the @xmath127 best fit of this cluster alone . there is a fundamental qualitative difference between the density profiles derived for the a586 cluster and the set of 13 chandra clusters : the later do not exhibit a transition between an inner region with approximately constant density and an outer region with decreasing density of approximately constant slope ( as embodied in the @xmath179-model ) . this drives the ensuing distinction between the fits for the a586 cluster obtained in this study : indeed , as fig . [ densities ] shows , the best fit @xmath127 obtained by considering only this cluster yielded an initial plateau for the mimicked dark matter density . this is to be expected , as most of the mass comes from the integration of the density in this higher , approximately constant region thus leading to a reduced mass difference @xmath180 . on the contrary , the mimicked dark matter profile of this cluster obtained in the global fit ( final graph of fig . [ mimic2 ] ) does not have an approximately flat density for the inner region , but compensates by more closely following the observed slope of the outer regime hence the proximity between the obtained best fit parameter @xmath176 and the value obtained from the scaling law @xmath76 , @xmath181 . this occurs because the minimization procedure of the global fit is mostly driven by the remaining thirteen chandra clusters : as it turns out , the best fit for this set ( excluding the a586 cluster ) is @xmath182 ( not shown in this study ) ; a balanced fitting solution with all fourteen clusters is thus obtained by relaxing the constraint of a flattened inner density region for the mimicked dark matter profile of the a586 cluster . the non - minimal coupling considered here may be regarded as an approximation to a more evolved form for @xmath42 , which would encompass couplings considered in previous studies , which perhaps hints at a laurent series , f_2(r ) = _ n = - ^ ( r r_n)^n . [ laurent ] each term would be valid in a particular regime : early _ vs. _ late time , central _ vs. _ long range , galactic _ vs. _ cluster size , _ etc_. in particular , this series should also include the aforementioned @xmath183 and @xmath184 terms , relevant to mimic dark matter at the galactic level , as well as a linear coupling @xmath185 considered in the context of preheating and inflationary dynamics ; the corresponding parameters are r_-1 & & ( r_-1)^-1/2 = 21.5 gpc , + r_-1/3 & & ( r_-1/3)^-1/2 = 1.69 10 ^ 6 gpc , + r_1 & & r_1 ^ -1/2 = 4.8 10 ^ -29 m , with @xmath186 @xcite . since the relevant coupling terms at galactic scales have negative exponents , it is trivial to conclude that their effect is negligible during inflation , when the curvature is much higher . conversely , the effect of the terms involved in preheating and inflation are irrelevant at astrophysical or cosmological scales at present . similarly , the cosmological effect of the best fit coupling for the considered sample of clusters ( with @xmath176 ) may be disregarded , as it is of order ( r_cosmor_n)^n ~(r_nr_h)^2n ( 10 ^ -13 kpc4.2 gpc)^0.2 ~10 ^ -4 . where @xmath187 is the hubble radius . thus , one is left with the possible conflict between astrophysical relevant terms . if one considers three power - law couplings with @xmath183 , @xmath188 , and @xmath189 , the resulting dark matter profile will not be a sum of the individual contributions , due to the non - linearity of . even if one disregards the resulting interaction between terms ( a reasonable approximation , as shown in ref . @xcite ) , the exponents simply yield the slope for each derived dark matter contribution , not for their relative values . indeed , when numerically integrating , one imposes boundary conditions on the overall dark matter density ( _ i.e. _ the dimensionless function @xmath169 ) the relative contributions of each dark matter component are then determined dynamically . given the scaling relation , a larger @xmath190 implies a steeper dark matter component @xmath191 . resorting to the @xmath192 behaviour of the visible mass density eq . ( [ generalized_beta ] ) and using the average value @xmath193 , this translates into n=-1 : _ dm ~r^-0.98 & , & n=-1 3 : _ dm ~r^-1.5 , + n=0.2 & : & _ dm ~r^-2.2 . thus , the @xmath168 merely gives rise to a quickly decaying component : the @xmath183 and @xmath184 contributions eventually dominate at large distances , even if the former dominates at short ones . in order for the contributions from the @xmath194 and @xmath188 to be neglected , it suffices that they are subdominant at cluster scales , but dominant at the long range of galactic scales . one does not pursue an explicit computation of the effect of these three power - law terms here , given that the main scope was to describe clusters . nevertheless , it is not difficult to see that the difference between the galactic and cluster typical densities do allow for a consistent disentangling of the effects of these two sets of terms . in this work , the possibility of mimicking the dark matter component of a galaxy cluster in the context of a model , , where the scalar curvature is non - minimally coupled with matter is discussed . for that , one first assesses the viability of the model using the relaxed abell a586 cluster , given the availability of quality data for most of its relevant dynamical parameters . using a power - law non - minimal coupling @xmath195 , one obtains the best fit for @xmath127 , with a strong constraint on the typical length scale @xmath196 arising from the requirement that the induced extra - force is smaller than the newtonian one . a weaker constraint @xmath197 arises from the requirement that the coupling is perturbative , @xmath173 . the obtained fit does not correspond to the analytical value for @xmath82 computed from the scaling relation between dark and visible matter , as the latter would give rise to a mimicked dark matter density that does not follow the observational curve ( although its outer behaviour exhibits the same slope ) . instead , a shift from this analytical value is enforced , whereas the mimicked density is slightly steeper than the observed profile , but with a negligible difference in total mass . furthermore , the exponent @xmath82 is large enough to keep the non - minimal coupling perturbative , with @xmath198 . this suppresses the destabilizing effect of the outward extra - force arising from the non - conservation of the energy - momentum tensor , as previously found in the context of galaxy rotation curves ( where two negative exponents @xmath183 and @xmath184 were shown to correspond to the best fit for the rotation curves ) @xcite . subsequently , this dark matter mimicking mechanism was successfully extended to a large sample of galaxy clusters using the available profiles for the visible matter density and temperature obtained from modifications of the usual @xmath25-model , eq . ( [ visible_profile ] ) , and the polytropic temperature profile , eq . ( [ temperature_profile ] ) @xcite . for this purpose , one has used the same power - law coupling function , obtaining an exponent @xmath176 from a simultaneous fit to all considered clusters . as in the detailed case of the abell a586 cluster , the fitting procedure is also independent of the length scale @xmath143 , as long as the coupling is perturbative , @xmath173 . by the same token , a more stringent upper bound of @xmath199 is obtained from the requirement that the extra force due to non - geodesic motion is much smaller than the newtonian force . the fitting procedure depends on the underlying model , but also on several other assumptions namely , sphericity , virialization , and thermal equilibrium . clearly , deviations from any of these hypotheses will introduce an undetermined bias into the best fit scenario : the choice of the abell 586 cluster as a initial test case was motivated by the wish to minimize these sources of error , as this cluster is very well characterized and has a good spherical morphology @xcite . in galaxy clusters with more evolved structures , one may observe deviations due to the presence of non - sphericity and non - thermal pressure support , and the density profile may be much rougher than in well - behaved clusters . however , an adequate description of the mimicked dark matter content was obtained for clusters with more evolved internal structures , namely rxj1159 + 5531 , mkw 4 and usgc s152 , thus supporting the universality and robustness of the proposed mimicking mechanism . studies addressing different scenarios have resorted to power - law couplings with different exponents than the best fit value @xmath168 here reported which may be regarded as another term in a putative series expansion of a more general non - minimal coupling . the specific form of the coupling function can only be obtained from its consideration in different scenarios , in order to reconstitute its full functional dependence on the ricci scalar curvature . thus , the relevance of each term in the several scenarios considered in previous works was also discussed . by the same reasoning , yet unprobed terms of this series may be charted through the study of other phenomena and environments ( ranging from astrophysical to cosmological scales ) , where distinct curvatures and densities are at play . d. n. spergel _ et al . _ [ wmap collaboration ] , , 377 ( 2007 ) [ http://arxiv.org/abs/astro-ph/0603449[arxiv:astro-ph/0603449 ] ] ; g. hinshaw _ et al . _ , , 225 ( 2009 ) [ http://arxiv.org/abs/0803.0732[arxiv:0803.0732 ] ] ; e. komatsu _ et al . _ , , 330 ( 2009 ) [ http://arxiv.org/abs/0803.0547[arxiv:0708.2750 ] ] ; , 18 ( 2011 ) [ http://arxiv.org/abs/1001.4538[arxiv:1001.4538 ] ] . m. tegmark _ et al . _ [ sdss collaboration ] , , 123507 ( 2006 ) [ http://arxiv.org/abs/astro-ph/0608632[arxiv:astro-ph/0608632 ] ] ; m. sako _ et al . _ , , 348 ( 2008 ) [ http://arxiv.org/abs/0708.2750[arxiv:0708.2750 ] ] . m. cirelli , to appear in the proceedings of the 25th international symposium conference [ http://arxiv.org/abs/1202.1454[arxiv:1202.1454 ] ] ; g. bertone , particle dark matter ( cambridge university press , cambridge , uk , 2010 ) . o. bertolami , f. gil pedro and m. le delliou , 165 ( 2007 ) [ http://arxiv.org/abs/astro-ph/0703462[arxiv:astro-ph/0703462 ] ] ; , 2839 ( 2009 ) [ http://arxiv.org/abs/0705.3118[arxiv:0705.3118 ] ] ; ( 2011 ) [ http://arxiv.org/abs/arxiv:1105.3033[arxiv:1105.3033 ] ] .
low - energy nonperturbative effects in quantum chromodynamics ( qcd ) , especially at nonzero densities , can only be studied by approximate methods within the framework of various effective models . it is well known that light meson physics is described by four - fermion models , such as the nambu jona - lasinio ( njl ) model , which was successfully used to deal with dynamical chiral symmetry breaking ( d@xmath0sb ) both in the vacuum and in hot / dense baryonic matter ( see , e.g. @xcite , as well as the reviews @xcite and references therein ) . recently , much attention has been paid to the effects of diquark condensation and color superconductivity ( csc ) . the first studies of the gap equations and the ginzburg - landau free energy for a system of relativistic fermions led to the conclusion that superconductive and color superconductive states may arise in baryonic matter @xcite ( see also the recent papers @xcite ) . another interesting phenomenon , the condensation of charged pions , which may appear in dense hadronic matter due to an asymmetry of its isospin composition , has been investigated in the framework of qcd - like effective models , including the njl model , as well @xcite . note that all these phenomena might be inherent to physics of compact stars , where rather strong magnetic as well as gravitational fields are present . therefore , investigations of the influence of an external ( chromo-)magnetic field on the properties of the d@xmath0sb phase transition @xcite , color superconductivity @xcite and pion condensation @xcite effects are quite appropriate . in particular , it was shown in @xcite that external fields significantly change the properties of the chiral and csc phase transitions . in several papers , in the framework of the njl model , the influence of a gravitational field on the d@xmath0sb due to the creation of a finite quark condensate @xmath1 has been investigated at zero values of temperature and chemical potential @xcite . the study of the combined influence of curvature and temperature has been performed in @xcite . recently , the dynamical chiral symmetry breaking and its restoration for a uniformly accelerated observer due to the thermalization effect of acceleration was studied in @xcite at zero chemical potential . further investigations of the influence of the unruh temperature on the phase transitions in dense quark matter with a finite chemical potential , and especially on the restoration of the broken color symmetry in csc were made in @xcite . one of the widely used methods of accounting for gravitation is based on the expansion of the fermion propagator in powers of small curvature @xmath2 @xcite . for instance , in @xcite , the three - dimensional gross - neveu model in a spacetime with a weakly curved two - dimensional surface was investigated , using an effective potential at finite curvature and nonzero chemical potential . in paper @xcite , this weak curvature expansion was used in considering the d@xmath0sb at non - vanishing temperature and chemical potential . it should , however , be mentioned that near the phase transition point , one can not consider the critical curvature @xmath3 to be small and therefore the weak curvature expansion method can not be applied . hence , in the region near the critical regime different nonperturbative methods or exact solutions with finite values of @xmath2 should be used . this kind of solution with consideration for the chemical potential and temperature in the gravitational background of a static einstein universe has been considered in @xcite . there it was demonstrated that chiral symmetry is restored at large values of the space curvature . analogous studies of diquark condensation and the related color symmetry breaking under the influence of a gravitational field have been performed recently in the model of a static einstein universe in @xcite . recall that this model is widely discussed in literature either as a solution of the einstein equations with a given cosmological constant and a nonvanishing energy - momentum tensor of an ideal fluid as a source , or as an initial state in inflationary cosmology with a scalar field , and the cosmological constant as its vacuum energy ( see , for instance , @xcite ) . moreover , the einstein universe and other suitably generalized compact curved spacetimes were extensively employed for studying the phenomenon of bose - einstein condensation ( see , e.g. @xcite and references therein ) . as is well known , one of the possible models for the expanding universe is the closed friedmann model . since the formation of quark condensates is expected to take place considerably faster than the expansion rate of the universe , its radius can be considered in our calculations as constant . in this sense , the chosen model of a static einstein universe can be considered as a simple cosmological model for a space with positive curvature . on the other hand , the same form of the metric does also hold for the interior of a ( collapsing ) spherically symmetric star , thereby admitting also a non - cosmological interpretation of the chosen model . as was mentioned above , in quark matter there may arise another remarkable phenomenon , pion condensation . up to now little information about the properties of pion condensation in the presence of external fields has been obtained . in particular , there arises the interesting question , what influence gravitational fields can produce on pion condensation . the main aim of the present paper is just to give a detailed investigation of this issue . in order to be able to consider the effects of gravitation on pion condensation in a rigorous nonperturbative way , we will investigate here the phase structure of isotopically asymmetric quark matter in the framework of an extended njl model with dynamical breaking of chiral and flavor symmetries in the static einstein universe . in particular , our calculations are performed in the most simple case , restricting ourselves , for simplicity , to the flavor @xmath4 group . by using the mean field approximation , we will then derive an analytical expression for the thermodynamic potential of the njl model that will enable us to calculate the chiral and pion condensates of quarks moving in a space with constant positive curvature . on this basis , we study , in particular , the phase portraits of the model . our paper is organized as follows . section ii , iii contain the nambu - jona - lasinio model in curved spacetime and the general formalism for the derivation of the thermodynamical potential . numerical calculations and conclusions are given in sections iv , v. finally , some estimates of the role of quantum fluctuations in the case of a closed universe are given in the appendix . suppose that dense , isotopically asymmetric quark matter ( in this case the densities of @xmath5 and @xmath6 quarks are different ) in curved spacetime is described by an extended njl model with the following action : @xmath7 and the lagrangian @xmath8q+ g\big [ ( \bar qq)^2 + ( \bar qi\gamma^5\vec\tau q)^2\big ] , \label{2}\end{aligned}\ ] ] where the quark field @xmath9 is a flavor doublet ( @xmath10 or @xmath11 ) and color triplet ( @xmath12 ) as well as a four - component dirac spinor ( the summation in ( [ 2 ] ) over flavor , color and spinor indices is implied ) ; @xmath13 ( @xmath14 ) are pauli matrices . in 4-dimensional curved spacetime with signature @xmath15 , the line element is written as @xmath16 the gamma - matrices @xmath17 , the metric @xmath18 and the vierbein @xmath19 , as well as the definitions of the spinor covariant derivative @xmath20 and spin connection @xmath21 are given by the following relations @xcite : @xmath22 , \nonumber \\ & & \omega^{\hat{a}\hat{b}}_{\mu}=\frac{1}{2}e^{\hat{a}\lambda } e^{\hat{b}\rho}[c_{\lambda\rho\mu}-c_{\rho\lambda\mu}-c_{\mu\lambda \rho } ] , \quad c_{\lambda\rho\mu}= e^{\hat{a}}_{\lambda}\partial_{[\rho}e_{\mu]\hat{a}}. \label{3_3}\end{aligned}\ ] ] here , the index @xmath23 refers to the flat tangent space defined by the vierbein at spacetime point @xmath24 , and the @xmath25 are the usual dirac gamma - matrices of minkowski spacetime . moreover @xmath26 , is defined , as usual ( see , e.g. , @xcite ) , i.e. to be the same as in flat spacetime and thus independent of spacetime variables . in order to describe the quark composition of matter we introduced @xmath27 in ( [ 2 ] ) as the quark number chemical potential . since the generator @xmath28 of the third component of isospin is equal to @xmath29 , the quantity @xmath30 in ( [ 2 ] ) is half the isospin chemical potential , @xmath31 . if the bare quark mass @xmath32 is equal to zero , then at @xmath33 , apart from the trivial color su(@xmath34 ) symmetry , the lagrangian ( [ 2 ] ) is invariant under the chiral group @xmath35 transformations . however , at @xmath36 this symmetry is reduced to the subgroup @xmath37 ( here and above the subscripts @xmath38 imply that the corresponding group acts only on the left , right handed spinors , respectively ) . it is convenient to present this symmetry as @xmath39 , where @xmath40 and @xmath41 are the isospin and axial isospin subgroup of the chiral @xmath35 group . quarks are tranformed under these subgroups in the following way @xmath42 where @xmath43 , @xmath44 are independent group parameters . at nonzero bare quark mass , @xmath45 , and nonzero isotopic chemical potential , i.e. @xmath46 , the lagrangian ( [ 2 ] ) is still invariant under the isospin subgroup @xmath40 , but invariance with respect to @xmath41 does no longer hold in this case . the linearized version of lagrangian ( [ 2 ] ) , that contains collective bosonic fields @xmath47 and @xmath48 @xmath49 , has the following form @xmath50q -\frac{1}{4g}\big [ \sigma\sigma+\pi_k\pi_k\big ] . \label{3}\end{aligned}\ ] ] from the lagrangian ( [ 3 ] ) , one can find the equations of motion for the bosonic fields , @xmath51 it is clear from these relations that under isospin @xmath40 and axial isospin @xmath52 transformations ( [ 2010 ] ) the bosonic fields ( [ 200 ] ) are changed in the following way : @xmath53 in the fermion one - loop ( mean field ) approximation , the effective action for the boson fields is expressed through the path integral over quark fields : @xmath54[dq]\exp\bigl(i\int d^4 x\sqrt{-g } ~\tilde { \mathcal l}\,\bigr),\ ] ] where @xmath55+\tilde { \cal s}_{\rm { eff } } , \label{4}\ ] ] @xmath56 is a normalization constant . the quark contribution to the effective action , i.e. the term @xmath57 in ( [ 4 ] ) , is given by : @xmath58[dq]\exp\bigl(i\int d^4x\sqrt{-g } ~\bar q{\cal d}q\,\bigr)=\mbox { det}~{\cal d}. \label{5}\ ] ] in ( [ 5 ] ) , we have used the following notation @xmath59 clearly , @xmath60 is an operator in the coordinate , spinor and flavor spaces . apart from this , it is proportional to the unit operator @xmath61 in the color space . thus , it can be presented in the flavor@xmath62color space in the following matrix form : @xmath63 where operators @xmath64 act in the coordinate and spinor spaces only , and @xmath65 due to the trivial color structure , it follows from ( [ 7 ] ) that @xmath66 now , suppose that @xmath67 and @xmath68 are quantities independent of coordinates .. if the fields @xmath69 do not depend on coordinates , then the effective action is a function of the invariants @xmath70 and @xmath71 only ( this fact is due to the symmetry of the model with respect to the transformations ( [ 201 ] ) ) . therefore , without loss of generality one can put @xmath67 . ] in what follows , we shall denote the pion condensate @xmath72 by @xmath73 for convenience . then , using the general formula @xmath74=\det [ \bar aa-\bar ab\bar a^{-1}\bar b ] , \label{10}\end{aligned}\ ] ] we obtain @xmath75 - 3i\ln { \rm det}d \equiv -\omega(\sigma,\delta;\mu,\delta\mu)\int d^4x\sqrt{-g } , \label{11}\ ] ] where we have introduced the thermodynamic potential @xmath76 of the system at zero temperature and @xmath77[i\gamma^\nu \nabla_\nu + ( \mu-\delta\mu)\gamma^0 -\sigma - m ] \big\}. \label{12}\ ] ] the line element in the static einstein universe is defined by the following relation : @xmath78 where @xmath79 is the radius of the einstein universe ( this quantity is related to the scalar curvature , @xmath80 ) ; @xmath81 , @xmath82 , @xmath83 , @xmath84 . clearly , @xmath85 in this case anticommutes with all @xmath86 and commutes with all @xmath87 , where @xmath88 and @xmath89 . moreover , @xmath90 . so , starting from ( [ 12 ] ) , we have ( @xmath88 ; @xmath91 ) : @xmath92 + \delta\mu^2\big\}. \label{18}\end{aligned}\ ] ] let @xmath93 , @xmath94 . it is evident that @xmath95 . hence , we have from ( [ 18 ] ) @xmath96 evidently , @xmath97 is an operator in the hilbert space of functions , depending on the space coordinates @xmath98 . as it is well - known ( see , e.g. @xcite ) , each of the eigenvalues @xmath99 of this operator is @xmath100-fold degenerate , @xmath101 thus , one can write @xmath102 where the eigenfunctions @xmath103 ( @xmath104 ) of the operator @xmath97 are also known ( see , e.g. , @xcite ) , and @xmath105 ( see ( [ 17 ] ) ) . now , let us choose in the hilbert space of functions a basis of the form @xmath106 @xmath107 , where @xmath108 . since each element @xmath109 of this basis is an eigenfunction both of @xmath110 and @xmath97 , one can easily conclude that the operator @xmath111 from ( [ 19 ] ) is diagonal in this basis , i.e. each @xmath112 is an eigenfunction of @xmath113 . the corresponding eigenvalues @xmath114 of @xmath113 have the following form : @xmath115 it is clear from ( [ 192 ] ) that eigenvalues @xmath116 of the operator @xmath113 do not depend on the quantum number @xmath117 , being @xmath100-fold degenerate . taking into account in ( [ 11 ] ) the relations ( [ 19 ] ) and det@xmath118tr@xmath119 , as well as the results of appendix [ apb ] , where the quantity tr@xmath120 is calculated ( see ( [ b11 ] ) ) , one finds @xmath121=-3i{\rm tr}\ln\hat o=-3i { \cal t}\sum_{l\eta}\int\frac{dp_0}{2\pi}d_l\ln\big [ \delta^2-(p_0+\mu)^2+(\eta e_l+\delta\mu)^2\big ] . \label{193}\ ] ] here @xmath122 , where @xmath123 stands for an infinite time interval and @xmath124 is the space volume of the einstein universe ( the last relations are due to the fact that @xmath125 depends only on @xmath98 ) . now , using the definition ( [ 11 ] ) of the thermodynamic potential ( tdp ) and summing in ( [ 193 ] ) over @xmath126 , we have for the zero temperature case @xmath127 + \ln\big [ \delta^2-(p_0+\mu)^2+(e_l+\delta\mu)^2\big ] \big\}. \label{194}\ ] ] to find the tdp @xmath128 in the case of nonzero temperature @xmath129 , one should use the imaginary time technique , where , after summation over matsubara frequencies ( see , e.g. , @xcite ) , the following expression can be found @xmath130 + \ln\big [ 1+e^{-\beta ( e_l^{(+)}-\mu)}\big ] \big\ } \nonumber\\&-&\frac{3t}{{\cal v } } \sum_{l=0}^\infty d_l\big\{\ln\big [ 1+e^{-\beta ( e_l^{(-)}+\mu)}\big ] + \ln\big [ 1+e^{-\beta ( e_l^{(-)}-\mu)}\big ] \big\ } \label{195}\end{aligned}\ ] ] with @xmath131 and @xmath132 . it is clear that @xmath128 is an even function with respect to each of the transformations @xmath133 or @xmath134 . thus , one can deal only with non - negative values of the chemical potentials , @xmath135 , @xmath136 . from this moment on , we will consider only the case of nonzero isospin chemical potential @xmath137 , whereas the baryon chemical potential is set equal to zero , @xmath138 , since its presence is not of principle importance for us . so , at @xmath138 two particular cases can be investigated on the basis of the tdp ( [ 195 ] ) . first , let us choose @xmath139 and @xmath138 , but @xmath46 . then we obtain from ( [ 195 ] ) the expression : @xmath140 secondly , at @xmath141 , @xmath142 , @xmath46 we obtain : @xmath143 + \ln\big [ 1+e^{-\beta e_l^{(-)}}\big ] \big\}. \label{198}\end{aligned}\ ] ] next , let us consider the limit of zero curvature or infinitely large radius of the universe . it is clear that the metric ( [ 17 ] ) never coincides with that of the flat minkowsky spacetime because these two spacetimes have different topologies . however , in the limit @xmath144 and @xmath145 one can obtain from ( [ 198 ] ) the usual expression for the tdp in flat spacetime ( see , e.g. , @xcite ) by the following substitution : @xmath146 in this case , the tdp looks as follows : @xmath147 + 2 t \ln[1+e^{-\beta e_k^{(-)}}]\right\ } , \label{199}\end{aligned}\ ] ] where @xmath148 . it should be noted that one more particular case , when @xmath141 , @xmath149 , but @xmath33 , can easily be reduced to the investigation of the one - flavored njl model at @xmath141 , @xmath149 in the einstein universe @xcite , and hence we shall not consider it here . first of all , in order to normalize the tdp , we should subtract a corresponding constant from it , such that @xmath150 . the thermodynamic potential , normalized in this way , is still divergent at large @xmath151 , and hence , we should introduce a ( soft ) cutoff in the summation over @xmath151 by means of the multiplier @xmath152 @xcite , where @xmath153 is the cutoff parameter . can be specified according to the experimental results . however , in the case of a curved spacetime , in order to fix the cutoff @xmath154 , we need new theoretical / experimental inputs for chiral qcd in ( strong ) gravitational background fields , concerning , for instance , an effective gluon mass , known values of the quark condensate or even experimentally measured characteristics of a pion . due to the lack of experimental knowledge , our aim is here to perform only a qualitative study of gravitational effects on the quark and pion condensates by investigating the respective phase portraits of the njl model . for this reason it is convenient to scale the thermodynamic potential and all relevant quantities like condensates , curvature , chemical potential and temperature by the cutoff @xmath153 . ] for convenience , we shall multiply all dimensional quantities that enter the thermodynamic potential by the corresponding power of @xmath153 to make them dimensionless , i.e. , @xmath155 , and denote them using the same letters as before : @xmath156 . then the regularized thermodynamic potential is written as @xmath157 + \ln\big [ 1+e^{-\beta e_l^{(-)}}\big ] \big\}. \label{198_1}\end{aligned}\ ] ] in the following section , we shall perform a numerical calculation of the points of the global minimum of the finite regularized thermodynamic potential @xmath158 ( they should of course coincide with the minima of the potential @xmath159 ) , and with the use of them , consider quark matter phase transitions in the gravitational field of the einstein universe . in this section , on the basis of the thermodynamic potentials ( [ 197])-([198 ] ) , we will study numerically phase transitions in quark matter and consider only the case of nonzero isospin chemical potential @xmath137 , whereas the baryon chemical potential is set equal to zero , @xmath138 . in order to obtain the values of condensates , one should find the global minimum point ( gmp ) of the thermodynamic potentials over the variables @xmath160 , @xmath73 from the corresponding gap equations @xmath161 formally , there are four different expressions for the gmp : i ) @xmath162 , ii ) @xmath163 , iii ) @xmath164 , iv ) @xmath165 . the first two gmps correspond to the isotopically invariant phases of the model , whereas the gmps of the form iii ) and iv ) correspond to the phases , in which the ground state is no more @xmath40 invariant . in these phases the pion condensation phenomenon occurs . for simplicity , we take for the numerical calculations of the gmp the value of the coupling constant @xmath166 ( in our dimensionless choice of parameters ) . let us first consider phase transitions at zero temperature , @xmath139 and choose the current quark mass to be equal to zero , @xmath167 . the thermodynamic potential in this case is described by formula ( [ 197 ] ) . the detailed investigation of the gmp properties vs external parameters @xmath2 and @xmath30 results in the phase portrait shown in fig . [ pp1 ] . for the points of symmetric phase 1 , the gmp is at @xmath168 and @xmath169 . in the phase 3 , the minimum is at @xmath168 and @xmath170 , and this indicates that the isospin @xmath40 symmetry of the model is dynamically broken in this phase . we note that at the same time the chiral @xmath52 symmetry remains unbroken in this phase , since in the gmp we have @xmath168 , as it should be in the case of zero current quark mass , when @xmath170 . @xmath171 as one can see from fig . [ pp1 ] , the critical curve , which separates the phases 1 and 3 , has an oscillating character . this phenomenon is explained by the discreteness of the fermion energy levels ( [ 190 ] ) in compact space . moreover , as it is clearly seen from the curves in [ cond23 ] , the pion condensate @xmath73 vs @xmath2 also oscillates in the phase 3 . this effect resembles the van alphen - de haas oscillations of different physical quantities in the magnetic field @xmath172 , where fermion levels are also discrete ( the landau levels ) @xcite ( see also @xcite , where a similar influence of a magnetic field on the oscillation behavior of the compton scattering and photoproduction cross - sections was demonstrated ) . indeed , the corresponding magnetic oscillations of the critical curve in the @xmath173-@xmath172 phase portrait of dense cold quark matter with four - fermion interactions were found in papers @xcite . there , the existence of the standard van alphen - de haas magnetic oscillations of some thermodynamical quantities , including magnetization , pressure and particle density of cold dense quark matter was also demonstrated . the behavior of the pion condensate @xmath73 as a function of @xmath2 at fixed @xmath30 and as a function of @xmath30 at fixed @xmath2 is shown in fig . [ cond23 ] ( left and right pictures respectively ) . the phase portrait at finite current quark mass , @xmath45 , is depicted in fig . [ pp2 ] . in phase 2 the chiral symmetry is now broken due to a finite value of the current quark mass , and the global minimum of tdp is at @xmath174 and @xmath175 . in the mixed phase 4 both condensates are nonzero , i.e. @xmath174 and @xmath170 . the behavior of the chiral condensate @xmath160 in the case of finite quark mass , @xmath45 , is shown in fig . [ cond5 ] as a function of @xmath2 at @xmath176 ( left picture ) and as a function of @xmath30 at @xmath177 ( right picture ) . @xmath178 one can see oscillations of @xmath160 on both pictures , although they are rather strong in the left picture for the dependence on @xmath2 , while they are weakly seen in the high @xmath179 tail of the curve in the right picture . one should note that the right picture resembles , except for these oscillations , the corresponding curve in @xcite ( fig . 1 ) for the flat case . using formula ( [ 198_1 ] ) for the thermodynamic potential , we can also study the influence of finite temperature on phase transitions . the phase portrait at @xmath180 and zero current quark mass , @xmath181 , is shown in fig . [ pp3 ] in terms of @xmath182 . it is seen from this figure that growing temperature leads to a smoothing of oscillations of the phase curve . for comparision , in fig . [ pp4 ] , the phase portraits in the @xmath183 and @xmath182 planes ( the latter now at another value of temperature @xmath184 ) are depicted . first of all , it is clear from fig . [ pp4 ] that the isospin symmetry is restored due to the vanishing of the pion condensate both at high temperature and high curvature . the similarity between these two plots leads to the conclusion that curvature and temperature effects play a similar role in the restoration of isospin symmetry . @xmath185 the phase portrait in fig . [ pp5 ] for @xmath186 corresponds to the case of flat minkowski spacetime . it looks very similar to the one obtained , for instance , in @xcite ( see upper panel of their corresponding fig . let us compare the phase portraits at zero ( fig . [ pp5 ] ) and finite curvature ( fig . [ pp4 ] , left picture ) . in the first case the pion condensation appears at arbitrary small nonzero values of the isospin chemical potential , while in the second case the isospin symmetry becomes dynamically broken only at some finite value of the chemical potential @xmath30 . this phenomenon may be explained by the existence of a gap in the quark spectrum ( [ 190 ] ) , which is proportional to the inverse radius of the einstein universe . in this case the effect of curvature is similar to the effect of a nonzero current quark mass ( for comparison see lower panel of fig . 12 in @xcite and our fig . [ pp4 ] ) . in the framework of an extended nambu jona - lasinio model , we have studied the influence of a gravitational field on pion condensation in isotopically asymmetric quark matter at finite temperature and isospin chemical potential . as a particular model of a gravitational field configuration we have taken the static einstein universe . this particular choice enabled us to investigate phase transitions of the system with an exact consideration of the role of the gravitational field in the formation of the quark and pion condensates and thus to demonstrate its influence on the phase portraits . in particular , we have found that thermodynamic quantities such as quark and pion condensates as well as the corresponding phase boundaries ( critical curves ) oscillate as functions of curvature . this oscillating behavior is smoothed out with growing temperature . there exists also an interesting similarity between the behavior of phase portraits when considered as functions of curvature and as functions of finite temperature ( compare figs . moreover , we have shown that for massless quarks and for some values of @xmath2 the isospin symmetry becomes in curved spacetime dynamically broken only at some finite value of the chemical potential @xmath30 ( look at fig . 6 ( left picture ) for rather small values of @xmath129 ) . this is contrary to the case of flat spacetime , where the pion condensation appears at arbitrary small nonzero values of the isospin chemical potential ( see fig . this effect resembles the pion condensation but at nonzero bare quark mass and may be explained by the presence of a gap in the energy spectrum of quarks in the gravitational field . finally , let us add some remarks on the possible role of quantum fluctuations of the collective fields and finite size effects in our approach . in fact , since the volume of the space region modelled by the closed einstein universe adopted in this paper is limited , finite size effects could eventually change the character of phase transitions ( see e.g. discussions in @xcite ) . eventually , this might even lead to particular situations , where no phase transition can occur . however , as it is clear from physical considerations , the finite size in itself may not practically forbid the dynamical symmetry breaking , if the characteristic length of the region of space occupied by the system is much greater than the compton wavelength of the excitation responsible for tunneling and restoration of symmetry ( see , e.g. , @xcite ) . it should be noted that our results , obtained with the use of the mean field approximation in the framework of the njl model are consistent with corresponding estimates of the role of quantum fluctuations ( see appendix b ) . in conclusion , we emphasize that the results of this paper are evidently only of a qualitative nature . they do not allow us to find an exact value of the critical radius , and hence , further studies with more realistic models of gravitational fields should be undertaken . two of us ( v.zh . and thank m. mueller - preussker for hospitality during their stay at the institute of physics of humboldt - university , where part of this work has been done , and also daad for financial support . d.e . is grateful to r. seiler for discussion on the role of quantum fluctuations and finite size effects . this work was supported in part by the dfg - grant 436 rus 113/477 . in section [ eu ] we have introduced two operators , @xmath110 and @xmath97 , acting in the hilbert space @xmath187 of all quadratically integrable functions @xmath188 defined on the spacetime manifold of the einstein universe . now suppose that there is an abstract hilbert space @xmath189 of vectors @xmath190 . let @xmath191 and @xmath192 be the operators of the space and time coordinates , correspondingly , defined on @xmath193 . moreover , let @xmath194 be the complete set , or basis , of eigenvectors of @xmath191 and @xmath195 , i.e. @xmath196 , @xmath197 . the set @xmath198 is usualy called the coordinate basis in @xmath193 . obviously , the completeness and normalization conditions for the coordinate basis @xmath198 are valid : @xmath199 where @xmath200 is the unit operator in @xmath193 and @xmath201 . due to ( [ b1 ] ) , it is possible to expand any vector @xmath202 in terms of the basis @xmath198 , namely @xmath203 . the quantity @xmath204 is called _ @xmath24- ( or coordinate ) representation for the vector @xmath190_. identifying @xmath205 with @xmath188 , we see that the hilbert space @xmath187 of all quadratically integrable functions is simply the coordinate representation of the above introduced abstract hilbert space @xmath193 . furthermore , if in the hilbert space @xmath193 there is an arbitrary operator @xmath206 , then the matrix @xmath207 , whose matrix elements in the coordinate basis are just the quantities @xmath208 , is called the @xmath24- ( or coordinate ) representation of the operator @xmath206 . obviously , in order to define an operator @xmath206 in the abstract hilbert space @xmath193 , it is sufficient to define its matrix @xmath209 , which acts in the hilbert space @xmath187 of all quadratically integrable functions , in the coordinate basis . it is clear that tr@xmath210@xmath211 . now , let us consider in @xmath193 the two commuting operators @xmath212 and @xmath213 such that in the @xmath24-representation they look like @xmath214 and latexmath:[$\hat{\cal h}\equiv \langle x ' operators @xmath110 and @xmath97 have a common set of eigenfunctions @xmath112 defined in section [ eu ] , from which it follows that @xmath216 the eigenfunctions @xmath112 are the coordinates of the corresponding eigenvectors @xmath217 of the operators @xmath212 and @xmath213 ( recall , @xmath218 ) : @xmath219 ( clearly , @xmath220 and @xmath221 . ) it follows from ( [ b4 ] ) that @xmath222 . using this relation in the normalization condition ( [ b3 ] ) and then integrating there over @xmath24 with the help of ( [ b1 ] ) , we obtain @xmath223 which is the analogue of the normalization condition ( [ b2 ] ) . it is possible to show that the completness condition for the basis @xmath224 follows from ( [ b5 ] ) : @xmath225 let us construct in @xmath193 the following operator @xmath226 which is diagonal in the basis ( [ b4 ] ) , i.e. each vector @xmath224 is its eigenvector with corresponding eigenvalue @xmath227 ( [ 192 ] ) . in the coordinate representation its matrix @xmath228 coincides with the operator @xmath113 ( [ 19 ] ) . so , @xmath229 where the last equality was obtained by employing the completeness relation ( [ b6 ] ) . now , by using in this formula the eigenvalue condition @xmath230 , the normalization condition ( [ b5 ] ) , and , finally , by performing in the obtained expression the integration and summation over primed indices , we have @xmath231 since in ( [ b9 ] ) the quantities @xmath232 and @xmath233@xmath234 ( the last relation is due to the notations accepted in formula ( [ 191 ] ) and below ) do not depend on the time coordinate , the expression ( [ b9 ] ) is proportional to the infinite time interval @xmath235 . the remaining @xmath98-integration in ( [ b9 ] ) gives simply unity due to the relation ( [ 191 ] ) . so we have @xmath236 , \label{b10}\end{aligned}\ ] ] where the fact that each eigenvalue @xmath237 is @xmath100-fold degenerated is taken into account and the notations from ( [ 190])-([192 ] ) are used . in a similar way it is possible to obtain the quantity tr@xmath120 : @xmath238 . \label{b11}\end{aligned}\ ] ] it is well known that spontaneous symmetry breaking in low dimensional quantum field theories may become impossible due to strong quantum fluctuations of fields . the same is also true for systems that occupy a limited space volume . however , as it is clear from physical considerations , the finite size in itself may not practically forbid the spontaneous symmetry breaking , if the characteristic length of the region of space occupied by the system is much greater than the compton wavelength of the excitation responsible for tunneling and restoration of symmetry . ( indeed , one may recall here well known physical phenomena such as the superfluidity of helium or superconductivity of metals that are observed in samples of finite volume ) . this idea has been discussed for some scalar field theories in the einstein universe for instance , in @xcite . in this appendix , we shall demonstrate that , under certain conditions , dynamical symmetry breaking in njl - type models is indeed possible in the closed einstein universe . in particular , we will show that , if the radius of the universe is large enough such that the fluctuations of quantum fields are comparatively small , the symmetry breaking obtained in the mean field approximation is not forbidden . for illustrations , let us confine to the analogous case of the chiral condensate and consider the simplified case of the linearized lagrangian ( [ 3 ] ) with @xmath239 , @xmath240 and the corresponding partition function @xmath241 with @xmath242 now , supposing that @xmath243 , where @xmath244 is the vacuum expectation value of the field @xmath160 and @xmath245 denotes its quantum fluctuation , we obtain : @xmath246 where @xmath247 . thus @xmath248 , where @xmath249 , @xmath250 , and @xmath251 is the effective potential at @xmath252 . the contribution of quantum fluctuations up to the @xmath253-term to the effective action @xmath254 is given by @xmath255 it is evident , that in the above expansion the term linear in @xmath245 corresponds to the so - called tadpole diagram with one external @xmath245-line and the term quadratic in @xmath245 corresponds to the `` polarization operator '' diagram of the @xmath245 field with one fermion loop . from ( [ v01 ] ) we can write the stationarity condition and find the gap equation , @xmath256 , @xmath257 and the linear terms in @xmath245 , corresponding to the tadpole diagram , cancel out in @xmath258 . thus , the contribution of fluctuations to the field action is given by @xmath259 next , we shall calculate the contribution of fluctuations @xmath245 to the effective action , taking into account the quark loop in the gravitational field . ( note that this corresponds to the integral ( 11 ) and fig.1a in @xcite . ) for our purpose of making estimates of the role of fluctuations , it is sufficient to limit ourselves to the consideration of fluctuations depending only on time . in this case , we can extract the necessary kinematical factor for the meson fluctuation field and then integrate over the time - component of the loop momentum in the limit of vanishing external momenta . finally , after going to the basis for the dirac equation in the einstein universe ( see ( [ 19])([191 ] ) ) we obtain a sum over fermion loop quantum numbers @xmath151 instead of an integration over momenta of free quarks made in @xcite . the sum is divergent and we regularize it by the cut off @xmath153 . in this way we obtain the effective action @xmath260 , \label{phiaction}\ ] ] where @xmath261 and @xmath262 are given in ( [ 190 ] ) . the summation over @xmath151 in the first term in parenthesis in the above equation cancels out by the term @xmath263 , due to the stationarity condition ( [ gap ] ) @xmath264 after this we obtain @xmath265 or @xmath266 with the lagrange function @xmath267 where @xmath268 is the space volume , and the renormalization @xmath269-factor is defined as @xmath270 for comparision , let us consider the limiting case of flat space which is reached by the replacements @xmath271 in ( [ major ] ) . then the @xmath269-factor takes the form of the integral in the euclidean spacetime @xmath272 with @xmath273 being the quark energy . this expression evidently corresponds to the similar formula for the @xmath274-factor in the flat space case considered in @xcite . next , let us perform the field renormalization @xmath275 in ( [ lagrange ] ) . quantum fluctuations of the boson field near the ground state @xmath244 can now be estimated , if we consider the renormalized expression ( [ lagrange ] ) as the lagrange function for a harmonic oscillator ( here , we follow the idea of @xcitemodel of a self - interacting scalar field was considered and the scalar loop contribution to the fluctuation lagrangian was calculated . ] ) with the mass @xmath32 and frequency @xmath276 , formally given here by the relations @xmath277 then we can estimate the quantum fluctuations as @xmath278 where m is the mass of the composite @xmath160-meson . thus , we obtain @xmath279 the estimate ( [ estimate ] ) gives a criterion for the role of quantum fluctuations for a system with finite volume . clearly , quantum fluctuations can be considered negligible , if @xmath280 . this is in agreement with the physical requirement that quantum fluctuations should be negligible if @xmath281 , i.e. , if the radius of the universe is much greater than the compton wavelength @xmath282 of the @xmath160-meson ( quarks ) ( see also @xcite ) . , which are of order @xmath283 @xcite . all these corrections are surely suppressed in the case of large numbers of colors @xmath284 , where the ( induced ) coupling constants become small . ] the above estimates are certainly of a qualitative nature , and hence they do not allow us to find an exact value of a critical radius such that symmetry breaking for lower values of the curvature radius of the einstein universe is forbidden . m.k . volkov , fiz . atom.yadra * 17 * , 433 ( 1986 ) ; m.k . volkov and a.e . radzhabov , phys . * 49 * , 551 ( 2006 ) ; a.a . andrianov , d. espriu , and r. tarrach , nucl . b * 533 * , 429 ( 1998 ) ; s.v . molodtsov , a.n . sissakian , a.s . sorin , and g.m . zinovjev , arxiv : hep - ph/0702178 . klevansky , rev . phys . * 64 * , 649 ( 1992 ) ; t. hatsuda and t. kunihiro , phys . rept . * 247 * , 221 ( 1994 ) . m. alford , k. rajagopal , and f. wilczek , phys . b 422 , 247 ( 1998 ) ; nucl . phys . b * 537 * , 443 ( 1999 ) ; s.v . molodtsov and g.m . zinovjev , phys . atom . nucl . * 66 * , 1349 ( 2003 ) . g. nardulli , riv . nuovo cim . * 25n3 * , 1 ( 2002 ) ; m. buballa , phys . rep . * 407 * , 205 ( 2005 ) ; i.a . shovkovy , found . phys . * 35 * , 1309 ( 2005 ) ; t. ohsaku , phys . b * 634 * , 285 ( 2006 ) ; m.g . alford , a. schmitt , k. rajagopal , and t. schafer , arxiv:0709.4635 . d. blaschke , d. ebert , k.g . klimenko , m.k . volkov , and v.l . yudichev , phys . d * 70 * , 014006 ( 2004 ) ; d. ebert , k.g . klimenko , and v.l . yudichev , phys . c * 72 * , 015201 ( 2005 ) ; eur . j. c * 53 * , 65 ( 2008 ) ; d. ebert and k.g . klimenko , theor . phys . * 150 * , 82 ( 2007 ) . d.t . son and m.a . stephanov , phys . atom . * 64 * , 834 ( 2001 ) ; j.b . kogut , and d. toublan , phys . d * 64 * , 034007 ( 2001 ) ; j.b . kogut , and d.k . sinclair , phys . d * 66 * , 014508 ( 2002 ) . andrianov and d. espriu , arxiv:0709.0049 ; m. loewe and c. villavicencio , braz . j. phys . * 37 * , 520 ( 2007 ) ; j. erdmenger , m. kaminski , and f. rust , arxiv:0710.0334 ; s. mukherjee , m.g . mustafa , and r. ray , phys . d * 75 * , 094015 ( 2007 ) ; h. abuki , m. ciminale , r. gatto , n.d . ippolito , g. nardulli , and m. ruggieri , arxiv:0801.4254 . k.g . klimenko , arxiv : hep - ph/9809218 ; d. ebert , k.g . klimenko , m.a . vdovichenko , and a.s . vshivtsev , phys . rev . d * 61 * , 025005 ( 2000 ) ; m.a . vdovichenko , a.s . vshivtsev , and k.g . klimenko , phys . * 63 * , 470 ( 2000 ) . vshivtsev , v.ch . zhukovsky , k.g . klimenko , and b.v . magnitsky , phys . * 29 * , 523 ( 1998 ) ; d. ebert and k.g . klimenko , nucl . phys . * a 728 * , 203 ( 2003 ) ; t. inagaki , d. kimura , and t. murata , prog . theor . phys . suppl . * 153 * , 321 ( 2004 ) ; a.a . osipov , b. hiller , a.h . blin , and j. da providencia , phys . b * 650 * , 262 ( 2007 ) ; arxiv:0802.3193 ; t.d . cohen , d.a . mcgady , and e.s . werbos , phys . c * 76 * , 055201 ( 2007 ) ; p. castelo ferreira and j. dias de deus , arxiv:0707.4200 ; e.s . werbos , arxiv:0711.2635 . d. ebert , k.g . klimenko , and h. toki , phys . d * 64 * , 014038 ( 2001 ) ; v.ch . zhukovsky , v.v . khudyakov , k.g . klimenko , and d. ebert , pisma zh . fiz . * 74 * , 595 ( 2001 ) ; d. ebert et al . d * 65 * , 054024 ( 2002 ) ; t. tatsumi , e. nakano , and k. nawa , arxiv : hep - ph/0506002 ; e.j . ferrer , v. de la incera , and c. manuel , phys . lett . * 95 * , 152002 ( 2005 ) ; nucl . b * 747 * , 88 ( 2006 ) ; j.l . noronha and i.a . shovkovy , phys . d * 76 * , 105030 ( 2007 ) . d. ebert , k.g . klimenko , v.c . zhukovsky , and a.m. fedotov , eur . j. c * 49 * , 709 ( 2007 ) ; vestn . astron . * 61n2 * , 69 ( 2006 ) . t. inagaki , s.d . odintsov , and t. muta , prog . . suppl . * 127 * , 93 ( 1997 ) ( see also further references in this review paper ) . d. ebert and v.ch . zhukovsky , phys . lett . b * 645 * , 267 , ( 2007 ) . ( the paper contains a misprint in the definition of @xmath285 matrices in the charge conjugation operator _ c _ , which should read like this @xmath286 . the results of the paper , however , were obtained with the above correct expression for this operator and do not depend on this misprint . ) t.s . bunch and l. parker , phys . d * 20 * , 2499 ( 1979 ) . kim and k.g . klimenko , j. phys . a * 31 * , 5565 ( 1998 ) . a. goyal and m. dahiya , j. phys . g * 27 * , 1827 ( 2001 ) . x. huang , x. hao , and p. zhuang , astropart . * 28 * , 472 ( 2007 ) . d. ebert , a.v . tyukov , and v.ch . zhukovsky , phys . d * 76 * , 064029 ( 2007 ) . klimenko , theor . phys . * 70 * , 87 ( 1987 ) . p. elmfors , d. persson , and b.s . skagerstam , phys . rev . lett . * 71 * , 480 ( 1993 ) ; j.o . andersen and t. haugset , phys . d * 51 * ( 1995 ) 3073 ; a.s . vshivtsev , k.g . klimenko , and b.v . magnitsky , j. exp . phys . * 80 * , 162 ( 1995 ) ; j. exp . phys . * 82 * , 514 ( 1996 ) . zhukovsky and j.herrmann , yad . fiz . * 14 * , 150 ( 1971 ) ; * 14 * , 1014 ( 1971 ) .
the gravitational interaction of two massive particles is described by newton s law @xmath0 which is a good approximation for the motion of nonrelativistic particles at large separations . einstein s theory of general relativity , however , predicts important corrections that have been verified experimentally , _ e.g. _ , in measurements of the precession of the perihelion of mercury @xcite . this effect can be calculated from the famous einstein - infeld - hoffmann lagrangian @xcite where the terms beyond newtonian physics stem from the relativistic @xmath1 corrections to the kinetic energy , from relativistic @xmath1 corrections to the newtonian potential and from a new piece of the potential proportional to @xmath2 which may be regarded as an @xmath3 correction to the leading newtonian potential . the two small expansion parameters here are @xmath4 and @xmath5 which for bound states are related due to the virial theorem . we , however , will examine scattering processes wherein the classical expansion parameters @xmath4 and @xmath5 are independent , and our focus will be on the components that vanish in the nonrelativistic limit @xmath6 . in the weak field limit gravitational dynamics can be described in terms of a quantum field theory which is based on expanding around flat minkowski spacetime and quantizing the metric fluctuation in terms of a massless spin-2 field the graviton . the resulting theory is an effective field theory whose predictions are trustworthy only for energies much smaller than the planck scale . nevertheless , one can in this picture calculate well - defined quantum predictions as well as classical corrections to newtonian physics . donoghue has pioneered the use of the effective field theory of gravity to extract the leading long distance effects in the nonrelativistic limit @xcite ( see @xcite and @xcite for reviews ) which yield corrections of the form @xmath7 where @xmath8 and @xmath9 are the coefficients of the classical and quantum corrections respectively and are evaluated below . note that while the classical expansion parameter @xmath10 can give measurable effects for macroscopic objects ( where a large factor multiplies a tiny factor ) , the quantum corrections which scale as @xmath11 are clearly tiny and phenomenologically irrelevant for macroscopic systems . the calculations of these leading long distance corrections are performed by evaluating the leading nonanalytic components of the scattering amplitude at one loop level , where @xmath12 effects first arise , and by defining a potential in terms of the fourier transform of this amplitude . once spin is introduced into the calculation , additional interaction structures arise , such as a spin - orbit coupling and a spin - spin coupling . the leading effects of order @xmath13 stemming from the tree level one - graviton exchange process give rise to the familiar geodetic precession and to the lense - thirring effect @xcite . at the two - graviton exchange level , classical @xmath14 and quantum @xmath15 corrections to these leading spin - dependent interactions arise and are calculated as part of our work . in earlier work on loop corrections to the form factors of graviton couplings it was found that the long distance corrections to the spin - independent interactions have the same form for scalars , spin-1/2 fermions and for spin-1 bosons @xcite . these nlo form factor interactions constitute one component of the calculation of the full scattering amplitude at the two - graviton exchange level , so the question arises as to whether the _ full _ scattering amplitude also exhibits such universalities , whereby the form of the corrections to the spin - independent newtonian potential is independent of the spins of the scattered particles in other words , are the coefficients @xmath8 and @xmath9 in eq . ( [ eq_pot_cor ] ) independent of spin ? this then is the goal of the present work to evaluate the gravitational scattering of two massive particles having various spins , in order to check previous work and to verify the universality hypothesis . in the next section then we review our calculational methods and reproduce the long range gravitational scattering amplitude in the case of a pair of spinless particles . in the following sections , we generalize these methods to the cases of spin-0 spin-1/2 , spin-0 spin-1 and spin-1/2 spin-1/2 scattering and demonstrate both the expected universality as well as novel spin - dependent interactions . we summarize our work and draw general implications in a brief concluding section . appendix [ app_iter ] encapsulates parts of our calculational methods while we refer to our companion paper on long distance effects in electromagnetic scattering @xcite for many of the details such as the required feynman integrals , the fourier transformations etc . in appendix [ app_eom ] we demonstrate how the classical equations of motion in the form of the einstein - infeld - hoffmann lagrangian can be extracted from our results . we first set the kinematic framework for our study . we consider the gravitational scattering of two non - identical particles particle @xmath16 with mass @xmath17 , and incoming four - momentum @xmath18 and particle @xmath19 with mass @xmath20 , and incoming four - momentum @xmath21 . after interacting the final four - momentum of particle @xmath16 is @xmath22 and that of particle @xmath19 is @xmath23_cf . [ fig_kinem ] . now we need to be more specific . we begin by examining the gravitational scattering of two spinless particles . the gravitational coupling of a spin-0 particle is found by expanding the minimally coupled scalar field matter lagrangian @xmath24 in terms of the gravitational field @xmath25 which is a small fluctuation of the metric about flat minkowski space defined as @xmath26 with @xmath27 . the inclusion of this factor @xmath28 in the definition of the graviton field @xmath29 gives this field a mass - dimension of unity and thus yields a kinetic term of standard normalization without a dimensionful parameter . for matter interactions , this choice is convenient since the order of @xmath28 keeps track of the number of gravitons involved in an interaction . once the action is written in terms of the expansion of the graviton field , all indices are understood to be lowered or raised using the minkowski metric @xmath30 . we also require the expansion of the inverse metric and square root of the determinant of the metric tensor @xmath31 then , expanding in powers of @xmath28 , we find @xmath32\nonumber\\ \sqrt{-g}{\cal l}_m^{(2)}&=&{\kappa^2\over 2 } \hspace*{0.5pt } \bigg[\frac{1}{4 } \big(h^2 - 2 h_{\mu \nu } h^{\mu \nu}\big ) \bigg(\frac{1}{2 } \partial_\alpha \phi \partial^\alpha \phi - \frac{1}{2 } m^2 \phi^2\bigg)\nonumber\\ & & \hspace*{13pt } + \bigg(h^{\mu \alpha } h_\alpha^\nu - \frac{1}{2 } h h^{\mu \nu}\bigg ) \partial_\mu \phi \partial_\nu \phi \bigg]\end{aligned}\ ] ] where @xmath33 represents the trace and the one- and two - graviton vertices are identified as @xmath34\nonumber\\ { } ^{0 } \tau^{(2)}_{\mu\nu,\rho\sigma}(p_2 , p_1 , m)\hspace*{-7.5pt}&=\hspace*{-7.5pt } & \frac{i\kappa^2}{2}\bigg[2 i_{\mu\nu,\kappa\zeta}{i^\zeta}_{\lambda,\rho\sigma}(p_1^\kappa p_2^\lambda+p_1^\lambda p_2^\kappa ) \nonumber\\ & & { } \hspace*{14pt } - \hspace*{-1pt } ( \hspace*{-0.5pt } \eta_{\mu\nu}i_{\kappa\lambda,\rho\sigma } \hspace*{-1.5pt } + \hspace*{-1.5pt } \eta_{\rho\sigma}i_{\kappa\lambda,\mu\nu}\hspace*{-0.5pt } ) p_1^\kappa p_2^\lambda \hspace*{-1pt } - \hspace*{-1pt } p_{\mu\nu , \rho\sigma } ( p_1 \hspace*{-1.5pt } \cdot \hspace*{-1.3pt } p_2 \hspace*{-1.8pt } - \hspace*{-1.5pt } m^2 ) \hspace*{-1pt } \bigg ] \label{eq : tp}\end{aligned}\ ] ] where we have defined @xmath35 the purely gravitational dynamics are derived from the einstein - hilbert action @xmath36 where higher dimensional operators such as @xmath37 , which are needed when performing renormalization , are not relevant here @xcite . since general relativity is invariant under _ local _ coordinate transformations , it is a gauge theory and one must deal with the subtleties that arise in the quantization of gauge theories we have to perform gauge fixing . the procedure works analogously to gauge fixing in yang - mills theories . furthermore , we will make use of the background field method , which provides a powerful organizational scheme for the quantization of the effective field theory of gravity : keeping the background metric general instead of restricting to the flat minkowski metric , we use the expansion @xmath38 where @xmath39 is the classical background metric ( or field ) and @xmath40 is the quantum field . a gauge fixing condition is only imposed on the quantum field @xmath40 , leaving the general covariance of the background unaffected . this procedure has the advantage of ensuring that the resulting theory can be renormalized , since the loop expansion then has the same symmetry properties as the action . moreover , the background field method greatly simplifies calculations involving graviton loops . gravitons running in loops must be derived from from an expansion involving the quantum field @xmath40 whereas gravitons that are _ not _ within a loop may be derived from expanding the background field @xmath41 where @xmath42 denotes an `` external '' graviton , _ i.e. _ , a graviton that is not inside a loop . at the one loop level , at most two gravitons involved in a vertex are propagating within a loop , which allows us to use a gauge fixing condition _ linear _ in the quantum field @xmath40 and greatly simplifies the derivation of both the triple graviton vertex and the vertex that couples one graviton to the ghost fields . in order to fix the gauge , we will use the harmonic gauge @xmath43 where @xmath44 denotes the covariant derivative on the background metric . this condition leads to an additional gauge fixing piece of the action @xmath45 as well as the ghost action @xmath46 where @xmath47 is the ghost field that annihilates a ghost particle while @xmath48 creates a ghost particle . for each closed ghost loop . ] now we are in the position to derive the feynman rules for the effective field theory of gravity . the complete quantum gravitational action consists of three components @xmath49 and we will illustrate the use of the background field method during the derivations of the feynman rules . when deriving the graviton propagator we expand the action to second order in the quantum fields @xmath40 and to zeroth order in the external gravitons @xmath42 and the ghosts , yielding @xmath50 we obtain the ghost propagator expanding the action to second order in the ghost fields and to zeroth order in the gravitons , whereby @xmath51 our triple graviton vertex is obtained by expanding to second order in the quantum fields @xmath40 , to first order in the background gravitons @xmath42 , and to zeroth order in the ghost fields and reads @xmath52\nonumber\\ & & \hspace*{17pt } + 2q_\lambda q_\sigma \hspace*{-1pt } \bigg[i^{\lambda\sigma,}{}_{\alpha\beta}i^{\mu\nu,}{}_{\gamma\delta } + i^{\lambda\sigma,}{}_{\gamma\delta}i^{\mu\nu,}{}_{\alpha\beta } \nonumber \\ & & \hspace*{54pt } -i^{\lambda\mu,}{}_{\alpha\beta}i^{\sigma\nu,}{}_{\gamma\delta } -i^{\sigma\nu,}{}_{\alpha\beta}i^{\lambda\mu,}{}_{\gamma\delta}\bigg]\nonumber\\ & & \hspace*{17pt } + \hspace*{-1pt } \bigg [ q_\lambda q^\mu(\eta_{\alpha\beta}i^{\lambda\nu,}{}_{\gamma\delta } \hspace*{-1pt } + \hspace*{-1pt } \eta_{\gamma\delta}i^{\lambda\nu,}{}_{\alpha\beta } ) \hspace*{-2pt } + \hspace*{-1pt } q_\lambda q^\nu(\eta_{\alpha\beta}i^{\lambda\mu,}{}_{\gamma\delta } \hspace*{-1pt } + \hspace*{-1pt } \eta_{\gamma\delta}i^{\lambda\mu,}{}_{\alpha\beta})\nonumber\\ & & \hspace*{24pt } - q^2 \hspace*{-0.6pt } ( \hspace*{-0.5pt } \eta_{\alpha\beta}i^{\mu\nu,}{}_{\gamma\delta } \hspace*{-1.5pt } + \hspace*{-1.5pt } \eta_{\gamma\delta } i^{\mu\nu,}{}_{\alpha\beta } \hspace*{-0.5pt } ) \hspace*{-2.3pt } - \hspace*{-1.5pt } \eta^{\mu\nu } q^\lambda q^\sigma \hspace*{-0.6pt } ( \hspace*{-0.5pt } \eta_{\alpha\beta } i_{\gamma\delta,\lambda\sigma } \hspace*{-2pt } + \hspace*{-1.5pt } \eta_{\gamma\delta } i_{\alpha\beta,\lambda\sigma}\hspace*{-0.5pt } ) \hspace*{-1.5pt } \bigg]\nonumber\\ & & \hspace*{17pt } + \hspace*{-1pt } \bigg[2q^\lambda \big(i^{\sigma\nu,}{}_{\gamma\delta } i_{\alpha\beta,\lambda\sigma}k^\mu + i^{\sigma\mu,}{}_{\gamma\delta}i_{\alpha\beta,\lambda\sigma}k^\nu \nonumber\\ & & \hspace*{50pt } -i^{\sigma\nu,}{}_{\alpha\beta}i_{\gamma\delta,\lambda\sigma}(k+q)^\mu - i^{\sigma\mu,}{}_{\alpha\beta}i_{\gamma\delta,\lambda\sigma}(k+q)^\nu \big ) \nonumber\\ & & \hspace*{24pt } + q^2(i^{\sigma\mu,}{}_{\alpha\beta}i_{\gamma\delta,\sigma}{}^\nu + i_{\alpha\beta,\sigma}{}^\nu i^{\sigma\mu,}{}_{\gamma\delta } ) \nonumber \\ & & \hspace*{24pt } + \eta^{\mu\nu}q^\lambda q_\sigma ( i_{\alpha\beta,\lambda\rho}i^{\rho\sigma,}{}_{\gamma\delta } + i_{\gamma\delta,\lambda\rho}i^{\rho\sigma,}{}_{\alpha\beta})\bigg]\nonumber\\ & & \hspace*{17pt } + \hspace*{-1pt } \bigg[\hspace*{-0.5pt } \big(k^2 \hspace*{-1pt } + \hspace*{-1pt } ( k \hspace*{-1pt } + \hspace*{-1pt } q)^2 \big ) \hspace*{-3pt } \left ( \hspace*{-2pt } i^{\sigma\mu,}{}_{\alpha\beta}i_{\gamma\delta,\sigma}{}^\nu \hspace*{-1pt } + \hspace*{-1pt } i^{\sigma\nu,}{}_{\alpha\beta}i_{\gamma\delta,\sigma}{}^\mu \hspace*{-1pt } - \hspace*{-1pt } { 1\over 2}\eta^{\mu\nu}p_{\alpha\beta,\gamma\delta } \hspace*{-2pt } \right ) \nonumber\\ & & \hspace*{24pt } - \big((k+q)^2\eta_{\alpha\beta}i^{\mu\nu,}{}_{\gamma\delta}+k^2\eta_{\gamma\delta } i^{\mu\nu,}{}_{\alpha\beta}\big)\bigg]\bigg\}\end{aligned}\ ] ] where the graviton with lorentz indices @xmath53 represents a background graviton , and therefore , it is _ not _ to be used within any loop ! in the amplitude . ] the final missing feynman rule from the purely gravitational action is the vertex coupling one graviton to two ghost fields which we derive by expanding to second order in the ghost fields and to first order in @xmath42 @xmath54\end{aligned}\ ] ] where the graviton with lorentz indices @xmath53 is again a background graviton and not to be used as part of any loop . we begin our calculation with the simplest tree level single - graviton exchange which leads to the amplitude . ] \nonumber\\ \quad\end{aligned}\ ] ] a convenient way to define the nonrelativistic potential is as the fourier transform of the nonrelativistic center of mass scattering amplitude . we utilize a symmetric center of mass frame with incoming momenta @xmath56 and @xmath57 and with outgoing momenta @xmath58 and @xmath59 . conservation of energy then requires @xmath60 so that @xmath61 for @xmath62 and @xmath63 . in the nonrelativistic limit @xmath64 the lowest order amplitude reads @xmath65 \nonumber \\ & + & g \pi \left[\frac{3(m_a^2 + m_b^2)}{2 m_a m_b } + \frac{\vec p^{\hspace*{1.4pt } 2}}{m_a m_b } \left(3 - \frac{5(m_a^2 + m_b^2)^2}{4 m_a^2 m_b^2}\right ) + \ldots \right ] + \ldots \nonumber \\ \label{eq_amplo_00}\end{aligned}\ ] ] yielding the potential @xmath66 \nonumber \\ & & + \ , g \pi \delta^3(\vec r ) \left[\frac{3(m_a^2 + m_b^2)}{2 m_a m_b } + \frac{\vec p^{\hspace*{1.4pt } 2}}{m_a m_b}\left(3 - \frac{5(m_a^2 + m_b^2)^2}{4 m_a^2 m_b^2}\right ) + \ldots \right ] \nonumber\\ \label{eq : po}\end{aligned}\ ] ] the leading component of eq . ( [ eq : po ] ) is recognized as the usual newtonian potential ( accompanied by a small kinematic correction ) while the second piece is a short range modification . our purpose in this paper is to study the long distance corrections to this lowest order potential which arise from the two - graviton exchange diagrams shown in fig . [ fig_loops ] . this problem has been previously studied by iwasaki using noncovariant perturbation theory @xcite , and by khriplovich and kirilin @xcite and by bjerrum - bohr , donoghue , and holstein @xcite using conventional feynman diagrams . our approach will be similar to that used in @xcite and @xcite . that is , using the ideas of effective field theory , we evaluate these feynman diagrams by keeping only the leading nonanalytic structure in @xmath67 , since it is these pieces that lead to the long range corrections to the potential while components analytic in @xmath67 only yield short range contributions , _ i.e. _ , delta functions or derivatives of delta functions . the leading nonanalytic behavior is of two basic forms * terms in @xmath68 which are @xmath69-independent and therefore classical * terms in @xmath70 which are @xmath69-dependent and therefore quantum mechanical . the former terms , when fourier transformed lead to corrections to the nonrelativistic potential of the form @xmath71 while the latter lead to @xmath72 corrections . for typical masses and separations the quantum mechanical forms are themselves numerically insignificant . however , they are intriguing in that their origin appears to be associated with zitterbewegung . that is , classically we can define the potential by measuring the energy when two objects are separated by distance @xmath73 . however , in the quantum mechanical case the distance between two objects is uncertain by an amount of order the compton wavelength due to zero point motion@xmath74 . this leads to the replacement @xmath75 which is the form found in our calculations . the diagrams to be evaluated are shown in fig . [ fig_loops ] where the `` blobs '' shown in diagrams ( f ) and ( g ) are explained in fig . [ fig_blobs ] . the calculational details including the feynman integrals are described in an appendix of our companion paper on electromagnetic scattering @xcite . here we present only the results . defining differs slightly from previous publications @xcite in that it does not include a factor of mass in the numerator . ] @xmath76 we have , from diagrams [ fig_loops](a)-(g ) respectively @xmath77\nonumber\\ { } ^0{\cal m}_{\ref{fig_loops}b}^{(2)}(q)&=&g^2m_am_b\left[28l + 8 m_a s\right]\nonumber\\ { } ^0{\cal m}_{\ref{fig_loops}c}^{(2)}(q)&=&g^2m_am_b\left[28l + 8 m_b s\right]\nonumber\\ { } ^0{\cal m}_{\ref{fig_loops}d}^{(2)}(q)&=&g^2 m_a m_b \left[\hspace*{-0.5pt } \left(\frac{4 m_a m_b}{q^2}+\frac{8(m_a^2 + m_b^2)}{m_a m_b } - 8\right ) \hspace*{-1.5pt } l + \hspace*{-0.5pt}4 ( m_a + m_b ) s \right]\nonumber\\ & -&i 4\pi g^2 m_a^2 m_b^2 \hspace*{1pt } { l\over q^2 } \sqrt{m_a m_b \over s - s_0}\nonumber\\ { } ^0{\cal m}_{\ref{fig_loops}e}^{(2)}(q)&=&g^2 m_a m_b \left[\hspace*{-0.5pt } \left(\hspace*{-1.5pt } - \frac{4 m_a m_b}{q^2}- \frac{8(m_a^2 + m_b^2)}{m_a m_b } - \frac{70}{3}\right ) \hspace*{-1.5pt } l - \hspace*{-0.5pt}4 ( m_a \hspace*{-0.5pt } + \hspace*{-0.5pt } m_b ) s \right]\nonumber\\ { } ^0{\cal m}_{\ref{fig_loops}f}(q)&=&g^2m_am_b\left[14 l - 2 ( m_a + m_b ) s\right]\nonumber\\ { } ^0{\cal m}_{\ref{fig_loops}g}(q)&=&g^2m_am_b\left[-{43\over 15 } l \right]\quad\end{aligned}\ ] ] where @xmath78 is the square of the center of mass energy and @xmath79 is its threshold value . the contributions from the form factor diagrams in fig . [ fig_loops](f ) have been calculated both directly and using the results from @xcite , and the contribution from vacuum polarization diagrams in fig . [ fig_loops](g ) have been obtained both by direct evaluation and by using previous results of t hooft and veltman @xcite for the divergences of the graviton self - energy that allow us to infer its nonanalytic contributions . here we should note that _ any _ massless particle would contribute to the nonanalytic terms of the graviton vacuum energy since gravitons couple to anything . we choose here not to include other massless particles besides the graviton in particular we do not include the contribution of the photon in the quantum piece of our potential . summing , we find the total result @xmath80-i4\pi g^2m_a^2m_b^2 \hspace*{1pt } { l\over q^2}\sqrt{m_am_b\over s - s_0}\label{eq : kp}\ ] ] and we observe that , in addition to the expected terms involving @xmath81 and @xmath82 , there arises a piece of the second order amplitude which is _ imaginary_. the origin of this imaginary piece is , of course , from the second born approximation to the newtonian potential , and reminds us that in order to define a proper correction to the first order newtonian potential we must subtract off such terms . for this purpose we will work in the nonrelativistic limit and the center of mass frame@xmath83 as defined above . we have then @xmath84 and @xmath85 where @xmath86 is the reduced mass and @xmath87-i4\pi g^2m_a^2m_b^2 \hspace*{1pt } { l\over q^2}{m_r\over p_0 } .\ ] ] for the iteration we shall use the simple potential @xmath88 which reproduces the long distance behavior of the lowest order amplitude for spin-0 spin-0 gravitational scattering eq . ( [ eq : po])in the nonrelativistic limit . we will employ the momentum space representation @xmath89 where we identify @xmath90 and @xmath91 , and the second born term is then @xmath92 where @xmath93 is the nonrelativistic propagator . note that in eq . ( [ eq : iteration00a ] ) we take both the leading order potential as well as the total energies @xmath94 and @xmath95 in the nonrelativistic limit . the remaining integration can be performed exactly , yielding ] @xmath96 which precisely reproduces the imaginary component of @xmath97 , as expected . in order to produce a properly defined second order potential @xmath98 we must then subtract this second order born term from the second order transition amplitude , yielding the result @xmath99\nonumber\\ & = & \int{d^3q\over ( 2\pi)^3}e^{-i\vec{q}\cdot\vec{r } } \hspace*{1pt } g^2 m_a m_b\left[-6s(m_a+m_b ) + { 41\over 5}l\right]\nonumber\\ & = & -{3g^2m_am_b(m_a+m_b)\over r^2 } -{41g^2m_am_b\hbar\over 10\pi r^3}\nonumber\\ \quad\label{eq : so}\end{aligned}\ ] ] the quantum mechanical@xmath100component of the second order potential given in eq . ( [ eq : so ] ) agrees with that previously given by bjerrum - bohr , donoghue , and holstein @xcite and by kirilin and khriplovich @xcite . however , the classical@xmath101contribution quoted by iwasaki @xmath102 differs from that quoted above in eq . ( [ eq : so ] ) and by bjerrum - bohr et al . in @xcite . the resolution of this issue was given by sucher , who pointed out that the classical term depends upon the precise definition of the first order potential used in the iteration @xcite . moreover , it depends and on whether one uses relativistic forms of the leading order potentials and the propagator @xmath103 in the iteration . in modern terms , the potential depends on how one performs the matching_e.g . _ , iwasaki @xcite performs an off - shell matching while we match on - shell . in our companion paper on electromagnetic scattering @xcite we have provided a more detailed discussion of these ambiguities . use of the simple lowest order form eq . ( [ eq_lopot0momsp ] ) within a nonrelativistic iteration yields our result for the iteration amplitude given in eq . ( [ eq : iteration00b ] ) and is sufficient to remove the offending imaginary piece of the scattering amplitude . in appendix [ app_eom ] we derive an alternative form of the @xmath12 classical potential which results from an iteration that includes the leading relativistic corrections and which reproduces the classical equations of motion . therefore , a unique definition of the potential does not exist . of course , the ambiguities in the form of the second order classical potential should not be a concern , since the potential is _ not _ an observable . what _ is _ an observable is the on - shell transition amplitude , which is uniquely defined in each case as @xmath104+{}^0{\rm amp}_i(\vec{q})\ ] ] where the index @xmath105 denotes differing possible definitions of the potentials and the iteration . thus we regard our potential as a nice way to display our resulting scattering amplitudes in coordinate space , but we emphasize that our main results are the long distance components of the scattering amplitude . having determined the form of the potential for the spinless scattering case we move on to the case of scattering of particles carrying spin . we begin with the scattering of a spinless particle @xmath16 from a spin-1/2 particle @xmath19 . for the case of spin-1/2 we require some additional formalism in order to extract the gravitational couplings , which is necessary because the dirac algebra @xmath106 is defined with respect to the minkoswki flat space metric . in this case the dirac matter lagrangian coupled to gravity reads @xmath107\psi\ ] ] and involves the vierbein @xmath108 which links global coordinates with those in a locally flat space . the vierbein is in some sense the `` square root '' of the metric tensor @xmath109 and satisfies the relations @xmath110 the covariant derivative is @xmath111 with @xmath112 $ ] and the partial derivative @xmath113 acts only on spinors and in such a way that @xmath114 putting everything together , we find then @xmath115\psi.\ ] ] the spin connection @xmath116 can be derived in terms of vierbeins by requiring @xmath117 and by antisymmetrization in @xmath118 in order to get rid of christoffel symbols . the result is : @xmath119 in order to derive the feynman rules we expand the ingredients in eq . ( [ lagrangian12 ] ) that contain graviton couplings , that is we need @xmath108 and @xmath116 expanded up to @xmath120 @xmath121 after these expansions are employed , we no longer need to distinguish between latin lorentz indeces and greek covariant indices and can use the minkowski metric to lower and raise all indices . the matter lagrangian then has the expansion(note here that our conventions are @xmath122 and @xmath123 ) @xmath124 and the corresponding one- and two - graviton vertices are found to be @xmath125\nonumber\\ { } ^{\frac{1}{2}}\tau_{\mu\nu,\rho\sigma}^{(2)}(p_2 , p_1 , m ) \hspace*{-8pt } & = \hspace*{-7.5pt } & i\kappa^2\bigg [ \hspace*{-1pt}-{1\over 2}\bigg({1\over 2}(\not \!{p}_1 \hspace*{1pt } + \ ! \not\!{p}_2)-m\bigg ) \hspace*{1pt } p_{\mu\nu,\rho\sigma}\nonumber\\ & & \hspace*{20.3pt } { } - { 1\over 16}\bigg[\eta_{\mu\nu } \big(\gamma_\rho(p_1+p_2)_\sigma+\gamma_\sigma(p_1+p_2)_\rho\big ) \nonumber\\ & & \hspace*{43.5pt } { } + \hspace*{-1pt } \eta_{\rho\sigma}\big(\gamma_\mu ( p_1+p_2)_\nu+\gamma_\nu(p_1+p_2)_\mu\big)\bigg ] \nonumber\\ & & \hspace*{20.3pt } { } + { 3\over 16}(p_1+p_2)^{\epsilon}\rho^{\xi}(i_{\xi\phi,\mu\nu}{i^{\phi}}_{\epsilon,\rho\sigma } + i_{\xi\phi,\rho\sigma}{i^{\phi}}_{\epsilon,\mu\nu } ) \nonumber\\ & & \hspace*{20.3pt } { } + { i\over 16}\epsilon^{\epsilon\phi\eta\lambda}\gamma_\lambda \gamma_5 \big(i_{\rho\sigma,\phi\xi } { i_{\mu\nu,\eta}}^\xi \hspace*{1pt } { k}_\epsilon - i_{\mu\nu,\phi\xi } { i_{\rho\sigma,\eta}}^\xi \hspace*{1pt } ( k+q)_\epsilon \big)\bigg ] . \nonumber \\ \quad\end{aligned}\ ] ] the tree level transition amplitude from one - graviton exchange is then @xmath126 . \quad\end{aligned}\ ] ] defining the spin vector as @xmath127 we find the identity @xmath128\label{eq : id}\ ] ] whereupon the nonanalytic part of the transition amplitude in the thereshold limit @xmath129 can be written in the form @xmath130.\ ] ] in order to define the potential we require the nonrelativistic amplitude in the symmetric center of mass frame ( @xmath131 ) where @xmath132 @xmath133 and @xmath134 so that @xmath135\ ] ] and the lowest order potential becomes @xmath136 where @xmath137 is the angular momentum the modification of the leading spin - independent potential has a spin - orbit character . in order to determine the second order potential , we must evaluate the one loop diagrams in fig . [ fig_loops ] . a subtlety that arises in the calculation involving spin is that _ two _ independent kinematic variables arise : the momentum transfer @xmath67 and @xmath138 , which is to leading order proportional to @xmath139 ( where @xmath140 ) in the center of mass frame . we find that our results differ if we perform an expansion first in @xmath138 and then in @xmath67 or vice versa . this ordering issue occurs only for the box diagram , diagram ( d ) of fig . [ fig_loops ] , where it stems from the reduction of vector and tensor box integrals . their reduction in terms of scalar integrals involves the inversion of a matrix whose gram determinant vanishes in the nonrelativistic threshold limit @xmath141 . more precisely , the denominators or the vector and tensor box integrals ( see appendix a in @xcite ) involve a factor of @xmath142 when expanded in the nonrelativistic limit . since @xmath143 with @xmath144 the scattering angle , we notice that @xmath145 unless we consider backward scattering where @xmath146 and where the scattering amplitude diverges . and since @xmath139 originates from the relativistic structure @xmath138 , it is clear that we must first expand our vector and tensor box integrals in @xmath67 and then in @xmath147 . with this procedure , the contributions of the loop diagrams in fig . [ fig_loops ] to the spin-0 spin-1/2 scattering amplitude read @xmath148\nonumber\\ { } ^{1\over 2}{\cal m}_{\ref{fig_loops}b}^{(2)}(q ) \hspace*{-5pt } & = \hspace*{-5pt}&g^2 m_a m_b \bigg[\bar{u}(p_4 ) u(p_3 ) ( - m_a s + 3 l ) \nonumber\\ & & \hspace*{41pt } + { 1\over m_a}\bar{u}(p_4 ) \ ! \not \!{p}_1u(p_3 ) ( 8 m_a s + 20 l ) \bigg]\nonumber\\ { } ^{1\over 2}{\cal m}_{\ref{fig_loops}c}^{(2)}(q ) \hspace*{-5pt } & = \hspace*{-5pt}&g^2 m_a m_b \bigg[\bar{u}(p_4 ) u(p_3 ) ( 4 m_b s + 2 l ) \nonumber\\ & & \hspace*{41pt } + { 1\over m_a}\bar{u}(p_4 ) \ ! \not \!{p}_1u(p_3 ) ( 4 m_b s + 16 l ) \bigg]\nonumber\\ { } ^{1\over 2}{\cal m}_{\ref{fig_loops}d}^{(2)}(q ) \hspace*{-5pt } & = \hspace*{-5pt}&g^2 m_a m_b \bigg[\bar{u}(p_4)u(p_3 ) \hspace*{-1pt } \bigg ( \ ! - \frac{4 m_a m_b}{q^2 } l- { m_am_b(3 m_a+4 m_b)\over s - s_0}s \nonumber\\ & & \hspace*{98pt } - \hspace*{0.9pt } ( 6 m_a \hspace*{-2pt } + \hspace*{-2pt } 11 m_b ) s \hspace*{-1pt } - \hspace*{-1pt } \frac{42 m_a^2 \hspace*{-2pt } + \hspace*{-2pt } 23m_am_b \hspace*{-2pt } - \hspace*{-2pt } 4 m_b^2}{6m_am_b } l \hspace*{-1pt}\bigg)\nonumber\\ & & \hspace*{41pt } + { 1\over m_a}\bar{u}(p_4 ) \ ! \not \!{p}_1u(p_3 ) \hspace*{-1pt}\bigg({8m_am_b\over q^2 } l + { m_am_b(3 m_a + 4 m_b)\over s - s_0 } s\nonumber\\ & & \hspace*{131pt } + ( 9 m_a+ 13 m_b ) s \nonumber\\ & & \hspace*{131pt } + \frac{153 m_a^2 - 44 m_a m_b + 88 m_b^2}{12 m_a m_b } l \bigg ) \bigg]\nonumber\\ & & - \ i 4\pi g^2 m_a^2 m_b^2 \hspace*{1pt } { l\over q^2}\sqrt{m_a m_b\over s - s_0 } \left(2\ , { 1\over m_a}\bar{u}(p_4 ) \ ! \not \!{p}_1u(p_3 ) - \bar{u}(p_4)u(p_3)\right)\nonumber\\ { } ^{1\over 2}{\cal m}_{\ref{fig_loops}e}^{(2)}(q ) \hspace*{-5pt } & = \hspace*{-5pt}&g^2 m_a m_b \bigg[\bar{u}(p_4)u(p_3 ) \hspace*{-1pt } \bigg ( \frac{4 m_a m_b}{q^2 } l\nonumber\\ & & \hspace*{98pt } + \frac{19 m_a}{4 } s + \frac{42 m_a^2 \hspace*{-1pt } + \hspace*{-1pt } 21m_am_b \hspace*{-1pt } - \hspace*{-1pt } 4 m_b^2}{6m_am_b } l \hspace*{-1pt}\bigg)\nonumber\\ & & \hspace*{41pt } + { 1\over m_a}\bar{u}(p_4 ) \ ! \not \!{p}_1u(p_3 ) \hspace*{-1pt}\bigg(- { 8m_am_b\over q^2 } l - \frac{33 m_a + 16 m_b}{4 } s \nonumber\\ & & \hspace*{131pt } - \frac{153 m_a^2 + 268 m_a m_b + 88 m_b^2}{12 m_a m_b } l \bigg ) \bigg]\nonumber\\ { } ^{1\over 2}{\cal m}_{\ref{fig_loops}f}^{(2)}(q ) \hspace*{-5pt } & = \hspace*{-5pt}&g^2 m_a m_b \bigg[\bar{u}(p_4 ) u(p_3 ) \bigg ( \ ! - \frac{3 m_a}{2 } s + 10 l\bigg ) \nonumber\\ & & \hspace*{41pt } + { 1\over m_a}\bar{u}(p_4 ) \ ! \not \!{p}_1u(p_3 ) \bigg ( \ ! - \frac{m_a + 4 m_b}{2 } s + 4 l \bigg ) \bigg]\nonumber\\ { } ^{1\over 2}{\cal m}_{\ref{fig_loops}g}^{(2)}(q ) \hspace*{-5pt } & = \hspace*{-5pt}&g^2 m_a m_b \bigg[\bar{u}(p_4 ) u(p_3 ) \hspace*{-0.5pt } \bigg(\hspace*{-2.5pt } - \hspace*{-1pt}\frac{1}{15 } l\bigg ) \hspace*{-1.5pt } + \hspace*{-0.5pt}{1\over m_a}\bar{u}(p_4 ) \ ! \not \!{p}_1u(p_3 ) \hspace*{-0.5pt } \bigg ( \hspace*{-2.5pt } - \hspace*{-1pt } \frac{42}{15}l\bigg ) \hspace*{-1pt}\bigg].\end{aligned}\ ] ] again , we have calculated the form factor diagrams in fig . [ fig_loops](f ) both directly and via the results from @xcite . summing , we find @xmath149\nonumber\\ & & - \ i 4\pi g^2 m_a^2 m_b^2 \hspace*{1pt } { l\over q^2}\sqrt{m_a m_b\over s - s_0 } \left(- \bar{u}(p_4)u(p_3 ) + 2\ , { 1\over m_a}\bar{u}(p_4 ) \ ! \not \!{p}_1u(p_3 ) \right)\nonumber\\ \quad \label{eq : su}\end{aligned}\ ] ] using the identity eq . ( [ eq : id ] ) and @xmath150 eq . ( [ eq : su ] ) becomes @xmath151\nonumber\\ & & -i4\pi g^2 m_a^2 m_b^2 \hspace*{1pt } { l\over q^2}\sqrt{m_a m_b\over s - s_0}\left(\bar{u}(p_4)u(p_3)+ { 2 i\over m_am_b^2}\epsilon_{\alpha\beta\gamma\delta } p_1^\alpha p_3^\beta q^\gamma s_b^\delta\right ) \nonumber\\ \quad \label{eq : ohla}\end{aligned}\ ] ] finally , working in the center of mass frame and taking the nonrelativistic limit , we find @xmath152 \chi_f^{b\dagger}\chi_i^b \nonumber\\ & + \hspace*{-5pt}&\bigg[g^2 \hspace*{-1pt } \left(\frac{12m_a^3 \hspace*{-1pt } + \hspace*{-1pt } 45 m_a^2 m_b \hspace*{-1pt } + \hspace*{-1pt } 56 m_a m_b^2 \hspace*{-1pt } + \hspace*{-1pt } 24 m_b^3}{2(m_a \hspace*{-1pt } + \hspace*{-1pt } m_b ) } \hspace*{1pt } s - \frac{87 m_a \hspace*{-1pt } + \hspace*{-1pt } 128 m_b}{10 } \hspace*{1pt } l\right)\nonumber\\ & & + \frac{g^2 m_a^2 m_b^2 ( 3 m_a + 4 m_b)}{(m_a + m_b ) } \left(- i \frac{2 \pi l}{p_0 q^2 } + \frac{s}{p_0 ^ 2 } \right)\bigg ] { i\over m_b}\vec{s}_b\cdot\vec{p}\times\vec{q}\label{eq : yo}\end{aligned}\ ] ] we note from eq . ( [ eq : yo ] ) that the scattering amplitude consists of two pieces a spin - independent component proportional to @xmath153 whose functional form @xmath154 is _ identical _ to that of spinless scattering together with a spin - orbit component proportional to @xmath155 whose functional form is @xmath156 we note in eq . ( [ eq : im ] ) the presence in the spin - orbit potential of an imaginary final state rescattering term proportional to @xmath157 , similar to that found in the case of spin - independent scattering , together with a _ completely new _ type of kinematic form , proportional to @xmath158 which diverges at threshold . the presence of _ either _ term would prevent us from writing down a well defined second order potential . the solution to this problem is , as before , to properly subtract the iterated first order potential @xmath159 where we now use the one - graviton exchange potential @xmath160 given in eq . ( [ eq : oj ] ) . splitting this lowest order potential into spin - independent and spin - dependent components @xmath161 where @xmath162 we find that the iterated amplitude splits also into spin - independent and spin - dependent pieces . the leading spin - independent amplitude is @xmath163 where we defined @xmath164 , and the leading spin - dependent term is @xmath165 ( in principle we would also have to iterate the leading order spin - orbit piece twice . however this procedure yields only terms higher order in @xmath139 . ) we observe that when the amplitudes eqs . ( [ eq : iteration0hso ] ) and ( [ eq : iteration0hsi ] ) are subtracted from the full one loop scattering amplitude eq . ( [ eq : yo ] ) both the terms involving @xmath158 and those proportional to @xmath157 disappear leaving behind a well - defined second order potential @xmath166\nonumber\\ & = & \int{d^3q\over ( 2\pi)^3}e^{-i\vec{q}\cdot\vec{r } } \bigg[g^2 m_a m_b \left(- 6(m_a+m_b)s+{41\over 5}l\right ) \chi_f^{b\dagger}\chi_i^b\nonumber\\ & & { } \hspace*{63pt } + g^2 \bigg ( \hspace*{-2pt } - \frac{12m_a^3 + 45 m_a^2 m_b + 56 m_a m_b^2 + 24 m_b^3}{2(m_a + m_b ) } \hspace*{1pt } s \nonumber\\ & & { } \hspace*{91pt } + \frac{87 m_a + 128 m_b}{10 } \hspace*{1pt } l\bigg ) \hspace*{1pt } { i\over m_b}\vec{s}_b\cdot\vec{p}\times\vec{q } \hspace*{1pt } \bigg]\nonumber\\ & = & \left[-{3g^2m_am_b(m_a+m_b)\over r^2 } - { 41g^2m_am_b\hbar\over 10\pi r^3}\right ] \chi_f^{b\dagger}\chi_i^b\nonumber\\ & + & { 1\over m_b}\vec{s}_b\cdot \vec{p}\times\vec{\nabla}\bigg [ \frac{g^2(12m_a^3 + 45 m_a^2 m_b + 56 m_a m_b^2 + 24 m_b^3)}{4(m_a + m_b)r^2 } \nonumber\\ & & \hspace*{66pt } + \frac{g^2(87 m_a + 128 m_b ) \hbar}{20 \pi r^3 } \bigg]\nonumber\\ & = & \left[-{3g^2m_am_b(m_a+m_b)\over r^2 } - { 41g^2m_am_b\hbar\over 10\pi r^3}\right ] \chi_f^{b\dagger}\chi_i^b\nonumber\\ & + & \bigg[{g^2(12 m_a^3 + 45 m_a^2 m_b + 56 m_a m_b^2 + 24 m_b^3)\over 2 m_b ( m_a+m_b)r^4 } \nonumber \\ & & + { 3 g^2(87 m_a+ 128m_b)\hbar\over 20 \pi m_b r^5}\bigg ] \vec{l}\cdot\vec{s}_b \label{eq : mn}\end{aligned}\ ] ] we observe then that the second order potential for long range gravitational scattering of a spinless and spin-1/2 particle consists of two components : one which is independent of the spin of particle b and is identical to the potential found for the case of spinless scattering , accompanied by a spin - orbit interaction involving a new form for its classical and quantum potentials . it is tempting to speculate that the form of this new spin - orbit potential is also universal . in order to check this hypothesis we consider the case of spin-0 spin-1 scattering . the dynamics of a neutral spin-1 field @xmath167 having mass @xmath168 is described by the proca lagrangian which , when coupled to gravity via minimal substitution , takes the form @xmath169 \label{eqn : la}\ ] ] where @xmath170 is the spin-1 field tensor . the last equality in eq . ( [ eq : fieldtens ] ) follows from the symmetry of the connection coefficients @xmath171 . expanded in terms of the graviton field , the matter lagrangian then has the form @xmath172 and the one- and two - graviton vertices are @xmath173 \nonumber \\ & + \hspace*{-7.5pt } & i \kappa \hspace*{1pt } i_{\mu \nu , \kappa \lambda } \bigg[(p_1 \cdot p_2 - m^2 ) { i_{\alpha \beta,}}^{\kappa \lambda } + \frac{1}{2 } \big(p_1^\kappa p_2^\lambda + p_1^\lambda p_2^\kappa\big ) \eta_{\alpha \beta } \nonumber \\ & & \hspace*{39pt } - \big(p_1^\kappa p_{2 \alpha } \delta_\beta^\lambda + p_2^\kappa p_{1 \beta } \delta_\alpha^\lambda \big)\bigg ] \nonumber \\ { } ^1 \tau^{(2)}_{\beta,\alpha,\mu\nu , \rho \sigma}(p_2,p_1 , m)\hspace*{-7.5pt } & = \hspace*{-7.5pt } & { i \kappa^2 \over 2 } \hspace*{1pt } p_{\mu \nu , \rho \sigma } \big[(p_1 \cdot p_2 - m^2 ) \eta_{\alpha \beta } - p_{1 \beta } p_{2 \alpha } \big ] \nonumber \\ & - \hspace*{-7.5pt } & i \kappa^2 \hspace*{-2.6pt } \left ( \hspace*{-1.8pt } { i_{\mu \nu,}}^{\kappa \delta } { i_{\rho \sigma , \delta}}^{\lambda } \hspace*{-1.9pt } + \hspace*{-1.9pt } { i_{\rho \sigma,}}^{\kappa \delta } { i_{\mu \nu , \delta}}^{\lambda } \hspace*{-1pt } - \hspace*{-1pt } \frac{1}{2 } \big(\eta_{\mu \nu } { i_{\rho \sigma,}}^{\kappa \lambda } \hspace*{-1pt } + \hspace*{-1pt } \eta_{\rho \sigma } { i_{\mu \nu,}}^{\kappa \lambda}\big ) \ ! \hspace*{-1pt } \right ) \nonumber \\ & & \hspace*{30pt } \times \bigg[(p_1 \cdot p_2 - m^2 ) { i_{\alpha \beta,}}^{\kappa \lambda } + \frac{1}{2 } \big(p_1^\kappa p_2^\lambda + p_1^\lambda p_2^\kappa\big ) \eta_{\alpha \beta } \nonumber \\ & & \hspace*{39pt } - \big(p_1^\kappa p_{2 \alpha } \delta_\beta^\lambda + p_2^\kappa p_{1 \beta } \delta_\alpha^\lambda \big)\bigg ] \nonumber \\ & - \hspace*{-7.5pt } & \frac{i \kappa^2 } { 2 } \hspace*{-1pt } \left({i_{\mu \nu,}}^{\eta \theta } { i_{\rho \sigma,}}^{\kappa \lambda } + { i_{\rho \sigma,}}^{\eta \theta } { i_{\mu \nu,}}^{\kappa \lambda } \right ) \nonumber \\ & & \hspace*{16pt } \times \hspace*{-0.5pt } \bigg[p_{1 \eta } \eta_{\alpha \kappa } ( p_{2 \theta } \eta_{\beta \lambda } \hspace*{-2.4pt } - \hspace*{-2pt } p_{2 \lambda } \eta_{\beta \theta}\hspace*{-1.3pt } ) \hspace*{-1.7pt } + \hspace*{-1.7pt } p_{2 \eta } \eta_{\beta \kappa } ( p_{1 \theta } \eta_{\alpha \lambda } \hspace*{-2.4pt } - \hspace*{-2pt } p_{1 \lambda } \eta_{\alpha \theta}\hspace*{-1.3pt } ) \hspace*{-1pt } \bigg ] . \nonumber \\\end{aligned}\ ] ] if we take the incoming spin-1 particle to have polarization vector @xmath174 satisfying @xmath175 and the outgoing particle to have polarization @xmath176 satisfying @xmath177 , then the one - graviton exchange amplitude can then be written as @xmath178\nonumber \\ & \simeq & - \frac{4 \pi g m_a m_b}{q^2}\bigg [ - \epsilon_f^{b*}\cdot\epsilon_i^b - \frac{2}{m_a m_b } \left(\epsilon_f^{b*}\cdot q\epsilon_i^b\cdot p_1-\epsilon_f^{b*}\cdot p_1\epsilon_i^b\cdot q \right ) \nonumber \\ & & \hspace*{72pt } + \hspace*{0.5pt } \frac{1}{m_b^2 } \hspace*{1pt } \epsilon_f^{b*}\cdot q\epsilon_i^b\cdot q \bigg ] \quad\end{aligned}\ ] ] where the first expression is exact while the second expression contains only the nonanalytic part in the threshold limit @xmath179 . now we rewrite this expression using the identity @xmath180\label{eq : kj}\ ] ] where we have defined the spin vector @xmath181 the leading one - graviton exchange amplitude can then be written as @xmath182\ ] ] in the nonrelativistic limit we have @xmath183 so that @xmath184 since @xmath185 eq . ( [ eq : ss ] ) becomes @xmath186 and the transition amplitude assumes the form @xmath187\end{aligned}\ ] ] the spin - independent and spin - orbit terms here are identical in form to those found in the spin-0 spin-1/2 case but now are accompanied by new terms which are quadrupole in nature , as can be seen from the identity @xmath188 the corresponding lowest order potential is then @xmath189 where we have defined @xmath190 and which agrees precisely with its spin-1/2 analog in eq . ( [ eq : oj ] ) up to quadrupole and tensor corrections . the calculation of the one loop corrections proceeds as before , but with increased complexity due to the unit spin . evaluating the diagrams in fig . [ fig_loops ] , we find then @xmath191 \nonumber\\ { } ^1{\cal m}^{(2)}_{\ref{fig_loops}b}(q ) \hspace*{-5pt } & = \hspace*{-5pt } & g^2 m_a m_b \bigg[(- 4 m_a s - 16 l ) \epsilon_f^{*b } \cdot \epsilon_i^b + \frac{8 m_a s + 16 l}{m_a^2 } \epsilon_f^{b*}\cdot p_1 \epsilon_i^b \cdot p_1 \nonumber\\ & & \hspace*{41pt } - \frac{8 l}{m_a m_b } ( \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot p_1 - \epsilon_f^{b * } \cdot p_1 \epsilon_i^b\cdot q ) \nonumber\\ & & \hspace*{41pt } - \frac{4 m_a s + 8 l}{m_a^2 } ( \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot p_1 + \epsilon_f^{b * } \cdot p_1 \epsilon_i^b\cdot q ) \nonumber\\ & & \hspace*{41pt } - \left(\left(2 m_a - \frac{3 m_b^2}{2 m_a}\right)s + \left(2 - \frac{2 m_b^2}{m_a^2}\right)l \right ) \frac{1}{m_b^2 } \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot q \bigg ] \nonumber\\ { } ^1{\cal m}^{(2)}_{\ref{fig_loops}c}(q ) \hspace*{-5pt } & = \hspace*{-5pt } & g^2 m_a m_b \bigg[(- 8 m_b s - 12 l ) \epsilon_f^{*b } \cdot \epsilon_i^b + \frac{16 l}{m_a^2 } \epsilon_f^{b*}\cdot p_1 \epsilon_i^b \cdot p_1 \nonumber\\ & & \hspace*{41pt } - \frac{4 m_b s + 8 l}{m_a m_b } ( \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot p_1 - \epsilon_f^{b * } \cdot p_1 \epsilon_i^b\cdot q ) \nonumber\\ & & \hspace*{41pt } - \frac{8 l}{m_a^2 } ( \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot p_1 + \epsilon_f^{b * } \cdot p_1 \epsilon_i^b\cdot q ) \nonumber\\ & & \hspace*{41pt } - \frac{m_b s + 8 l}{m_b^2 } \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot q \bigg ] \nonumber\\ { } ^1{\cal m}^{(2)}_{\ref{fig_loops}d}(q ) \hspace*{-5pt } & = \hspace*{-5pt } & g^2 m_a m_b \bigg [ \frac{4 m_a m_b } { q^2 } l \bigg ( \hspace*{-2.5pt } - \hspace*{-0.5pt } \epsilon_f^{*b } \hspace*{-1pt } \cdot \hspace*{-0.5pt } \epsilon_i^b \hspace*{-0.5pt } - \frac{2}{m_a m_b } ( \epsilon_f^{b * } \hspace*{-1pt } \cdot \hspace*{-0.5pt } q \epsilon_i^b \hspace*{-1pt } \cdot \hspace*{-0.5pt } p_1 \hspace*{-1pt } - \epsilon_f^{b * } \hspace*{-1pt } \cdot \hspace*{-0.5pt } p_1 \epsilon_i^b\cdot \hspace*{-0.5pt } q ) \nonumber \\ & & \hspace*{100pt } + \frac{1}{m_b^2 } \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot q\bigg ) \nonumber \\ & & \hspace*{41pt } + \frac{s}{s - s_0 } \bigg ( \hspace*{-2.5pt } -(3m_a + 4 m_b ) ( \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot p_1 - \epsilon_f^{b * } \cdot p_1 \epsilon_i^b\cdot q)\nonumber\\ & & \hspace*{88pt } + \frac{m_a ( 5 m_a + 7 m_b)}{2 m_b } \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot q \bigg)\nonumber\\ & & \hspace*{41pt } - \bigg((6 m_a + 4 m_b ) s + \frac{12 m_a^2 - 6 m_a m_b + 24 m_b^2}{3 m_a m_b } l \bigg ) \epsilon_f^{*b } \cdot \epsilon_i^b \nonumber\\ & & \hspace*{41pt } + \bigg(- 4 m_a s + \frac{22 m_a - 24 m_b}{3 m_b } l \bigg ) \frac{1}{m_a^2 } \epsilon_f^{b*}\cdot p_1 \epsilon_i^b \cdot p_1 \nonumber \\ & & \hspace*{41pt } - \bigg(13 ( m_a + m_b ) s + \frac{25 m_a^2 + 7 m_a m_b + 22 m_b^2}{3 m_a m_b } l \bigg ) \nonumber \\ & & \hspace*{59pt } \times \frac{1}{m_a m_b } ( \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot p_1 - \epsilon_f^{b * } \cdot p_1 \epsilon_i^b\cdot q ) \nonumber \\ & & \hspace*{41pt } + \bigg(\hspace*{-1pt } 2 m_a s \hspace*{-1.5pt } - \hspace*{-0.8pt } \frac{11 m_a \hspace*{-1.9pt } - \hspace*{-1.4pt } 12 m_b}{3 m_b } l \hspace*{-1pt } \bigg ) \frac{1}{m_a^2 } ( \epsilon_f^{b * } \hspace*{-1.5pt } \cdot \hspace*{-0.8pt } q \epsilon_i^b \hspace*{-1.5pt } \cdot \hspace*{-0.8pt } p_1 \hspace*{-1pt } + \hspace*{-0.7pt } \epsilon_f^{b * } \hspace*{-1.5pt } \cdot \hspace*{-0.8pt } p_1 \epsilon_i^b \hspace*{-1.5pt } \cdot \hspace*{-0.8pt } q)\nonumber \\ & & \hspace*{41pt } + \bigg(\!\bigg ( \hspace*{-0.5pt } 7 m_a \hspace*{-1pt } + \hspace*{-1pt } 9 m_b \hspace*{-1pt } - \hspace*{-1pt } \frac{3 m_b^2}{4 m_a}\bigg)s \hspace*{-0.5pt } + \hspace*{-0.5pt } \bigg ( \hspace*{-0.5pt } 5 \frac{m_a}{m_b } \hspace*{-1pt } + \hspace*{-1pt } 2 \hspace*{-1pt } + \hspace*{-1pt } 9 \frac{m_b}{m_a } \hspace*{-1pt } - \hspace*{-1pt } \frac{m_b^2}{m_a^2}\bigg ) l\bigg ) \nonumber \\ & & \hspace*{59pt } \times \frac{1}{m_b^2 } \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot q \bigg ] \nonumber \\ & - \hspace*{-5pt } & i 4\pi g^2 m_a^2 m_b^2 \hspace*{1pt } { l\over q^2}\sqrt{m_a m_b\over s - s_0 } \hspace*{-1pt } \bigg ( \hspace*{-5pt } - \hspace*{-1.5pt } \epsilon_f^{*b } \hspace*{-2.5pt } \cdot \hspace*{-1.5pt } \epsilon_i^b \hspace*{-1.5pt } - \hspace*{-1.5pt } \frac{2}{m_a m_b } ( \epsilon_f^{b * } \hspace*{-2.5pt } \cdot \hspace*{-1.5pt } q \epsilon_i^b \hspace*{-2.5pt } \cdot \hspace*{-1.5pt } p_1 \hspace*{-1.5pt } - \hspace*{-1pt } \epsilon_f^{b * } \hspace*{-2.5pt } \cdot \hspace*{-1.5pt } p_1 \epsilon_i^b \hspace*{-2.5pt } \cdot \hspace*{-1.5pt } q ) \nonumber \\ & & \hspace*{122pt } + \frac{1}{m_b^2 } \epsilon_f^{b * } \hspace*{-1pt } \cdot \hspace*{-0.5pt } q \epsilon_i^b \hspace*{-1pt } \cdot \hspace*{-0.5pt } q \bigg ) \nonumber \\ { } ^1{\cal m}^{(2)}_{\ref{fig_loops}e}(q ) \hspace*{-5pt } & = \hspace*{-5pt } & g^2 m_a m_b \bigg [ \frac{4 m_a m_b } { q^2 } l \bigg ( \epsilon_f^{*b } \cdot \epsilon_i^b + \frac{2}{m_a m_b } ( \epsilon_f^{b * } \cdot q \epsilon_i^b \cdot p_1 - \epsilon_f^{b * } \cdot p_1 \epsilon_i^b \cdot q ) \nonumber \\ & & \hspace*{100pt } - \frac{1}{m_b^2 } \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot q\bigg ) \nonumber \\ & & \hspace*{41pt } + \bigg((2 m_a + 4 m_b ) s + \frac{12 m_a^2 + 52 m_a m_b + 24 m_b^2}{3 m_a m_b } l \bigg ) \epsilon_f^{*b } \cdot \epsilon_i^b \nonumber\\ & & \hspace*{41pt } - \bigg(4 m_a s + \frac{22 m_a + 24 m_b}{3 m_b } l \bigg ) \frac{1}{m_a^2 } \epsilon_f^{b*}\cdot p_1 \epsilon_i^b \cdot p_1 \nonumber \\ & & \hspace*{41pt } + \bigg(\ ! \bigg(\frac{17}{4}m_a + 4 m_b\bigg ) s + \frac{25 m_a^2 + 49 m_a m_b + 22 m_b^2}{3 m_a m_b } l \bigg ) \nonumber \\ & & \hspace*{59pt } \times \frac{1}{m_a m_b } ( \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot p_1 - \epsilon_f^{b * } \cdot p_1 \epsilon_i^b\cdot q ) \nonumber \\ & & \hspace*{41pt } + \bigg(\hspace*{-1pt } 2 m_a s \hspace*{-1.5pt } + \hspace*{-0.8pt } \frac{11 m_a \hspace*{-1.9pt } + \hspace*{-1.4pt } 12 m_b}{3 m_b } l \hspace*{-1pt } \bigg ) \frac{1}{m_a^2 } ( \epsilon_f^{b * } \hspace*{-1.5pt } \cdot \hspace*{-0.8pt } q \epsilon_i^b \hspace*{-1.5pt } \cdot \hspace*{-0.8pt } p_1 \hspace*{-1pt } + \hspace*{-0.7pt } \epsilon_f^{b * } \hspace*{-1.5pt } \cdot \hspace*{-0.8pt } p_1 \epsilon_i^b \hspace*{-1.5pt } \cdot \hspace*{-0.8pt } q)\nonumber \\ & & \hspace*{41pt } - \bigg(\ ! \hspace*{-1pt } \bigg ( \hspace*{-0.8pt } \frac{13}{8 } m_a \hspace*{-2.5pt } + \hspace*{-2pt } \frac{31}{8 } m_b \hspace*{-2pt } + \hspace*{-2pt } \frac{3 m_b^2}{4 m_a } \hspace*{-0.8pt}\bigg ) \hspace*{-0.8pt}s \hspace*{-2.2pt } + \hspace*{-2.5pt } \bigg ( \hspace*{-2pt } 5 \frac{m_a}{m_b } \hspace*{-2pt } + \hspace*{-2pt } \frac{16}{3 } \hspace*{-2pt } + \hspace*{-2pt } 9 \frac{m_b}{m_a } \hspace*{-2pt } + \hspace*{-2pt } \frac{m_b^2}{m_a^2 } \hspace*{-0.5pt}\bigg ) \hspace*{-0.5pt}l \hspace*{-1.5pt}\bigg ) \nonumber \\ & & \hspace*{59pt } \times \frac{1}{m_b^2 } \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot q \bigg ] \nonumber \\ { } ^1{\cal m}^{(2)}_{\ref{fig_loops}f}(q ) \hspace*{-5pt } & = \hspace*{-5pt } & g^2 m_a m_b \bigg[(2(m_a + m_b ) s - 14 l ) \epsilon_f^{*b } \cdot \epsilon_i^b \nonumber\\ & & \hspace*{41pt } + \bigg(\!\bigg(\frac{1}{2 } m_a \hspace*{-2pt } + \hspace*{-1pt } 2 m_b\hspace*{-2pt}\bigg ) s \hspace*{-1.5pt}-\hspace*{-1.5pt } 4 l \hspace*{-2pt}\bigg ) \frac{1}{m_a m_b } ( \epsilon_f^{b * } \hspace*{-2pt } \cdot \hspace*{-1pt } q \epsilon_i^b \hspace*{-1pt } \cdot \hspace*{-1pt } p_1 \hspace*{-1.5pt } - \hspace*{-1pt } \epsilon_f^{b * } \hspace*{-2pt } \cdot \hspace*{-1pt } p_1 \epsilon_i^b \hspace*{-1pt}\cdot \hspace*{-1pt } q ) \nonumber\\ & & \hspace*{41pt } + \bigg ( \ ! \bigg ( \hspace*{-1.5pt } - \frac{1}{4 } m_a + \frac{1}{2 } m_b \bigg ) s + 4 l \bigg ) \frac{1}{m_b^2 } \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot q \bigg ] \nonumber\\ { } ^1{\cal m}^{(2)}_{\ref{fig_loops}g}(q ) \hspace*{-5pt } & = \hspace*{-5pt } & g^2 m_a m_b \bigg[\frac{43 l}{15 } \epsilon_f^{*b } \cdot \epsilon_i^b + \frac{14 l}{5 m_a m_b } ( \epsilon_f^{b * } \cdot q \epsilon_i^b \cdot p_1 - \epsilon_f^{b * } \cdot p_1 \epsilon_i^b \cdot q ) \nonumber\\ & & \hspace*{41pt } - \frac{7 l}{5 m_b^2 } \epsilon_f^{b*}\cdot q \epsilon_i^b \cdot q \bigg]\end{aligned}\ ] ] combining , we find the full one loop amplitude @xmath192\end{aligned}\ ] ] which , using the identity eq . ( [ eq : kj ] ) , becomes @xmath193\nonumber\\ & - \hspace*{-5pt } & i 4\pi g^2 m_a^2 m_b^2 \hspace*{1pt } { l\over q^2}\sqrt{m_a m_b\over s - s_0 } \bigg ( \hspace*{-3pt } - \epsilon_f^{*b } \cdot \epsilon_i^b + \frac{2i}{m_a m_b } \hspace*{1pt } \epsilon_{\alpha\beta\gamma\delta}\hspace*{1pt } p_1^\alpha p_3^\beta q^\gamma s_b^\delta \nonumber \\ & & \hspace*{125pt } - \hspace*{0.5pt } \frac{1}{m_b^2 } \hspace*{1pt } \epsilon_f^{b * } \cdot q \epsilon_i^b \cdot q \bigg ) \label{eq : m2spin1kir}\end{aligned}\ ] ] notice here that without the @xmath194 terms , eq . ( [ eq : m2spin1kir ] ) has an identical structure to that of the case of spin-0 spin-1/2 scattering eq . ( [ eq : ohla])provided we substitute @xmath195 . finally , taking the nonrelativistic limit we find @xmath196 \nonumber \\ & & \hspace*{15pt } \times \left(\hat{\epsilon}_f^{b * } \cdot \hat{\epsilon}_i^b - { 1\over m_b^2}\ , \vec p \hspace*{-0.5pt } : \hspace*{-1pt } t^b \hspace*{-2pt } : \hspace*{-1pt}\vec p\right ) \nonumber\\ & + \hspace*{-6pt}&\bigg[g^2 \hspace*{-1pt } \left(\frac{12m_a^3 \hspace*{-1pt } + \hspace*{-1pt } 45 m_a^2 m_b \hspace*{-1pt } + \hspace*{-1pt } 56 m_a m_b^2 \hspace*{-1pt } + \hspace*{-1pt } 24 m_b^3}{2(m_a \hspace*{-1pt } + \hspace*{-1pt } m_b ) } \hspace*{1pt } s - \frac{87 m_a \hspace*{-1pt } + \hspace*{-1pt } 128 m_b}{10 } \hspace*{1pt } l\right)\nonumber\\ & & + \frac{g^2 m_a^2 m_b^2 ( 3 m_a + 4 m_b)}{(m_a + m_b ) } \left(- i \frac{2 \pi l}{p_0 q^2 } + \frac{s}{p_0 ^ 2 } \right)\bigg ] { i\over m_b}\vec{s}_b\cdot\vec{p}\times\vec{q } \nonumber \\ & + \hspace*{-6pt}&\bigg[g^2 m_a m_b \left(- \frac{21 m_a^2 + 47 m_a m_b + 28 m_b^2}{4 ( m_a + m_b ) } \ , s + \frac{241}{60 } \hspace*{1pt } l \right ) \nonumber \\ & & + \frac{g^2 m_a^3 m_b^3}{2(m_a + m_b ) } \left(i \frac{6 \pi l}{p_0 q^2 } - \frac{s}{p_0 ^ 2}\right ) \bigg ] \frac{1}{m_b^2 } \ , \vec q : t^b : \vec q \label{eq : yo2}\end{aligned}\ ] ] as found in the earlier calculations , there exist terms involving both @xmath157 and @xmath158 which prevent the defining of a simple second order potential . the solution now is well known subtraction of the iterated first order potential . since the form of the spin - independent@xmath197and spin - orbit@xmath198terms is identical to that found for the case of spin-1/2 , it is clear that the subtraction goes through as before and that the corresponding pieces of the second order potential have the _ same _ form as found for spin-1/2 . in addition , there are now two new pieces of the amplitude , the quadrupole structure @xmath199 which multiplies terms involving both @xmath157 and @xmath158 and the tensor structure @xmath200 multiplying only @xmath157 . in order to remove these we must iterate the full first order potential including these quadrupole and tensor components . however , we find that our simple nonrelativistic iteration fails to remove them ! we suspect the reason to be the presence of the tensor structure @xmath200 in the lowest order potential which is in some sense a relativistic correction but which when iterated yields also quadrupole pieces @xmath199 . a fully relativistic iteration should thus be performed which is under study but is beyond the scope of this paper . it would be interesting to investigate if the requirement of canceling all @xmath157 and @xmath158 forms in the quadrupole and tensor pieces could clarify the ambiguity in the iteration of the leading order potential as discussed in @xcite . since we did not perform the proper iteration of the quadrupole and tensor pieces we include only the spin independent and spin - orbit pieces in the resulting second order potential @xmath201\nonumber\\ & = & \left[-{3g^2m_am_b(m_a+m_b)\over r^2 } - { 41g^2m_am_b\hbar\over 10\pi r^3}\right ] \hat{\epsilon}_f^{b * } \cdot \hat{\epsilon}_i^b \nonumber\\ & + & { 1\over m_b}\vec{s}_b\cdot \vec{p}\times\vec{\nabla}\bigg [ \frac{g^2(12m_a^3 + 45 m_a^2 m_b + 56 m_a m_b^2 + 24 m_b^3)}{4(m_a + m_b)r^2 } \nonumber\\ & & \hspace*{66pt } + \frac{g^2(87 m_a + 128 m_b ) \hbar}{20 \pi r^3 } \bigg ] + { } ^{1}v^{(2)}_t(\vec r ) \nonumber\\ & = & \left[-{3g^2m_am_b(m_a+m_b)\over r^2 } - { 41g^2m_am_b\hbar\over 10\pi r^3}\right ] \hat{\epsilon}_f^{b * } \cdot \hat{\epsilon}_i^b \nonumber\\ & + & \bigg[{g^2(12 m_a^3 + 45 m_a^2 m_b + 56 m_a m_b^2 + 24 m_b^3)\over 2 m_b ( m_a+m_b)r^4 } \nonumber \\ & & + { 3 g^2(87 m_a+ 128m_b)\hbar\over 20 \pi m_b r^5}\bigg ] \vec{l}\cdot\vec{s}_b + { } ^{1}v^{(2)}_t(\vec r ) \label{eq : nlopot01}\end{aligned}\ ] ] where @xmath202 denotes the tensor pieces not explicitly shown . comparison with the corresponding form of @xmath203 given in eq . ( [ eq : mn ] ) confirms the universality which we have suggested the spin - independent and spin - orbit terms have identical forms . the next task is to see whether this universality applies when both scattered particles carry spin . for this purpose we consider the case of spin-1/2 spin-1/2 scattering . in the case of scattering of a pair of spin-1/2 particles , the basic vertices have already been developed in the spin-0 spin-1/2 scattering section , so we can proceed directly to our calculation . the one - graviton exchange amplitude at tree level reads @xmath204 \times \sqrt{m_a^2m_b^2\over e_1e_2e_3e_4 } \quad \quad\label{eq : kh}\end{aligned}\ ] ] and with the spin identities eq . ( [ eq : id ] ) for system @xmath19 and @xmath205\label{eq : ida}\ ] ] for system @xmath16 , eq . ( [ eq : kh ] ) can be written as @xmath206.\end{aligned}\ ] ] in the symmetric center of mass frame we take the nonrelativistic limit yielding @xmath207 \label{eq : lomnonrelhh_gr}\end{aligned}\ ] ] whereby the lowest order potential for spin-1/2 spin-1/2 scattering becomes @xmath208 note that since the piece proportional to @xmath209 in eq . ( [ eq : lomnonrelhh_gr ] ) is analytic in @xmath210 it yields only a short distance contribution which is omitted in the potential eq . ( [ eq : mj ] ) . in this case when we evaluate the loop diagrams in fig . [ fig_loops ] , we notice that part of the spin - spin structure piece contains the form @xmath211 multiplying the nonanalytic structures @xmath81 and @xmath82 . due to the extra factor of @xmath67 in this form , we must expand all loop integrals to _ one order higher _ in @xmath67 than before in order to be consistent . this has been done and does make a difference in our results for the spin - spin interaction piece . the results for the individual diagrams are then found to be @xmath212\nonumber\\ { } ^{{1\over 2}{1\over 2}}{\cal m}_{\ref{fig_loops}b}^{(2)}(q)\hspace*{-5pt}&=&\hspace*{-5pt } g^2 m_a m_b \bigg[{\mathcal u}_a { \mathcal u}_b \left(15 l + 7 m_a s\right ) \nonumber\\ & & \hspace*{37pt } + i \frac{{\mathcal e}_a { \mathcal u}_b}{m_a^2 m_b } \left(\frac{21}{2 } l + 4 m_a s\right ) + i \frac{{\mathcal u}_a { \mathcal e}_b}{m_a m_b^2 } \left(13 l + 8 m_a s\right ) \nonumber \\ & & \hspace*{37pt } + \frac{s_a\cdot qs_b\cdot q}{m_a m_b } \left(\frac{17}{3 } l + \frac{5}{2 } m_a s\right ) \nonumber\\ & & \hspace*{37pt } - \frac{q^2 s_a \cdot s_b}{m_a m_b } \left(\frac{11}{3 } l + \frac{3}{2 } m_a s\right ) \hspace*{-1.5pt}\bigg]\nonumber\\ { } ^{{1\over 2}{1\over 2}}{\cal m}_{\ref{fig_loops}c}^{(2)}(q)\hspace*{-5pt}&=&\hspace*{-5pt } g^2 m_a m_b \bigg[{\mathcal u}_a { \mathcal u}_b \left(15 l + 7 m_b s\right ) \nonumber\\ & & \hspace*{37pt } + i \frac{{\mathcal e}_a { \mathcal u}_b}{m_a^2 m_b } \left(13 l + 8 m_b s\right ) + i \frac{{\mathcal u}_a { \mathcal e}_b}{m_a m_b^2 } \left(\frac{21}{2 } l + 4 m_b s\right ) \nonumber \\ & & \hspace*{37pt } + \frac{s_a\cdot qs_b\cdot q}{m_a m_b } \left(\frac{17}{3 } l + \frac{5}{2 } m_b s\right ) \nonumber\\ & & \hspace*{37pt } - \frac{q^2 s_a \cdot s_b}{m_a m_b } \left(\frac{11}{3 } l + \frac{3}{2 } m_b s\right ) \hspace*{-1.5pt}\bigg]\nonumber\\ { } ^{{1\over 2}{1\over 2}}{\cal m}_{\ref{fig_loops}d}^{(2)}(q)\hspace*{-5pt}&=&\hspace*{-5pt } g^2 m_am_b \bigg[{\mathcal u}_a { \mathcal u}_b \hspace*{-1.5pt}\bigg(\hspace*{-1.5pt}l \hspace*{-1pt}\left(\hspace*{-1.5pt } { 4m_am_b\over q^2}\hspace*{-1pt}+\hspace*{-1pt}\frac{31 m_a^2\hspace*{-1pt } - \hspace*{-1pt}16 m_a m_b\hspace*{-1pt } + \hspace*{-1pt } 31 m_b^2}{4m_am_b}\hspace*{-0.5pt}\right ) \nonumber \\ & & \hspace*{68pt } + s \hspace*{2pt } { 9\over 2}(m_a + m_b)\hspace*{-1pt } \bigg ) \nonumber\\ & & \hspace*{37pt}+ i \frac{{\mathcal e}_a { \mathcal u}_b}{m_a^2 m_b}\bigg(\hspace*{-1pt } l \hspace*{-1pt } \left(\hspace*{-1pt } \frac{8 m_a m_b}{q^2 } \hspace*{-1pt } + \hspace*{-1pt } \frac{124 m_a^2 - 11 m_a m_b + 177 m_b^2}{12 m_a m_b}\right ) \nonumber\\ & & \hspace*{82pt } + s \hspace*{-1pt } \left ( \hspace*{-1pt } \frac{m_a m_b(4m_a \hspace*{-1pt } + \hspace*{-0.5pt } 3 m_b)}{s - s_0 } \hspace*{-1pt } + \hspace*{-1pt } ( 13 m_a \hspace*{-1pt } + \hspace*{-0.5pt } 9 m_b ) \hspace*{-1pt}\right ) \ ! \hspace*{-1pt } \bigg)\nonumber\\ & & \hspace*{37pt}+ i \frac{{\mathcal u}_a { \mathcal e}_b}{m_a m_b^2}\bigg(\hspace*{-1pt } l \hspace*{-1pt } \left(\hspace*{-1pt } \frac{8 m_a m_b}{q^2}\hspace*{-1pt } + \hspace*{-1pt } \frac{177 m_a^2 - 11 m_a m_b + 124 m_b^2}{12 m_a m_b}\right ) \nonumber\\ & & \hspace*{82pt } + s \hspace*{-1pt } \left(\hspace*{-1pt } \frac{m_a m_b(3 m_a \hspace*{-1pt } \hspace*{-1pt } + \hspace*{-0.5pt } 4 m_b)}{s - s_0 } \hspace*{-1pt } + \hspace*{-1pt } ( 9 m_a \hspace*{-1pt } + \hspace*{-0.5pt } 13 m_b ) \hspace*{-1pt}\right ) \ ! \hspace*{-1pt } \bigg)\nonumber\\ & & \hspace*{37pt } + \frac{s_a\cdot qs_b\cdot q}{m_a m_b } \bigg ( \hspace*{-1pt } l \hspace*{-1pt } \left(\frac{4 m_a m_b}{q^2 } + \frac{39 m_a^2 \hspace*{-1pt } + \hspace*{-1pt } 25 m_a m_b \hspace*{-1pt } + \hspace*{-1pt } 39 m_b^2}{6 m_a m_b } \right ) \nonumber\\ & & \hspace*{99pt } + s \hspace*{-1pt } \left(\frac{m_a m_b ( m_a \hspace*{-1pt } + \hspace*{-1pt } m_b)}{s - s_0 } \hspace*{-1pt } + \hspace*{-1pt } \frac{39}{4 } \left(m_a \hspace*{-1pt } + \hspace*{-1pt } m_b \right ) \hspace*{-1pt } \right ) \hspace*{-3pt } \bigg ) \nonumber \\ & & \hspace*{37pt } - \frac{q^2 s_a \cdot s_b}{m_a m_b } \bigg ( \hspace*{-1.5pt } l \hspace*{-2pt } \left(\hspace*{-1pt } \frac{2 m_a m_b}{q^2 } \hspace*{-1.5pt } + \hspace*{-1.5pt } \frac{153 m_a^2 \hspace*{-1.2pt } + \hspace*{-1.2pt } 218 m_a m_b \hspace*{-1.2pt } + \hspace*{-1.2pt } 153 m_b^2}{24 m_a m_b } \hspace*{-1pt}\right)\nonumber\\ & & \hspace*{92pt } + s \hspace*{-1pt } \left(\frac{m_a m_b ( m_a \hspace*{-1pt } + \hspace*{-1pt } m_b)}{s - s_0 } \hspace*{-1pt } + \hspace*{-1pt } \frac{41}{4 } \left(m_a \hspace*{-1pt } + \hspace*{-1pt } m_b \right ) \hspace*{-1pt } \right ) \hspace*{-3pt } \bigg ) \nonumber \\ & & \hspace*{37pt } + \big(2 s_a \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } p_3 s_b \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } p_1 \hspace*{-0.5pt } + \hspace*{-0.5pt } s_a \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } q s_b \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } p_1 \hspace*{-0.5pt } - \hspace*{-0.5pt } s_a \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } p_3 s_b \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } q\big ) \frac{22 l}{3 m_a m_b}\bigg ] \nonumber\\ & - & i 4\pi g^2 m_a^2 m_b^2 \hspace*{1pt } { l\over q^2}\sqrt{m_a m_b \over s - s_0 } \ , \bigg({\mathcal u}_a { \mathcal u}_b + 2 \hspace*{0.5pt } i \frac{{\mathcal e}_a { \mathcal u}_b}{m_a^2 m_b } + 2 \hspace*{0.5pt } i \frac{{\mathcal u}_a { \mathcal e}_b}{m_a m_b^2}\nonumber\\ & & \hspace*{123pt } + \frac{s_a \cdot qs_b \cdot q - \frac{1}{2}q^2 s_a \cdot s_b}{m_a m_b}\bigg ) \nonumber\\ { } ^{{1\over 2}{1\over 2}}{\cal m}_{\ref{fig_loops}e}^{(2)}(q)\hspace*{-5pt}&=&\hspace*{-5pt } g^2 m_am_b \bigg[{\mathcal u}_a { \mathcal u}_b \hspace*{-1.5pt}\bigg(\hspace*{-1.5pt}l \hspace*{-1pt}\left(\hspace*{-1.5pt } - { 4m_am_b\over q^2 } \hspace*{-1pt } - \hspace*{-1pt}\frac{93 m_a^2\hspace*{-1pt } + \hspace*{-1pt}232 m_a m_b\hspace*{-1pt } + \hspace*{-1pt } 93 m_b^2}{12 m_am_b}\hspace*{-0.5pt}\right ) \nonumber \\ & & \hspace*{68pt } + s \left(- { 7\over 2}(m_a + m_b ) \right)\hspace*{-1pt } \bigg ) \nonumber\\ & & \hspace*{37pt}+ i \frac{{\mathcal e}_a { \mathcal u}_b}{m_a^2 m_b}\bigg(\hspace*{-1pt } l \hspace*{-1.5pt } \left(\hspace*{-1.5pt } - \frac{8 m_a m_b}{q^2 } \hspace*{-1.5pt } - \hspace*{-1.5pt } \frac{124 m_a^2 \hspace*{-1.5pt } + \hspace*{-1.5pt } 235 m_a m_b \hspace*{-1.5pt } + \hspace*{-1.5pt } 177 m_b^2}{12 m_a m_b } \hspace*{-1.4pt } \right ) \nonumber\\ & & \hspace*{82pt } + s \hspace*{-1pt } \left ( - 4 m_a - \frac{33}{4 } m_b \right ) \ ! \hspace*{-1pt } \bigg)\nonumber\\ & & \hspace*{37pt}+ i \frac{{\mathcal u}_a { \mathcal e}_b}{m_a m_b^2}\bigg(\hspace*{-1pt } l \hspace*{-1.5pt } \left(\hspace*{-1.5pt } - \frac{8 m_a m_b}{q^2 } \hspace*{-1.5pt } - \hspace*{-1.5pt } \frac{177 m_a^2 \hspace*{-1.5pt } + \hspace*{-1.5pt } 235 m_a m_b \hspace*{-1.5pt } + \hspace*{-1.5pt } 124 m_b^2}{12 m_a m_b } \hspace*{-1.4pt } \right ) \nonumber\\ & & \hspace*{82pt } + s \hspace*{-1pt } \left ( - \frac{33}{4 } m_a - 4 m_b \right ) \ ! \hspace*{-1pt } \bigg)\nonumber\\ & & \hspace*{37pt } + \frac{s_a\cdot qs_b\cdot q}{m_a m_b } \bigg ( \hspace*{-1pt } l \hspace*{-1.5pt } \left(\hspace*{-1.5pt } - \frac{4 m_a m_b}{q^2 } \hspace*{-1.5pt } - \hspace*{-1.5pt } \frac{39 m_a^2 \hspace*{-1.7pt } + \hspace*{-1.5pt } 23 m_a m_b \hspace*{-1.7pt } + \hspace*{-1.5pt } 39 m_b^2}{6 m_a m_b } \hspace*{-1.4pt } \right ) \nonumber\\ & & \hspace*{99pt } + s \hspace*{-1pt } \left ( \frac{39}{4 } \left(m_a \hspace*{-1pt } + \hspace*{-1pt } m_b \right ) \hspace*{-1pt } \right ) \hspace*{-3pt } \bigg ) \nonumber \\ & & \hspace*{37pt } - \frac{q^2 s_a \hspace*{-1.7pt } \cdot \hspace*{-1.6pt } s_b}{m_a m_b } \hspace*{-1pt } \bigg ( \hspace*{-2pt } l \hspace*{-2pt } \left(\hspace*{-2.3pt } - \frac{2 m_a m_b}{q^2 } \hspace*{-1.5pt } - \hspace*{-1.5pt } \frac{153 m_a^2 \hspace*{-1.8pt } + \hspace*{-2.2pt } 158 m_a m_b \hspace*{-1.8pt } + \hspace*{-2.2pt } 153 m_b^2}{24 m_a m_b } \hspace*{-1.4pt}\right)\nonumber\\ & & \hspace*{87pt } + s \left ( - \frac{3}{2 } \left(m_a \hspace*{-1pt } + \hspace*{-1pt } m_b \right ) \hspace*{-1pt } \right ) \hspace*{-3pt } \bigg ) \nonumber \\ & & \hspace*{37pt } + \big(2 s_a \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } p_3 s_b \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } p_1 \hspace*{-0.5pt } + \hspace*{-0.5pt } s_a \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } q s_b \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } p_1 \hspace*{-0.5pt } - \hspace*{-0.5pt } s_a \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } p_3 s_b \hspace*{-0.5pt } \cdot \hspace*{-0.5pt } q\big ) \frac{- 22 l}{3 m_a m_b}\bigg ] \nonumber\\ { } ^{{1\over 2}{1\over 2}}{\cal m}_{\ref{fig_loops}f}^{(2)}(q)\hspace*{-5pt}&=&\hspace*{-5pt } g^2 m_a m_b \bigg[{\mathcal u}_a { \mathcal u}_b \big(14 l - 2 ( m_a + m_b ) s\big ) \nonumber\\ & & \hspace*{37pt } + i \frac{{\mathcal e}_a { \mathcal u}_b}{m_a^2 m_b } \bigg(4 l + \left(- 2 m_a - \frac{1}{2 } m_b \right ) \ ! s \bigg ) \nonumber \\ & & \hspace*{37pt } + i \frac{{\mathcal u}_a { \mathcal e}_b}{m_a m_b^2 } \bigg(4 l + \left(- \frac{1}{2 } m_a - 2 m_b \right ) \ ! s \bigg ) \nonumber \\ & & \hspace*{37pt } + \frac{s_a\cdot qs_b\cdot q}{m_a m_b } \left(- 2 l - ( m_a + m_b ) s\right ) \nonumber\\ & & \hspace*{37pt } - \frac{q^2 s_a \cdot s_b}{m_a m_b } \left(- 2 l - ( m_a + m_b ) s\right ) \hspace*{-1.5pt}\bigg]\nonumber\\ { } ^{{1\over 2}{1\over 2}}{\cal m}_{\ref{fig_loops}g}^{(2)}(q)\hspace*{-5pt}&=&\hspace*{-5pt}g^2 m_a m_b \bigg[{\mathcal u}_a { \mathcal u}_b \hspace*{-1pt } \left(\hspace*{-1pt } - \frac{43}{15}l \hspace*{-0.5pt } \right ) \hspace*{-2pt } + i \frac{{\mathcal e}_a { \mathcal u}_b}{m_a^2 m_b } \hspace*{-1pt } \left(\hspace*{-1pt } - \frac{14}{5 } l \hspace*{-0.5pt } \right ) \hspace*{-2pt } + i \frac{{\mathcal u}_a { \mathcal e}_b}{m_a m_b^2 } \hspace*{-1pt } \left(\hspace*{-1pt } - \frac{14}{5 } l \hspace*{-0.5pt } \right ) \nonumber \\ & & \hspace*{37pt}+ \frac{s_a\cdot qs_b\cdot q}{m_a m_b } \left(- \frac{7}{5 } l\right ) - \frac{q^2 s_a \cdot s_b}{m_a m_b } \left(- \frac{7}{5 } l\right ) \hspace*{-1.5pt}\bigg].\end{aligned}\ ] ] where we have defined @xmath213 and @xmath214 with @xmath215 to keep our notation compact . the sum of all diagrams is found to be @xmath216\nonumber\\ & - \hspace*{-7.5pt } & i 4\pi g^2 m_a^2 m_b^2 { l\over q^2}\sqrt{m_a m_b \over s - s_0 } \ , \bigg({\mathcal u}_a { \mathcal u}_b + 2 i \frac{{\mathcal e}_a { \mathcal u}_b}{m_a^2 m_b } + 2 i \frac{{\mathcal u}_a { \mathcal e}_b}{m_a m_b^2}\nonumber\\ & & \hspace*{124pt } + \frac{s_a \cdot qs_b \cdot q - \frac{1}{2}q^2 s_a \cdot s_b}{m_a m_b}\bigg ) . \quad \label{eq : ohla2}\end{aligned}\ ] ] comparing with our finding for spin-0 spin-1/2 scattering in eq . ( [ eq : ohla ] ) , we notice the by now familiar universality of the amplitude : the form of the component proportional to @xmath217 of eq . ( [ eq : ohla ] ) is found here in the component proportional to @xmath218 , and the form of the structure proportional to @xmath219 of eq . ( [ eq : ohla ] ) is now found in the component proportional to @xmath220 in eq . ( [ eq : ohla2 ] ) . moreover , the amplitude is symmetric in @xmath221 and new gravitational spin - spin interaction corrections arise . the quantum part of the spin - spin component has been calculated previously by kirilin @xcite whose result disagrees with our result for the numerical prefactors . for the quantum terms in the spin - orbit components however , we fully agree with kirilin s results in @xcite . in the nonrelativistic limit we obtain the expression @xmath222 \hspace*{-1.8pt } \chi_f^{a\dagger}\chi_i^a \hspace*{1.3pt } \chi_f^{b\dagger}\chi_i^b \nonumber\\ & + \hspace*{-7.5pt}&\bigg[\hspace*{-0.8pt } g^2 \hspace*{-1pt } \left(\frac{24m_a^3 \hspace*{-1pt } + \hspace*{-1pt } 56 m_a^2 m_b \hspace*{-1pt } + \hspace*{-1pt } 45 m_a m_b^2 \hspace*{-1pt } + \hspace*{-1pt } 12 m_b^3}{2(m_a \hspace*{-1pt } + \hspace*{-1pt } m_b ) } \hspace*{1pt } s - \frac{128 m_a \hspace*{-1pt } + \hspace*{-1pt } 87 m_b}{10 } \hspace*{1pt } l\right)\nonumber\\ & & + \frac{g^2 m_a^2 m_b^2 ( 4 m_a + 3 m_b)}{(m_a + m_b ) } \left(- i \frac{2 \pi l}{p_0 q^2 } + \frac{s}{p_0 ^ 2 } \right)\bigg ] { i\over m_a}\vec{s}_a\cdot\vec{p}\times\vec{q } \ , \chi_f^{b\dagger}\chi_i^b \nonumber \\ & + \hspace*{-7.5pt}&\bigg[\hspace*{-0.8pt } g^2 \hspace*{-1pt } \left(\frac{12m_a^3 \hspace*{-1pt } + \hspace*{-1pt } 45 m_a^2 m_b \hspace*{-1pt } + \hspace*{-1pt } 56 m_a m_b^2 \hspace*{-1pt } + \hspace*{-1pt } 24 m_b^3}{2(m_a \hspace*{-1pt } + \hspace*{-1pt } m_b ) } \hspace*{1pt } s - \frac{87 m_a \hspace*{-1pt } + \hspace*{-1pt } 128 m_b}{10 } \hspace*{1pt } l\right)\nonumber\\ & & + \frac{g^2 m_a^2 m_b^2 ( 3 m_a + 4 m_b)}{(m_a + m_b ) } \left(- i \frac{2 \pi l}{p_0 q^2 } + \frac{s}{p_0 ^ 2 } \right)\bigg ] \chi_f^{a\dagger}\chi_i^a \ , { i\over m_b}\vec{s}_b\cdot\vec{p}\times\vec{q } \nonumber \\ & + \hspace*{-7.5pt } & g^2 m_a m_b \ , \frac{19 m_a^2 + 36 m_a m_b + 19 m_b^2}{2(m_a + m_b)}\ , s \ , \frac{\vec s_a \cdot \vec q \ , \vec s_b \cdot \vec q - \vec q^{\hspace*{1.4pt}2 } \vec s_a \cdot \vec s_b}{m_a m_b}\nonumber\\ & -\hspace*{-7.5pt } & g^2 m_a m_b \ , l \ , \frac{11 \vec s_a \cdot \vec q \ , \vec s_b \cdot \vec q - 16 \hspace*{1pt } \vec q^{\hspace*{1.4pt}2 } \vec s_a \cdot \vec s_b}{15 m_a m_b}\nonumber\\ & + \hspace*{-7.5pt } & \frac{g^2 m_a^3 m_b^3}{m_a + m_b } \ , \frac{s}{p_0 ^ 2 } \ , \frac{\vec s_a \cdot \vec q \ , \vec s_b \cdot \vec q - \vec q^{\hspace*{1.4pt}2 } \vec s_a \cdot \vec s_b}{m_a m_b}\nonumber\\ & + \hspace*{-7.5pt } & \frac{g^2 m_a^3 m_b^3}{m_a + m_b } \left(-i \frac{4 \pi l}{p_0 q^2}\right ) \frac{\vec s_a \cdot \vec q \ , \vec s_b \cdot \vec q - \frac{1}{2}\vec q^{\hspace*{1.4pt}2 } \vec s_a \cdot \vec s_b}{m_a m_b } \label{eq : yohh}\end{aligned}\ ] ] which obviously has to exhibit the universalities of the spin - independent and the spin - orbit pieces . therefore , we know that the subtraction of the second born iteration successfully removes the unwanted @xmath157 and @xmath158 structures for the spin - independent and the spin - orbit components . the leading spin - spin term of the second born iteration amplitude is new , however , and we compute @xmath223\nonumber\\ & = \hspace*{-10pt } & \frac{g^2m_a^2m_b^2}{m_a + m_b } \ , \frac{s}{p_0 ^ 2 } \ , ( \vec s_a \cdot \vec q \ , \vec s_b \cdot \vec q - \vec q^{\hspace*{1.4pt}2 } \vec s_a \cdot \vec s_b)\nonumber\\ & + \hspace*{-10pt } & \frac{g^2m_a^2m_b^2}{m_a + m_b } \left(-i \frac{4 \pi l}{p_0 q^2}\right ) ( \vec s_a \cdot \vec q \ , \vec s_b \cdot \vec q - \frac{1}{2}\vec q^{\hspace*{1.4pt}2 } \vec s_a \cdot \vec s_b).\end{aligned}\ ] ] where we again defined @xmath164 . with this , the full second born iteration amplitude becomes @xmath224 and we observe that when this amplitude is subtracted from the full one loop scattering amplitude eq . ( [ eq : yohh ] ) , _ all _ terms involving @xmath158 and @xmath157 disappear leaving behind a well - defined second order potential @xmath225\nonumber\\ & \simeq \hspace*{-5pt } & \left[-{3g^2m_am_b(m_a+m_b)\over r^2 } - { 41g^2m_am_b\hbar\over 10\pi r^3}\right ] \chi_f^{a\dagger}\chi_i^a \ , \chi_f^{b\dagger}\chi_i^b \nonumber\\ & + \hspace*{-5pt } & { 1\over m_a}\vec{s}_a\cdot \vec{p}\times\vec{\nabla}\bigg [ \frac{g^2(24m_a^3 + 56 m_a^2 m_b + 45 m_a m_b^2 + 12 m_b^3)}{4(m_a + m_b)r^2 } \nonumber\\ & & \hspace*{68pt } + \frac{g^2(128 m_a + 87 m_b ) \hbar}{20 \pi r^3 } \bigg ] \chi_f^{a\dagger}\chi_i^a \nonumber\\ & + \hspace*{-5pt } & \chi_f^{a\dagger}\chi_i^a \ , { 1\over m_b}\vec{s}_b\cdot \vec{p}\times\vec{\nabla}\bigg [ \frac{g^2(12m_a^3 + 45 m_a^2 m_b + 56 m_a m_b^2 + 24 m_b^3)}{4(m_a + m_b)r^2 } \nonumber\\ & & \hspace*{97pt } + \frac{g^2(87 m_a + 128 m_b ) \hbar}{20 \pi r^3 } \bigg]\nonumber\\ & + \hspace*{-5pt } & \frac{\vec s_a \cdot \vec{\nabla } \vec s_b \cdot \vec{\nabla } - \vec{\nabla}^2 \vec s_a \cdot \vec s_b}{m_a m_b } \left[\frac{g^2 m_a m_b(19 m_a^2 + 36 m_a m_b + 19 m_b^2)}{4(m_a + m_b)r^2}\right]\nonumber\\ & + \hspace*{-5pt } & \frac{11 \vec s_a \cdot \vec{\nabla } \vec s_b \cdot \vec{\nabla } - 16 \vec{\nabla}^2 \vec s_a \cdot \vec s_b}{15 m_a m_b } \left[\frac{g^2 m_a m_b \hbar}{2 \pi r^3}\right]\nonumber\\ & \simeq \hspace*{-5pt } & \left[-{3g^2m_am_b(m_a+m_b)\over r^2 } - { 41g^2m_am_b\hbar\over 10\pi r^3}\right ] \chi_f^{a\dagger}\chi_i^a \ , \chi_f^{b\dagger}\chi_i^b\nonumber\\ & + \hspace*{-5pt } & \bigg[{g^2(24 m_a^3 + 56 m_a^2 m_b + 45 m_a m_b^2 + 12 m_b^3)\over 2 m_a ( m_a+m_b)r^4 } \nonumber \\ & & + { 3 g^2(128 m_a + 87 m_b)\hbar\over 20 \pi m_a r^5}\bigg ] \vec{l}\cdot\vec{s}_a \ , \chi_f^{b\dagger}\chi_i^b \nonumber \\ & + \hspace*{-5pt } & \bigg[{g^2(12 m_a^3 + 45 m_a^2 m_b + 56 m_a m_b^2 + 24 m_b^3)\over 2 m_b ( m_a+m_b)r^4 } \nonumber \\ & & + { 3 g^2(87 m_a+ 128m_b)\hbar\over 20 \pi m_b r^5}\bigg ] \chi_f^{a\dagger}\chi_i^a \ , \vec{l}\cdot\vec{s}_b \nonumber \\ & + \hspace*{-5pt } & \left[\frac{2 g^2(19 m_a^2 + 36 m_a m_b + 19 m_b^2)}{(m_a + m_b ) r^4}\right ] \ ! \left(\vec s_a \cdot \vec r \ , \vec s_b \cdot \vec r \ , / r^2 - \frac{1}{2 } \hspace*{1pt } \vec s_a \cdot \vec s_b\right)\nonumber\\ & + \hspace*{-5pt } & \frac{11 g^2 \hbar}{2 \pi r^5 } \left(\vec s_a \cdot \vec r \ , \vec s_b \cdot \vec r \ , / r^2 - \frac{43}{55 } \vec s_a \cdot \vec s_b\right)\end{aligned}\ ] ] which besides the universal spin - independent and spin - orbit components displays a new ( and presumably universal ) gravitational spin - spin interaction . above we have analyzed the gravitational scattering of two particles having nonzero mass . in lowest order the interaction arises from one - graviton exchange and leads at threshold to the well known newtonian interaction @xmath226 . inclusion of two - graviton exchange effects means adding the contribution from box , cross - box , triangle , and bubble diagrams , which have a rather complex form . the calculation can be simplified , however , by using ideas from effective field theory . the point is that if one is interested only in the leading long - range behavior of the interaction , then one need retain only the leading nonanalytic small momentum - transfer piece of the scattering amplitude . specifically , the terms which one retains are those which are nonanalytic and behave as either @xmath68 or @xmath70 . when fourier transformed , the former leads to classical ( @xmath69-independent ) terms in the potential of order @xmath227 while the latter generates quantum mechanical ( @xmath69-dependent ) corrections of order @xmath228 . ( of course , there are also shorter range nonanalytic contributions than these that are generated by scattering terms of order @xmath229 or @xmath230 . however , these pieces are higher order in momentum transfer and thus lead to shorter distance effects than those considered above and are therefore neglected in our discussion . ) specific calculations were done for particles with spin @xmath231 , @xmath232 , @xmath233 , and @xmath234 and various universalities were found . in particular , we found that in each case there was a spin - independent contribution of the form @xmath235 \nonumber \\ & & \times \left < s_a , m_{af}|s_a , m_{ai}\right > \left < s_b , m_{bf}|s_b , m_{bi}\right>\end{aligned}\ ] ] where @xmath236 and @xmath237 and with @xmath238 the spin of particle @xmath16 and @xmath239 the spin of particle @xmath19 with projections @xmath17 and @xmath20 on the quantization axis . the imaginary component of the amplitude , which would not , when fourier - transformed lead to a real potential , is eliminated when the iterated lowest order potential contribution is subtracted , leading to a well defined spin - independent second order potential of universal form @xmath240 \nonumber\\ & & \times \left < s_a , m_{af}|s_a , m_{ai}\right > \left < s_b , m_{bf}|s_b , m_{bi}\right>\end{aligned}\ ] ] whose classical component depends on the way the iteration of the leading order potential is performed . this ambiguity shows that the second order potential itself is not an observable , but we use it as a nice way to display the long distance components of the scattering amplitude in coordinate space . if either scattering particle carries spin then there exists an additional spin - orbit contribution , whose form is also universal @xmath241 \times \vec{l}\cdot\vec{s}_a \left < s_b , m_{bf}|s_b , m_{bi}\right > \nonumber \\ & + \hspace*{-5pt } & \bigg[{g^2(12 m_a^3 + 45 m_a^2 m_b + 56 m_a m_b^2 + 24 m_b^3)\over 2 m_b ( m_a+m_b)r^4 } \nonumber \\ & & + { 3 g^2(87 m_a+ 128m_b)\hbar\over 20 \pi m_b r^5}\bigg ] \times \left < s_a , m_{af}|s_a , m_{ai}\right > \vec{l}\cdot\vec{s}_b \quad \quad\end{aligned}\ ] ] where we have defined @xmath242 in this case a well defined second order potential required the subtraction of infrared singular terms behaving as both @xmath157 _ and _ @xmath158 which arise from the iterated lowest order potential . in the calculation of spin-0 spin-1 scattering we encountered new tensor structures including a quadrupole interaction . unfortunately , the subtraction of the @xmath157 and @xmath158 tensor pieces in the two - graviton exchange amplitude was not successful with our simple nonrelativistic iteration of the leading order potential so that we can not at this time give the form of the quadrupole component of the potential . further work is needed to clarify this issue . the corrections to the spin - spin interaction have only been calculated in spin-1/2 spin-1/2 scattering where we found their contributions to the scattering amplitude and to the potential . since we verified these forms only for a single spin configuration we have not confirmed its universality which we , however , strongly suspect . of course , for higher spin configurations , there also exist quadrupole - quadrupole interactions , spin - quadrupole interactions , etc . however , the calculation of such forms becomes increasingly cumbersome as the spin increases , and the phenomenological importance becomes smaller . thus we end our calculations here . one point of view to interpret the universalities of the long distance components of the scattering amplitudes and the resulting potentials is that if we increase the spins of our scattered particles , all we do is to add additional multipole moments . the spin - independent component can then be viewed as a monopole - monopole interaction , the spin - orbit piece as a dipole - monopole interaction etc . as long as we do not change the quantum numbers that characterize the lower multipoles , an increase in spin of the scattered particles merely adds new interactions that are less important at long distances . the same kind of universalities were also found in long distance effects in electromagnetic scattering @xcite and in the long range components of mixed electromagnetic - gravitational scattering @xcite . * acknowledgements * we would like to thank john donoghue for many clarifying discussions and walter goldberger for important comments . this work was supported in part by the national science foundation under award phy05 - 53304 ( brh and ar ) and by the the us department of energy under grant de - fg-02 - 92er40704 ( ar ) . in this appendix we give the integrals @xmath243 = i\int { d^3\ell\over ( 2\pi)^3 } { - 4\pi gm_am_b\over |\vec{\ell } - \vec{p}_f|^2+\lambda^2 } { i[1;\ell_r;\ell_r\ell_s]\over { p_0 ^ 2\over 2m_r}-{\ell^2\over 2m_r}+i\epsilon } { - 4\pi gm_am_b\over |\vec{p}_i - \vec{\ell}|^2+\lambda^2}\ ] ] which are needed in order to perform the iteration of the lowest order newton potentials . here we list only the results ; for a more detailed derivation , albeit with a different prefactor , see @xcite . the leading expressions for the iteration integrals read @xmath244 above we have argued that the scattering amplitude which is ultimately related to observables in quantum field theory is a physical result while the potential we have given is not an observable and depends both on the gauge , _ i.e. _ , the choice of coordinates used , and on the way we perform the iteration , _ i.e. _ , on the way we perform the matching . while the classical component of our potential is in fact plagued by these ambiguities , the quantum part is unique since it is unaffected by how we perform the matching and since a quantum field theory calculation in any gauge would result in the same result @xcite . in this appendix we will demonstrate how we can recover the classical equations of motion from our scattering amplitudes by setting up the einstein - infeld - hoffmann ( eih ) lagrangian @xcite . the eih lagrangian is itself dependent on the choice of coordinates , but can be expressed in the center of mass frame ( @xmath245 , @xmath246 ) in a general way as @xcite @xmath247 where the kinetic energy to nlo in the nonrelativistic expansion reads @xmath248 and the potential is @xmath249 with @xmath250 \frac{\vec p^{\hspace*{1.4pt } 2}}{m_a m_b } \right . \nonumber \\ & & \hspace*{69.5pt } \left.+ \left[\frac{1}{2 } + \alpha \ , \frac{(m_a + m_b)^2}{m_a m_b } \right ] \frac{\left(\vec p \cdot \hat r\right)^2}{m_a m_b}\right\ } \label{eq_eihv1 } \\ v^{(2 ) } & = & \left(1 - 2 \alpha\right ) \frac{g^2 m_a m_b ( m_a + m_b)}{2 r^2}. \label{eq_eihv2}\end{aligned}\ ] ] the parameter @xmath251 parameterizes the choice of coordinates used , where @xmath252 was the gauge of the original eih result . the coordinate change @xmath253 which implies @xmath254\ ] ] brings the original eih lagrangian into the form above , which is the most general result . since we perform our matching on - shell , _ i.e. _ , we use the on - shell one - graviton exchange amplitude to define the leading order @xmath255 potential , terms proportional to @xmath256 would never arise . clearly , our result must be in a gauge such that the coefficient of the structure @xmath257 in eq . ( [ eq_eihv1 ] ) vanishes . that is the case if and only if the gauge parameter is @xmath258 whereby the eih potential becomes @xmath259 \frac{\vec p^{\hspace*{1.4pt } 2}}{m_a m_b } \right\ } \label{eq_eihhrv1 } \\ v^{(2 ) } & = & \left(1 + \frac{m_a m_b}{(m_a + m_b)^2}\right ) \frac{g^2 m_a m_b ( m_a + m_b)}{2 r^2}. \label{eq_eihhrv2}\end{aligned}\ ] ] comparing the eih potential @xmath260 in this gauge of eq . ( [ eq_eihhrv1 ] ) with the long distance component of the leading order spin - independent potential in eq . ( [ eq : po ] ) we find full agreement for the relativistic corrections to the @xmath255 potential . however , comparing the eih potential @xmath261 in this gauge of eq . ( [ eq_eihhrv2 ] ) with the classical component of our spin - independent potential in eq . ( [ eq : so ] ) we see that the two do not agree ! the reason for this discrepancy is that we elected to use a nonrelativistic iteration when we performed the second born iteration of the leading order potential in eq . ( [ eq : iteration00b ] ) . this procedure , however , is not self - consistent when we are interested in equations of motion at nlo , for which we must account for the leading relativistic corrections in the iteration . in particular , we must use expressions for the potential and the propagator in eq . ( [ eq : iteration00a ] ) which include the leading relativistic corrections in this sections refers to the iteration being performed at nlo in the relativistic expansion . ] @xmath262\ ] ] @xmath263\ ] ] which yields a second born iteration amplitude @xmath264\nonumber\\ & \simeq \hspace*{-5pt } & h+{1\over m_am_b}\left(p_0 ^ 2h+\delta_{rs}h_{rs}\right ) \left(\frac{1}{4 } + \frac{7}{4 } \frac{m_a m_b}{m_r^2}\right ) \nonumber\\ & \simeq \hspace*{-5pt } & - i \hspace*{0.6pt } 4 \pi g^2 m_a^2 m_b^2 \hspace*{1pt } { l\over q^2 } { m_r\over p_0 } + \frac{g^2 m_a^2 m_b^2}{m_a + m_b } \left(1 + \frac{7 ( m_a + m_b)^2}{m_a m_b } \right)s . \label{eq : jk}\end{aligned}\ ] ] subtracting this iterated amplitude which includes all corrections to nlo from the scattering amplitude @xmath265 of eq . ( [ eq_amplo_00 ] ) we find then the second order potential @xmath266\nonumber\\ & = \hspace*{-5pt}&\int{d^3q\over ( 2\pi)^3 } e^{-i\vec{q}\cdot\vec{r } } \ , g^2 m_a m_b \left [ \left((m_a + m_b ) + { m_am_b\over m_a+m_b}\right ) s + { 41\over 5}l\right]\nonumber\\ & = \hspace*{-5pt } & \left(1 + \frac{m_a m_b}{(m_a + m_b)^2}\right ) \frac{g^2 m_a m_b ( m_a + m_b)}{2 r^2 } - { 41 g^2 m_a m_b \hbar \over 10 \pi r^3 } \label{eq : sf}\end{aligned}\ ] ] and observe that now the classical component agrees with the @xmath12 eih potential of eq . ( [ eq_eihhrv2 ] ) . thus we have shown that if we consistently take into account the @xmath4 and @xmath5 corrections beyond newtonian physics we reproduce the eih lagrangian in a certain gauge . from the resulting eih lagrangian we could evaluate observables such as the precession of the perihelion of mercury which must clearly be independent of the gauge used . the inclusion of the @xmath4 corrections is required since the equations of motion can be used to describe bound states where @xmath267 by the virial theorem . however , our methods are clearly clumsy for the calculation of classical observables . recently , goldberger and rothstein have developed an effective field theory of gravity which is optimized for calculating classical observables of bound states called nrgr @xcite . here the external particles are static sources so that no loops are to be calculated in their theory when calculating classical observables since the only propagating particles present are gravitons which are massless and thus the loop expansion in nrgr corresponds to an expansion in @xmath69 . in the nrgr framework the spin - dependent classical equations of motion were calculated recently to nlo by porto and rothstein @xcite so that we will not continue here to evaluate the corresponding spin - dependent classical potentials consistently taking into account all relativistic @xmath1 effects in the iteration . a. einstein , l. infeld and b. hoffmann , annals math . * 39 * , 65 ( 1938 ) . j. f. donoghue , phys . lett . * 72 * , 2996 ( 1994 ) [ arxiv : gr - qc/9310024 ] . j. f. donoghue , phys . d * 50 * , 3874 ( 1994 ) [ arxiv : gr - qc/9405057 ] . j. f. donoghue , arxiv : gr - qc/9512024 . c. p. burgess , living rev . * 7 * , 5 ( 2004 ) [ arxiv : gr - qc/0311082 ] . n. e. j. bjerrum - bohr , j. f. donoghue and b. r. holstein , phys . d * 68 * , 084005 ( 2003 ) [ erratum - ibid . d * 71 * , 069904 ( 2005 ) ] [ arxiv : hep - th/0211071 ] . b. r. holstein , phys . d * 74 * , 084030 ( 2006 ) . + b. r. holstein , arxiv : gr - qc/0607051 . holstein , a. ross , arxiv:0802.0715 [ hep - ph ] . y. iwasaki , prog . phys . * 46 * , 1587 ( 1971 ) . i. b. khriplovich and g. g. kirilin , j. exp . phys . * 95 * , 981 ( 2002 ) [ zh . eksp . * 95 * , 1139 ( 2002 ) ] [ arxiv : gr - qc/0207118 ] . i. b. khriplovich and g. g. kirilin , econf * c0306234 * , 1361 ( 2003 ) [ j. exp . phys . * 98 * , 1063 ( 2004 zetfa,125,1219 - 1228.2004 ) ] [ arxiv : gr - qc/0402018 ] . n. e. j. bjerrum - bohr , j. f. donoghue and b. r. holstein , phys . d * 67 * , 084033 ( 2003 ) [ erratum - ibid . d * 71 * , 069903 ( 2005 ) ] [ arxiv : hep - th/0211072 ] . g. t hooft and m. j. g. veltman , annales poincare phys . a * 20 * , 69 ( 1974 ) . j. sucher , phys . d * 49 * , 4284 ( 1994 ) . s. deser and p. van nieuwenhuizen , phys . d * 10 * , 411 ( 1974 ) . g. g. kirilin , nucl . b * 728 * , 179 ( 2005 ) [ arxiv : gr - qc/0507070 ] . holstein , a. ross , arxiv:0802.0717 [ hep - ph ] . k. hiida and h. okamura , prog . * 47 * , 1743 ( 1972 ) . b. m. barker and r. f. oconnell , phys . d * 12 * , 329 ( 1975 ) . w. d. goldberger and i. z. rothstein , phys . d * 73 * , 104029 ( 2006 ) [ arxiv : hep - th/0409156 ] . w. d. goldberger and i. z. rothstein , gen . * 38 * , 1537 ( 2006 ) [ int . j. mod . d * 15 * , 2293 ( 2006 ) ] [ arxiv : hep - th/0605238 ] . w. d. goldberger , arxiv : hep - ph/0701129 . w. d. goldberger and i. z. rothstein , phys . d * 73 * , 104030 ( 2006 ) [ arxiv : hep - th/0511133 ] . b. kol and m. smolkin , arxiv:0712.2822 [ hep - th ] . b. kol and m. smolkin , arxiv:0712.4116 [ hep - th ] . r. a. porto , phys . d * 73 * , 104031 ( 2006 ) [ arxiv : gr - qc/0511061 ] . r. a. porto and i. z. rothstein , phys . lett . * 97 * , 021101 ( 2006 ) [ arxiv : gr - qc/0604099 ] . r. a. porto , arxiv:0710.5150 [ hep - th ] . r. a. porto and i. z. rothstein , arxiv:0712.2032 [ gr - qc ] .
ground states of quantum systems with many constituents are typically entangled . in a similar manner as one can identify characteristic length scales of correlation functions , quantum correlations are expected to exhibit some general scaling behavior , beyond the details of a fine - grained description . such characteristic features provide a physical picture that goes beyond the specifics of the underlying microscopic model . a central question of this type is the following : if one distinguishes a certain collection of subsystems , representing some spatial region , of a quantum many - body system in a pure ground state , the state of this part will typically have a positive entropy , reflecting the entanglement between this region and the rest of the system @xcite . this degree of entanglement is certainly expected to depend on the size and also on the shape of the region . yet , how does the degree of entanglement specifically depend on the size of the distinguished region ? in particular , does it scale as the volume of the interior which is meant to be the number of degrees of freedom of the interior ? or , potentially as the area of the boundary , i.e. , the number of contact points between the interior and the exterior ? this paper provides a detailed answer to the scaling behavior of the entanglement of regions with their exterior in a general setting of harmonic bosonic lattice systems and provides a comprehensive treatment of upper bounds on these quantities . we find that in arbitrary spatial dimensions the degree of entanglement in terms of the von neumann entropy scales asymptotically as the area of the boundary of the distinguished region . this paper significantly extends the findings of ref.@xcite on harmonic bosonic lattice systems : there , the area - dependence of the geometric entropy has been proven for cubic regions in non - critical harmonic lattice systems of arbitrary dimension with nearest - neighbor interactions , corresponding to discrete versions of klein - gordon fields . in this paper we extend our analysis to a general class of finite - ranged harmonic interactions and also take regions of arbitrary shape into account . for thermal gibbs states , the entropy of a reduction is longer a meaningful measure of entanglement . instead , an area - dependence for an appropriate mixed - state entanglement measure , the distillable entanglement , is established . also , an analogous statement holds for classical correlations in classical systems . the area - dependence is even found in certain cases where one can prove the divergence of the two - point correlation length . this demonstrates that this previously conjectured dependence between area and entanglement is valid under surprisingly general conditions . the presented analysis will make use of methods from the quantitative theory of entanglement in the context of quantum information science @xcite . it has been become clear recently that on questions about scaling of entropies and degrees of entanglement albeit often posed some time ago new light can be shed with such methods @xcite . in this language , quantum correlations are sharply grasped in terms of rates that can be achieved in local physical transformations . to assess quantum correlations using novel powerful tools from quantum information and to relate them to information - theoretical quantities constitutes an exciting perspective . in the context of quantum field theory , such questions of scaling of entropies and entanglement have a long tradition under the keyword of geometric entropy . in particular , work on the geometric entropy of free klein - gordon fields was driven in part by the intriguing suggested connection @xcite to the bekenstein - hawking black hole entropy @xcite . in seminal works by bombelli et al . @xcite and srednicki @xcite the relation between the entropy and the boundary area of the region has been suggested and supplemented with numerical arguments . this connection has been made more specific using a number of different methods . in particular , for half spaces in general and intervals in the one - dimensional case , the problem has been assessed employing methods from conformal field theory , notably @xcite , based on earlier work by cardy and peschel @xcite , and by cardy and calabrese @xcite . in one - dimensional non - critical chains , one observes a saturation of the entanglement of a distinguished block , as was proven analytically for harmonic chains @xcite and was later observed for non - critical spin chains numerically @xcite and analytically @xcite . in turn , in critical systems systems , one often but not always finds a logarithmically diverging entropy @xcite . in case of a model the continuum limit of which leads to a conformal field theory , the factor of the logarithmically diverging term is related to the central charge of the conformal field theory . the findings of the present paper and of the above - mentioned results motivate further questions concerning the general area - dependence of the degree of entanglement , for example in fermionic systems @xcite . in particular , the connection between the geometric entropy fulfilling no area law and the correlated quantum many - body system being critical is not fully understood yet . this is particularly true for the interesting case of more than one - dimensional quantum systems . this paper is structured as follows . we start , in section [ major ] , with a presentation of the major results of the paper . in section [ prelims ] we define our notation and recall some basics on harmonic lattice systems . section [ bounds ] provides a general framework of upper and lower bounds for entanglement measures , expressed in terms of the spectrum of the hamiltonian and two - point correlation functions . this analysis is performed both for the case of the ground state as well as for mixed gibbs states . of particular interest are hamiltonians with finite - ranged interactions in section [ sectionfiniteranged ] . for such hamiltonians , we first study the behavior of the two - point correlation functions , then the entanglement bounds . in section [ nomore ] , discussing the gibbs state case , we determine temperatures above which there is no entanglement left . a class of examples of hamiltonians which exhibit a divergent two - point correlation length in their ground state , but an area - dependence of the entanglement is presented in section [ divergent ] and expressed in analytical terms . the specific case of hamiltonians the interaction part of which can be expressed as a square of a banded matrix is discussed in section [ thasquared ] . in this case , very explicit expressions for entanglement measures can be found . we then consider , in section [ classical ] , the case of classical correlations in classical harmonic systems with arbitrary interaction structure . interestingly , this case is related to the quantum case for squared interactions in the sense of section [ thasquared ] . finally , we summarize what has been achieved in the present paper , and present a number of open questions in this context . throughout the whole paper we will consider harmonic subsystems on a @xmath0-dimensional cubic lattice @xmath1^{\times d}$ ] . the system thus has @xmath2 canonical degrees of freedom . the central question of this work will be : how does the degree of entanglement of a distinguished region @xmath3 with the rest @xmath4 scale with the size and shape of @xmath5 ? ^{\times 2}$ ] with a distinguished ( shaded ) region @xmath5 , consisting of @xmath6 degrees of freedom . oscillators representing the exterior @xmath4 are shown as , whereas pairs belonging to the surface @xmath7 of the region are marked by lines ( ) . ] we define the volume @xmath6 and surface @xmath7 of the distinguished region @xmath5 as @xmath8 where @xmath9 defines the @xmath10-norm distance for vectors @xmath11 that specify the position of oscillators on the @xmath0-dimensional lattice . more specifically , @xmath7 is the number of contact points , that is , the number of pairs of sites of @xmath5 and @xmath12 that are immediately adjacent . note that the distinguished region @xmath5 may have arbitrary shape and does not have to be contiguous . for the pure ground state @xmath13 , we will study the entropy of entanglement of @xmath3 given by @xmath14 where @xmath15 $ ] is the von neumann entropy of a state @xmath16 and @xmath17 $ ] denotes the reduced state associated with the degrees of freedom of the interior @xmath5 . for the pure ground state , this entropy of entanglement is identical to both the distillable entanglement and the entanglement cost . for pure states , it is indeed the unique asymptotic measure of entanglement . for gibbs states , in turn , we have to study a mixed - state entanglement measure , as the entropy of a subsystem no longer meaningfully quantifies the degree of entanglement . we will bound the rate at which maximally entangled pairs can be distilled , i.e. , the distillable entanglement @xmath18 . clearly , @xmath19 for zero temperature . we will subsequently suppress the index @xmath20 for notational clarity . we derive the following properties of @xmath21 and @xmath22 : _ _ * for @xmath0-dimensional harmonic lattice systems , we derive general upper and lower bounds to @xmath23 for pure ground states and to @xmath24 for gibbs states with respect to a temperature @xmath25 . these bounds are expressed entirely in terms of the potential matrix and stated in eqs . ( [ upperboundfinitet ] ) , ( [ upperboundzerot ] ) , and ( [ lowerbound ] ) . a necessary condition for the following results to hold is that the spectral condition number @xmath26 of the coupling matrix @xmath27 satisfies @xmath28 for some @xmath29 independent of @xmath5 and @xmath12 . _ _ * for nearest - neighbor interactions and the ground state , the entropy of entanglement @xmath21 scales as the surface area @xmath7 of @xmath5 . more specifically , there exist numbers @xmath30 independent of @xmath12 and @xmath5 such that @xmath31 this is stated in eqs . ( [ finiteupperbound ] ) and ( [ nearestlowerbound ] ) . note that the specific case of cubic regions @xmath32^{\times d}$ ] has already been proven in the shorter paper ref.@xcite . * for general finite - ranged harmonic interactions meaning arbitrary interactions which are strictly zero after a finite distance the entropy of entanglement @xmath21 of the ground state scales at most linearly with the surface area of @xmath5 : there exists a @xmath29 independent from @xmath12 and @xmath5 such that @xmath33 this is expressed in eq . ( [ finiteupperbound ] ) . * for gibbs states with respect to a temperature @xmath25 and general finite - ranged interactions the distillable entanglement @xmath34 scales at most linearly with the surface area of @xmath5 , so @xmath35 with @xmath36 being independent of @xmath12 and @xmath5 , see again eq . ( [ finiteupperbound ] ) , which applies for any temperature . we also determine temperatures above which the entanglement is strictly zero , see eq . ( [ nomoretemp ] ) . * we construct a class of hamiltonians the ground state of which exhibit an infinite two - point correlation length , for which the entropy of entanglement @xmath21 scales provably at most linearly in the boundary area . this is stated in eq . ( [ divergenceexample ] ) . * for interactions that are specified by potential matrices @xmath27 ( see eq . ( [ potmat ] ) ) that can be written as @xmath37 with @xmath38 corresponding to a finite - ranged interaction , we determine the entropy of entanglement . this is made specific in particular for @xmath39 , see eq . ( [ squaredonedim ] ) . * for classical systems of harmonic oscillators with nearest - neighbor interactions prepared in a thermal state , the mutual information @xmath40 measuring classical correlations scales linearly with the surface area of the region . here , @xmath41 , @xmath42 , and @xmath43 are the discrete classical entropies in phase space of the interior , the exterior , and the entire system with respect to an arbitrary coarse graining as defined in eq . ( [ boltzmann ] ) , for a classical gibbs state at temperature @xmath25 . the mathematical formulation of the problem is intimately related to case ( vi ) . + the remainder of the paper is now concerned with the detailed discussion and rigorous proofs of these statements . we consider quantum systems on @xmath0-dimensional lattices , where each site is associated with a physical system . the starting point is the following hamiltonian which is quadratic in the canonical coordinates , @xmath44 the matrix @xmath27 is the potential matrix . @xmath45 denotes the potential energy contained in the degree of freedom labeled with @xmath46 . for @xmath47 the element @xmath48 describes the coupling between the two oscillators at @xmath46 and @xmath49 . we assume that @xmath50 is a real symmetric positive matrix . in this paper , we will consider finite - ranged and nearest - neighbor interactions . furthermore , interactions for which the correlation length diverges are considered . we will subsequently study properties of both the ground state @xmath51 as well as of gibbs states @xmath52}\ ] ] corresponding to some nonzero temperature @xmath53 . note that we have set the boltzmann constant @xmath54 . the questions we ask are essentially those of the geometric entropy ( @xmath55 ) , and those on the distillable entanglement ( @xmath25 ) with respect to a distinguished region @xmath5 of the lattice @xmath20 . the specific case of nearest - neighbor interactions , considered in refs.@xcite corresponds to the one of a discrete version of free klein - gordon fields in flat space - time . here , we consider regions of arbitrary shape and allow for more general interactions . the system we consider embodies @xmath2 canonical degrees of freedom , associated with a phase space @xmath56 , where the @xmath57-matrix @xmath58 , @xmath59,\ ] ] specifies the symplectic scalar product , reflecting the canonical commutation relations between the @xmath60 canonical coordinates @xmath61 of position and momentum . instead of considering states we will refer to their moments , considering that both the ground state as well as the gibbs states are quasi - free ( gaussian ) states . the second moments can be collected in the @xmath57 real symmetric covariance matrix @xmath62 ( see , e.g. , ref . @xcite ) , defined as @xmath63)= 2 \re \langle o_j o_k\rangle_{\varrho}.\ ] ] here , it is assumed that the first moments vanish , which does not restrict generality as they can be made to vanish by local unitary displacements that do not affect the entanglement properties . following ref . @xcite , we find from symplectic diagonalization that for the ground state at zero temperature @xmath64 i.e. , there is no mutual correlation between position and momentum , and the two - point vacuum correlation functions are given by the entries of @xmath65 respectively @xmath65 , @xmath66_{{\boldsymbol{i}},{\boldsymbol{j } } } , \label{twopoint1}\\ h_{{\boldsymbol{i}},{\boldsymbol{j } } } & { \ensuremath{\;\raisebox{.06ex}{:}\!\!=}}\langle\text{gs}|p_{{\boldsymbol{i}}}p_{{\boldsymbol{j}}}|\text{gs}\rangle=[v^{1/2}]_{{\boldsymbol{i}},{\boldsymbol{j}}}. \label{twopoint2}\end{aligned}\ ] ] here , @xmath67 denotes the state vector of the ground state . for the thermal gibbs state at finite temperature the covariance matrix takes the form @xmath68 where we define @xmath69 note that the additional term @xmath70 is the same for the position as well as for the momentum canonical coordinates . as in the zero temperature case position and momentum are not mutually correlated and @xmath71 ( @xmath72 ) are the two - point correlation functions of position ( momentum ) , now with respect to the gibbs state . the entropy of entanglement can be expressed in terms of the symplectic eigenvalues of the covariance matrix corresponding to a reduction . let the @xmath73-matrix @xmath74 denote the covariance matrix associated with the interior @xmath5 ; this is the principle submatrix of @xmath75 associated with the degrees of freedom of the interior . the symplectic spectrum of @xmath74 is then defined as the spectrum @xmath76 of the matrix @xmath77 for the situation at hand we find that @xmath78 where we denote by @xmath79 the principle submatrix of @xmath80 associated with the interior and we index the entries of the submatrix as @xmath81_{{\boldsymbol{i}},{\boldsymbol{j}}}$ ] with vectors @xmath82 . given the direct - sum structure of @xmath83 and counting doubly degenerate eigenvalues only once , we find that the entropy of entanglement can be evaluated as @xmath84,\ ] ] where @xmath85 , and the @xmath86 are the square roots of the regular eigenvalues @xmath87 of the @xmath88 matrix @xmath89 , @xmath90 this reflects the fact that the entropy of a harmonic system is the sum of the entropies of uncoupled degrees of freedom after symplectic diagonalization . + using vectors @xmath82 to label the entries of @xmath89 , we find due to @xmath91 , @xmath92_{{\boldsymbol{i}},{\boldsymbol{j}}}&=\left[\left.v^{-1/2}\right|_i\left.v^{1/2}\right|_i\right]_{{\boldsymbol{i}},{\boldsymbol{j}}}\\ & = \delta_{{\boldsymbol{i}},{\boldsymbol{j}}}-\sum_{{\boldsymbol{k}}\in o}\left[v^{-1/2}\right]_{{\boldsymbol{i}},{\boldsymbol{k } } } \left[v^{1/2}\right]_{{\boldsymbol{k}},{\boldsymbol{j}}}. \end{split}\ ] ] [ lb]@xmath93 [ lb]@xmath94 [ lb]@xmath95 [ lb]@xmath96 in one dimension , @xmath39 , for a finite - ranged coupling matrix @xmath27 , @xmath97 $ ] and @xmath98\cup[65, ... ,70]$ ] , yielding @xmath99 . bars show @xmath100 , color - encoded surface depicts @xmath101 . all units are arbitrary . note that the entries decay exponentially away from the boundary of @xmath5.,title="fig : " ] this form of @xmath102 hints at an area theorem in the following way ( see fig . [ r ] , where we depict the entries of @xmath103 ) : if the entries of @xmath80 decay fast enough away from the main diagonal , i.e. , if the correlation functions decay sufficiently fast , the main contribution to the entropy comes from oscillators inside a layer around the surface of @xmath5 as for all the others the product @xmath81_{{\boldsymbol{i}},{\boldsymbol{k } } } [ v^{1/2}]_{{\boldsymbol{k}},{\boldsymbol{j}}}$ ] will be very small ( here vectors @xmath104 are inside @xmath5 , @xmath105 outside the region @xmath5 ) . thus , the matrix @xmath103 has an effective rank proportional to @xmath7 , and the number of symplectic eigenvalues contributing to the sum in eq . ( [ entropy ] ) is approximately proportional to @xmath7 . much of the remainder of the paper aims at putting this intuition on rigorous grounds . for mixed states , such as thermal gibbs states , the von neumann entropy no longer represents a meaningful measure of the present quantum correlations , and has to be replaced by concepts such as the distillable entanglement . the distillable entanglement is the rate at which one can asymptotically distill maximally entangled pairs , using only local quantum operations assisted with classical communication . for pure states , it coincides with the entropy of entanglement the entropy of a reduction giving an operational interpretation to this quantity @xcite . from entanglement theory we know that an upper bound for the distillable entanglement is provided by the logarithmic negativity @xcite . note that this is the case even in this infinite - dimensional context for gaussian states can be immediately verified on the level of second moments and a single - mode description . moreover , the entropy of entanglement indeed still has the interpretation of a distillable entanglement , albeit the fact that distillation protocols leave the gaussian setting . this is true as long as one includes an appropriate constraint to the mean energy in the distillation protocol @xcite . the logarithmic negativity @xcite is defined as @xmath106 where @xmath107 denotes the trace norm , and @xmath108 is the partial transpose of @xmath109 with respect to the split @xmath5 and @xmath4 . again following ref . @xcite , we find after a number of steps @xmath110,\ ] ] where @xmath111 labels the @xmath2 eigenvalues of @xmath112 , @xmath113 and matrices @xmath114 are defined as @xmath115 which become @xmath116 for zero temperature following eq . ( [ wt ] ) . the diagonal matrix @xmath117 , defined as @xmath118 is the matrix that implements time reversal in the subsystem corresponding to the inner part @xmath5 , reflecting partial transposition @xmath119 on the level of states . in this section we will derive upper and lower bounds for the entropy of entanglement and the distillable entanglement of the distinguished region @xmath5 with respect to the rest of the lattice . these bounds only depend on the geometry of the problem , i.e. , the region @xmath5 and on properties of @xmath27 , namely its minimal and maximal eigenvalue , which we define as @xmath120 and @xmath121 , respectively , its condition number @xmath122 and the entries of @xmath123 respectively the two - point correlation functions @xmath124 and @xmath125 as in eqs . ( [ twopoint1],[twopoint2 ] ) . in later sections these bounds will be made specific for a wide range of interaction matrices @xmath27 . an upper bound for the entropy of entanglement and the distillable entanglement is provided by the logarithmic negativity as in eq . ( [ logneg ] ) . utilizing this fact , we derive upper bounds using @xmath126-norms @xcite in this section . for brevity and clarity , we will present the case for @xmath55 and finite temperature in a single argument . a direct calculation shows that the matrix @xmath112 is given by @xmath127 introducing the matrix @xmath128 with entries @xmath129 therefore , we can bound the eigenvalues of @xmath112 according to @xmath130 where we denote by @xmath131 the smallest eigenvalue of @xmath70 which is given by @xmath132 hence , we can write @xmath133 \nonumber \\ & \le&\frac{1}{\ln(2 ) } \sum_{{\boldsymbol{i}}\in c } \max\left\{0,\lambda_{\text{min}}(w(t))^{-2}-1 + 2\left| \lambda_{{\boldsymbol{i}}}(x\omega^+)\right|\right\},\end{aligned}\ ] ] i.e. , for @xmath134 , there is no longer any bi - partite entanglement in the system . we will later see that there is a temperature @xmath135 above which this happens . but for now , we use the fact that @xmath136 to bound the logarithmic negativity , and relate it to the @xmath126 norm @xmath137 @xcite of @xmath138 . we have @xmath139 and therefore @xmath140 the @xmath126 norm is defined as the sum of the absolute values of all matrix entries . inserting the definition of @xmath128 , we find @xmath141 for finite temperature , and @xmath142_{{\boldsymbol{i}},{\boldsymbol{j}}}\right| \label{upperboundzerot}\ ] ] for zero temperature . these constitute upper bounds on the entropy of entanglement and the distillable entanglement . both only depend in the distinguished region @xmath5 , the maximum eigenvalue of @xmath27 , and the entries of @xmath65 , i.e. , the two - point correlation function with entries @xmath143 . this will be the starting point to derive explicit upper bounds for special types of interaction matrices @xmath27 . to achieve lower bounds for the zero temperature case , we consider the entropy of entanglement directly . starting from the general expression eq . ( [ entropy ] ) , we use the fact that the symplectic eigenvalues are never smaller than @xmath10 , i.e. , that the eigenvalues of @xmath89 are contained in the interval @xmath144 $ ] , with @xmath145 being the maximal eigenvalue of @xmath89 . using the pinching inequality @xcite , we find @xmath146 . thus we can bound the entropy of entanglement as follows , @xmath147 = \sum_{i=1}^{v(i ) } \frac{\log [ \lambda_i]}{2 } \ge \frac{\log[\alpha_i]}{2(\alpha_i-1)}\sum_{i=1}^{v(i)}\left(\lambda_i-1\right)\\ & \ge & \frac{\log[\sqrt{\kappa}]}{2(\sqrt{\kappa}-1)}\operatorname{tr}\left[a_i-{{\mathbbm{1}}}\right].\end{aligned}\ ] ] using vectors @xmath82 to label the entries of the @xmath88 matrix @xmath89 , we finally arrive at the lower bound @xmath148}{2(\sqrt{\kappa}-1 ) } \sum_{\substack { { \boldsymbol{i}}\in i \\ { \boldsymbol{j}}\in o } } \left[v^{-1/2}\right]_{{\boldsymbol{i}},{\boldsymbol{j } } } \left[v^{1/2}\right]_{{\boldsymbol{j}},{\boldsymbol{i}}}.\;\;\;\ ; \label{lowerbound}\end{aligned}\ ] ] this lower bound depends only of the geometry of @xmath5 , the spectral condition number @xmath149 of @xmath27 and the entries of @xmath123 , i.e. , the two - point correlation functions . ( [ lowerbound ] ) is difficult to evaluate in general but for special cases of the interaction matrix @xmath27 it can nevertheless be made specific . in section [ nearestneighbor ] , for example , we give an explicit expression for the important case of nearest - neighbor interactions . we expect eq . ( [ lowerbound ] ) to be a convenient starting point to derive such lower bounds also for more general cases of @xmath27 . a lower bound for the finite temperature case is difficult to obtain . generally speaking , a lower bound to the distillable entanglement ( two - way distillable entanglement in our case ) is given by the hashing inequality @xcite , @xmath150 where @xmath151 $ ] and @xmath152 $ ] . yet , naively applied , this inequality will vanish , as generally @xmath153 and @xmath154 : this can be made intuitively clear from the following argument . the above analysis demonstrates that both the interior @xmath5 and the exterior @xmath12 can be approximately disentangled with local unitaries up to a layer of the thickness of the two - point correlation length . that is , degrees of freedom associated with @xmath155 for which @xmath156 is sufficiently large for every @xmath157 , can to a very good approximation be decoupled and unitarily transformed into a thermal gibbs state . each such degree of freedom will therefore contribute a constant number to @xmath158 , such that @xmath153 for sufficiently large @xmath159 . similarly , one can argue to arrive at @xmath154 . in order to establish a lower bound to the distillable entanglement , however , we may start with any protocol involving only local quantum operations and classical communication , and apply the hashing inequality on the resulting quantum states . this first step can include in particular local filterings . a bound linear in the boundary area is expected to become feasible if one first applies an appropriate unitary both in @xmath5 and @xmath12 , and then performs a local filtering involving degrees of freedom associated with @xmath160 and @xmath161 for which @xmath162 . this option will be explored elsewhere . in this section we will make the upper bound on the entropy of entanglement and the distillable entanglement explicit for symmetric finite - ranged interaction matrices @xmath27 , i.e. , matrices for which @xmath163 where @xmath156 denotes again the @xmath10-norm distance . denoting as before the maximum and minimum eigenvalue of @xmath27 as @xmath164 and @xmath165 respectively , we require that the spectral condition number @xmath166 be strictly less than infinity independent of @xmath20 , i.e. , independent of @xmath159 . note that we do not require any further assumptions on the matrix @xmath27 . we will make use of a result of ref . @xcite concerning the exponential decay of entries of matrix functions . after generalizing this result to matrices @xmath27 with the properties specified above it enables us to bound the entries of @xmath167 as follows ( see appendix [ decay ] ) @xmath168 and @xmath169 where @xmath170 for zero temperature we then find @xmath171 this shows that off - diagonal terms of @xmath65 decay exponentially , cf . [ sqrtinvv ] . [ cl]@xmath172 [ cb]@xmath172 [ cc]@xmath172 [ cc]@xmath173 [ cc]@xmath174 [ cc]@xmath175 [ c]@xmath176 _ { { \boldsymbol{i}},{\boldsymbol{j}}}$ ] for a finite - ranged coupling matrix @xmath27 in two dimensions , @xmath177 , @xmath178 . bars show @xmath81_{{\boldsymbol{i}},{\boldsymbol{j}}}$ ] and the color - encoded surface shows @xmath179_{{\boldsymbol{i}},{\boldsymbol{j}}})$ ] . the inset depicts the same for @xmath180 . all units are arbitrary . note the exponential decay away from the main diagonal @xmath181 . ] substituting eqs . ( [ entrybounds ] ) into the general result in ( [ upperboundfinitet ] ) , we find @xmath182 where @xmath183 note that @xmath184 coincides with the definition of the surface area of @xmath5 . we find @xmath185 ( see appendix [ abzaehlerei ] ) , i.e. , we have @xmath186 thus we finally arrive at the desired upper bound linear in the surface area of @xmath5 for both the entropy of entanglement and the distillable entanglement , @xmath187 which becomes @xmath188 for zero temperature , i.e. , these upper bounds depend solely on the maximal and minimal eigenvalue of the interaction matrix @xmath27 and the surface area of the region @xmath5 . this result demonstrates that indeed , an area - bound of the degree of entanglement holds in generality for bosonic harmonic lattice systems . this shows that the previously expressed intuition can indeed be made rigorous in form of an analytical argument . notably , the derived results hold also for systems in which the coupling coefficients are not identical , but independent realizations of random variables . if the coefficients @xmath189 of the real symmetric matrix @xmath27 are taken from a distribution with a carrier @xmath190 $ ] , such that the interaction is ( i ) finite - ranged , and ( ii ) the carrier is chosen such that @xmath191 then the same result holds true . this follows immediately from the considerations in appendix [ decay ] , where it is not assumed that the hamiltonian exhibits translational symmetry . ( [ finiteupperbound ] ) is thus valid also for disordered harmonic lattice systems . in section [ sectionupperbound ] we found that for @xmath192 there is no entanglement between the regions @xmath5 and @xmath12 . for finite - range interactions @xmath27 we found in section [ sectionfiniteranged ] @xmath193 i.e. , we have @xmath194 on the right - hand side only @xmath195 and @xmath196 depend on the temperature , both are decreasing in @xmath53 and go to zero as @xmath53 goes to infinity , i.e. , @xmath197 gives an implicit equation for the temperature @xmath135 above which there is no bi - partite entanglement left in the system . [ nearestneighbor ] in this section we consider nearest - neighbor interactions and periodic boundary conditions in @xmath0 spatial dimensions , i.e. , block - circulant matrices @xmath27 . for @xmath198 this is a special case of the matrices considered in section [ sectionfiniteranged ] as boundary conditions become irrelevant in this limit , i.e. , the upper bound coincides with the one derived for finite - ranged interactions . for a tighter bound see appendix [ nnentries ] . we will now make use of the circulant structure of @xmath27 to show that for matrices of this kind it is possible to also derive a lower bound on the entropy of entanglement that it is proportional to the surface area of @xmath5 . we write @xmath199 for the circulant matrix @xmath200 whose first block column is specified by the tupel of matrices @xmath201 . we can then recursively define @xmath202 via @xmath203 where @xmath204 . we choose an energy scale in which @xmath205 and we demand @xmath29 @xcite . this circulant structure leads to the following properties of @xmath27 : the eigenvalues of @xmath27 are given by @xmath206 in particular the maximum eigenvalue is given by @xmath207 , where equality holds for @xmath159 even . the minimum eigenvalue reads @xmath208 , i.e. , positivity of @xmath27 demands @xmath209 . note that the assumption that @xmath209 independent of @xmath5 and @xmath12 is essential for the argument . if we allow for an @xmath159-dependence of @xmath210 as it arises , e.g. , in the field - limit where @xmath211 , then in one spatial dimension one will encounter an area law up to a logarithmically divergent correction . this behavior resembling the behavior of critical spin chains and quadratic fermionic models will be studied in more detail elsewhere . furthermore the circulant nature of @xmath27 yields the following explicit expressions for the entries of @xmath123 @xmath212_{{\boldsymbol{i}},{\boldsymbol{j}}}= \frac{1}{n^d}\sum_{{\boldsymbol{k}}\in c}{\mathrm{e}}^{2\pi{\mathrm{i}}{\boldsymbol{k}}({\boldsymbol{i}}-{\boldsymbol{j}})/n}\left(\lambda_{{\boldsymbol{k}}}(v)\right)^{\pm 1/2}.\ ] ] this reasoning leads to the following properties for @xmath47 that are crucial for the present proof ( see appendix [ nnentries ] ) , @xmath213_{{\boldsymbol{i}},{\boldsymbol{j}}}[v^{1/2}]_{{\boldsymbol{j}},{\boldsymbol{i } } } = \left|[v^{-1/2}]_{{\boldsymbol{i}},{\boldsymbol{j}}}[v^{1/2}]_{{\boldsymbol{j}},{\boldsymbol{i}}}\right|\\ \ge \left(\frac{c}{2}\right)^{2d({\boldsymbol{i}},{\boldsymbol{j } } ) } \big(\frac{d({\boldsymbol{i}},{\boldsymbol{j}})!}{(2d({\boldsymbol{i}},{\boldsymbol{j}})-1)\prod_{\delta=1}^d|i_\delta - j_\delta|!}\big)^2.\end{gathered}\ ] ] substituting these results into the general lower bound ( [ lowerbound ] ) and keeping only terms with @xmath162 immediately yields @xmath214c^2}{8(\sqrt{\kappa}-1)}s(i ) , \ ] ] where @xmath215 in this case . this generalizes the result of ref.@xcite to regions @xmath5 of arbitrary shape . we expect that these lower bounds can be generalized to other interactions @xmath27 and numerical results suggest that these bounds hold quite generally . a simple special case is related to a certain kind of interaction : this the one where the interaction matrix @xmath27 can be written as @xmath216 with a real symmetric matrix @xmath200 corresponding to a finite - range interaction . then , the covariance matrix associated with the interior is nothing but @xmath217 in this case , the symplectic spectrum can be determined in a fairly straightforward manner . we mention this case also as it appears to be an appropriate toy model as a starting point for studies aiming at assessing the symplectic spectrum of the reduction itself and therefore the spectrum of the reduced density matrix @xcite . we are looking for the spectrum of the @xmath88 matrix @xmath89 as in eq . ( [ amatrix ] ) which now takes the form ( recall that we label the entries of @xmath89 by vectors @xmath218 ) @xmath219_{{\boldsymbol{i}},{\boldsymbol{j}}}=\delta_{{\boldsymbol{i}},{\boldsymbol{j}}}-\sum_{{\boldsymbol{k}}\in o}[m^{-1}]_{{\boldsymbol{i}},{\boldsymbol{k } } } m_{{\boldsymbol{k}},{\boldsymbol{j } } } = \delta_{{\boldsymbol{i}},{\boldsymbol{j}}}-r_{{\boldsymbol{i}},{\boldsymbol{j}}}.\ ] ] it is now the central observation that the number of rows of the matrix @xmath103 that are nonzero and therefore also the number of eigenvalues that are nonzero is proportional to the surface area of @xmath5 , as @xmath200 is a banded matrix . that is , @xmath220 for @xmath221 , meaning that @xmath222 if @xmath223 for all @xmath224 . as the eigenvalues of @xmath89 are bounded through the pinching inequality by @xmath10 and @xmath225 , an upper bound linear in the surface area of the interior follows immediately from the fact that the number of eigenvalues entering the sum in eq . ( [ entropy ] ) is proportional to @xmath7 . this argument demonstrates that in this simple case , one immediately arrives at bounds that are linear in the number of contact points . the task of finding the eigenvalues of @xmath89 explicitly is now reduced to finding the eigenvalues of the sparse matrix @xmath103 , @xmath226 this case is particularly transparent in the one - dimensional case , @xmath39 , and for @xmath200 being a circulant matrix with first row @xmath227 for @xmath228 . the potential matrix @xmath27 corresponds then to nearest - neighbor interactions , together with next - to - nearest - neighbor interactions . in this one - dimensional setting , we set @xmath32 $ ] . the matrix @xmath103 then takes the simple form ( cf . [ sq_r ] ) @xmath229_{i , k}m_{k , j}\\ & = & -c\left([m^{-1}]_{i , m+1}\delta_{j , m}+[m^{-1}]_{i , n}\delta_{j,1}\right).\end{aligned}\ ] ] [ cl]@xmath172 [ bc]@xmath172 [ lt]@xmath230 [ lt]@xmath231 [ lt]@xmath232 [ lt]@xmath233 [ lt]@xmath234 [ cl]@xmath235 for the case @xmath39 , @xmath97 $ ] , @xmath236 $ ] , @xmath237 , where @xmath200 is a nearest - neighbor circulant matrix . note that for this particular form of @xmath27 the entries @xmath100 are exactly zero for @xmath238 . this is in contrast to the general form of @xmath103 depicted in fig . the inset shows the entries @xmath239 as bars and @xmath240 in blue ( cf . ( [ rdecay])).,title="fig : " ] to find its eigenvalues , we calculate @xmath241 , which is straightforward for a matrix of this form . @xmath242 reflecting the fact that @xmath243 eigenvalues are zero , i.e , the number of nonzero eigenvalues is @xmath244)=2 $ ] . we find for the non - vanishing eigenvalues @xmath245 symmetry of @xmath246 yields @xmath247_{1,n}\pm [ m^{-1}]_{m , n}\right).\ ] ] from the circulant structure of @xmath200 we have @xmath248_{i , j}&= & \frac{1}{n}\sum_{k=1}^n\frac{{\mathrm{e}}^{2\pi{\mathrm{i}}k(i - j)/n}}{1 - 2c\cos(2\pi k / n)}\\ & = & \frac{1}{n}\sum_{k=1}^n\frac{\cos(2\pi k(i - j)/n)}{1 - 2c\cos(2\pi k / n)},\\ \ ] ] i.e. , for large @xmath159 we arrive at @xmath248_{1,n}&=&\frac{1}{2c\sqrt{1 - 4c^2}}-\frac{1}{2c},\nonumber\\\ [ m^{-1}]_{m , n}&=&\frac{1}{\sqrt{1 - 4c^2}}\left(\frac{1-\sqrt{1 - 4c^2}}{2c}\right)^m.\label{rdecay}\end{aligned}\ ] ] note that these expressions are asymptotically independent of @xmath249 as shown in the inset of fig . [ sq_r ] . in this limit we finally arrive at @xmath250 this expression specifies the symplectic spectrum of the reduction in a closed form . in this section we will analytically demonstrate that there exist hamiltonians for which the ground state two - point correlation functions diverge whereas the geometric entropy is still bounded by an area - law . in spin systems with a long - range ising interaction such a behavior has been observed in the one - dimensional case and sketched for higher dimensions @xcite . here , we present a class of examples of this type valid in arbitrary dimension @xmath0 , for which one can prove the validity of an upper bound linear in the boundary area . moreover , this set of examples is not restricted to cubic regions . the interaction is here a suitable harmonic long - range interaction . this analysis shows that a divergent two - point correlation function alone is no criterion for a saturating block entropy in the one - dimensional case , and for an area - dependence in higher dimensions . consider the matrix @xmath200 with entries @xmath251 for some @xmath252 . now set @xmath253 . this choice implies that the correlation function @xmath254_{{\boldsymbol{i}},{\boldsymbol{j}}}$ ] decays only algebraically . we will show that despite this fact one still has an upper bound linear in the boundary area of @xmath5 for appropriate values of @xmath255 . firstly , we have to make sure that the maximum eigenvalue of @xmath27 can be bounded from above independent of @xmath159 . from gershgorin s theorem ( see , e.g. , ref . @xcite ) we know that for every eigenvalue @xmath256 there exists an @xmath257 such that @xmath258 i.e. , @xmath259=1,\ ] ] and therefore @xmath260 substituting this and the specific form of @xmath200 into the general expression for the entropy , we obtain @xmath261^\alpha } = \sum_{l=1}^\infty \frac{1}{l^\alpha}n_l\\ & \le & 2^{2d}s(i)\sum_{l=1}^\infty l^{2d-\alpha-1},\end{aligned}\ ] ] which converges for @xmath262 to @xmath263 where @xmath264 is the riemann zeta function . note that these bounds are not necessarily tight in the sense that even for smaller values of @xmath255 , such a behavior can be expected . steps towards tightening these bounds seem particularly feasible in case of cubic regions , where we conjecture that for @xmath265 we arrive at an area - dependence . this analysis shows that for long - range interactions , an area - law in the degree of entanglement can be concomitant with divergent two - point correlation functions . in previous sections we have considered the entanglement between some region and the rest of a lattice of interacting quantum harmonic oscillators with hamiltonians that are at most quadratic in position and momentum . we have demonstrated that both for the ground state and the thermal state of the entire lattice the quantum correlations , i.e. , the amount of entanglement , between the region and the remainder of the lattice is bounded by quantities proportional to their boundary surface area . this suggests similar questions concerning the classical correlations between a region and the rest of the lattice in the corresponding classical systems when the lattice system as a whole is prepared in a thermal state . in this section we will demonstrate that indeed analogous area laws hold . we will note furthermore that there is a quite striking intimate relation between the _ classical system _ with potential matrix @xmath266 and the _ quantum system _ with a potential matrix @xmath267 in this context . for the following considerations we use the classical equivalent of the quantum mechanical hamiltonian ( [ potmat ] ) , namely again @xmath268 where now @xmath269 and @xmath270 are vectors of classical position and momentum variables , respectively , and @xmath27 denotes again the potential matrix . the state of the classical system is characterized by a phase space density @xmath271 , a classical probability distribution , where @xmath272 denotes all the canonical coordinates in phase space . for nonzero temperatures @xmath25 this phase space distribution is given by the boltzmann distribution @xmath273 where @xmath274 ( as before , @xmath54 ) , and @xmath275 given this density in phase space we will encounter a familiar ambiguity when defining the entropy of the system . using the discrete classical entropy @xcite , we split the phase space into cubic cells each with a volume @xmath276 , where @xmath277 and @xmath278 being an arbitrary constant . from the phase space density we obtain the probability associated with each of these cells which in turn can be used to determine the entropy function of this probability distribution . we will now make use of the multiple indices @xmath279 and @xmath280 , assuming that the cell corresponding to @xmath281 is centered around @xmath282 and @xmath283 . that is , for each degree of the @xmath2 degrees of freedom , the phase space is discretized . the contribution of each cell to the discretized probability density is then given by @xmath284 as usual , the discrete classical entropy is then defined as the corresponding shannon entropy @xmath285 we will denote the discrete classical entropy with respect to the degrees of freedom of the interior with @xmath286 , the entropy of the exterior with @xmath287 . the value for the entropy will depend on the choice of @xmath288 and in the limit @xmath289 this entropy definition will diverge due to a term proportional to @xmath290 . in classical statistical mechanics this problem is avoided with the help of the third law of thermodynamics . the entropy itself is however not the quantity that we wish to compute but rather the mutual information @xmath291 between the interior @xmath5 and the rest of a lattice , denoted as before by @xmath12 . this classical mutual information meaningfully quantifies the classical correlations between the inner and the outer . in that case we find that the limit @xmath289 exists and that the mutual information @xmath5 can be defined as @xmath292 following these preparations we are now in a position to determine the mutual information between a region and the rest of the lattice explicitly when the lattice as a whole is in a thermal state . for the evaluation of the mutual information we need to determine the entropy of the total lattice @xmath20 , as well as the entropy determined by the reduced densities describing the two regions @xmath5 and @xmath12 . to this end we carry out the partial summation over all degrees of freedom of region @xmath12 in order to find the reduced phase space density @xmath293 describing region @xmath5 only . employing the schur complement we find that the reduced density @xmath293 is described by the boltzmann distribution corresponding to the same temperature and the hamiltonian @xmath294_{{\boldsymbol{i}},{\boldsymbol{j } } } x_{{\boldsymbol{j } } } + \sum_{{\boldsymbol{i}}\in i } p_{{\boldsymbol{i}}}^2\right ) \ ] ] an analogous result holds for the reduced phase space density of region @xmath12 . for a thermal phase space distribution eq . ( [ density ] ) corresponding to a classical hamiltonian function of the form eq . ( [ classicalham ] ) we can compute the entropy straightforwardly , to find @xmath295 @xmath296 , which increases with temperature as expected . for the mutual information we find @xmath297 which is , perhaps surprisingly , independent of temperature . using jacobi s determinant identity @xmath298 this expression can be rewritten as @xmath299= \frac{1}{2 } \log \det\left [ { { \mathbbm{1}}}-r\right].\ ] ] it is now advantageous to notice the close connection of this expression , in particular of the matrix @xmath103 @xmath300_{{\boldsymbol{i}},{\boldsymbol{k}}}v_{{\boldsymbol{k}},{\boldsymbol{j}}}\ ] ] with those that arise in the quantum mechanical problem that we have treated previously in section [ thasquared ] . indeed , the classical problem for a system with potential matrix @xmath301 is related to the quantum mechanical system with the squared potential matrix @xmath302 . this formal similarity arises because in section [ thasquared ] we have shown that for a lattice of quantum harmonic oscillators with potential matrix @xmath303 in its ground state the symplectic eigenvalues of the covariance matrix describing region i alone are exactly the standard eigenvalues of the matrix @xmath304 . the properties of these eigenvalues have already been discussed in detail in section [ thasquared ] . this allows us now straightforwardly to establish the area theorem for the mutual information in a classical system employing the result for the corresponding quantum system . this establishes in particular that the classical correlations as measured by the mutual information @xmath305 between the distinguished region @xmath5 and the rest of the lattice @xmath4 satisfy @xmath306 for large @xmath6 and appropriate constants @xmath307 independent of @xmath12 and @xmath5 . ( compare also the assessment of the thermodynamical entropy of parts of classical fluids in ref.@xcite . ) in summary we have seen that the area - dependence of correlations is not restricted to quantum systems , as long as one replaces the notion of entanglement representing quantum correlations by the notion of classical correlations in a classical system . in this paper , we have considered the question of the area - dependence of the geometric entropy and the distillable entanglement in general bosonic harmonic lattice systems of arbitrary dimension . the question was the general scaling behavior of these measures of entanglement with the size of a distinguished region of a lattice . such an analysis generalizes assessments of block entropies in the one - dimensional case . using methods from entanglement theory , we established bounds that allow for a conclusion that may be expressed in a nutshell as : _ in surprising generality , we find that the degree of entanglement scales at most linearly in the boundary area of the distinguished region . _ this analysis shows that the intuition that both the interior and the exterior can be approximately disentangled up to a layer of the thickness of the two - point correlation length by appropriate local unitaries carries quite far indeed . for cubic regions @xmath308^{\times d}$ ] the area law can be formulated as @xmath309 where @xmath310 is the landau theta . such area - laws are expected to have an immediate implication on the accuracy to which ground states can be approximated with matrix - product states and higher dimensional analogues in classical simulations of the ground states of quantum many - body systems @xcite . after all , the failure of dmrg algorithms close to critical points can be related to the logarithmic divergence of the block entropy in the one - dimensional case . the findings of the present paper raise a number of interesting questions . notably , in general quantum many - body systems on a lattice ( fermionic or bosonic ) , what are necessary and sufficient conditions for an area - law in the above sense to hold ? clearly , as we have seen above , the divergence of two - point correlation functions alone is not in one - to - one correspondence with an area law . it would be interesting to consider and possibly decide the conjecture that a one - to - one relationship between a system being critical and not satisfying a law of the form as in eq . ( [ thearea ] ) holds if one ( i ) restricts attention to systems in arbitrary dimension with nearest - neighbor interactions , and ( ii ) grasps criticality in terms of two - point correlation functions with algebraic decay , concomitant with a vanishing energy gap . note that the latter two criteria of criticality have not been rigorously related to each other yet and may indeed not be simultaneously satisfied in lattice systems . the general relationship is still awaiting rigorous clarification . as steps towards such an understanding of a relationship between criticality and properties of ground state entanglement in more than one - dimensional fermionic and bosonic systems , it seems very interesting to study models different from the ones considered in this paper . for example , complementing our bosonic analysis , area - laws in fermionic critical systems have been addressed @xcite , where logarithmic corrections have been found also in higher - dimensional settings . other settings in the bosonic case , corresponding to field theories beyond this quasi - free setting , are also still not clarified . in particular , the scaling behavior of the geometric entropy in general bosonic theories in higher dimensions is far from clear . then , the case of finite - size effects in harmonic lattice systems where the correlation length is larger than the full system , resembling the critical case , will be presented elsewhere . in less generality , it seems also feasible to identify the prefactors of the leading and next - to - leading order terms in an area - law of the geometric entropy . it is the hope that the present work can contribute to an understanding of genuine quantum correlations in quantum many - body systems and inspire such further considerations . we would like to thank c. h. bennett , h. j. briegel , j. i. cirac , w. dr , m. fleischhauer , b. reznik , t. rudolph , n. schuch , r. g. unanyan , r. f. werner , and m. m. wolf for discussions . this work has been supported by the epsrc qip - irc ( gr / s82176/0 ) , the eu thematic network quprodis ( ist-2002 - 38877 ) , the dfg ( spp 1078 and spp 1116 ) , the leverhulme trust ( f/07 058/u ) and the european research councils ( euryi ) . the result concerning the exponential decay of entries of matrix functions of ref.@xcite relies on the fact that the @xmath311-th power @xmath312 of a @xmath313-banded matrix @xmath314 is @xmath315-banded , i.e. , @xmath316_{i , j}=0 $ ] for @xmath317 for a matrix @xmath318 with @xmath319 for @xmath320 . for the purposes of the present paper , we will need a generalization of the result of ref . @xcite to block banded matrices . we refer to @xmath321 as being @xmath313-banded , if @xmath322 for @xmath221 . it can be proven by induction over @xmath311 that the @xmath311-th power of @xmath27 is @xmath315 banded in this sense . this enables us to formulate the general form of ref.@xcite as follows . let @xmath321 be a @xmath313-banded symmetric matrix , i.e. , @xmath322 for @xmath323 . define @xmath164 , @xmath165 , @xmath166 , @xmath324 and @xmath325 as an ellipse with foci in @xmath326 and @xmath10 and half axes @xmath255 , @xmath327 , @xmath328 . + now let @xmath329 be such that @xmath330 is analytic in the interior of the ellipse @xmath325 , @xmath331 , and continuous on @xmath325 . furthermore suppose @xmath332 for @xmath333 . then there exist constants @xmath334 and @xmath335 , @xmath336 , @xmath337 such that @xmath338_{{\boldsymbol{i}},{\boldsymbol{j}}}\right| \le k q^{\sum_{\delta=1}^d|i_\delta - j_\delta|},\ ] ] where @xmath339 to bound the entries of @xmath167 , we apply the above theorem to the function @xmath340 for @xmath341 , we have that @xmath342 is analytic in the interior of the ellipse @xmath325 and continuous on @xmath325 , i.e. , @xmath167 satisfy the assumptions of the above theorem . to make @xmath334 and @xmath335 specific , we choose @xmath343 which yields @xmath344 where @xmath345 and for zero temperature @xmath195 reduces to @xmath346 these findings explicitly relate the spectral properties of the hamiltonian to the two - point correlation functions . in this appendix , we make the evaluation of the entries of @xmath347 and @xmath65 of the important case of nearest - neighbor interactions specific . the power series expansion of the square root is given by @xmath348 which is valid for @xmath349 , i.e. , the positivity constraint @xmath350 allows us to write @xmath351 using the multinomial theorem , we have @xmath352 where the sum runs over all @xmath353 with @xmath354 . now , applying the binomial theorem @xmath355 substituting all the above into eq . ( [ entries ] ) , we find @xmath356_{{\boldsymbol{i}},{\boldsymbol{j}}}= \delta_{{\boldsymbol{i}},{\boldsymbol{j}}}\mp\sum_{l=1}^\infty a_l^\pm c^l \sum l!\prod_{\delta=1}^df(d_\delta , n_\delta),\ ] ] where @xmath357 and @xmath358 to the sum over @xmath359 only terms with @xmath360 for some @xmath361 contribute . we thus arrive at the following expression for the entries of @xmath123 @xmath356_{{\boldsymbol{i}},{\boldsymbol{j}}}=\\ \delta_{{\boldsymbol{i}},{\boldsymbol{j}}}\mp\sum_{l=1}^\infty a_l^\pm c^l\!\!\!\ ! \sum_{\sum_\delta n_\delta = l}\!\!\ ! l!\prod_{\delta=1}^d\!\!\sum_{\substack{r=0\\ d_\delta+n_\delta-2r = zn}}^{n_\delta } \!\!\!\!\!\!\!\!\!\!\!\!\!\frac{1}{(n_\delta - r)!r!},\ ] ] i.e. , for nearest - neighbor interactions the product @xmath362_{{\boldsymbol{i}},{\boldsymbol{j}}}[v^{+1/2}]_{{\boldsymbol{i}},{\boldsymbol{j}}}$ ] is always positive for @xmath47 . this does not hold in general and makes it difficult to obtain explicit bounds on the entries of @xmath123 for more general interactions @xmath27 . to obtain a lower bound we keep only the term @xmath363 and @xmath364 . the restriction on @xmath359 is then fulfilled for @xmath365 ( @xmath366 ) if @xmath367 ( @xmath368 ) , yielding @xmath369 it is also possible to obtain an upper bound that is tighter than the one derived for finite - ranged interactions : the elements of @xmath123 are symmetric under @xmath370 , where @xmath371 is a unit vector along dimension @xmath372 . thus , we can demand @xmath373 . then we find that @xmath353 has to be larger or equal to @xmath374 otherwise the restriction on @xmath359 can not be fulfilled . this in turn means that @xmath375 has to be larger or equal to @xmath156 . we then obtain an upper bound by summing all terms in the sum over @xmath359 regardless of the given restriction , yielding @xmath376_{{\boldsymbol{i}},{\boldsymbol{j}}}| & \le & \sum_{l\ge d({\boldsymbol{i}},{\boldsymbol{j } } ) } a_l^\pm c^l \sum_{\sum_\delta n_\delta = l } l!\prod_{\delta=1}^d \frac{2^{n_\delta}}{n_\delta ! } \\ & = & \sum_{l\ge d({\boldsymbol{i}},{\boldsymbol{j } } ) } a_l^\pm ( 2cd)^l\\ & = & \sum_{l=0}^\infty a_{l+d({\boldsymbol{i}},{\boldsymbol{j}})}^\pm ( 2cd)^{l+d({\boldsymbol{i}},{\boldsymbol{j}})}\\ & \le & a_{d({\boldsymbol{i}},{\boldsymbol{j}})}^\pm ( 2cd)^{d({\boldsymbol{i}},{\boldsymbol{j } } ) } \sum_{l=0}^\infty ( 2cd)^l\\ & = & \frac{a_{d({\boldsymbol{i}},{\boldsymbol{j}})}^\pm}{1 - 2cd } ( 2cd)^{d({\boldsymbol{i}},{\boldsymbol{j}})}.\end{aligned}\ ] ] as in appendix [ abzaehlerei ] . as before , oscillators belonging to the distinguished ( shaded ) region @xmath5 are marked , the outside ones are shown as . @xmath377 is shown as the orange shaded area . @xmath378 for a certain oscillator is shaded green , its surface oscillators @xmath379 are depicted by . ] we start by identifying the set of oscillators that can contribute to @xmath380 , @xmath381 , @xmath382 oscillators @xmath383 can only contribute if their distance to the boundary @xmath377 is not larger than @xmath384 , @xmath385 thus , we can restrict the sum over @xmath12 to the set @xmath386 @xmath387 i.e. , we can write @xmath388 where @xmath389 is the number of surface oscillators of a ball with radius @xmath375 within the metric @xmath390 , i.e. , @xmath391 for @xmath392 . using the fact that @xmath393 , @xmath394 can now be bounded from above in the following way @xmath395 where @xmath396 is the volume of a ball with radius @xmath375 within the metric @xmath390 , i.e. , @xmath397 . to summarize , we have @xmath398 g. vidal , j. i. latorre , e. rico , and a. kitaev , phys . * 90 * , 227902 ( 2003 ) , http://arxiv.org/abs/quant-ph/0211074[quant-ph/0211074 ] ; j. i. latorre , e. rico , and g. vidal , quant . comp . * 4 * , 048 ( 2004 ) , http://arxiv.org/abs/quant-ph/0304098[quant-ph/0304098 ] . k. zyczkowski , p. horodecki , a. sanpera , and m. lewenstein , phys . a * 58 * , 883 ( 1998 ) ; j. eisert and m. b. plenio , j. mod . opt . * 46 * , 145 ( 1999 ) ; j. eisert ( phd thesis , potsdam , february 2001 ) ; g. vidal and r. f. werner , phys . a * 65 * , 032314 ( 2002 ) ; k. audenaert , m. b. plenio , and j. eisert , phys . * 90 * , 027901 ( 2003 ) ; m. b. plenio , http://arxiv.org/abs/quant-ph/0505071[quant-ph/0505071 ] . note that the above choice of @xmath29 does not restrict generality : clearly , there exists a diagonal matrix @xmath399 , the elements of which are @xmath400 , such that @xmath401 and @xmath402 . hence , there exists a local symplectic transformation @xmath403 relating the hamiltonian with @xmath404 to the one with @xmath29 .
the modern approach to the study of a dynamical system identifies solving the problem with finding the global structure of phase space . on the other hand , bifurcation theory has mostly concentrated on _ local behavior in the parameter space _ ( so even when global phase space bifurcations are considered , they are always considered in a small neighborhood of a given parameter value ) . here we suggest that global analysis with respect to parameters leads to understanding of the phase space structure in parameter regimes which are unreachable by the currently known analysis techniques . in particular , we examine some features of the global dependence of periodically forced systems with finite forcing amplitude on the forcing frequency . the study of the effect of periodic forcing on homoclinic loops has been thoroughly investigated since the times of poincar . analytical understanding of such systems has been attained in three possible limits ( see @xcite,@xcite ) : the fast and slow oscillations limits and the small forcing limit ( see figure [ fig : spanfam ] ) . for fast oscillations averaging methods apply , and together with kam theory , these are used to obtain upper bounds on the separatrix splitting . these bounds are exponentially small in the frequency ( @xcite,@xcite ) . lower bounds on the separatrix splitting involve delicate analysis which has been proved , under some structural assumptions on the flow , only in recent years ( see @xcite and references therein ) . the slow oscillations regime corresponds to the region where adiabatic theory applies . two approaches have been applied to this limit - the first one corresponds to extending classical adiabatic theory to separatrix crossing ( @xcite,@xcite ) and the other corresponds to examining the geometrical properties of stable and unstable manifolds to hyperbolic manifolds with slowly varying motion on them ( see @xcite,@xcite,@xcite,@xcite ) . most of the analytical studies have been performed in the small oscillations regime in which perturbation methods apply ( see ( @xcite ) and references therein ) . the most important ingredient from these studies to the current paper is the existence of a flux mechanism via lobes ( see @xcite,@xcite,@xcite,@xcite,@xcite ) . here we show that for a large class of systems , the phase space structure and its associated transport properties have some common non - monotone dependence on the frequency of the forcing term . namely , we establish that in some cases , even for finite size oscillations , there exists a function which depends on the frequency continuously . after some natural scaling , this function is simply the sum of the areas of the incoming lobes per period , normalized by the forcing period . hence , results obtained in the fast and slow limits supply information regarding the behavior in the intermediate regime where no analytical methods apply ( see figure [ fig : spanfam ] ) . furthermore , this proposed viewpoint leads to a non - traditional scaling of the flow which is relevant for studying its behavior near homoclinic tangles . this paper supplies mathematical formulation and generalization of the common work of the author with a. poje @xcite , in which similar issues were considered in the context of fluid mixing . the paper is organized as follows : in the first section we define the class of forced hamiltonian families which we consider . roughly , these are forced hamiltonian families with split separatrices , for which the splitting exists for all positive parameter values and as the family parameter varies the scaled frequency includes the semi - infinite line . in the second section we construct a labelling scheme for primary homoclinic orbits , which relies on the concepts of unstable ordering introduced by easton @xcite . this labelling scheme is valid for all parameter values and behaves well across homoclinic bifurcations . in the third section we prove the main result of this paper - that the flux through the homoclinic loop depends continuously ( in fact it is @xmath3 ) in the spanning parameter of the family ( in @xcite we considered a restricted class of systems for which the continuity of the flux followed immediately ) . to prove this result several properties of the lobes are studied , the details are included in appendix 2 . in the fourth section we recall the proof of rk and poje , showing that the flux is non - monotone in the scaled frequency and recall the implications of these results , especially in view of their implications regarding the stochastic zone width ( see @xcite ) . section five includes a demonstration of these concepts for the forced center - saddle bifurcation problem . conclusions are followed by two appendices - in the first the homoclinic scaling is discussed whereas in the second detailed proofs of some properties of the lobes are included , the relation between the flux and the areas of the exit and entry sets is explained , and the adiabatic limit is discussed . consider a one - and - a - half degree of freedom hamiltonian @xmath5 which is @xmath6 periodic in its last argument and depends smoothly on the parameter @xmath7 where @xmath8 may be finite or infinite . this hamiltonian may be written in the form @xmath9 where the second term has zero time average : @xmath10 and may be as large as the first one . assume the hamiltonian system of ( [ ham ] ) : @xmath11 satisfies the following structural assumptions : a1 : : @xmath12 and @xmath13 are @xmath14 functions of all their arguments , @xmath13 is @xmath6 periodic with zero mean in its last argument . a2 : : for all @xmath0 values in @xmath15 $ ] , ( [ hamsys ] ) possesses a hyperbolic periodic orbit which depends smoothly on @xmath0 . a3 : : the family of hamiltonians spans the frequency parameter : the frequency @xmath16 is a smooth @xmath14 function , and it maps the interval @xmath17 onto the semi - infinite line : @xmath18 furthermore , @xmath19 only at one of the interval s boundaries and at no interior point . a4 : : for @xmath20 , for any phase @xmath21 , at least one branch of the stable and unstable manifolds of that periodic orbit coincide to form a homoclinic loop . furthermore , the manifolds and the homoclinic loops created at @xmath20 depend smoothly on @xmath21 , and the phase space area enclosed by these homoclinic loops , @xmath22 , is non - constant and has a finite number of extrema : @xmath23\backslash\{\theta_{1},\theta_{2}, ... ,\theta_{n_{0}}\}\text{.}\ ] ] for concreteness , with no loss of generality , we assume that at @xmath20 the manifolds topology is of a lying figure eight ( closed geometry ) or of a fish swimming to the left ( open geometry ) , so the left branches always coincide at @xmath20 , see figure [ fig1 ] . . a. closed geometry ( lying 8) b. open geometry ( fish).,width=415 ] for @xmath24 , consider the standard global poincar map @xmath25 of the extended phase space @xmath26 which is simply the time-@xmath27 map of the time dependent flow ( here @xmath21 is introduced to account for phase variations ) : @xmath28 the two dimensional symplectic map @xmath25 has , by the above assumptions , a hyperbolic fixed point @xmath29 with its associated stable and unstable manifolds . @xmath30 is a _ primary _ homoclinic point of @xmath29 iff the open segments of the unstable ( respectively stable ) manifold connecting @xmath29 with the homoclinic point @xmath30 , @xmath31 ( respectively @xmath32 ) do not intersect ( @xcite,@xcite ) : @xmath33 the points @xmath34 in figure [ fig2 ] are all primary homoclinic points , whereas the intersection points of the manifolds inside the shaded region in figure [ fig5 ] are homoclinic points which are not primary . we assume : a5 : : for all @xmath35 there is a finite number of primary homoclinic orbits . furthermore , for each intersecting branches of the manifolds there exist at least two primary homoclinic orbits which are topologically transverse . a6 : : the number of _ primary _ homoclinic bifurcations is bounded in any bounded open interval of @xmath0 for all @xmath35 . the order of the _ primary _ homoclinic bifurcations is uniformly bounded by the integer @xmath36 + for convenience of notation , let us assume that @xmath37 finally , to establish asymptotic behavior in the frequency , we assume there exists an appropriate scaling so that the scaled separatrix length scales and the maximal velocity along the separatrix are of order one : a7 : : there exist smooth ( @xmath38 in @xmath0 ) non - vanishing scaling functions , @xmath39 @xmath40 such that in the scaled variables : @xmath41 the scaled system : @xmath42 with the two non - dimensional parameters @xmath43 : @xmath44 satisfies assumptions a1-a6 . _ furthermore , for the scaled system , the width and length of the separatrix and the maximal velocity along it are all of order one for all _ @xmath45__. _ _ @xmath46 measures the relative strength of the temporal oscillations to the mean flow whereas @xmath1 compares the oscillation s time scale with the travel time along the separatrix loop ( outside of the saddle orbit neighborhood ) . we assert that the magnitude of these two parameters supplies _ complete information on the qualitative behavior _ of the system ( [ schamsys ] ) and hence on the original system . notice that assumption a3 is now made with respect to the scaled frequency . a family of hamiltonian systems depending on a parameter @xmath0 which satisfies assumptions a1-a7 is called a _ _ frequency - spanning homoclinic family , _ _ and @xmath0 is called a _ frequency - spanning parameter . _ a few remarks are now in order : * assumptions a1-a7 clearly hold in the standard near integrable case ( small @xmath46 ) where the steady flow has a homoclinic loop of fixed size ( independent of @xmath0 ) , the perturbation is generic ( so a5 and a6 are satisfied ) , and its frequency , @xmath47 spans the half real line . * a5 refers only to _ primary _ homoclinic bifurcations - otherwise it would have been violated generically , see @xcite and references therein . * in the near - integrable case the scaling functions may be simply extracted from the integrable flow , see for example section [ sec : sadcen ] . in appendix 1 , after some notation is established in the next section , we propose an algorithm for defining the characteristic scales in the finite amplitude size forcing case . * the assumptions here are slightly weaker than the ones in @xcite : in particular a5 and a6 replace the stronger assumption of @xcite that there exists one topologically transverse primary homoclinic orbit which depends continuously on @xmath0 , a property which may be easily violated in applications . relaxing this assumption requires the introduction of a new labelling scheme for the homoclinic points which takes into account the possible annihilation of primary homoclinic points . hereafter we assume the forced system is in its scaled form and we drop all the over bars when non - ambiguous , as described next . denote by @xmath48 @xmath49 the _ topologically transverse primary _ homoclinic _ orbits _ of @xmath29 belonging to the left branch ( for simplicity of notation we consider hereafter only the left branch of the loop ) , where @xmath50 , namely the discrete index @xmath51 denotes iterations under the map and the index @xmath52 denotes the label of the specific orbit . each such transverse homoclinic point , @xmath53 defines a region , @xmath54 enclosed by the closed segments of the stable and unstable manifolds which emanate from the fixed point and meet at @xmath53 namely @xmath55\cup s[\gamma ( \nu),p_{i}^{r_{n}}(\nu ) ] , \label{rdef}\ ] ] where , clearly @xmath56 denote by @xmath57 the ordered parameter values at which primary homoclinic bifurcations occur ( by a6 this may be done ) , and by @xmath58 the set of all primary homoclinic bifurcation values in @xmath17 : @xmath59 @xmath60 . in figures [ fig2 ] and [ fig3 ] we illustrate a possible behavior of the manifolds before and after a primary homoclinic bifurcation occurs . we say that @xmath61 undergoes a homoclinic bifurcation at @xmath62 if @xmath63 is a tangent homoclinic orbit . next we construct a labelling of the orbits in @xmath64 which is convenient even across homoclinic bifurcations . consider a frequency spanning homoclinic family . then , there exists a labelling scheme for the primary homoclinic orbits satisfying the following properties : 1 . @xmath65 is piecewise constant , changing only at values of @xmath0 at which @xmath66 undergoes a homoclinic bifurcation . the labelling scheme respects the unstable ordering , so that @xmath67 iff @xmath68 , namely ( see @xcite ) : @xmath69\subset u[\gamma(\nu),p_{0}^{r_{m}}(\nu)]\text { \ \ for \ \ } r_{n}<r_{m}. \label{orderhom}\ ] ] by construction . we construct such a labelling scheme inductively , in the intervals @xmath70 starting with the interval @xmath71 by a5 , @xmath72 for all @xmath35 . since the poincar . map is orientation preserving and since only topologically transverse intersections are counted , @xmath73 is even . by assumption a6 , there exists a well defined limit of the number of primary homoclinic orbits as @xmath74 , so that @xmath75 is well defined and is finite , and @xmath76 . for @xmath77 , let @xmath78 the labelling is chosen to obey the unstable ordering @xmath79 for @xmath80 where we identify : @xmath81 these rules determine the labelling for @xmath82 up to a cyclic permutation , and do not define the origin ( which homoclinic point is @xmath83 ) . to remove the first ambiguity , define the labelling so that as @xmath74 , the first homoclinic orbit defines the region of maximal area for all @xmath84 : @xmath85 . to remove the second ambiguity , we define the origin ( @xmath86 as the point for which the boundaries of @xmath87 are of minimal length ; denote the arc length of the boundary of @xmath88 by @xmath89 . notice that @xmath90 as @xmath91 . let @xmath83 , satisfy @xmath92 . when there are no degeneracies ( here , when the maximal area is achieved at a unique orbit and the minimal boundary length is achieved at a unique homoclinic point ) this procedure determines uniquely the labelling of the topologically transverse homoclinic orbits for @xmath93 if a degeneracy occurs , choose any one of the finite number of the maximizing ( respectively minimizing ) orbits ( respectively points ) . of figure [ fig4].,width=415 ] given the labelling @xmath94 , for @xmath95 , so that @xmath96 for @xmath97 , we construct @xmath98 as follows : if the associated homoclinic orbit , @xmath99 , depends continuously on @xmath0 at @xmath100 leave @xmath101 unchanged . if @xmath99 cease to exist for @xmath102 , delete @xmath101 from @xmath103 if @xmath104 additional homoclinic orbits are created between ( along the unstable ordering ) @xmath99 and @xmath105 insert the @xmath104 labels , @xmath106 , @xmath107 , so that @xmath108 by construction the labelling scheme respects the unstable ordering and does not change labels unless homoclinic bifurcation occur , as claimed . notice that the first case in the proof includes two possibility which are topologically equivalent - the first is that the homoclinic orbit @xmath109 is not involved in the homoclinic bifurcation and the second possibility is that @xmath109 is a tangent periodic orbit with odd order tangency so that no topological bifurcation occurs at @xmath100 . for any @xmath0 we have constructed a finite , ordered set of labels @xmath110 ( so @xmath111 ) with corresponding homoclinic orbits which satisfy ( [ orderhom ] ) . the ordering @xmath112 enumerates the primary homoclinic orbits in @xmath64 by their unstable ordering . while the @xmath113 depend smoothly on @xmath0 as long as @xmath114 has not bifurcated , @xmath115 can be discontinuous in @xmath65 due to bifurcations of orbits @xmath116 with some @xmath117 . for example , in the simplest , generic case of quadratic tangency , if at @xmath118 two primary homoclinic orbits are created between ( in the sense of the unstable ordering ) the @xmath119th and the @xmath120 homoclinic orbits which existed for @xmath121 , then these will be labeled as @xmath122 so that @xmath123 @xmath124 as demonstrated in figure [ fig3 ] : @xmath125 where @xmath126@xmath127 see figure [ fig4].,width=415 ] by construction , for @xmath82 , @xmath128 or equivalently , @xmath129 . on the other hand , for @xmath130 , for the generic case of ( [ pwexm ] ) we have @xmath131 so @xmath132{l}r\text { \ \ \ \ \ \ \ \ \ for } r\leq j\\ \lbrack r]+[(r-[r])\cdot10]\text { \ \ for } j < r < j+1\\ r+2\text { \ \ for } r\geq j+1 , \end{array } \right . \text { \ \ \ \ for \ } \nu_{1}<\nu<\nu_{2}\ ] ] and we see that @xmath133 changes at @xmath118 for all @xmath134 hence , as opposed to the @xmath101 s , it does not serve as a good labelling system for the homoclinic orbits across bifurcations . represents natural parametarization along the unstable manifold.,width=415 ] consider now a fixed @xmath0 value , with @xmath135 where @xmath136 , and @xmath137 . henceforth we omit the explicit dependence on @xmath0 unless needed . @xmath138 @xmath139 are called neighbors if @xmath140 @xmath141 and @xmath142 ( see ( [ rdef ] ) ) are called neighboring regions if @xmath138 are neighbors . the difference between neighboring regions @xmath141 and @xmath142 are lobes - regions which are enclosed by the segments @xmath143\cup$ ] @xmath144 $ ] : a lobe @xmath145 is the region enclosed by @xmath146\cup$ ] @xmath147 $ ] where @xmath148 are neighboring primary homoclinic points . let @xmath149 where ( [ cycllab ] ) is used to define @xmath150 , see figure [ fig2 ] . next we prove that this definition indeed corresponds to the figure , namely that the choice of the labelling is such that @xmath151 is exterior to @xmath152 whereas @xmath153 is inside the region @xmath154 : [ lem : ekdk]for all @xmath155 and all @xmath156 the lobes @xmath153 and @xmath157 satisfy denotes a union of sets with disjoint interior . ] : @xmath158 first we prove that for @xmath82 , the lemma follows from the definition of the labelling and claim [ uinout ] which is proved in appendix 2 ; recall that for these @xmath159s @xmath160 and @xmath161 has maximal area . by the lobes and regions definitions , it follows immediately that neighboring regions defer from each other by lobes , so proving the lemma amounts to proving that : 1 . the lobes boundaries can not intersect the corresponding boundaries of the regions , hence they can be either completely interior or completely exterior to the regions . the definition of the exiting ( @xmath162 s ) and entering ( @xmath163 s ) lobes is consistent with ( [ nghbrreg ] ) . the first part is contained in claim [ uinout ] of appendix 2 , where it is proved that:@xmath164@xmath165@xmath166 where @xmath167 denotes topologically transverse intersection . these results , together with the observation that @xmath168\subset$ ] @xmath169 , implies that the lobe @xmath170 , which is enclosed by the segments @xmath171\cup$ ] @xmath168 $ ] is either contained in @xmath161 or is completely outside of it . hence , either @xmath172 or @xmath173 . however , the latter contradicts the choice of @xmath161 as the region with maximal area as @xmath74 . another manifestation of the labelling scheme may be formulated by looking at the manifold s orientation : @xmath174 iff @xmath171 $ ] is interior to @xmath168 $ ] as shown in figure [ fig2 ] . the manifolds orientation is clearly preserved for @xmath175 . furthermore , by topological transversality of the points in @xmath64 it follows immediately that all odd ( respectively even ) indexed points have the same manifolds orientation as of @xmath83 ( respectively @xmath176 ) , so that for @xmath82 the lemma is proved for all @xmath156 . for @xmath177 , the lemma is proved by induction , noticing that at the bifurcations , topologically transverse homoclinic points are inserted / deleted in neighboring pairs , hence the orientation of the manifolds at @xmath178 ( respectively at @xmath179 ) is always as that of @xmath180 ( respectively as of @xmath181 ) . notice that tangent , non topologically transverse homoclinic points are not labeled . for all @xmath155 the regions with even order have smaller areas than their neighbors , namely @xmath182 are regions of locally maximal area whereas @xmath152 are regions of locally minimal areas . finally , note the following property of the lobes : [ simplecon]the lobes @xmath183 @xmath184 are simply connected . the lobe s boundaries , by definition , are given by @xmath185 , which are both connected arcs which are joined at their end - points , hence the lobes are connected . each component of the boundary can not intersect itself , and by lemma [ uinout ] they can not intersect transversely each other either , so the only source of non - trivial topology may stem from tangencies . the appearance of primary homoclinic tangency ( e.g. at @xmath186 of figure [ fig3 ] ) does not change the simple connectedness property of a lobe . finally , it follows from eq . ( [ eq : trans ] ) that the lobe can not reconnect and form a ring ( even in the closed geometry case ! ) . for every @xmath0 there exist a collection of incoming and outgoing lobes . their areas changes with @xmath0 , and , at homoclinic bifurcation points these changes can be discontinuous ( at even order bifurcations where lobes split or coincide , as shown in figure [ fig3 ] ) . nonetheless , [ prop : c1areas]consider a frequency - spanning homoclinic family . the sums of the areas of the outgoing and incoming lobes : @xmath187 are @xmath3 in @xmath4 it follows from lemma [ simplecon ] that away of the primary homoclinic bifurcation values @xmath188 the area of each lobe depends smoothly ( @xmath38 ) on @xmath0 , hence so do @xmath189 and @xmath190 . by orientation preservation , near the bifurcation values an even number of topological transverse primary homoclinic points are added or eliminated . consider first a tangential bifurcation of even order occurring at @xmath100 so that @xmath191 new topologically transverse homoclinic points are created at @xmath192 these new homoclinic points , appearing , for example , between the homoclinic points @xmath193 , are denoted by @xmath194 , @xmath195 see figure 5 , is similar : simply change the corresponding indices and interchange the letters @xmath163 and @xmath162 in this paragraph . ] . the bifurcation splits the lobe @xmath196 ( enclosed by @xmath197\cup s[p^{2j+1},p^{2j}]$ ] ) to the @xmath198 disjoint lobes @xmath199 created by the new homoclinic points . as @xmath200 by the smooth dependence of the manifolds on @xmath0 ( recall that only a finite extension of the local stable and unstable manifolds is considered here ) , the areas of the interior , newly created lobes ( @xmath201 ) vanish , whereas @xmath202 . the area of the diminishing lobes near @xmath100 is of the form @xmath203 where @xmath204 is the order of the tangency at @xmath100 . clearly the same argument applies to the case where pairs of homoclinic orbits annihilate at @xmath205 it follows that generically @xmath206 and @xmath207 are @xmath3 in @xmath0 ( i thank b. fiedler for this observation ) . intuitively , we think of the @xmath157 lobes as existing lobes and the @xmath153 as entering lobes . a more elegant framework for discussing transport in phase space is achieved by introducing the notion of exit and entry sets ( see @xcite , @xcite ) : for any region @xmath208 , the exit ( @xmath209 ) and entry ( @xmath210 ) sets of @xmath208 are defined as : @xmath211 in appendix 2 we study the relation between entry and exit sets of the regions @xmath212 and the corresponding entering and exiting lobes . in particular , we define , for each region @xmath212 the set of turnstile lobes @xmath213 in terms of union of entering and exiting lobes ( see equations ( [ lobed ] ) in the appendix ) . we conclude that in case there are no intersections between these two sets they correspond exactly to the entry and exit sets of @xmath212 . however , if such intersections exist , the two sets are different . we provide an example which demonstrates that when such intersections exist the measure of the entry and exit sets depends on @xmath214 _ we therefore conclude that while the notion of entry and exit sets appears more elegant , it has a major flaw from a dynamic point of view - it is not invariant with respect to the regions definition . _ another problem arising from the definition of the entry and exit sets has to do with self intersection of the lobes ( for close geometry , in the limit of highly stretched lobes ) . indeed , the boundaries of the entry and exit sets may depend on the appearance of non - primary homoclinic orbits . the appearance of such orbits is expected to depend sensitively on parameter values . equivalently , the proposition regarding the smooth dependence of the lobes area on parameters is proved for the sum of their areas , @xmath190 ( see [ sumed ] ) , and not for the area of the lobes union . in fact , any definition of flux which depends on homoclinic orbits of higher order ( namely not primary homoclinic orbits ) must be carefully examined due to the recent results on richness of chaos @xcite in the neighborhood of a homoclinic tangency . hence , we define the flux into a region using the concept of the sum of lobes areas : consider a frequency spanning homoclinic family . the _ scaled flux function _ of this family is : @xmath215 [ cor : conflux]the flux function of a frequency spanning homoclinic family depends continuously and has a continuous derivative ( is @xmath3 ) in the frequency spanning parameter . by proposition [ prop : c1areas ] and assumptions a3 and a7 . notice that the scaled system defines a scaled flux function . substituting the scaling function , we observe that the flux function is scaled by the typical area and time scale associated with travelling along the homoclinic loop , as appropriate : @xmath216 consider a frequency - spanning homoclinic family . then , in the scaled system ( [ schamsys ] ) the scaled flux function ( eq ( [ fluxdef ] ) ) depends non - monotonically on the scaled frequency @xmath217 recall that by corollary [ cor : conflux ] the flux function depends continuously on @xmath0 and that @xmath218 as @xmath219 , the system ( [ schamsys ] ) approaches the adiabatic limit . then , by corollary [ melimit ] of appendix 2 , @xmath220 , which implies , by assumption a3 and a7 , that @xmath221 . furthermore , since @xmath118 ( the first primary homoclinic bifurcation value ) is bounded away from zero , it follows from a3 that there exists a @xmath222 such that @xmath223 and @xmath224 is monotone in the interval @xmath225 $ ] . on the other hand , as @xmath226 , averaging of the scaled system ( [ schamsys ] ) implies that the separatrix splitting is exponentially small in @xmath227 ( see @xcite,@xcite,@xcite ) . it follows that @xmath228 exponentially as @xmath229 ( where @xmath230 , and by a3 @xmath231 ) , and so does @xmath224 . in particular , there exists a @xmath232 @xmath233 such that for @xmath234 @xmath159 , @xmath235 @xmath236 . this proves that @xmath224 is non - monotone in @xmath4 we now discuss some of the implications of the non - monotone behavior of the flux function , see @xcite for a fluid mixing application of these results . first , we notice that in the near - integrable case , when the number of primary homoclinic orbits is fixed , the flux function is proportional to the amplitude of the melnikov integral . hence the above theorem proves that the melnikov function amplitude is non - monotone in this case , which indeed explains the non - monotone figure one typically gets in numerous calculations of the melnikov integral ( e.g. forced duffing s , forced cubic potential , forced pendulum ) . it is clear that very small ( large ) flux corresponds to a small ( large ) chaotic region near the separatrix , hence , it was suggested that the melnikov function amplitude gives a good characteristic to the amount of chaos in the system . however , we proved that the flux function is non - monotone in the spanning parameter . it is therefore natural to compare the dynamics , and the properties of the chaotic region at equi - flux parameter values . let @xmath237 denote equi - flux parameter values such that : @xmath238 it follows immediately , by the flux definition , that @xmath239 namely , the sum of the lobes areas of the lower frequency is larger than that of the higher frequency . this observation , which is well reflected when one examines poincar maps of equi - flux frequencies may lead to the wrong conclusion that adiabatic mixing leads to a larger chaotic zone . in fact , at least for small amplitudes , we can use @xcite to show that the converse result hold : @xmath240 where @xmath241 denotes the area of the mixing zone ( the area enclosed by the kam tori which are closest to the separatrix on either side of the loops ) . a more precise statement of these observations , with numerical demonstrations , may be found in @xcite . the seemingly contradicting statements are well understood when one realizes that for small @xmath242 the lobes overlap considerably - in the adiabatic limit , the measure of overlap between the turnstile lobes may approach the full lobe s area ( see @xcite for a proof ) , namely there exists a @xmath119 such that @xmath243 where @xmath244 are defined by eq . ( [ lobed ] ) in appendix 2 . as an example consider the forced saddle center bifurcation : @xmath245 for @xmath246 . near @xmath247 a hyperbolic periodic orbit appears . its separatrix length scales ( width and height ) in the @xmath248 directions are @xmath249 , so @xmath250 the maximal velocity along the loop scales like @xmath251 . finally , since @xmath252{\nu}})$ ] . we obtain the scaling : @xmath253{\nu}t,\ \ \\ \ \ \ \overline{h } & = \nu^{-3/2}h\end{aligned}\ ] ] hence , the effective homoclinic forcing amplitude and frequency are : @xmath254{\nu } } \label{sadnodpar}\ ] ] and the scaled system ( with over bars dropped where non ambiguous ) is : @xmath255 namely , the effective frequency and damping increases inversely with the bifurcation parameter . for fixed @xmath256 and fixed @xmath257 we recover the well known result that the separatrix splitting is exponentially small as the bifurcation is approached . for fixed @xmath258 we see that the exact dependence on @xmath259 matters . for large @xmath260 an appropriate ( see @xcite ) melnikov calculation for ( [ eq : ressadcen ] ) will lead to a function of the form : @xmath261{\nu}})\frac{b}{\nu^{1-n/2}}\exp(-\frac{\pi\omega}{\sqrt[4]{\nu}})\cos\frac{\omega t}{\sqrt[4]{\nu}}\ ] ] where @xmath262 is , up to exponential small corrections , a polynomial . if @xmath263 and @xmath264 , then @xmath265{\nu}})=k\frac{\omega } { \sqrt[4]{\nu}}$ ] ( see @xcite ) . let @xmath266 , and consider a frequency @xmath16 which vanishes at some finite distance from the bifurcation value and is non decreasing as the bifurcation value is approached . for example , take @xmath267 where @xmath268 , @xmath269 . then , the family of forced saddle - center bifurcation with @xmath270 $ ] is a spanning family , satisfying assumptions a1-a4 . verifying assumptions a5-a6 amounts to obtaining a lower bound on the separatrix splitting for equation ( [ eq : ressadcen ] ) as @xmath271 , a task which has been achieved in @xcite for the case @xmath263 , and we will assume here that a similar lower bound may be found for @xmath272 so that a5 and a6 are satisfied . a7 is clearly satisfied with the scaled parameters [ sadnodpar ] . furthermore , the scaled frequency @xmath273 is monotone for @xmath270 $ ] , and @xmath274)=[0,\infty)$ ] . in particular , taking @xmath275 shows that the system : @xmath276 is transformed to the scaled system ( [ eq : ressadcen ] ) with @xmath277 and @xmath278{\nu}}$ ] . we therefore conclude that the homoclinic structure of this family behaves in a non - monotone fashion - there exists at least one value of @xmath279 @xmath280 such that the scaled homoclinic flux function : @xmath281 has a local extrema at @xmath282 . in particular , this implies that the corresponding melnikov function amplitude is non - monotone in @xmath4 one expects that @xmath283 , hence @xmath284 , namely the maxima location scales with the size of the interval on which the family is spanning . we have demonstrated that the concept of spanning families , in which one considers global dependence on parameters , leads to non - trivial results regarding the properties of forced systems . this approach leads to the development of a new tool - a novel way of labelling of primary homoclinic points across bifurcations . using this tool we are able to prove several results regarding the global dependence of the lobes on the spanning parameter . we proved that lobes are simply connected . we proved that under natural assumptions on the spanning families the flux is continuous and non - monotone in the spanning parameter . we demonstrated that under the same conditions the area of the exit and entry sets may depend on the spanning parameter discontinuously . we demonstrated that this approach leads to non - standard view of the forced saddle - center bifurcation . the relation of this work to higher dimensional systems , in which the spanning parameter is in fact a slowly varying variable is yet to be explored . discussions with b. fiedler and d. turaev are greatly appreciated . below we supply an algorithm for determining the length and time scales characteristics for the finite amplitude forcing case . consider the original system . define : @xmath285 where the regions @xmath88 are the regions defined by the primary homoclinic points @xmath99 as in section [ sec : label ] . the characteristic width and length of the separatrix loop are given by : @xmath286 where @xmath287 and @xmath288 is some finite number . let @xmath289 the choice of the labelling so that @xmath290 is minimal among all @xmath291 and the oscillatory nature of the boundary for large @xmath51 , implies that @xmath292 is expected to be small , and therefore independent of @xmath288 if @xmath288 is sufficiently large . the characteristic velocities along the separatrix loops are similarly defined : @xmath293 where @xmath294 denotes the primary homoclinic orbit with the initial condition @xmath295 . since @xmath296 are all primary homoclinic orbits ( in particular , bounded orbits ) the functions @xmath297 are clearly well defined and are independent of the choice of the labelling . away from the homoclinic bifurcation points @xmath298 depend smoothly on @xmath0 and @xmath299 depend continuously on @xmath0 . at @xmath300 these functions may be discontinuous . let @xmath301 denote @xmath38- functions , which are close in the @xmath302 topology to the corresponding hatted functions , uniformly in @xmath0 . these are the proposed scaling functions for assumption a7 . [ uinout]for all neighboring homoclinic points @xmath303 either @xmath304 or @xmath305 , namely @xmath306 and similarly@xmath307@xmath308 ( where @xmath167 denotes topologically transverse intersection ) . in particular , @xmath309 recall that by definition @xmath310\cup s[\gamma , p_{0}^{r_{j}}]$ ] . since the unstable segments can not intersect each other clearly @xmath311=\emptyset.$ ] so , to prove ( [ uisout ] ) we need to prove that @xmath312=\emptyset$ ] . notice that @xmath313=s[\gamma , p_{0}^{r_{j+1}}]\cup s[p_{0}^{r_{j+1}},p_{0}^{r_{j}}]$ ] . since @xmath314 are primary homoclinic points @xmath315 @xmath316 and in particular @xmath317 and @xmath318 hence , it is left to prove ( [ uands ] ) . assume the contrary : denote by @xmath319 all the topologically transverse homoclinic points in this intersection . denote by @xmath320 the homoclinic point which is closest to @xmath321 by the unstable ordering , namely @xmath322 . then , it follows from ( [ usinters ] ) and the observation that @xmath323 that @xmath324 namely @xmath325 is a topologically transverse _ primary _ intersection point , which contradicts the statement that @xmath326 are neighbors in @xmath64 . hence ( [ uisout ] ) and ( [ uands ] ) are proven . to prove ( [ sisout ] ) notice that @xmath327\cup s[\gamma , p_{0}^{r_{j+1}}]$ ] , hence , to prove the claim we need to prove that @xmath328=$ ] @xmath329\cup u(p_{0}^{r_{j}},p_{0}^{r_{j+1}}))=\emptyset.$ ] the intersection with the first term is empty by ( [ usinters ] ) and the transverse intersection with the second term is empty by ( [ uands ] ) consider first the case @xmath333 where only two topologically transverse primary homoclinic orbits exist . from ( [ nghbrreg ] ) we immediately get that : @xmath334 hence @xmath335 and@xmath336 therefor , we see that @xmath337 , @xmath338 and @xmath339 iff @xmath340 . in particular , in figure [ fig5 ] we demonstrate that if @xmath341 and @xmath342 we obtain : @xmath343 so the measure of the entry and exit sets depend on the region one chooses , while the measure of the entry and exit lobes are independent of this arbitrary choice . furthermore , we see that the entry and exit sets depend on the nature of non - primary homoclinic orbits . the appearance of such orbits may depend sensitively on parameter values . it follows ( after some manipulation of ( [ nghbrreg ] ) ) that in the open geometry case ( where @xmath345 for all @xmath346 ) the exit and entry sets of the regions @xmath212 are given by : @xmath347{c}e_{r_{0}^{r_{j}}}=e^{j}\backslash\left ( e^{j}\cap d^{j}\right ) \\ d_{r_{0}^{r_{j}}}=f^{-1}\left ( d^{j}\backslash\left ( e^{j}\cap d^{j}\right ) \right ) \end{array } \text { \ if \ } e_{i}\cap e_{j}=d_{i}\cap d_{j}=\emptyset\text { \ for all } i\neq j.\text{\ \ \ \ \ \ \ \ \ } \ ] ] in particular , in the near - integrable case , where @xmath348 for all @xmath349 the exit and entry sets are given by the entering and exiting lobes . indeed , for @xmath350 we find:@xmath351 and therefore the first example follows this general rule . the limit of @xmath352 is called the adiabatic limit . it corresponds to a forcing with frequency which is much smaller than the time scale associated with the flow along the separatrix . the traditional way of analyzing this limit is by using the adiabatic approximation . in this approximation one considers solutions to the frozen hamiltonian @xmath353 , namely the time - dependent oscillations are now fixed at the frozen phase @xmath21 . one then studies adiabatic variables ( to leading order , the action ) , and asserts that in the nonlinear case , as long as the solutions are bounded away from the separatrix , such variables can not change by much due to the persistence of kam tori . furthermore , when the solution crosses the separatrix one can estimate the resulting change in the adiabatic invariances and the phases by constructing an adiabatic separatrix mapping ( see for example @xcite,@xcite,@xcite,@xcite ) . another point of view which is applicable to the same limit ( see @xcite,@xcite and @xcite ) , is concerned with the geometry of the manifolds and lobes . consider the extended system @xmath354 , and realize that the manifold @xmath355 is a normally hyperbolic manifold for @xmath356 , hence it persists with its stable and unstable manifolds for sufficiently small @xmath357 . the solutions belonging to these manifolds can be well approximated , for semi - infinite time intervals by the extended system at @xmath358 denoting the homoclinic solution to this extended frozen system by @xmath359 and the orbits belonging to the stable and unstable manifolds of the extended system by @xmath360 we see that as @xmath361@xmath362\\ q^{u}(t;\tau_{0},\nu ) & = q_{h}(t;\tau_{0}+\overline{\omega}(\nu ) t)+ ... ,\text { \ \ \ \ } t\in\lbrack - t,\infty)\end{aligned}\ ] ] where dots stand for higher order terms in @xmath1 , @xmath363 is a positive constant and the higher order terms are small for @xmath364 @xmath365 . recall that in the limit @xmath74 ( which , by a3 corresponds to the limit @xmath366 ) , there are @xmath75 topologically transverse primary homoclinic orbits , @xmath367 . by definition , these orbits belong to both @xmath368 and @xmath369 for some phase @xmath370 , so that @xmath371 @xmath372 . we conclude that these phases , @xmath373 , are given , to leading order in @xmath374 by the phases at which the area enclosed by the frozen homoclinic loop achieves its maxima or minima : ( kaper and kovacic ) let @xmath375 denote the @xmath376 s at the minima and maxima of the homoclinic loop area @xmath377 , where @xmath378 , and @xmath379 . then , as @xmath361 , the primary homoclinic orbits @xmath380 and : @xmath381 where @xmath382 denotes the region enclosed by @xmath383 $ ] and @xmath384 $ ] . the structure of the lobes in this limit is more complicated than the commonly seen graphs of homoclinic tangles . in particular , on one hand , generically , only one of the primary homoclinic points @xmath389 is bounded away from the fixed point @xmath29 , leading to the creation of large lobes with area of @xmath390 ( see @xcite ) . on the other hand the lobes get stretched and elongated , creating a web of small intersecting regions ( see @xcite,@xcite ) . ( neishtadt @xcite ) let @xmath392 denote the phase at which @xmath377 is minimal ( @xmath393 ) . then , as @xmath352 , the area enclosed by the largest kam torus inside @xmath394 ( for any @xmath259 ) , @xmath395 asymptotes @xmath396 : @xmath397
there is no more fundamental physical quantity in cosmology than the expansion rate of the universe . in recent years the present value of the expansion rate , hubble s constant , has been measured with increasing accuracy @xcite . with the exploration of the universe at redshifts of order unity , we now have information about the time evolution of the expansion rate @xcite . a most surprising result is that the time evolution of the expansion rate does not seem to be described by a matter - dominated friedmann - lematre - robertson - walker ( flrw ) cosmological model . the usual explanation for the discrepancy is that there is a new component of the energy density of the universe , known as dark energy , that determines the recent evolution of the expansion rate . of course all indications for dark energy are _ indirect _ ; they all involve some form of the time evolution of the expansion rate . since the expansion rate of the universe is of such fundamental importance , we must understand any possible effects that would result in an expansion rate different from the flrw prediction . in this paper we study the change in the expansion rate due to perturbations of a homogeneous , isotropic , flrw model . in particular , we perform a second - order calculation of the effect of inhomogeneities on the expansion rate . we only consider modifications to the expansion rate of a matter - dominated universe , although our results can be extended to a universe containing a mixture of matter and a cosmological constant . we find that the mean corrections are a few parts in @xmath0 . the expansion rate of a perturbed flrw cosmology has been discussed in many works ( although we believe ours is the first complete second - order calculation ) . hui and seljak @xcite estimated the order of magnitude of the effect by considering a representative second - order term . they arrived at the correct mean order of magnitude of the result . some recent works @xcite suggested that small - scale contributions could give a large correction , producing an apparent accelerated expansion of the universe . in particular rsnen @xcite suggested that due to ultraviolet sensitivity , one of the second - order terms could give large corrections depending on boundary conditions . as we will discuss in the conclusions , we find that if one employs the correct averaging procedure the result ( at least for the terms we could compute exactly using the second - order formalism ) is well behaved in the ultraviolet and the term identified by rsnen does not result in a large correction . in addition to calculating the mean value of the expansion rate , we calculate the variance about the mean value . in calculating the variance we uncover an interesting infrared effect . at second order the expansion rate has a term proportional to @xmath1 , where @xmath2 is the peculiar gravitational potential , related to the density perturbations through the cosmological poisson equation . if inflation is the origin of perturbations , then @xmath3 should be a gaussian random variable with zero mean . however , if the perturbation spectrum on super - hubble - radius scales is no bluer than a harrison zeldovich spectrum , then the variance of @xmath3 formally has an infrared singularity . now if the value of @xmath3 in our local hubble volume is sufficiently large to modify the expansion rate , our perturbative expansion fails . nevertheless , our results suggest that if the super - hubble modes of @xmath3 have physical significance , then a nonperturbative extension of our calculation could yield a most important modification to the friedmann equation . one might even speculate that a complete treatment of the effect could explain the observed time dependence of the expansion rate on its own and obviate the need for the dark - energy assumption . since that conjecture is beyond the perturbative calculation of this paper , we will postpone discussion of this point to a subsequent communication @xcite . in the next section we discuss the general perturbative expansion . in section [ thetasynchronous ] we calculate the corrections to the expansion rate in terms of metric fluctuations , and then express the results in terms of the density perturbation spectrum . in section [ numbercrunch ] we present the numerical results . we then conclude , followed by two technical appendices . in this section we describe how to treat the average properties of a perturbed universe up to second order in the metric variables , including the effect of the inhomogeneous gravitational field on the homogeneous background field . we then discuss the proper definition of the average expansion rate and express the effect of inhomogeneities to second order as averages of density perturbations . by `` average , '' we mean the average over a spatial hypersurface at a given time . clearly an average defined in this way depends on the chosen coordinate system , _ i.e. , _ the gauge . we will consider a universe filled only by irrotational dust and choose the coordinates of an observer at rest with the dust ( _ i.e. , _ comoving coordinates ) , and with the same time coordinate for every point of the spatial hypersurface ( _ i.e. , _ synchronous coordinates ) . this system of coordinates can be chosen if the universe is filled by a single pressureless component . since the pressure vanishes , the only force acting on the particles is gravity , and the comoving world lines coincide with geodesics . ( see ref . @xcite for a full discussion of this point . ) we will call @xmath4 the ( conformal ) time in this gauge , and @xmath5 the spatial coordinates , so that the metric has the form @xmath6 \label{metric}\ ] ] we will perform the averages on constant-@xmath4 spatial hypersurfaces . the goal of this paper is a second - order calculation , but before embarking , let us recall the familiar linearized first - order result . the energy - momentum tensor and metric are expanded to first - order as [ subonee ] @xmath7 where the superscript @xmath8 denotes the @xmath9-th order perturbation . by definition @xmath10 and @xmath11 are homogeneous and isotropic . in the synchronous gauge , to first order the metric may be written in terms of a set of perturbation variables consisting of two scalars ( @xmath12 ) , a vector ( @xmath13 ) , and a tensor ( @xmath14 ) . the vector @xmath15 is transverse ( @xmath16 ) and the tensor @xmath14 is symmetric , transverse , and traceless ( @xmath17 , @xmath18 , @xmath19 ) . in terms of these variables the metric is @xmath20 where @xmath21 . the metric perturbations @xmath22 and @xmath23 may be expressed in terms of the peculiar gravitational potential @xmath24 , which is related to @xmath25 , the first - order density perturbation , by the poisson equation , @xmath26 ignoring metric perturbations that decay in time , for a matter - dominated universe @xmath22 and @xmath23 are given by ( see ref . @xcite ) in ref . @xcite corresponds to our variable @xmath27 , and vice versa . ] [ fomp ] @xmath28 to first order we will neglect vector modes , as they do not arise in conventional perturbation - generation mechanisms such as inflation . we will also assume the tensor mode amplitude is small . in calculating spatial averages we will also require @xmath29 , where @xmath30 is the determinant of the spatial metric . to first order @xcite , @xmath31 whether calculating to first order or second order , when calculating the spatial average @xmath32 of a quantity @xmath33 one must fix a system of coordinates in which to express @xmath34 , and then integrate with the proper integration measure over @xmath35 at fixed @xmath4 . for all orders we adopt the definition which is a scalar under the residual gauge freedom . ] @xmath36 where the domain of integration is some large volume . if @xmath37 is a homogeneous quantity ( _ i.e. , _ it does not depend on @xmath5 , but only on @xmath4 ) , then simply @xmath38 . in the first - order calculation , if @xmath39 is already a first order quantity we may set @xmath40 , and @xmath41 becomes @xmath42 to first order , the einstein equations are @xmath43 which yield upon averaging the @xmath44 component @xmath45 here we see that @xmath46 may be interpreted as an extra component to the stress - energy tensor . for the unperturbed flrw cosmology , the expansion rate @xmath47 is found from the @xmath48 component of the einstein equations . for the first - order perturbed flrw model @xmath49 where @xmath50 . this is the basic point . in a perturbed flrw cosmology , @xmath51 is _ not _ @xmath52 . , where @xmath53 is related to @xmath4 by @xmath54 . ] however , one must be careful to find the true variables that describe the evolution of the averaged background . it is not clear that the quantity @xmath47 describes the physical hubble flow . the correct quantity to describe the hubble flow may be found from the evolution of @xmath55 . we know that for a homogeneous isotropic universe there are two independent equations that govern the dynamics of the expansion . so we have to find the two independent equations that describe the evolution of averaged physical quantities such as like @xmath56 . for the unperturbed model we may augment the @xmath48 equation with the continuity equation , @xmath57 . this gives for a perfect pressureless fluid ( dust ) @xmath58 where @xmath59 and @xmath60 is the fluid four velocity . ( [ continuity ] ) is true to all orders . for an unperturbed flrw universe with scale factor @xmath61 , this immediately gives @xmath62 , which results in @xmath63 . for the perturbed model , we want to expand eq . ( [ continuity ] ) to first order and then average it . this will give us the effective scale factor for the dilution of the matter density . in addition to the expansion of @xmath64 , we expand @xmath65 as @xmath66 and obtain @xmath67 using the fact that @xmath68 we are left with @xmath69 this tells us that in a first - order perturbed universe the average matter density is diluted with expansion according to @xmath70 , which in general is _ not _ equivalent to @xmath71 . an alternative way to express the result is to define a new scale factor , @xmath72 , such that @xmath73 . it is clear that @xmath72 must be defined by @xmath74 and so the _ true _ scale factor is @xmath72 and not @xmath61 , at least for the dilution of matter ( see , _ @xcite ) . the physical quantity of interest is @xmath75 . it is given by @xmath76 we will define a @xmath77 , and express @xmath65 as @xmath78 two important notational points : we have defined @xmath79 ( _ not _ as @xmath47 ) ; and we have defined @xmath80 as the difference between @xmath75 and @xmath81 ( _ not _ as the difference between @xmath75 and @xmath82 ) . to calculate @xmath83 , we can express eq . ( [ backeinstein001 ] ) in the form @xmath84 where of course the second equality holds if the corrections are small . combining eqs . ( [ combo11 ] ) and ( [ combo21 ] ) , we find @xmath85 it will turn out that in the first - order calculation @xmath86 will appear only as a spatial gradient in the final expression for @xmath80 . this means that the physical result is insensitive to the choice of the normalization of @xmath87 , since we could always add a constant to @xmath86 to make @xmath88 anything we please . the cancellation of the non - gradient first - order terms in eq . ( [ combo31 ] ) seems accidental , but in appendix [ alternative ] we derive an expression for @xmath80 that at first order only contains spatial derivatives of @xmath86 , which do not change if we shift @xmath86 by a constant ( or even a time - dependent renormalization ) . the spatial metric @xmath89 is expanded up to second order ( neglecting first order vector and tensor perturbations ) as . ] @xmath90 the functions @xmath91 , @xmath92 , @xmath93 , and @xmath94 represent the @xmath9th - order perturbation of the metric . vector and tensor modes have been included here at second order as they are dynamically generated by the non - linear evolution of purely scalar perturbations @xcite . it will turn out that @xmath95 is the only second - order term for which we will require the explicit form . for a matter - dominated universe the metric perturbation @xmath95 is obtained similarly to ref . @xcite @xmath96 the first term arises from a primordial epoch of inflation and can be computed as follows . there exists a second - order extension of the well - known gauge - invariant variable @xmath97 , the curvature perturbation on uniform density hypersurfaces . in terms of @xmath98 , to first order it is given by @xmath99 . to second order , @xmath100 , where @xmath101 remains _ constant _ on superhorizon scales in the case in which only adiabatic perturbations are present . in standard single - field inflation , @xmath101 is generated during inflation and its value is given by @xmath102 @xcite . since in the synchronous gauge and on superhorizon scales @xmath103 and @xmath104 , one readily concludes that the primordial contribution to the second - order metric perturbation @xmath105 is given by the constant term @xmath106 and it also propagates to the coefficients of the term proportional to @xmath107 . in the second - order calculation care must be taken in defining the spatial average . again , if @xmath108 is a homogeneous quantity , then simply @xmath109 . if @xmath110 is already a second - order quantity , then we can take @xmath40 , both in the numerator and in the denominator . but in the second - order calculation of @xmath111 we must remember to include @xmath29 . using eq . ( [ fosd ] ) , @xmath112 where we have introduced the notation @xmath113 for a first - order term to denote the spatial average with the factor @xmath114 , _ i.e. , _ @xmath115 . so the correct second - order averaging procedure gives @xmath116 as an illustration of the subtleties in averaging , if at first order @xmath117 , where the first - order averaging prescription is defined in eq . ( [ ffo ] ) , at second order the averaging prescription for a first - order quantity is defined by eq . ( [ avgone ] ) , and @xmath118 need not vanish . now we perform a second - order expansion . we first expand the energy - momentum tensor and the metric as [ ourway ] @xmath119 where again the superscript @xmath8 represents the @xmath9th - order perturbation . again we will take @xmath120 and @xmath121 to be homogeneous and isotropic , so @xmath122 and @xmath123 . note that this is _ not _ equivalent to the statement that @xmath124 and @xmath125 . expanding the einstein tensor @xmath126 to second order yields several types of terms . the first type are the zeroth - order terms , denoted @xmath127 , which only involve @xmath121 . @xmath128 will be homogeneous since @xmath121 is homogeneous by definition . then there are first - order terms , @xmath129,which involve a single power of @xmath130 , possibly combined with @xmath121 . there are two types of second - order terms . the first type , denoted by @xmath131 , involves squares of @xmath130 . the second type of second - order terms , denoted @xmath132 , involves @xmath133 . now we consider the einstein equations @xmath134 we proceed by averaging eq . ( [ pee ] ) . this yields @xmath135 from eq . ( [ backeinstein ] ) we see that now @xmath136 may be interpreted as an extra component to the stress - energy tensor . to second order , the @xmath48 component of the perturbed flrw model gives @xmath137 again we see that @xmath51 is _ not _ @xmath138 . now let us consider the second - order expression for the evolution of @xmath139 . we again use the continuity equation , eq . ( [ continuity ] ) , and expand @xmath64 and @xmath65 as @xmath140 the result is a bit more complicated than the first - order result : @xmath141 using the fact that @xmath142 , the left - hand side of eq . ( [ coneq ] ) becomes @xmath143 we may also express the right - hand side of eq . ( [ coneq ] ) as @xmath144 equating eqs . ( [ lefths ] ) and ( [ righths ] ) we obtain @xmath145 since to first order a simple calculation yields @xmath146 , we are left with @xmath147 this tells us that as in a first - order perturbed universe , in a second - order perturbed universe the average matter density is diluted with expansion according to @xmath70 , which in general is _ not _ equivalent to @xmath71 . note that it was crucial to define the averages with the @xmath29 factor ; otherwise we would not have discovered the right quantity to describe the scaling of @xmath56 . again , the physical quantity of interest is @xmath75 , given by @xmath148 we emphasize again that @xmath79 ( and _ not _ @xmath47 ) . the goal of this paper is to calculate @xmath83 . to do so , we can express eq . ( [ backeinstein00 ] ) in the form @xmath149 where of course the second equality holds if the corrections are small . using eq . ( [ combo2 ] ) , we find @xmath150 it will turn out that @xmath151 will appear only as a spatial gradient in the final expression for @xmath80 . this means that now the physical result is insensitive to the choice of the normalization of @xmath152 , since we could always add a term that is spatially constant . the cancellation of the non - gradient second - order terms in eq . ( [ combo3 ] ) seems accidental , but again , in appendix [ alternative ] we derive an expression for @xmath80 that explicitly only contains spatial derivatives of @xmath151 , which do not change if we shift @xmath151 by a term that is spatially constant ( or even a time - dependent normalization ) . the relevant quantity which we want to calculate is @xmath83 . the simplest way to proceed is to perform the computation directly in the synchronous gauge . the synchronous coordinates are more physical for our purposes . if the calculation is performed in the poisson gauge it is necessary to perform a complex calculation ( as we do in appendix [ thetapoisson ] ) to express the result in synchronous coordinates ( see also refs . @xcite and @xcite ) . moreover , in general for second - order calculations , the synchronous gauge turns out to be very convenient , since here the scalar perturbations do not have nonlocal terms that appear in other gauges such as the poisson gauge . also , for matter @xmath153 in this gauge takes the trivial form @xmath154 , so the only terms that have to be computed are the christoffel symbols in the covariant derivative @xmath155 moreover , in the synchronous gauge , to second order @xmath156 @xcite . finally , the computation of the perturbation of @xmath65 with respect to its background value consists of finding just the trace of @xmath157 . from @xcite , @xmath158 in the last equation we expressed the result in terms of the initial perturbation @xmath3 , using the time evolution of the relevant perturbation variables in the synchronous gauge given in eqs.([fomp ] ) and ( [ somp ] ) . the simplest way to find @xmath159 is to realize that @xmath160 , and use the explicit solutions for @xmath161 and @xmath162 from ref . @xcite , but using for @xmath105 the expression augmented with the @xmath163 as in eq . ( [ somp ] ) . in the matter - dominated universe , @xmath164 the value of @xmath165 is found by summing eqs . ( [ fundamentalsyn ] ) and ( [ fundamentalsyng ] ) . we must now perform the appropriate spatial average . in @xmath77 there is only one first - order term ; the rest of the terms are second order . the first - order term must be averaged using the procedure of eq.([soa ] ) , while the other terms are averaged using @xmath114 . the result is @xmath166 . \label{risultatowithbt}\end{aligned}\ ] ] we now proceed to express the averages in terms of the matter power spectrum . the procedure is to fix a spherical domain of radius @xmath167 with volume @xmath168 . from eq . ( [ average ] ) and the definition of @xmath113 , all of the averages in eq.([risultatowithbt ] ) involve integrals of the form @xmath169 for calculational convenience we will employ a gaussian window function and assume @xmath168 is a spherically symmetric volume with volume element @xmath170 and volume @xmath171 . the fourier transform of the window functions is @xmath172 of course as @xmath173 , @xmath174 . we wish to evaluate the typical expected value of @xmath65 averaged over this sphere . by `` typical expected value '' we mean the ensemble average . the metric fluctuation @xmath3 is treated as a gaussian variable with zero mean ( of which we know the @xmath175-point correlation functions ) that takes random values over different `` realizations '' of volumes @xmath168 . in other words , we calculate the typical value of a quantity for a region of radius @xmath167 as the statistical mean over many different similar regions . we will indicate this statistical average with a bar : @xmath176 . we will express @xmath3 and its derivatives in terms of a fourier integral , so @xmath177 the fourier components @xmath178 satisfy @xmath179 \ p_{\varphi}(k_1)p_{\varphi}(k_2 ) \right\ } \label{fourpoint } , \end{aligned}\ ] ] where @xmath180 . from eq.([nabla2varphi ] ) we can express @xmath181 in terms of the matter power spectrum as @xmath182 where @xmath183 is the ( dimensionless ) power spectrum of the matter density fluctuations . let us first consider @xmath184 and @xmath185 . clearly from eq . ( [ onepoint ] ) , @xmath186 and @xmath187 . however this does not imply that @xmath188 or @xmath189 over any individual volume of radius @xmath167 . the question is the magnitude of typical departures from the mean values , which corresponds to the statistical variance of our quantities . as we will show , it is intuitively clear that if the radius @xmath167 is big enough , this variance will go to zero . so the effect of variance could be important . we will return to the calculation of the variances of the different terms after completing the calculation of the mean values . next , consider @xmath190 . passing to fourier space , we have @xmath191 \nonumber \\ & = & - \int \frac{d^3\!k_1}{(2\pi)^3 } \frac{d^3\!k_2}{(2 \pi)^3 } \left ( \frac{k_1 ^ 2 + k_2 ^ 2}{2 } \right ) \overline{\varphi_{\vec{k}_1 } \varphi_{\vec{k}_2 } } \ w\left(\left|\vec{k_1 } + \vec{k_2}\right| r\right ) .\end{aligned}\ ] ] making use of eq . ( [ twopoint ] ) , we find @xmath192 using eq . ( [ pphipdelta ] ) , we can express @xmath193 in its final form . in the same manner we find the means for the other terms . the result is @xmath194 of course in the limit @xmath195 , eqs . ( [ ggg ] ) and ( [ hhh ] ) vanish . therefore , the entire second - order calculation gives @xmath196 where we have used @xmath197 appropriate for a matter - dominated universe . we will give numerical results in the next section . now for the variances of selected terms . the variance is defined as @xmath198 = \overline { \left ( \left\langle\cdots\right\rangle - \overline{\left\langle\cdots\right\rangle } \right)^2 } .\ ] ] for instance , @xmath199 & = & \frac{9}{4 } a^4 h^4 \int_0^\infty \frac{dk}{k^5 } \delta^2(k , a ) \ w^2(kr ) \label{varphi } \\ \textrm{var}\left[\left\langle\nabla^2\varphi\right\rangle_1\right ] & = & \frac{9}{4 } a^4 h^4 \int_0^\infty \frac{dk}{k } \delta^2(k , a ) \ w^2(kr ) . \label{varlinear}\end{aligned}\ ] ] the variances of other terms are more complicated , but straightforward to derive . for instance , @xmath200 = \frac{1}{(2 \pi)^6 } \int d^3\!k_1 d^3\!k_2 \frac{1}{2}\left [ k_1 ^ 2+k_2 ^ 2+\frac{13}{9}\vec{k}_1\cdot\vec{k}_2 \right]^2 w^2(|\vec{k_1}+\vec{k_2}| r ) \ p_{\varphi}(k_1 ) p_{\varphi}(k_2 ) . \label{twopointvar}\ ] ] the angular integrals can be expressed in terms of a filter function , defined in general as @xmath201 for a gaussian window function the filter expression can be expressed in terms of incomplete @xmath202 functions as @xmath203 \left[\gamma(l+1,-\,2r^2 { k_1 } { k_2 } ) - \gamma(l+1 , 2r^2{k_1}{k_2})\right ] } { { 2^{l+1}r^{2(l+1)}{k_1}}^{l+1}{{k_2}}^{l+1 } } .\end{aligned}\ ] ] we will make use of the fact that for @xmath204 and @xmath205 , @xmath206 . using the gaussian filter , eq . ( [ twopointvar ] ) becomes @xmath207 & = & \left ( \frac{9}{8 } a^4 h^4 \right)^2 \int_0^{\infty } \frac{d k_1}{k_1 ^ 3 } \delta^2(k_1,a ) \int_0^{\infty } \frac{d k_2}{k_2 ^ 3 } \delta^2(k_2,a ) \left [ \left(\frac{k_1 ^ 2}{k_2 ^ 2}+\frac{k_2 ^ 2}{k_1 ^ 2}+2\right ) j^{(0)}(k_1,k_2,r)\right.\nonumber\\ & & \left . + \frac{26}{9}\left ( \frac{k_1}{k_2}+\frac{k_2}{k_1}\right ) j^{(1)}(k_1,k_2,r)+\frac{169}{81}j^{(2)}(k_1,k_2,r ) \right ] . \label{phi2k2}\end{aligned}\ ] ] it is interesting that the integral is not well behaved in the infrared . as discussed in the next section , for a harrison zeldovich spectrum @xmath208 , so the term proportional to @xmath209 has an infrared divergence : @xmath210_\infty & = & \left ( \frac{9}{8 } a^4 h^4 \right)^2 \int_0^{\infty } \frac{d k_1}{k_1 ^ 3 } \delta^2(k_1,a ) \int_0^{\infty } \frac{d k_2}{k_2 ^ 3 } \delta^2(k_2,a ) \left(\frac{k_1 ^ 2}{k_2 ^ 2}+\frac{k_2 ^ 2}{k_1 ^ 2}\right ) j^{(0)}(k_1,k_2,r ) \nonumber \\ & = & \left ( \frac{9}{8 } a^4 h^4 \right)^2 4 \int\frac{dk_2}{k_2}\delta^2(k_2,a ) \int\frac{dk_1}{k_1 ^ 5}\delta^2(k_1,a ) .\end{aligned}\ ] ] we will discuss the importance of this term below . now we turn to the variances of @xmath211 and @xmath212 . they are given by [ uytr ] @xmath213 & = & \left(\frac{9}{4}a^4h^4\right)^2 \left[\int \frac{dk_1}{k_1}\delta^2(k_1,a)\,w^2(k_1r ) \int \frac{dk_2}{k_2 ^ 5 } \delta^2(k_2,a)w^2(k_2r ) \right . \nonumber \\ & & \left . + \left ( \int\frac{dk}{k^3}\delta^2(k , a)w^2(kr)\right)^2 \right ] \label{another } \\ \textrm{var}\left [ \left\langle\nabla^2\varphi\right\rangle \left\langle\nabla^2\varphi\right\rangle \right ] & = & 2\left(\frac{9}{4}a^4h^4\right)^2 \left ( \int\frac{dk}{k}\delta^2(k , a)w^2(kr)\right)^2 .\end{aligned}\ ] ] note that eq . ( [ another ] ) also has an infrared divergence @xmath214_\infty = \left ( \frac{9}{8 } a^4 h^4 \right)^2 4 \int\frac{dk_2}{k_2}\delta^2(k_2,a ) \int\frac{dk_1}{k_1 ^ 5}\delta^2(k_1,a ) .\ ] ] of course we will be interested in the variance of the total expression eq.([risultatowithbt ] ) , not the individual terms . of particular interest is the cross term of the infrared - singular parts . this will be the only one for which we will include the cross terms . the infrared - singular pieces appear in eq . ( [ risultatowithbt ] ) proportional to @xmath215 . the infrared part of the variance of this term is @xmath216_\infty = \left(\frac{9}{4 } a^4 h^4 \right)^2 4 \int\frac{dk_2}{k_2}\delta^2(k_2,a ) \int\frac{dk_1}{k_1 ^ 5}\delta^2(k_1,a ) . \label{blowup}\ ] ] it is straightforward to obtain the variance for the second - order , four - derivative terms . ( the sum of these terms has zero mean . ) it is given by @xmath217 & = & \left(\frac{9}{4 } a^4 h^4 \right)^2 \int_0^{\infty } \frac{d k_1}{k_1 } \delta^2(k_1,a ) \int_0^{\infty } \frac{d k_2}{k_2 } \delta^2(k_2,a ) \left [ j^{(0)}(k_1,k_2,r ) -2 j^{(2)}(k_1,k_2,r ) \right . \nonumber \\ & & \left.+j^{(4)}(k_1,k_2,r ) \right ] . \label{phi2k4}\end{aligned}\ ] ] there is another potential contribution that will result in contributions to the variance similar to the terms we have found . suppose we expand @xmath77 to _ third _ order in perturbation theory . we can express @xmath77 in the general form @xmath218 where @xmath219 , @xmath220 , and @xmath221 are operators which contain derivatives . then the variance of @xmath77 will contains terms like @xmath222 , @xmath223 , and @xmath224 . the first term is the usual cosmic variance term ; it is well behaved in the infrared . the second term is singular in the infrared ; it is given in eq . ( [ blowup ] ) . the third term will be present , in principle it also will have an infrared singular part , and its value requires a relativistic third - order perturbation calculation . however there is no reason for the infrared singular part of the @xmath225 term to cancel exactly the infrared singular part of the @xmath226 term . finally we remark on the ultraviolet behavior of the corrections to the expansion rate . the second - order result for the mean , eq . ( [ thisisit ] ) , should be well behaved in the ultraviolet . as discussed in the next section , in the linear regime @xmath183 increases logarithmically with @xmath227 for a harrison - zeldovich spectrum , so the first term should be well behaved in the ultraviolet . the ultraviolet behavior of the last two terms are regulated by the filter function @xmath228 . the contributions to the variance of the terms we have calculated , eqs . ( [ varlinear ] ) , ( [ phi2k2 ] ) , ( [ uytr ] ) , and ( [ phi2k4 ] ) , all involve filter functions that regulate the ultraviolet behavior . however , we expect there to be terms that do not involve filter functions . for instance , if one performs the relativistic third - order calculation , one expects to find contributions to @xmath80 from terms like @xmath229 . the variance would then include evaluation of terms like @xmath230 ( an example of the aforementioned @xmath225 terms ) . these terms would include parts with a momentum integration unregulated by a filter function . if there are third - order terms with large numbers of derivatives bringing down large powers of momentum , then the variance might be sensitive to the ultraviolet behavior . in the next section we show the numerical results for the mean and variance of the corrections to the expansion rate . in this section we present the numerical results for @xmath83 for a matter - dominated universe . we will give the mean values , as well as the variances . for both the mean and the variances we express the power spectrum @xmath183 in terms of the transfer function @xmath231 . for a harrison zeldovich spectrum , the power spectrum is @xmath232 where @xmath233 is the dimensionless amplitude , @xmath234 . we will discuss the implications of other spectra . for our purposes the bardeen , bond , kaiser , szalay ( bbks ) transfer function @xcite will be adequate . the bbks transfer function may be expressed in terms of a dimensionless parameter @xmath235 where @xmath202 is the shape factor , defined for a flat universe in terms of the baryon fraction @xmath236 and the total value of @xmath237 as @xmath238 . in terms of @xmath239 , @xmath240^{-1/4 } .\ ] ] of course at small @xmath239 , @xmath241 , while at large @xmath239 , @xmath242 . also , in all expressions we make use of the fact that in a matter - dominated universe @xmath243 , where @xmath244 is the present value of the scale factor . consider the mean , given by eq . ( [ thisisit ] ) : @xmath245 where @xmath9 is the dimensionless size of the region , @xmath246 . we will present results for @xmath247 ( so @xmath248 ) and the last two terms are negligible . while we have indicated the range of integration from @xmath249 to @xmath250 , in reality to employ eq . ( [ powerspectrum ] ) there is a cutoff on the maximum value of the integral . the ultraviolet cutoff arises because density perturbations become nonlinear . the mean value is very insensitive to the ultraviolet cutoff . the integral for the mean value receives most of the contribution in the decade between @xmath251 . the result for the present mean value of @xmath252 is shown in fig . [ meaneds ] for various choices of @xmath202 . it scales as @xmath253 . the present mean value , @xmath252 , in the matter - dominated universe as a function of the ultraviolet cutoff @xmath254 for various values of the shape factor @xmath202.,scaledwidth=90.0% ] now consider the variance about the mean value . first consider the variance of the linear term , eq.([varlinear ] ) . this term contributes to the variance in @xmath255 an amount @xmath256}}{3h } \right|_\textrm{linear term}= \frac{5}{9}\frac{1}{a^2h^2}\sqrt{\textrm{var } \left[\left\langle\nabla^2\varphi\right\rangle_1\right ] } = \frac{5}{6 } \left[\int_0^\infty \frac{dk}{k } \delta^2(k ) \ w^2(kr ) \right]^{1/2}.\ ] ] defining a dimensionless wavenumber @xmath257 , the term becomes @xmath256}}{3h } \right|_\textrm{linear term } = \frac{5}{6}a \left(\frac{h\ , \textrm{mpc}^{-1}}{h_0}\right)^2\frac{a}{a_0 } r^{-2}\left [ \int_0^{x_\textrm{max } } dx \ x^3 \ t^2(x/\gamma r ) \ e^{-x^2 } \right]^{1/2 } \simeq \frac{a}{a_0}\frac{100}{r^2}\quad ( \gamma r\gg1),\ ] ] where the last expression holds for @xmath258 . the result is shown in fig . [ variance ] . because of the window function , the results do not depend on @xmath254 , the ultraviolet cutoff ( so long as it is greater than about @xmath259 ) . the result is also well behaved in the infrared . of course as @xmath195 , the variance of the linear term approaches the mean of the linear term , which is zero . we can see comparing fig . [ meaneds ] and fig . [ variance ] that the variance in the linear term dominates the mean value out to distances of about @xmath260 mpc . the present contribution to @xmath261}/3h$ ] from the indicated terms in a matter - dominated universe as a function of @xmath167 , the volume sampled . the results are for @xmath262 and @xmath263 . on the logarithmic scale , the dependence on @xmath202 and @xmath264 is not noticeable . we emphasize the point made in the text that the calculation for the the second - order , two - derivative term has an infrared cutoff of the present hubble radius.,scaledwidth=90.0% ] next , consider the contribution to the variance of a typical second - order , four - derivative term , eq . ( [ phi2k4 ] ) . this term contributes to the variance in @xmath255 an amount @xmath256}}{3h } \right|_{\textrm{second - order , } k^4}= \frac{4}{27}\frac{1}{a^4h^4}\sqrt { \textrm{var } \left [ \left\langle\left(\nabla^2\varphi\right)^2\right\rangle - \left\langle\varphi_{,ij}\varphi^{,ij}\right\rangle\right]}.\ ] ] again , with the same dimensionless variables @xmath265}}{3h } \right|_{\textrm{second - order , } k^4 } & = & \frac{1}{3}a^2 \left(\frac{h\,\textrm{mpc}^{-1}}{h_0}\right)^4\left(\frac{a}{a_0}\right)^2 r^{-4 } \left [ \int_0^{x_\textrm{max } } dx_1\int_0^{x_\textrm{max } } dx_2 \ x_1 ^ 3 \ x_2 ^ 3 \right . \nonumber \\ & & \left . t^2(x_1/\gamma r ) \ t^2(x_2/\gamma r ) \ e^{-x_1 ^ 2}e^{-x_2 ^ 2}\int_{-1}^1 d\mu \ e^{-2x_1x_2\mu}\ \left(1-\mu^2\right)^2 \right]^{1/2 } .\end{aligned}\ ] ] the results are also shown in fig . [ variance ] . this term scales as @xmath266 . the result is well behaved in the infrared and an infrared cutoff does not need to be introduced . finally , consider the variance of a typical second - order two - derivative term , eq . ( [ phi2k2 ] ) . as mentioned in the previous section this term is not well behaved in the infrared . we first calculate the variance by introducing an infrared cutoff which we will take to be the wavenumber of the present hubble radius . the second - order @xmath267 term contributes to the variance in @xmath255 an amount @xmath256}}{3h } \right|_{\textrm{second - order , } k^2}= \frac{25}{27}\frac{1}{a^2h^2 } \sqrt{\textrm{var}\left [ \left\langle\varphi\nabla^2\varphi\right\rangle + \frac{13}{18}\left\langle\varphi_{,i}\varphi^{,i}\right\rangle\right ] } .\ ] ] in terms of the dimensionless variables @xmath268 and @xmath9 , this term contributes to the variance an amount @xmath265}}{3h } \right|_{\textrm{second - order , } k^2 } & = & \frac{25}{24 } a^2 \left(\frac{h\,\textrm{mpc}^{-1}}{h_0}\right)^2\frac{a}{a_0 } r^{-2 } \left [ \int_{x_\textrm{min}}^{x_\textrm{max } } dx_1 \int_{x_\textrm{min}}^{x_\textrm{max } } dx_2 \ t^2(x_1/\gamma r ) \ t^2(x_2/\gamma r ) \ e^{-x_1 ^ 2}e^{-x_2 ^ 2}\right . \nonumber \\ & & \times \left . \int_{-1}^1 d\mu \ e^{-2x_1x_2\mu } \ \left(\frac{x_1 ^ 3}{x_2}+\frac{x_2 ^ 3}{x_1 } + 2x_1 x_2 + \frac{26}{9}x_1 ^ 2\mu + \frac{26}{9}x_2 ^ 2\mu + \frac{169}{81}x_1x_2\mu^2\right ) \right]^{1/2 } .\end{aligned}\ ] ] as a result of the infrared behavior , for this term we must introduce another parameter , @xmath269 , as an infrared cutoff . its value is shown in fig . [ variance ] as a function of @xmath167 . as expected , it has a @xmath270 behavior . now we turn to the infrared - singular part of the second - order two - derivative term . the contribution is @xmath256}}{3h } \right|_{\textrm{second - order , } k^2}= \frac{25}{27}\frac{1}{a^2h^2 } \sqrt{\left [ \left\langle\varphi\nabla^2\varphi\right\rangle -3 \left\langle\varphi\right\rangle_1 \left\langle\nabla^2\varphi\right\rangle_1\right]_\infty } .\ ] ] we will evaluate the integral by defining some @xmath271 such that @xmath272 . we will be interested in large values of @xmath9 , so if @xmath273 , then @xmath274 , and @xmath275 can be set to unity . using eq . ( [ blowup ] ) , this term may be written as @xmath256}}{3h } \right|_{\textrm{second - order , } k^2}=\frac{25}{12 } a^2 \left(\frac{h\,\textrm{mpc}^{-1}}{h_0}\right)^2\frac{a}{a_0 } r^{-2}\left [ x_*^4\ln \frac{k_*}{k_\textrm{min}}\right]^{1/2 } .\ ] ] if we take @xmath276 where @xmath277 is the hubble wavenumber , then evaluating the term for @xmath9 corresponding to the hubble radius we find @xmath256}}{3h } \right|_{\textrm{second - order , } k^2 } = \frac{25}{24}a^2\frac{a}{a_0 } \left[\ln\frac{k_h}{k_\textrm{min}}\right]^{1/2 } = \frac{25}{36}a\frac{a}{a_0}\sqrt{\textrm{var}[\langle\varphi\rangle_1]}. \label{logterm}\ ] ] we started with a perturbative expansion , and for that expansion to be valid requires @xmath278 , so our perturbative analysis will break down for @xmath279}/3h \sim 10^{-5}$ ] . of course the total variance of @xmath83 includes cross terms from the various contributions . however we have seen that the linear term will dominate . our results are illustrated in figs [ meaneds ] and [ variance ] . the mean corrections to the expansion rate are a few parts in @xmath0 . if the correct prescription for dealing with the infrared singular nature of @xmath280 $ ] is to employ a cutoff of the order of the hubble radius , then the variance about the mean is small , and dominated by the first - order term . if , however , the super - hubble modes are physical , then the variance is dominated by the infrared singular part of the second - order corrections . restricting ourselves to @xmath280<1 $ ] where the perturbative calculation is valid , then the variance will be small , of order @xmath281 . in ref . @xcite , rsnen considered what amounts to the variance associated with the second - order term proportional to @xmath282 and suggested that due to unknown boundary conditions it may lead to a large contribution to the expansion rate . we claim that when properly averaged , this term is subdominant . in closing , let us speculate that corrections to the expansion rate will include a term like @xmath1 even when @xmath283 . returning to the infrared - singular part of the second - order variance , eq.([logterm ] ) , @xmath256_\infty } } { 3h } \right|_{\textrm{second - order , } k^2 } = 3.76\times10^{-10 } \frac{a}{a_0}\ln\frac{k_h}{k_\textrm{min } } . \label{logterm2}\ ] ] for the variance to be of order unity , the perturbation spectrum would have to extend to a factor of @xmath284 ( @xmath285 @xmath286-folds ! ) times the present hubble radius . rather than a harrison zeldovich spectrum , if we assume a slightly red spectrum so that @xmath287 with @xmath288 , . ] then the logarithmic term in eq . ( [ logterm2 ] ) is replaced by @xmath289 . now this will give unit variance if @xmath290 . for instance , if @xmath291 on super - hubble - radius scales , then a variance of order unity is obtained if the perturbation spectrum extends 676 @xmath286-folds beyond the hubble radius . since the present hubble radius corresponds to a scale that crossed the hubble radius about 60 @xmath286-folds before the end of inflation , if inflation lasted more than 736 @xmath286-folds with a super - hubble - radius spectral index of @xmath291 , then the effect of super - hubble - radius perturbations would have a large effect on the expansion rate of our hubble volume . speculation about the significance of this result will follow in a separate paper @xcite . e.w.k . is supported in part by nasa grant nag5 - 10842 and by the department of energy . we would like to thank enrico barausse , riccardo catena , scott dodelson , lam hui , luigi pilo , syksy rsnen , and dominik schwarz for helpful conversations . the aim of this appendix is to demonstrate that @xmath75 does not depend on the value of @xmath294 since second - order terms will only enter as spatial derivatives . for the demonstration , we first introduce @xmath295 , the extrinsic curvature of constant-@xmath4 hypersurfaces : @xmath296 in either the synchronous or comoving gauge one can see that ( see ref . @xcite ) @xmath297 one can now write the @xmath298 term in a convenient form ( see eq.(4.3 ) of @xcite ) @xmath299 where @xmath300 is the intrinsic curvature of the three space with metric @xmath89 . at second order @xmath300 is @xmath301 using eq . ( [ vartheta ] ) , eq . ( [ eqr ] ) also can be written as @xmath302 . \label{usefulrel}\ ] ] recall now that in any of the definitions adopted , eq . ( [ combo3 ] ) is always valid . so independent of @xmath294 , @xmath303 using eq . ( [ usefulrel ] ) , we obtain @xmath304 . \label{complete}\ ] ] we note now that the only piece that depends on @xmath305 is the one contained in @xmath306 . and , as we can see from its explicit expression @xmath307 this is a spatial gradient , so independent on @xmath308 . from this formula it is also clear that @xmath65 has no term with zero powers of @xmath227 , since the spatial curvature @xmath300 always involves spatial derivatives ( and the same is true for the @xmath309 terms ) . if one computes @xmath75 using eq . ( [ complete ] ) , one obtains the same result as eq . ( [ risultatowithbt ] ) , thus our results are independent of @xmath294 . by the same argument , in a first - order calculation we could add a constant to @xmath310 . in this appendix we repeat the calculation in the poisson gauge in order to check our results and compare with other results in the literature for @xmath65 in the that gauge . the poisson gauge is a second - order generalization of the longitudinal ( or newtonian ) gauge . since this is a very involved way to do the computation , and since it is only intended to be a check , we do it only in a partial way by just computing the corrections @xmath311 [ see eqs . ( [ combo11 ] ) and ( [ fundamentalsyn ] ) ] . furthermore we compute exactly the terms with two spatial derivatives , while as for the four derivatives terms , we only check that they have zero statistical mean ( that is , we disregard total spatial gradients ) . the metric is given at second order in this gauge by and the tensor @xmath312 will never enter in the computations . ] @xmath313 dy^i dy^j \right\ } .\ ] ] here @xmath314 is a pure symmetric , transverse , traceless tensor degree of freedom . we will denote the derivative with respect to conformal time @xmath315 as @xmath316 . we will also introduce here a cosmic time @xmath317 , linked to @xmath315 through the relation @xmath318 , and we will denote the derivative with respect to @xmath317 with a @xmath319 symbol . the spatial derivatives with respect to @xmath5 will be denoted as before , with a simple @xmath320 or @xmath321 symbol . the derivatives with respect to @xmath322 instead will be explicitly written as @xmath323 or with the subscript @xmath324 . recall also , as well known , that the @xmath325 component of einstein s equations ( in the absence of anisotropic stress ) imposes @xmath326 , and moreover , the evolution equations in matter domination give @xmath327 the four velocity of the fluid @xmath153 here has a more involved form where @xmath329 . here @xmath330 is the second - order contribution to the three velocity . we will need only its divergence ( which is obtained taking the divergence of the @xmath331 einstein equation ) : @xmath332 now , taking these expressions and taking christoffel symbols from ref.@xcite , we must compute @xmath333 this gives us @xmath334 here @xmath335 is defined as @xmath336 . the piece coming from @xmath337 in eq.([thetapoiss ] ) contains first order ( @xmath338 ) and second order ( @xmath339 ) terms . the first order term [ see eq . ( [ ui1 ] ) ] is equal to @xmath340 and it will be important in the next steps of the computation , since it will produce second - order quantities . the term @xmath341 [ eq . ( [ ui2 ] ) ] is a spatial gradient of second - order quantities . for the purposes of this appendix we have to keep the first two pieces in eq . ( [ ui2 ] ) , and moreover we have to consider also the two intrinsically second - order terms @xmath342 and @xmath105 . note also that , in their time evolution they contain non - local quantities ( @xmath343 and @xmath344 of eqs . ( 6.8 ) of ref.@xcite ) , _ i.e. , _ they are defined through @xmath345 . \label{theta_0}\end{aligned}\ ] ] so the spatial derivatives of these quantities will produce not only spatial gradients , but also terms who do not have zero statistical mean . in the end , we keep them in the calculation as @xmath346 where `` @xmath347 '' indicates terms with four spatial gradients ( that we will omit ) . so , the relevant expression for @xmath348 is obtained by eqs.([thetaexpl ] ) , ( [ u1 ] ) , and ( [ u2 ] ) @xmath349 this is the result in the poisson gauge in coordinates @xmath350 . note that here there are terms which do not vanish in the infrared limit , proportional to @xmath3 and in @xmath163 ( they will disappear going back to the synchronous and comoving coordinates , as noted in ref . @xcite ) . in order to compare with the results already obtained in the synchronous gauge , we have to express our quantity @xmath348 as @xmath351 and average it over a volume in coordinates @xmath5 at constant @xmath4 as in section [ thetasynchronous ] . first , we have to change time from @xmath317 to @xmath53 in all the quantities . this is relevant only for the zeroth - order and first - order terms , and not for the ones which are already second order . so the quantities in which the time has to be changed are @xmath335 , @xmath352 and @xmath353 . the first two , @xmath354 and @xmath355 , have to be expanded up to second order in @xmath356 , which is defined by @xmath357 and expressed in terms of the scale factor and the hubble rate defined by a comoving observer : @xmath358 , @xmath359 . instead , for the variable @xmath360 , @xmath361 holds since @xmath362 . in order to find the change of the time coordinate from @xmath317 to @xmath53 one knows that @xmath363 , and that @xmath364 . in this way one obtains an iterative equation @xmath365 where @xmath366 . for zeroth , first , and second order , the equation becomes @xmath367 which lead to @xmath368 now using these equations , one can change the time in eq.([thetatx ] ) , obtaining @xmath369 as said before , this transformation ( _ i.e. , _ passing to synchronous time @xmath53 ) has eliminated the two terms in @xmath3 and @xmath163 that do not vanish in the infrared . we wrote it in this form in order to compare with a previous result in literature ( eq.(10 ) of ref . @xcite ) . as we are going to show , we agree with that result , but we added here the contribution of intrinsic second - order terms @xmath342 and @xmath105 . this is crucial if we want to recover the result obtained in sec . [ thetasynchronous ] in the other gauge . here @xmath370 contains the terms @xmath371 note that after changing time , a new term ( which is not a spatial gradient ) appears , namely @xmath372 . finally , one has to change also spatial variables from @xmath322 to @xmath5 . the change of coordinates is found by imposing that the metric @xmath373 in the coordinates @xmath374 has @xmath375 . using the transformation given by eq . ( [ deltatau ] ) , one finds that to first order the change in spatial coordinates is given by @xmath376 since it is already first order , this change of coordinates is irrelevant for quantities that are already second - order . moreover , the only first order term of eq . ( [ thetataux ] ) is @xmath377 . here , the change of spatial coordinates in the argument of @xmath378 is not relevant for our purposes , since it produces only an extra term which is a four derivative spatial gradient of something . but passing from @xmath379 to @xmath380 produces a ( relevant ) extra term with four spatial derivatives , @xmath381 in all the other terms we can safely put @xmath382 . after doing this , the result is @xmath383 using eqs . ( 6.8 ) of ref . @xcite , together with eqs.([theta_0 ] ) given in the present section , one obtains for the time evolution of the intrinsic second order terms @xmath342 and @xmath384 the result @xmath385 now this is the first check we wanted to do . we can compare with eq.([fundamentalsyn ] ) and see that deriving by parts and using @xmath386 , they coincide ( up to four - derivative total spatial gradients ) . an important observation here is that the presence of the intrinsically second - order terms is necessary to obtain not only the correct result , but even a consistent calculation . in fact if one ignores them , _ i.e. , _ ignores @xmath105 and @xmath387 in the synchronous gauge and ignores @xmath342 and @xmath105 in the poisson gauge , and compares the two results , one sees that they do not coincide . in other words they are not really `` intrinsic '' second order , since what appears as intrinsic second order in one gauge , can appear as first order times first order in another gauge . at this point we may also average expression eq . ( [ thetapf ] ) over a volume @xmath35 as we did in sec . [ thetasynchronous ] . this has to be done with the appropriate measure of integration , which is @xmath388 , where @xmath389 . up to first order the spatial metric is given by @xmath390 = a^2\left(1-\frac{4}{3}\varphi\right ) \left[\delta_{ij}(1 - 2\varphi ) -\frac{4}{3 a^2 h^2}{\partial_i\partial_j \varphi } \right ] \ , , \ ] ] where @xmath391 is the metric in the @xmath392 coordinates , and where we used the transformation of eq . ( [ transformation ] ) . using this metric , the measure becomes @xmath393 , \label{measure2}\ ] ] and this is the same as eq . ( [ fosd ] ) , and so the final result is the same . finally , we can check the computation of ref . @xcite , which used only the first order metric in this gauge . that is , we have to put @xmath342 and @xmath105 to zero in eq . ( [ thetatauy ] ) , @xmath394 and average it with the measure of eq . ( [ measure2 ] ) . doing so , we recover the result @xmath395 as in eq . ( 11 ) of ref . @xcite . l. hui and u. seljak , in _ proceedings of the 173rd iau symposium _ , edited by c. s. kochanek and j. n. hewitt ( kluwer academic publishers , dordrecht , 1995 ) ; u. seljak and l. hui , in _ proceedings of clusters , lensing and the future of the universe _ ( asp conference series vol . 88 , v. trimble and a. reisenegger , eds . , 1995 ) , p 267 .
gravitational wave ( gw ) observations are expected to provide a new view of relativistic phenomena in the universe . one of the most promising candidates for the direct detection of gws is the merger of compact binaries such as binary neutron stars ( nss ) . the second generation of ground - based gw detectors , such as advanced ligo @xcite , advanced virgo @xcite and kagra @xcite , will reach the sensitivity required to detect gws from the inspiral and coalescence of compact binary systems including binary nss within a few hundred mpc . statistical studies suggest that a few tens of merger events should be observed per year @xcite . electromagnetic counterparts of gw emitters have been recently focused on to maximize a scientific return from the expected detection of gws ( e.g. , * ? ? ? follow - up observations of these electromagnetic counterparts are important to confirm a gw detection and to investigate progenitors and environments . the electromagnetic detection also improves the localization of gw sources because the localization accuracy by photons is much better than that by the ground - based gw detectors @xmath1 deg@xmath2 ( e.g. , * ? ? ? sophisticated simulations have revealed mass ejection associated with the mergers of binary nss by several mechanisms . significant mass is dynamically ejected by gravitational torques and hydrodynamical interactions during the mergers , called dynamical ejecta ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? general relativistic simulations show that these ejecta distribute nearly isotropic compared to newtonian simulations in the cases of binary nss @xcite , while they are anisotropic for ns - black hole ( bh ) mergers @xcite . mass may be also ejected through winds driven by neutrinos @xcite , magnetic fields of and/or amplified by the merged objects @xcite , viscous heating and nuclear recombination @xcite . a traditional electromagnetic counterpart is short - hard gamma - ray bursts ( grbs ; * ? ? ? recent simulations have revealed that a hypermassive ns is formed from the merger of a ns binary ( e.g. , * ? ? ? * ) , which is believed to collapse into a bh at later time . non - collapsed matter and some ejecta falling back to the bh form a torus around the bh ( e.g. , * ? ? ? then , a relativistic jet may be launched from the bh - torus system , which is believed to be the central engine of short - hard grbs . another interesting possibility is a so - called macronova / kilonova , which is thermal emission from ejecta ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the radiative energy of a macronova is estimated between that of a classical nova and supernova . ejecta can also produce non - thermal emission at later time similarly to supernova remnants @xcite . ejecta may accompany an advanced relativistic part , producing early emission ( @xmath3 hours ; * ? ? ? * ; * ? ? ? emission from macronovae and ns binary merger remnants is almost isotropic and hence different from that of short grbs which depends on the directions of their relativistic jets . moreover , macronovae are closer in time to mergers than emission from merger remnants and do not depend on the properties of circumburst environments . therefore , macronovae are expected to play a crucial role to localize a large sample of gw events @xcite . recently , a macronova candidate following grb 130603b was discovered @xcite . this candidate is widely interpreted as the results of the radioactive decay of @xmath0-process elements produced in the ejecta of a compact binary merger @xcite . we call this scenario _ an @xmath0-process model _ throughout this paper . the ejecta from a merger of binary nss is primarily neutron - rich . then , heavy radioactive elements ( mass number @xmath4 ) are expected to form through neutron - capture onto nuclei ( @xmath0-process nucleosynthesis ) ( e.g. , * ? ? ? although the @xmath0-process nucleosynthesis ends a few hundred millisecond after a merger , synthesized elements release energy due to nuclear fission and beta decays up to @xmath5 days ( e.g. , * ? ? ? * ) . a schematic picture for this model is shown in the left panel of figure [ figure : model ] . if this scenario is correct , the observations also give important insights into the enrichment of @xmath0-process elements in the galaxy evolution ( e.g. , * ? ? ? although the @xmath0-process model explains the observed light curve of the macronova , it is based on the limited observational data and the nuclear heating rate with large uncertainties . required mass of dynamical ejecta to explain the observations is relatively large compared with the simulation results @xcite . in addition , the occurrence of @xmath0-process nucleosynthesis needs the ejecta with low electron fraction ( @xmath6 ) . however , relatively high electron fraction ( @xmath7 ) can be also realized , which has been discussed for neutrino - driven wind ( e.g. , * ? ? ? it is worth considering other possibilities such as the scenarios of an external shock between ejecta and surrounding medium @xcite , a supramassive magnetar @xcite and dust grains @xcite . in this study , we consider another power source of macronovae , i.e. , energy injection from the activity of the central engine , in addition to the radioactive decay of @xmath0-process elements . this is similar to the early evolution of core - collapse supernovae ( e.g. , * ? ? ? * ; * ? ? ? we call this model _ an engine model _ throughout this paper . there are several motivations to consider that the activity of the central engine contributes to the heating of ejecta . one observational motivation is the extended emission following the prompt emission of short grbs . the origin of extended emission is considered to be the activity of the central engine @xcite because the sharp drop of its light curve is difficult to be reproduced by afterglow emission @xcite . after the merger , a stable ns or a bh is formed . in the case that a bh with a torus ( or disk ) is formed , the energy injection to the ejecta is expected as a form of the jet and/or disk wind ( e.g. , * ? ? ? * ) . in the case that a ns with strong poloidal magnetic field is formed as a result of a merger , the wind of relativistic particles is ejected @xcite . then , the wind collides with the ejecta , and about half of the wind energy converts to the internal energy by the shock - heating . a schematic picture is shown in the right - hand side of figure [ figure : model ] . the ejecta emission powered by a stable magnetar has already been discussed @xcite . they suggest that the magnetar - powered ejecta emit the brighter optical and x - ray emissions than that of the @xmath0-process model . however , they did not show that the magnetar - powered ejecta explain the detected infrared excess in grb 130603b @xcite . the engine model can provide energy enough to reproduce the detected macronova candidate , grb 130603b . we do not specify the specific heating sources . alternatively , to estimate the luminosity and temperature , we assume that the internal energy @xmath8 erg is injected to the ejecta at the time @xmath9 s after the merger . these values are consistent with typical isotropic energy @xmath10 erg and duration @xmath11 s of the extended emission @xcite . using the velocity of the ejecta @xmath12 , the temperature at @xmath13 is @xmath14^{1/4}$ ] , where @xmath15 is the radiative constant . if we only consider the adiabatic cooling for the cooling process of the ejecta , the evolution of the internal energy @xmath16 and temperature @xmath17 is scaled as @xmath18 and @xmath19 . the luminosity is described as @xmath20 . adopting the ejecta velocity @xmath21 cm s@xmath22 @xcite , the luminosity @xmath23 and the temperature @xmath17 at @xmath24s are @xmath25 and @xmath26 the observations of macronova of grb 130603b give j - band luminosity @xmath27 erg s@xmath22 and the difference between j - band and b - band @xmath28 mag which corresponds to the temperature @xmath29 k at @xmath30 days after grb 130603b in the source rest frame @xcite . therefore , in this estimate , the luminosity and temperature for the engine model is consistent with the observation of the macronova following grb130603b . we model the evolution of luminosity and temperature of a macronova . unlike the previous studies , we treat the model in an analytical manner and formulate a light curve including the early phase ( @xmath31 s ) , which is important for the search of electromagnetic counterparts of gw emitters . we consider shock - heating due to the activity of a central engine as a heating mechanism of the ejecta . for comparison , the @xmath0-process model , which has been discussed in most papers ( e.g. , * ? ? ? * ) , is also formulated . then , we compare the results of our models with observations to constrain the model parameters such as the ejected mass and the velocity of the ejecta . although our models are simplified , it is valuable to make comparison between two heating models . in section [ model ] , we introduce our model assumptions . we describe the analytical models for the evolution of luminosity and temperature in section [ evolution ] . then , we compare our results with observations in section [ discuss ] . implications for the discrimination between two models are also discussed . we summarize our results in section [ summary ] . in appendix a , we summarize the formulae for the observed temperature and bolometric luminosity . significant mass of material @xmath32 is ejected during a binary merger . we model ejecta by following the results of the general relativistic simulations of ns - ns mergers in @xcite . the simulations show that ejecta expand in a nearly homologous manner ( see also * ? ? ? the morphology of the ejecta is quasi - spherical in the case of a merger of binary nss . according to these results , we assume an isotropic and homologous expansion for the ejecta . then , the velocity of ejecta @xmath12 is @xmath33 where the radius @xmath0 originates the central engine and the time @xmath34 is measured from the time when a compact binary merges . note that in the case of a merger of ns - bh binary , the ejected mass expands with significant anisotropy @xcite . we do not consider such anisotropic ejecta in this work . @xcite found that the profile of ejecta obtained from simulations by @xcite can be well fitted by a power - law function @xmath35 . the power - law index of snapshot density @xmath36 is more or less independent on the dynamics of mergers , which is in the range of @xmath374 for @xmath38 , where @xmath39 and @xmath40 are the velocities of the outer and inner edges of the ejecta , respectively . we choose the middle of this range @xmath41 in this study . we also fix the maximum velocity @xmath42 from simulation results @xcite . the maximum velocity @xmath39 is comparable with the escape velocity of the system . the minimum velocity @xmath43 is mainly determined by complicated dynamics at the initial stage of the merger @xmath44 . for the mass density profile at the front of the ejecta , we assume the discrete boundary and the mass density @xmath45 at the region @xmath46 . although this profile may be far from the actual one , ) . such a profile is difficult to calculate precisely with current numerical calculations . ] our main aim is to compare two models for energy sources , so that our conclusions are not affected . in section [ outer ] , we discuss the dependence on the mass density profile at the outer region of the ejecta for the observed light curve . here , we only consider the evolution after the initial stage of the merger ( @xmath47 ) and treat @xmath43 as a model parameter . because of homologous expansion , the density decreases as @xmath48 . then , the density profile is described by @xmath49 where @xmath50 and @xmath51 are normalization factors . the factor @xmath52 is related to the total mass of the ejecta @xmath53 as following , @xmath54,\end{aligned}\ ] ] where we use @xmath55 from equation ( [ sec2:v ] ) . we also introduce the radius of ejecta outer edge @xmath56 and their inner edge @xmath57 the inner part of the ejecta is optically thick , and therefore the propagation of radiation in the ejecta can be regarded as a diffusion process . photons can diffusively escape from the region which satisfies that the diffusion time , @xmath58 , is smaller than the dynamical time @xmath34 , @xmath59 the medium in this region is called to be effectively thin @xcite . for convenience , we introduce a diffusion radius @xmath60 which is the radius satisfying the condition @xmath61 . furthermore , we divide ejecta into two regions called the effectively thin ( @xmath62 ) and effectively thick ( @xmath63 ) regions . near the diffusion radius , the optical depth is @xmath64 . we consider random walk for photons so that the mean number of scatterings to propagate for the distance @xmath65 is @xmath66 , where @xmath67 is the mean free path for a photon . hence , the diffusion time @xmath58 for the propagation distance @xmath65 is @xmath68 in the right hand of equation ( [ sec2 - 1:t_diff ] ) , we use @xmath69 . we calculate the diffusion radius @xmath70 from the condition @xmath71 . since the mass density profile of the ejecta is described by a decreasing power - law function ( equation [ sec2 - 1:rho ] ) , the diffusion time @xmath58 is negligible in an outer part . thus , in order to calculate the diffusion radius @xmath70 , it is a good approximation to only consider scatterings near @xmath70 ( @xmath72 ) . however , in the early phase , the distance from the outer edge of the ejecta @xmath73 to the diffusion radius @xmath70 is smaller than the diffusion radius @xmath74 . therefore , we should take the propagation distance as @xmath75 we call the first the thin - diffusion phase and the second the thick - diffusion phase throughout this paper . we schematically show these two phases in figure [ figure : diffuse ] . note that in the thin - diffusion phase , since the size of the effectively thin region is much smaller than the size of the ejecta ( @xmath76 ) , the calculation of the radiative transfer using monte carlo technique ( e.g. , * ? ? ? * ; * ? ? ? * ) requires a large number of realizations to follow the temporal evolution , which do not seem to have been considered properly so far . to obtain the diffusion radius , we need to calculate the optical depth @xmath77 of photons which propagate a distance @xmath65 . using equations ( [ sec2 - 1:rho ] ) and ( [ sec2 - 1:deltar ] ) , the optical depth @xmath77 is described as , @xmath78^{-1 } \nonumber \\ & & \times { \displaystyle \left[\left(\frac{r_{\rm diff}}{v_{\min}t}\right)^{1-\beta}-\left(\frac{v_{\max}}{v_{\min}}\right)^{1-\beta}\right]},\end{aligned}\ ] ] in the thin - diffusion phase , and @xmath79^{-1 } \nonumber \\ & & \times { \displaystyle \left(\frac{r_{\rm diff}}{v_{\min}t}\right)^{1-\beta}(1 - 2^{1-\beta})},\end{aligned}\ ] ] in the thick - diffusion phase , where @xmath80 is the opacity of the ejecta . for simplicity , we use a grey approximation and a spatially uniform value of the opacity @xmath80 . from the results of @xcite ( see also * ? ? ? * ) which consider the contribution from all @xmath0-process elements to the opacity of merger ejecta , the evolution of the bolometric luminosity can be approximately described by the constant value of the opacity , @xmath81 cm@xmath2g@xmath22 . following their results , we use this value for the opacity of the ejecta . note that the exact value of the opacity of the ejecta has some uncertainties in the production efficiency of @xmath0-process elements and its spatial distribution . moreover , if the ejecta temperature is low enough for dust formation ( @xmath83 k ) , the opacity significantly increases @xcite . from these reasons , we consider the dependence on @xmath80 in section [ evolution ] . our model is based on the formulation of the light curves of supernovae ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but there are several differences . in the case of type ii supernovae , the opacity is significantly reduced due to hydrogen recombination ( e.g. , * ? ? ? however , since the ionization potentials of the lanthanides included in the @xmath0-process elements are generally lower than that of hydrogen and the iron group , the opacity remains high at relatively low temperature @xcite . therefore , we do not consider the recombination effects for the opacity . as far as we know , the supernova studies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) have not taken into account the thin - diffusion phase , which is necessary for treating the thickness of the diffusion length appropriately and estimating the physical quantities by the values at the outer edge of the ejecta in the analytical formulae . this phase may be also important for the case of supernovae . some supernova studies consider the planar phase @xcite in which the evolution of the ejecta is approximately planar as long as its radius do not double . in the case of the ns - ns merger , since the initial length scale of the merger system is small @xmath84 cm and the velocity of the merger ejecta is subrelativistic , the planar phase is irrelevant for the observations . one of the two heating mechanisms we consider is nuclear heating by @xmath0-process elements . since the beta decay products of @xmath0-process elements produced in ns binary mergers naturally heat ejecta , this mechanism is considered to power the emission of a macronova ( e.g. , * ? ? ? the nuclear heating rate is calculated in several works @xcite . the derived heating rates per unit mass @xmath85 are described by the following formula @xmath86 in this study , we use @xmath87 and @xmath88 obtained by @xcite . the value of @xmath89 has been obtained by simulations under some simplified assumptions with only limited parameter regions . thus , we should note that the value of @xmath90 has uncertainties . the injected internal energy by the nuclear decay is @xmath91 in the region @xmath63 . on the other hand , the injected energy in this region is decreased by adiabatic cooling . the time - evolution of internal energy due to the adiabatic cooling is proportional to @xmath92 . comparing the two temporal evolution , the index of the adiabatic cooling is smaller than that of the increase of internal energy due to the nuclear decay for @xmath93 . since we use @xmath87 , we neglect the injected internal energy in the region @xmath63 . unlike the @xmath0-process model , energy injection occurs only within the time @xmath13 in the engine model . we only consider adiabatic cooling as a cooling process of ejecta after @xmath13 , and therefore , the temperature distribution at time @xmath34 is , @xmath94 where the index @xmath95 is a parameter for a snapshot distribution and @xmath96 is a normalization factor described later . the time dependence of @xmath92 is the effect of adiabatic expansion . the normalized value @xmath96 is determined by using the relation of total injected internal energy @xmath97 as @xmath98,\end{aligned}\ ] ] where we use @xmath99 . for the temperature index @xmath100 , the innermost region of ejecta has dominant internal energy . as will be shown in section [ evolution ] , since the luminosity and temperature always depend on the product of @xmath97 and @xmath13 , we treat @xmath101 as a parameter . thus , the engine model has two parameters , @xmath95 and @xmath101 instead of @xmath90 and @xmath102 in the @xmath0-process model . energy injection is not always a single event and the shock does not always get through the whole ejecta . it is considered that the activity of the central engine accompanies violent time variability . in this case , multiple shocks propagate into the ejecta . some of the shock may not catch up with the outer edge of the ejecta . current general relativistic simulations can not calculate the evolution of ejecta for such a long time after merger ( @xmath103 s ) , so that the index @xmath95 of temperature distribution is highly uncertain . therefore , we treat the temperature index @xmath95 as a parameter . unlike the case of core - collapse supernova @xcite , it is difficult to determine the temperature distribution of heated ejecta by the activity of a central engine . in the case that the activity of a central engine injects the energy into the ejecta , the radiation - dominated shock ( where the internal energy behind the shock is dominated by radiation ) is formed in the ejecta . the ejecta are heated during the propagation of the shock . this situation is similar to the initial phase of core - collapse supernovae ( e.g. , * ? ? ? * ; * ? ? ? in the cases of core - collapse supernovae , the kinetic energy of ejecta before the shock heating is much smaller than the injected internal energy . in such ejecta , the relation between velocity and mass density was obtained by @xcite ( in the non - relativistic case for the velocity of the ejecta ) . using sakurai s ( 1960 ) solution and the equipartition between the kinetic energy after the shock heating and the internal energy @xcite , the distribution of the temperature distribution is derived . however , in the case of compact binary mergers , the merger ejecta have a large velocity ( @xmath104 ) before the shock heating @xcite . then , injected internal energy is not always larger than the kinetic energy of ejecta so that it is not clear whether we can use the equipartition to estimate the distribution of internal energy or not . the kinetic energy of the ejecta @xmath105 is described as @xmath106}{(5-\beta)\left[1-\left(\frac{v_{\max}}{v_{\min}}\right)^{3-\beta}\right]}}.\end{aligned}\ ] ] note that if the injected internal energy @xmath97 is larger than the kinetic energy of the ejecta , it is expected that some of the internal energy converts to the kinetic energy of the ejecta . as a result , the internal energy and the kinetic energy are equal as in the case of core - collapse supernovae . then , the mass density distribution and the maximum velocity of the ejecta derived from simulations may be changed because the injection time may be long @xmath107s compared to that calculated by simulations @xmath108s @xcite . for simplicity , we only consider the case @xmath109 . clccc + symbol & & fiducial model & minimum mass model & hot interior model + @xmath53 & ejecta mass & @xmath110 & @xmath111 & @xmath112 + @xmath43 & minimum velocity & @xmath113 & @xmath114 & @xmath115 + @xmath39 & maximum velocity & @xmath116 & @xmath116 & @xmath116 + @xmath36 & index of the density profile & 3.5 & 3.5 & 3.5 + @xmath80 & opacity & 10 cm@xmath2 g@xmath22 & 30 cm@xmath2 g@xmath22 & 10 cm@xmath2 g@xmath22 + @xmath90 & nuclear heating rate at 1 day & @xmath117 erg s@xmath22 g@xmath22 & @xmath118 & @xmath117 erg s@xmath22 g@xmath22 + @xmath102 & index of nuclear heating rate & 1.3 & @xmath118 & 1.3 + @xmath97 & internal energy at @xmath13 & @xmath119 erg & @xmath120 erg & @xmath121 erg + @xmath13 & injection time & @xmath122 s & @xmath122 s & @xmath122 s + @xmath95 & index of the temperature profile & 1.6 & 1.1 & 2.7 + [ tab : parameter ] in this section , we present the evolution of the observed temperature and luminosity of a macronova using our model introduced in the previous section . in sections [ thin ] [ transparent ] , we focus on the parameter dependence of the evolution using some approximations . in section [ fiducial ] , we calculate the temperature and luminosity using the fiducial model with parameters summarized in the first column of table 1 . to calculate the luminosity and temperature , we assume that the emission is well described by the blackbody radiation ( e.g. , * ? ? ? * ) . for simplicity , we assume that the observed temperature equals to the temperature at the diffusion radius @xmath70 . we also assume that the temperature is not so different from the diffusion radius @xmath70 to @xmath123 so that in the thick - diffusion phase ( @xmath124 ) , we only consider the emission from @xmath70 to @xmath123 to calculate the observed luminosity for both the @xmath0-process and engine models . in some studies ( e.g. , * ? ? ? * ) , the observed temperature is approximated by the temperature at the radius of the photosphere @xmath125 where the optical depth is unity . since the velocity of the ejecta is near the light speed , the optical depths at the diffusion radius @xmath70 and its twice @xmath123 are @xmath126 . therefore , our assumed temperature approximately equals to the temperature at the photosphere . in section [ diffusion ] , we introduced two phases , the thin- and thick - diffusion phases ( figure [ figure : diffuse ] ) , depending on the size of the region where photons make the diffusion in the ejecta @xmath65 . we also introduce another phase @xmath127 , the transparent phase , in which photons can diffuse out from the entire of the ejecta . thus , we divide the evolution into these three phases for the values of the diffusion radius @xmath70 as described below . the size of the effectively thin region gets larger with time . at the early phase of a macronova , the diffusion radius @xmath70 , which is the inner radius of the effectively thin region , is near the outer edge of the ejecta @xmath73 . in this early phase , we take the propagation distance @xmath65 of a photon as @xmath128 . since we assume that the density is a homologous function of the velocity @xmath35 , the density can be approximated as @xmath129 in the region @xmath130 . using the escaping condition for the diffusing photons @xmath131 , equation ( [ sec2 - 1:t_diff ] ) and approximation on the optical depth @xmath132 , the propagation distance @xmath65 can be estimated as @xmath133 in the discussion of parameter dependence ( sections [ thin ] [ transparent ] ) , we only consider the dominant term . for example , we neglect the second term in the right - hand side of equation ( [ sec2 - 1:m_ej ] ) to derive the parameter equation ( [ sec2 - 2 - 1:deltar ] ) because the index of the mass density is @xmath134 in our model . in section [ fiducial ] , we include the subdominant terms to calculate the light curves numerically . first we consider the @xmath0-process model . the evolution of temperature @xmath135 is obtained by the internal energy density @xmath136 at the radius @xmath137 . using equations ( [ sec2 - 1:rho ] ) and ( [ sec2 - 1:m_ej ] ) , the parameter dependence of the density is @xmath138 . the observed temperature is @xmath139 for @xmath87 , the observed temperature evolves as @xmath140 . this is because in the thin - diffusion phase the ejecta is effectively a single expanding shell with @xmath141 and the injected energy @xmath142 is almost constant so that the observed temperature approximately follow adiabatic cooling @xmath19 . note that in this phase the observed temperature does not depend on the opacity . the bolometric luminosity @xmath143 for the radioactivity is described as the product of the mass within the thickness @xmath65 in equation ( [ sec2 - 2 - 1:deltar ] ) and the nuclear heating rate @xmath89 in equation ( [ sec2 - 1:dotepsilon ] ) so that @xmath144 for @xmath87 , the evolution of the bolometric luminosity is @xmath145 . next we consider the engine model . we should take into account the freedom of the temperature index @xmath95 in the temperature distribution ( equation [ sec2 - 1:t ] ) . since we only consider a dominant term in the right - hand side of equation ( [ sec2 - 1:e_int ] ) ( the first term for @xmath146 or the second term for @xmath147 ) in this subsection , the parameter dependence of the temperature @xmath96 is described as @xmath148 substituting @xmath149 into equation ( [ sec2 - 1:t ] ) , the observed temperature is described as @xmath150 since the observed temperature @xmath135 approximately equals to the temperature at the outer edge of the ejecta , @xmath151 , the evolution of the observed temperature and the luminosity are also determined by the adiabatic cooling . the bolometric luminosity in the effectively thin region is equal to the total radiation created by thermal emission in this region @xcite . using equation ( [ sec2 - 2 - 1:t_obs , sh ] ) , the bolometric luminosity is described as @xmath152 the time evolution of bolometric luminosity for the engine model does not depend on the temperature index @xmath95 in the thin - diffusion phase . comparing the engine model with the @xmath0-process model in the thin - diffusion phase , the bolometric luminosity and the observed temperature decrease faster in the engine model than those in the @xmath0-process model . these time - dependence do not depend on the indices of the density and temperature . note that the light curve may depend on the detailed profile of the front of the ejecta in this thin - diffusion phase . the profile of the ejecta front is difficult to calculate by the numerical simulation due to its low density , and hence has large uncertain @xcite . we discuss its dependence in section [ discuss ] . we consider diffusion to evaluate the diffusion radius in the thick - diffusion phase . we take the propagation distance @xmath72 after the time when the difference between the radius of the outer edge of the ejecta @xmath73 and the diffusion radius @xmath70 is larger than the diffusion radius , @xmath153 , since the optical depth of the outer part is negligible for the density profile in equation ( [ sec2 - 1:rho ] ) . in this thick - diffusion phase , mass density significantly deviates from @xmath154 . substituting equations ( [ sec2 - 1:t_diff ] ) and ( [ sec2 - 1:tau2 ] ) into @xmath61 , the diffusion radius @xmath70 is calculated as @xmath155^{\frac{1}{\beta-2 } } \nonumber \\ & \propto&\kappa^{\frac{1}{\beta-2}}m_{\rm ej}^{\frac{1}{\beta-2}}v_{\min}^{\frac{\beta-3}{\beta-2}}t^{\frac{\beta-4}{\beta-2}},\end{aligned}\ ] ] where we use the relations @xmath72 and @xmath156 . the latter is obtained from the assumption of the homologous expansion . regarding the optical depth @xmath77 , the second term is neglected in the right - hand side of equation ( [ sec2 - 1:tau ] ) to focus only on the dominant term to study parameter dependence in sections [ thin ] [ transparent ] . for @xmath41 , the diffusion radius decreases with time ( @xmath157 ) . then , emission from the region with relatively high mass density can be observed progressively in this phase ( @xmath158 ) . we introduce the transition time @xmath159 between thin- and thick - diffusion phases , which satisfies the relation @xmath160 . substituting equation ( [ sec2 - 2 - 2:r_diff ] ) and @xmath161 into the relation @xmath160 , we can obtain the transition time @xmath159 as @xmath162 where @xmath163 cm@xmath2 g@xmath22 , @xmath164 , @xmath165 and @xmath166 . as seen above , the transition time @xmath159 is typically several days . this timescale is expected to allow the follow - up observations @xcite . thus , we should consider both phases to predict something useful for follow - up observations . if we fix @xmath80 and @xmath39 and use @xmath41 , equation ( [ sec2 - 2 - 2:t_times ] ) gives @xmath167 . if we increase the total mass of the ejecta @xmath53 and the velocity at the inner edge of the ejecta @xmath43 , the mass density of the ejecta @xmath168 and hence the optical depth are increased . as a result , the transition time @xmath159 becomes large . first , we consider the @xmath0-process model . here , we introduce the velocity @xmath169 based on the homologous relation . using the velocity @xmath170 and equation ( [ sec2 - 2 - 2:r_diff ] ) for the mass density ( equation [ sec2 - 1:rho ] ) , the evolution of temperature is @xmath171 for @xmath87 and @xmath41 , the evolution is described by @xmath172 so that the observed temperature increases with time . the bolometric luminosity is described as the product of the mass between @xmath70 and @xmath123 and the nuclear heating rate @xmath89 . using equations ( [ sec2 - 1:dotepsilon ] ) and ( [ sec2 - 2 - 2:r_diff ] ) , we obtain the bolometric luminosity as , @xmath173 for @xmath87 and @xmath41 , the evolution of the bolometric luminosity is @xmath174 . next , we consider the engine model . using equations ( [ sec2 - 1:t ] ) and ( [ sec2 - 2 - 2:r_diff ] ) , the evolution of the observed temperature is described as @xmath175 for @xmath41 , the value @xmath176 is the boundary whether the observed temperature increases with time ( @xmath100 ) or not ( @xmath146 ) . the evolution of the luminosity equals to the total radiation created by thermal emission in the sphere with radius @xmath70 . using the relation @xmath177 and equations ( [ sec2 - 2 - 2:r_diff ] ) and ( [ sec2 - 2 - 2:t_obs , sh ] ) , we obtain @xmath178 if we take @xmath41 and @xmath179 , the evolution of the bolometric luminosity is @xmath180 . this is almost the same dependence as in the @xmath0-process model . note that even if the inner part of the ejecta has the larger internal energy ( @xmath100 ) , the bolometric luminosity does not always increase with time . using the relation @xmath181 , from the adiabatic cooling , the evolution of bolometric luminosity for a given mass shell with @xmath12 is @xmath182 , where @xmath183 is the total internal energy for the mass shell with a given expanding velocity @xmath12 . since @xmath184 and @xmath185 , the bolometric luminosity increases with time for the value of the temperature index @xmath186 . once the diffusion radius reaches the inner edge of the ejecta @xmath187 , all photons emitted from the ejecta can diffuse out within dynamical timescale . if energy is not injected into the ejecta in this transparent phase , the internal energy in the ejecta runs out immediately . the transition time from the thick - diffusion phase to the transparent phase @xmath188 is described as @xmath189 where we use the diffusion radius @xmath190 . first we consider the @xmath0-process model . the observed temperature equals to the temperature at the inner edge of the ejecta , @xmath191^{1/4}$ ] . using equation ( [ sec2 - 1:rho ] ) , we obtain @xmath192 since the energy is continuously injected due to the nuclear heating in the @xmath0-process model , the bolometric luminosity from the entire ejecta is described as @xmath193 . however , the outer part of the ejecta emits photons with lower temperature and/or x - rays and @xmath194-rays produced directly in radioactive decays . although such emission contributes to the bolometric luminosity , we here focus only on the optical and infrared emissions . in the thick - diffusion phase , the observed emission comes from the region between @xmath195 and @xmath196 . in the transparent phase , we assume that the time evolution of the diffusion radius @xmath70 is the same as the thick - diffusion phase until @xmath197 and the observed luminosity comes from the region from @xmath198 to @xmath123 for simplicity . then , the bolometric luminosity is described as @xmath199 although it appears that this time evolution directly reflects the nuclear decay rate , when we calculate the mass between @xmath198 and @xmath123 the evolution of the upper limit of the integration @xmath123 makes the decrease of the luminosity faster than @xmath200 ( see a dashed line in the middle panel of figure [ figure : evolution ] ) . in addition , the evolution of @xmath70 depends on the index @xmath36 ( see equation [ sec2 - 2 - 2:r_diff ] ) , so that the mass between @xmath198 and @xmath123 also depends on the index @xmath36 . next we consider the engine model . we assume that the internal energy is exhausted when the diffusion radius reaches @xmath197 . for the observed temperature @xmath135 , we assume the relation @xmath201 and use equation ( [ sec2 - 1:t ] ) , @xmath202 the bolometric luminosity is described as @xmath203 since the internal energy at the innermost region almost equals to the total internal energy @xmath204 and determines the bolometric luminosity @xmath205 for the temperature index @xmath100 , the bolometric luminosity does not depend on the mass @xmath53 and velocities @xmath39 and @xmath43 . this luminosity always corresponds to the maximum luminosity for @xmath147 , so that we can impose the lower limit on the parameter @xmath101 . -process model and the engine model , respectively . for comparison , we also plot the bolometric luminosity from the whole ejecta for the @xmath0-process model after the transparent phase ( @xmath206 in equation [ sec2 - 2 - 3:t_tr ] ) as a blue long - dashed line in the middle panel.,width=264 ] we show the temporal evolution of the diffusion radius @xmath70 , the bolometric luminosity @xmath143 and the observed temperature @xmath135 in figure [ figure : evolution ] under the fiducial parameter set . the parameters are summarized in the first column of table 1 . here , we do not use approximations @xmath207 and @xmath208 at the thin - diffusion phase as in section [ thin ] . instead , the diffusion radius @xmath70 is calculated from equations ( [ sec2 - 1:t_diff ] ) ( [ sec2 - 1:tau ] ) without approximations . using the obtained diffusion radius @xmath70 and the relation @xmath169 , we calculate the observed temperatures in the thin- and thick - diffusion phases , @xmath209^{1/4}$ ] ( equation [ sec2 - 2 - 2:t_obs , nuc ] ) , and @xmath210 ( equation [ sec2 - 2 - 2:t_obs , sh ] ) for the @xmath0-process and the engine models , respectively . in the transparent phase , the temperature in equations ( [ sec2 - 2 - 3:t_obs , nuc ] ) and ( [ sec2 - 2 - 3:t_obs , sh ] ) are evaluated with @xmath211 . equations on observed temperature and bolometric luminosity for both models are summarized in appendix . the set of parameters we choose here explains the observed optical and infrared light curves of grb 130603b ( see next section ) . the vertical dash - dotted lines in figure [ figure : evolution ] show the time @xmath212 ( equation [ sec2 - 2 - 2:t_times ] ) . the diffusion radius is plotted only up to the transition time @xmath213 ( equation [ sec2 - 2 - 3:t_tr ] ) . in the thick - diffusion phase , the diffusion radius ( @xmath214 for @xmath41 ) moves inward in the ejecta ( @xmath215 ) . since the observed luminosity and temperature are determined at the diffusion radius @xmath70 , the time evolution of luminosity and temperature strongly depends on the indices of the profile , @xmath36 and @xmath95 . for the @xmath0-process model , the bolometric luminosity decreases with time ( @xmath216 ) in the thick - diffusion phase ( see equation [ sec2 - 2 - 2:l_bol , nuc ] ) , which is more rapid than that in the thin - diffusion phase ( @xmath217 , see equation [ sec2 - 2 - 1:l_bol , nuc ] ) . since the index of the mass density @xmath41 is close to 3 , in which the mass of each shell with a certain size @xmath218 is the same value in logarithmic scale , the mass between the diffusion radius @xmath70 and its doubled value @xmath123 does not significantly change with time . the luminosity is mainly determined by that mass , so that the evolution of the luminosity is slow compared with the evolution of nuclear heating rate ( @xmath200 ) in the thick - diffusion phase . on the other hand , bolometric luminosity and observed temperature increase with time in the engine model with the parameter set of the fiducial model . these mainly reflect the profile of the temperature distribution ( @xmath219 ) . in fact , using equation ( [ sec2 - 2 - 2:l_bol , sh ] ) , the index of the time @xmath34 for the bolometric luminosity is @xmath220 for the engine model . after the transition time @xmath221 , the luminosity and temperature are almost determined by the quantities at the inner edge of the ejecta . then , the evolution of the luminosity and temperature does not significantly depend on the indices of profile @xmath36 and @xmath95 as in the case of the thin - diffusion phase ( except for the case @xmath146 of the engine model , equation [ sec2 - 2 - 3:l_bol , sh ] ) . since our used profile of mass density has an artificially steep cut - off at the inner edge of the ejecta ( figure [ figure : diffuse ] ) , bolometric luminosity in both models rapidly declines after the time @xmath221 . in the bottom panel of figure [ figure : evolution ] , the observed temperature in both models has a steep cutoff at @xmath197 . for comparison , we also consider the time evolution of bolometric luminosity from the whole ejecta @xmath222 in the @xmath0-process model . time evolution is shown in the middle panel of figure [ figure : evolution ] as a blue long - dashed line . this luminosity evolution ( @xmath223 ) is significantly slower than that of the engine model in the transparent phase . in section [ implication ] , we discuss the implication for discriminating the @xmath0-process model and the engine model using these temporal behaviors . -process model ) . since two regions are overlapped , the color looks like purple for the @xmath0-process model . these regions are only a schematic view . we fix @xmath42 , @xmath41 , @xmath224 erg s@xmath22 g@xmath22 and @xmath87 . for the opacity @xmath80 , we use the range @xmath225 cm@xmath2 g@xmath22 . for the engine model , we treat @xmath101 and @xmath95 as free parameters to fit the light curve . the circle , square and triangle denote the case of the fiducial model , minimum mass model and hot interior model , respectively ( see table 1).,width=377 ] we compare the results with the optical and infrared observations of short grb 130603b in figure [ figure : observation ] . the fiducial parameter set in table 1 is adopted . the @xmath0-process model and the engine model result in similar light curves at the optical and infrared bands . both of them satisfy the observational data of grb 130603b . note that the detection point at f606w band at @xmath226 s is consistent with the afterglow of grb 130603b modeled as a smoothly broken power law ( blue dashed line , * ? ? ? we regard this detected value as an upper limit for the luminosity of emission from the ejecta . the detection point at f160w band at @xmath84 s exceeds the extrapolation of the afterglow emission ( red dashed line , * ? ? ? * ) , so that we regard this detected emission as a thermal radiation from the ejecta . the range of the model parameters @xmath43 and @xmath53 to satisfy the constraints obtained from the observation of grb 130603b is shown in figure [ figure : parameter ] as colored areas ( red area for the @xmath0-process model and blue area for the engine model ) . note that the red area has a completely overlap with the blue area . we fix the other model parameters @xmath42 , @xmath41 , @xmath224 erg s@xmath22 g@xmath22 and @xmath87 as in the fiducial model . we take into account the uncertain range of the opacity , @xmath225 cm@xmath2 g@xmath22 to constrain the parameters , @xmath43 and @xmath53 . in the engine model , @xmath95 and @xmath101 are additionally treated as free parameters to derive the allowed area in figure [ figure : parameter ] . in the @xmath0-process model , the luminosity becomes smaller for smaller ejecta mass @xmath53 . the small ejected mass @xmath227 can not reproduce the infrared excess of grb 130603b ( figure [ figure : observation ] ) . the required ejecta mass is relatively large compared to the mass indicated by recent numerical simulations for a merger of binary nss ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? note that in @xcite , @xmath228 is required to explain the observed infrared excess , which is a factor @xmath32 smaller than our results . their theoretical light curves are based on the study of @xcite . in @xcite , a broken power - law mass density profile with the index @xmath229 for the inner layer and @xmath230 for the outer layer of ejecta is adopted , in which the mass of the ejecta is efficiently concentrated at the transition point of the density index . therefore , the luminosity of ejecta is evaluated as the heating rate multiplied by the total ejecta mass at the moment when the diffusion radius reaches the transition point . on the other hand , the index of our mass density profile of the ejecta is @xmath231 , which is indicated by general relativistic simulations by @xcite . this profile is quite different from the profile adopted in @xcite ; the index is close to 3 , in which the mass in each logarithmic radius is constant . then , at the time @xmath232 ( equation [ sec2 - 2 - 3:t_tr ] ) , the mass contributing to the luminosity is about @xmath360% of the total ejecta mass in the case @xmath233 . this profile predicts luminosity dimmer than that other studies . in fact , @xcite tried to explain the observed infrared excess using the mass profile which is almost the same with ours . in the case of a binary ns merger with ejecta mass @xmath30.02 @xmath234 , even if they use a larger nuclear heating rate ( larger by a factor of 2 ) , their predicted luminosity is slightly smaller than the observed infrared excess ( in the left panel of their figure 3 ) . this result is consistent with our model , i.e. , our model requires larger mass than most of previous studies . in the engine model , the injected internal energy which determines the luminosity does not depend on the ejecta mass ( except for the limit in equation [ sec2 - 1:e_int ] ) . however , the luminosity declines rapidly after the transition time @xmath235 which depends on the ejecta mass as in equation ( [ sec2 - 2 - 3:t_tr ] ) . the condition @xmath236 s in the observer frame is required to reproduce the excess observed from grb 130603b in the near - infrared band . this condition gives the lower limit for the ejecta mass in the engine model , @xmath237 with the opacity @xmath238 cm@xmath2 g@xmath22 . note that the observed upper limit on the infrared luminosity at @xmath239 s in the observer frame ( figure [ figure : observation ] ) , which corresponds to @xmath240 s , gives the upper limit on the ejecta mass for both models . however , this limit is not important for the range @xmath241 in the range of the opacity @xmath225 cm@xmath2 g@xmath22 . the smaller minimum velocity @xmath43 gives the smaller bolometric luminosity at certain time in the @xmath0-process model ( see equations [ sec2 - 2 - 1:l_bol , nuc ] and [ sec2 - 2 - 2:l_bol , nuc ] ) . the small minimum velocity enlarges the size of ejecta ( when we fix the maximum velocity @xmath39 ) . then , the diffusion time @xmath58 of photons emitted from the inner region of the ejecta becomes long for the small velocity @xmath43 ( equation [ sec2 - 2 - 3:t_tr ] ) . the mass between @xmath70 and @xmath123 ( or @xmath73 ) increases toward inner region of the ejecta ( as long as @xmath134 ) so that the mass is reduced for the small minimum velocity @xmath40 at certain time . in fact , the dependence of the mass on the minimum velocity is @xmath242 . as a result , smaller minimum velocity gives smaller luminosity to reproduce the observed infrared excess of grb 130603b . moreover , smaller minimum velocity gives larger temperature @xmath135 at certain time ( equations [ sec2 - 2 - 2:t_obs , nuc ] and [ sec2 - 2 - 3:t_obs , nuc ] ) because mass density at a shell with small velocity is large . the difference between the detected luminosity at f160w band and the upper limit on the luminosity at f606w band at @xmath84 s in the observer frame gives the upper limit on the observed temperature ( @xmath243 k ) . to satisfy the observed upper limit on the temperature from grb 130603b , a lower limit of @xmath244 is obtained for @xmath245 . the smaller minimum velocity @xmath43 gives higher temperature @xmath135 in the engine model ( equations [ sec2 - 2 - 2:t_obs , sh ] and [ sec2 - 2 - 3:t_obs , sh ] ) . the observational limit for the temperature at @xmath84 s in the observer frame indicates that the range of the minimum velocity @xmath40 is limited in the engine model ( @xmath246 for @xmath245 ) . we discuss the dependence on the value of @xmath80 . as mentioned in section 2.2 , we use the temperature - independent opacity @xmath80 with the grey approximation . in general , the @xmath0-process line opacity depends on frequency and changes with temperature and ionization state of the ejecta @xcite . the indicated grey opacity is @xmath225 cm@xmath2 g@xmath22 . in the case of the @xmath0-process model , the luminosity significantly depends on opacity @xmath80 . the larger opacity causes larger diffusion time @xmath58 , so that larger time is required to observe the inner region of the ejecta for given ejecta mass @xmath53 and minimum velocity @xmath43 . in fact , two transition times @xmath159 and @xmath188 are proportional to @xmath247 ( equations [ sec2 - 2 - 2:t_times ] and [ sec2 - 2 - 3:t_tr ] ) . then , the mass around the diffusion radius @xmath70 is small at certain time , so that the luminosity is reduced . as a result , in order to explain the infrared excess observed in grb 130603b , larger mass @xmath53 is required for the larger value of opacity @xmath80 . for the opacity @xmath248 cm@xmath2 g@xmath22 , total ejecta mass @xmath249 is required to reproduce the observed excess , which is much larger than the simulation results of mergers of binary nss ( e.g. , * ? ? ? * ) . on the other hand , the transition time @xmath188 is smaller for the smaller value of the opacity . then , the luminosity significantly increases at @xmath226 s. for the opacity @xmath250 cm@xmath2 g@xmath22 , there is no parameter set which gives smaller luminosity than the detection at f606w band ( @xmath226 s in the observer frame ) and the luminosity comparable to the observed excess at f160w band simultaneously in the @xmath0-process model . in the case of the engine model , a larger value of opacity @xmath80 reduces the lower limit for the mass @xmath53 to explain the observed excess . for certain temperature and luminosity , the opacity @xmath80 and the ejecta mass @xmath53 always degenerate in the form @xmath251 ( see equations [ sec2 - 2 - 1:t_obs , sh ] , [ sec2 - 2 - 1:l_bol , sh ] , [ sec2 - 2 - 2:t_obs , sh ] , [ sec2 - 2 - 2:l_bol , sh ] , [ sec2 - 2 - 3:t_obs , sh ] and [ sec2 - 2 - 3:l_bol , sh ] ) . this dependence comes from the optical depth ( equation [ sec2 - 1:tau ] ) because the internal energy in the ejecta does not depend on the opacity and the ejecta mass , contrary to the @xmath0-process model . we present a parameter set to give the minimum ejecta mass @xmath53 in table 1 as the minimum mass parameter set . we also plot the value of @xmath53 and @xmath43 of this model in figure [ figure : parameter ] as a square . this ejecta mass is naturally realized in general relativistic simulations ( e.g. , * ? ? ? although the larger value of the opacity @xmath80 reduces the lower limit for the ejecta mass @xmath53 , the kinetic energy is @xmath252 erg ( equation [ sec2 - 1:e_kin ] ) which is close to the initial injected energy @xmath253 erg for the minimum mass parameter set . the lower ejecta mass @xmath53 reduces the kinetic energy of the ejecta , @xmath254 in equation ( [ sec2 - 1:e_kin ] ) , so that the required energy @xmath97 may exceed the kinetic energy of the ejecta for the larger opacity . for the small value of the opacity , larger mass and smaller minimum velocity is required to satisfy the condition @xmath236 s ( @xmath255 in equation [ sec2 - 2 - 3:t_tr ] ) . for the opacity @xmath256 cm@xmath2 g@xmath22 , the condition corresponds to @xmath257 . the observational constraint for the temperature also requires a large value of the minimum velocity @xmath43 . then , there is no solution to explain the observed excess within the parameter range shown in figure [ figure : parameter ] for the opacity @xmath258 cm@xmath2 g@xmath22 . therefore , for the small opacity @xmath259 cm@xmath2 g@xmath22 the engine model can not explain the observed excess . since the engine model has additional free parameters , @xmath95 and @xmath101 , the allowed region of the parameters is larger than that of the @xmath0-process model . we can impose the lower limit on the parameter @xmath101 by regarding the infrared luminosity @xmath27 erg s@xmath22 at @xmath260 day as bolometric luminosity in the source rest frame with equation ( [ sec2 - 2 - 3:l_bol , sh ] ) . the derived limit is @xmath261 . to satisfy the optical upper limit at @xmath226 s and the detected luminosity at @xmath84 s in the observer frame ( figure [ figure : observation ] ) , we find the lower limit on the index of the temperature profile @xmath262 . for a smaller value of the index @xmath95 , emission from the ejecta with relatively high temperature can be observed at time @xmath226 s , so that luminosity at f606w band is larger than the observed upper limit of grb 130603b . in addition , the smaller value of @xmath95 decreases relative internal energy in the inner edge of the ejecta . to reproduce the luminosity at time @xmath263 s in the observer frame when observed emission comes from the inner ejecta , the smaller @xmath95 requires the larger initial internal energy @xmath97 which exceeds the kinetic energy of the ejecta @xmath105 in some cases . several deep optical observations of short grbs give stringent upper limits on the luminosity of macronovae @xcite . we compare the results with two deep optical observations of short grbs , grb 050509b and grb 080905a . for the fiducial parameter set , the luminosity exceeds the observational upper limits on these two observations . in the engine model , we can reduce the luminosity in the early phase @xmath264 s without reducing the luminosity in the late phase @xmath84 s by utilizing the steep temperature profile ( large @xmath95 ) . here , we introduce the hot interior parameter set with larger value of index @xmath95 than that of the fiducial parameter set . since emission from the inner part of the ejecta is observed at the later time , the luminosity at the early phase decreases and avoids the observational limits if most of the internal energy is injected to the inner part of the ejecta . we show the light curve of the hot interior parameter set in figure [ figure : observation2 ] . we choose the parameters as @xmath265 and @xmath266 ( the right column of table 1 ) . from figure [ figure : observation2 ] , the light curves are consistent with all three observations using the same model parameters . a possible scenario for the hot interior parameter set is that the shock produced by the activity of the central engine may not be able to catch up with the outer part of the ejecta because the velocity of the ejecta is close to the light speed ( @xmath42 ) . then , only the inner part of the ejecta will be heated . for comparison , we also show the light curves in the @xmath0-process model with the parameter set of the hot interior in figure [ figure : observation2 ] as dashed lines . the luminosity of the @xmath0-process model exceeds the observed upper limits in two observations , grb 050509b and grb 080905a ( left and middle panels of figure [ figure : observation2 ] ) if we choose the parameter set of the hot interior model . we are not able to find any parameter set in the @xmath0-process model , which simultaneously satisfies the observed limits of the three observations . note that we do not argue that the @xmath0-process model is excluded from these results because we need to take into account the variations of the model parameters for each event . note that the extended emission was not detected in three short grbs . however , it is not unreasonable to miss the extended emission of these bursts . one possibility is a selection effect . observationally , the fraction of short grbs with extended emission is significantly larger at softer energy bands : @xmath325% in the swift bat samples ( @xmath26715 kev ; * ? ? ? * ) and @xmath37% in the batse samples ( @xmath26720 kev ; * ? ? ? this suggests that observations with a low energy threshold may dramatically increase short grbs with extended emission @xcite . this will be further tested by future soft x - ray survey facilities such as wide - field maxi ( 0.7 - 10 kev ) @xcite . the three referred short grbs were detected by swift bat and therefore swift bat could not detect extended emission by chance . alternatively , the outflow following the main short grb jet could not breakout the ejecta . @xcite and @xcite investigated the propagation of jets in merger ejecta . they found the cases that relativistic jets can penetrate merger ejecta and produce the prompt emission of short grbs , but in the late energy injection cases , outflow fails to breakout the ejecta . therefore , some extended emission may not be observed , although the central engine works actively . -process model on the shape of the front of the ejecta . we plot the case of grb 130603b . thick - dashed lines denote the fiducial model . thin - solid , thin - dashed , and thin - dot - dashed lines correspond to the exponential profiles of mass density ( equation [ sec3:rho ] ) with @xmath268 and @xmath269 , respectively . we also plot the observational data from @xcite.,width=302 ] in the thin - diffusion phase , the light curve strongly depends on the density profile of the ejecta surface . the density profile is determined by the complex merger dynamics @xcite , so that the density profile of the ejecta can not be analytically derived as mentioned in section 2.3 . since the outer part of the density profile is difficult to calculate precisely , little attention has paid on the mass profile at the outer region in current numerical simulations . in order to investigate the dependence of the light curve on the mass profiles in the thin - diffusion phase , we consider other forms of the mass profile and compare the light curve with that of equation ( [ sec2 - 1:rho ] ) . we adopt an exponential profile @xmath270 we introduce an additional free parameter @xmath271 and the ejecta expand @xmath272 . the calculations are the same except for using @xmath273 . we fix the mass with velocity larger than @xmath274 and calculate three models for @xmath275 . for the other parameters , we adopt from the fiducial parameter set in table 1 . we show the results in the @xmath0-process model in figure [ figure : exp_comp ] . since we fix the mass with velocity larger than @xmath274 , the bolometric luminosities at the time @xmath276 are almost the same values . in the thin - diffusion phase ( @xmath277 ) , both the luminosity and the temperature are smaller than the fiducial model . this is because the density at the front of the ejecta is reduced in this mass profile . since the maximum velocity effectively becomes large and the adiabatic cooling becomes efficient , these effects for luminosity and temperature should be also seen in the engine model . we conclude that the luminosities in the thin - diffusion phase ( @xmath277 ) have uncertainties at least with @xmath278 mag , which originates from the uncertainty of the outermost mass profile . note that the emission from the ejecta with the mass profile discussed here reduces the tension between the light curve in the @xmath0-process model and the upper limits of the deep optical observations ( grb 050509 and grb 080905a ) as discussed in section [ other ] . especially , the optical luminosity in the case @xmath279 ( thin - dot - dashed line ) significantly decreases after @xmath280 s. therefore , the ejecta with relatively shallow mass distribution at the front of the ejecta is able to explain the current optical follow - up observations in the @xmath0-process model . in the fiducial parameter set , the light curves for two models in the optical and infrared bands are similar ( figure [ figure : observation ] ) . in figure [ figure : parameter ] , the allowed parameter region to explain the observation of grb 130603b for both models are also overlapped . recall that emission from ejecta is described as blackbody radiation for the two models , whose spectrum is narrow in bands . therefore , no excess at other wavelengths is expected and also the prediction is consistent with radio observations @xcite . therefore , it is difficult to discriminate two models from the currently available observational data . in the @xmath0-process model , the light curve with mass profile @xmath35 ( @xmath281 ) and a parameter set which explains the infrared excess detected from grb 130603b can not explain the upper limits obtained from the deep optical observations of some short grbs . therefore , if both stringent optical upper limits at @xmath226 s and bright infrared emission at @xmath84 s are simultaneously obtained from a single event ( with a difference larger than two magnitudes @xmath282 mag ) , the @xmath0-process model is significantly restricted . for the engine model , these observations give a constraint for the temperature distribution in the ejecta , which may give new insights into the activity of the central engine . as shown in the middle panel of figure [ figure : evolution ] , the bolometric luminosity in the @xmath0-process model from the whole ejecta ( blue long - dashed line ) , including low temperature and/or x - rays and @xmath194-rays produced directly in radioactive decays @xcite , declines more gradually than that for the engine model . this is because there is no energy injection after the time @xmath283 for the engine model . then , the luminosity significantly decreases when photons at the inner edge of the ejecta begin to diffuse out ( see the middle panel of figure [ figure : evolution ] and figure [ figure : observation ] ) . the luminosity from the whole ejecta can be described as @xmath193 in the transparent phase . the index of time @xmath34 is determined by the nuclear heating rate , @xmath284 . therefore , the two models are distinguishable by observing the temporal evolution of bolometric luminosity from the whole ejecta in this phase . we calculated the light curves of macronovae by developing analytical models . we modeled the ejecta based on the results of numerical simulations for a merger of binary nss . in addition to the nuclear decay of @xmath0-process elements ( the @xmath0-process model which is often discussed ) , we considered another heating mechanism for the ejecta , the engine - driven shock ( engine model ) . we compared the results with the optical and infrared observations of the first macronova candidate associated with grb 130603b , and showed that both models can explain the observations . in order to reproduce the observed light curve , the @xmath0-process model requires relatively large ejecta mass @xmath285 which is mainly determined by the observed infrared luminosity @xmath27 erg s@xmath22 at @xmath84 s. in the engine model , the internal energy of ejecta , which mainly determines the observed luminosity , does not depend on the ejecta mass . then , unless the entire of the ejecta is effectively thin ( the diffusion time is smaller than the dynamical time , @xmath286 , at the inner edge of the ejecta ) , s in the engine model , the luminosity rapidly decreases after the time @xmath287 when photons can diffuse out from the inner edge of the ejecta ( section [ transparent ] ) . ] the required ejecta mass is @xmath237 , which is comparable to the recent numerical simulation results . the initial internal energy @xmath97 and the injection time @xmath13 are required as @xmath288 , which is consistent with the observed extended emission of short grbs , @xmath10 erg and @xmath11 s. the required minimum velocity is about @xmath289 for both models , which is mainly determined by the constraint for the observed temperature @xmath29 k at @xmath84 s. for the range of the opacity @xmath250 cm@xmath2 g@xmath22 , it is difficult for both models to explain the observations of macronova associated with grb 130603b by the ejecta mass less than @xmath241 . if macronovae are identical , the upper limits on the luminosity obtained in the deep optical observations of other short grbs give stringent constraints on the @xmath0-process model . on the other hand , the engine model satisfies these constraints if the temperature profile is centrally concentrated in the ejecta ( large @xmath95 ) . thus , if the difference between the optical magnitude at @xmath226 s and the infrared magnitude at @xmath84 s is larger than @xmath290 mag in a single event , the @xmath0-process model is difficult to explain the observations unless the front of the ejecta has much shallow mass distribution . another difference in the light curves between two models is the bolometric luminosity at the transparent phase when dynamical time is smaller than the diffusion time at the inner edge of the ejecta @xmath198 . although the optical and infrared luminosities rapidly decrease in the transparent phase , the bolometric luminosity from the whole ejecta , including lower frequency than near - infrared band and/or x - rays and @xmath194-rays produced directly in radioactive decays , is determined by the energy injection rate of nuclear decay , @xmath291 . for the engine model , the bolometric luminosity decreases rapidly in this phase ( faster than @xmath292 ) . therefore , we expect that the light curve of the bolometric luminosity from the whole ejecta can distinguish between two heating mechanisms . our results show that early light curves depend on the density profile of the outermost edge of the ejecta . it is necessary to develop a method to calculate the low - density region of the ejecta in either the analytical or numerical ways in order to precisely predict the early light curves of macronovae . we thank k. asano , k. kashiyama , k. kiuchi , h. nagakura , t. nakamura , y. sekiguchi , m. shibata for fruitful discussions . this work is supported by kakenhi 24103006 ( s.k . , k.i . ) , 24.9375 ( h.t . ) , 24000004 , 26247042 , 26287051 ( k.i . ) . we summarize the formula for the observed temperature and bolometric luminosity . the detailed derivation of equations in this section is described in sections 2 and 3 . since we assume that the observed luminosity and temperature approximate the luminosity and temperature at the diffusion radius @xmath70 ( section 3 ) , we need to calculate the diffusion radius . for the dynamics of the ejecta , we assume an isotropic and homologous expansion . then , the velocity of ejecta @xmath12 is described by equation ( [ sec2:v ] ) @xmath293 where the radius @xmath0 originates the central engine and the time @xmath34 is measured from the time when a compact binary merges . as in section [ fiducial ] , we calculate the diffusion radius @xmath70 from the condition that the diffusion time equals the dynamical time , @xmath71 . the diffusion time is described by equation ( [ sec2 - 1:t_diff ] ) as @xmath294 where @xmath295 is the speed of the light , @xmath77 is the optical depth described by equation ( [ sec2 - 1:tau ] ) as @xmath296 and @xmath65 is the width of the diffusion region described by the equation ( [ sec2 - 1:deltar ] ) as @xmath297 in the calculation of the optical depth @xmath77 , we use the spatially uniform value of the optical depth @xmath80 with grey approximation and the ejecta mass density @xmath298 described by equation ( [ sec2 - 1:rho ] ) as , @xmath299 where @xmath50 and @xmath51 are normalized factors , and @xmath43 is the velocity at the inner edge of the ejecta . the radius @xmath73 is the outer edge of the ejecta , described by equation ( [ sec2 - 1:r_out ] ) as @xmath300 where the velocity @xmath39 is at the outer edge of the ejecta . the radius at the inner edge of the ejecta @xmath198 is described by equation ( [ sec2 - 1:r_in ] ) as @xmath301 the normalization factor @xmath52 in the profile of the mass density is determined by the ejecta mass @xmath53 ( in equation [ sec2 - 1:m_ej ] ) as @xmath302 we consider two heating sources of the ejecta . in the @xmath0-process model , the internal energy of the ejecta is determined by the nuclear heating rate of the @xmath0-process element described in equation ( [ sec2 - 1:dotepsilon ] ) as @xmath303 for the engine model , we assume the temperature profile of the ejecta @xmath304 as a result of the activity of the central engine , described in equation ( [ sec2 - 1:t ] ) as @xmath305 where @xmath96 is the normalization factor . this factor is determined by injected internal energy @xmath97 at the time @xmath13 , which is described in equation ( [ sec2 - 1:e_int ] ) as @xmath306 observed temperatures in the @xmath0-process model and the engine model are given by @xmath307^{1/4 } } \\ ~~~~~~~~~~~~~~~(r_{\rm diff}>r_{\rm in } ) \\ \\ { \displaystyle \left[\frac{\dot{\epsilon}t\rho(t , v_{\min})}{a}\right]^{1/4 } } \\ ~~~~~~~~~~~~~~~(0.5r_{\rm in } < r_{\rm diff } \le r_{\rm in } ) \\ \\ { \displaystyle 0 } \\ ~~~~~~~~~~~~~~~(r_{\rm diff } \le 0.5r_{\rm in } ) , \\ \end{array } \right.\end{aligned}\ ] ] and by @xmath308 respectively . note that we do not use the approximation @xmath309 and @xmath310 in the thin - diffusion case . the bolometric luminosities for the @xmath0-process model and the engine model are given by @xmath311 and @xmath312 respectively . an example of the calculated result is shown in figure [ figure : evolution ] . we present the numerical values with the parameter dependence for later use . unlike equations ( [ sec2 - 2 - 2:t_times ] ) and ( [ sec2 - 2 - 3:t_tr ] ) , we include the contribution from subdominant terms to the numerical values when we integrate equations . some of the subdominant terms include the ratio @xmath313 . hereafter , the value @xmath314 in subdominant terms are fixed and are not included in the parameter dependence . we introduce the normalized quantities @xmath164 , @xmath315 , @xmath166 , @xmath316 cm@xmath2 g@xmath22 , @xmath317 erg and @xmath318 s. for other parameters , we fix the index of the mass density profile @xmath41 and the parameters of the nuclear heating rate @xmath224 erg s@xmath22 g@xmath22 and @xmath87 . we also introduce the normalized time @xmath319 s and @xmath320 s. the values of observed temperature and bolometric luminosity in the thin - diffusion phase are @xmath321 and @xmath322 respectively . the transition time from the thin - diffusion phase to the thick - diffusion phase @xmath159 is @xmath323cv_{\max}}\left(\frac{v_{\max}}{v_{\min}}\right)^{3-\beta } } \nonumber \\ \nonumber \\ & \sim&4.53\times10 ^ 5~\kappa_{10}^{0.5}m_{\rm ej,0.1}^{0.5}v_{\min,0.1}^{0.25}v_{\max,0.4}^{-0.75}~{\rm s}.\end{aligned}\ ] ] the values of the observed temperature and bolometric luminosity in the thick - diffusion phase are @xmath324 and @xmath325 respectively . note that since the diffusion radius @xmath70 can not be analytically described in the thin - diffusion phase , we use the approximations @xmath309 and @xmath310 in equations ( [ app : t_obs , thin ] ) and ( [ app : l_bol , thin ] ) . these approximations make discontinuity at the transition time @xmath159 . the ratios of the temperature in the thick - diffusion phase to the temperature in the thin - diffusion phase for the @xmath0-process @xmath326 and the engine model @xmath327 at the time @xmath159 are @xmath328 and @xmath329 respectively . the ratios of the luminosity in the thick - diffusion phase to the luminosity in the thin - diffusion phase for the @xmath0-process model @xmath330 and the engine model @xmath331 at the time @xmath159 are @xmath332 and @xmath333 respectively . the transition time from the thick - diffusion phase to the transparent phase @xmath188 is @xmath334cv_{\min } } } \nonumber \\ \nonumber \\ & \sim&7.62\times10 ^ 5~\kappa_{10}^{0.5}m_{\rm ej,0.1}^{0.5}v_{\min,0.1}^{-0.5}~{\rm s}.\end{aligned}\ ] ] we introduce another transition time @xmath335 when the upper limit of the integral for the luminosity @xmath123 reaches the inner edge of the ejecta @xmath198 , @xmath336 the values of the observed temperature and bolometric luminosity in the transparent phase ( @xmath337 ) are @xmath338 and @xmath339 } \\ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~(r-{\rm process } ) \\ \\ { \displaystyle 1.33\times10^{41}~~{\rm erg~s}^{-1 } } \\ ~~{\displaystyle \times e_{\rm int0,51}t_{\rm inj,2}t_6^{-2}\left[1-\left(\frac{t}{t_{\rm tr2}}\right)^{1.333}\right ] } \\ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~({\rm engine},~\xi=1 ) , \\ \\ { \displaystyle 1.00\times10^{41}~~{\rm erg~s}^{-1 } } \\ ~~{\displaystyle \times e_{\rm int0,51}t_{\rm inj,2}t_6^{-2}\left[1-\left(\frac{t}{t_{\rm tr2}}\right)^{6.667}\right ] } \\ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~({\rm engine},~\xi=2 ) , \\ \\ { \displaystyle 1.00\times10^{41}~~{\rm erg~s}^{-1 } } \\ ~~{\displaystyle \times e_{\rm int0,51}t_{\rm inj,2}t_6^{-2}\left[1-\left(\frac{t}{t_{\rm tr2}}\right)^{12}\right ] } \\ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~({\rm engine},~\xi=3 ) , \\ \end{array } \right.\end{aligned}\ ] ] respectively .
since the discovery of the first known pulsating x - ray sources her x-1 and cen x-3 more than three decades ago ( giacconi et al . 1971 ; tananbaum et al . 1972 ) , over 50 new sources have been detected in the galaxy and the magellanic clouds , with luminosities in the range @xmath0 and pulsation periods @xmath1 . x - ray pulsars include a variety of objects powered by rotation or accretion , as well as several anomalous x - ray pulsars whose energy source is currently unclear . the emission from x - ray pulsars in binary systems is fueled by the accretion of material from the `` normal '' companion onto the neutron star , with the flow channeled onto one or both of the magnetic poles by the strong field . during the accretion process , gravitational potential energy is converted into kinetic energy , which escapes from the column in the form of x - rays as the gas decelerates through a radiative shock before settling onto the stellar surface . in the accretion - powered sources , which are the focus of this paper , the x - ray spectra are often well fitted using a combination of a power - law spectrum plus a blackbody component with a temperature in the range @xmath2k ( e.g. , coburn et al . 2002 ; di salvo et al . 1998 ; white et al . 1983 ) . most spectra also display quasi - exponential cutoffs at @xmath3kev , and there are indications of cyclotron features and iron emission lines in a number of sources . the observations suggest typical magnetic field strengths of @xmath4 g . although accretion - powered x - ray pulsars are among the most luminous sources in the galaxy , previous attempts to calculate their spectra based on static or dynamic theoretical models have generally yielded results that do not agree very well with the observed profiles ( e.g. , mszros & nagel 1985a , b ; nagel 1981 ; yahel 1980 ; klein et al . hence there is still no clear understanding of the basic spectral formation mechanism in x - ray pulsars ( see the discussion in coburn et al . 2002 ) . given the lack of a viable theoretical model , x - ray pulsar spectral data have traditionally been fit using multicomponent forms that include absorbed power - laws , cyclotron features , iron emission lines , blackbody components , and high - energy exponential cutoffs . the resulting parameters are sometimes difficult to relate to the physical properties of the source . motivated by the lack of a comprehensive theoretical model for x - ray pulsar spectral formation , we reconsider here the physical picture originally proposed by davidson ( 1973 ) , in which the accreting gas passes through a radiative , radiation - dominated shock before settling onto the surface of the star . we illustrate the accretion / emission geometry schematically in figure 1 . most of the photons emitted from the accretion column are produced in the dense `` thermal mound '' located inside the column , just above the stellar surface . the blackbody photons created in the mound are upscattered in the shock , and eventually diffuse through the walls of the column . the escaping photons carry away the kinetic energy of the gas , thereby allowing the plasma to settle onto the surface of the star . the strong compression that occurs as the plasma crosses the radiative shock renders it an ideal site for first - order fermi energization ( i.e. , `` bulk '' or `` dynamical '' comptonization ) of the photons produced by the thermal mound . in the bulk comptonization process , particles experience a mean energy gain if the scattering centers they collide with are involved in a converging flow ( e.g. , laurent & titarchuk 1999 ; turolla , zane , & titarchuk 2002 ) . by contrast , in the thermal comptonizaton process , particles gain energy due to the _ stochastic _ motions of the scattering centers via the second - order fermi mechanism ( e.g. , sunyaev & titarchuk 1980 ; becker 2003 ) . in the x - ray pulsar application , the scattering centers are infalling electrons , and the energized `` particles '' are photons . since the inflow speed of the electrons in an x - ray pulsar accretion column is much larger than their thermal velocity , bulk comptonization dominates over the stochastic process except at the highest photon energies ( titarchuk , mastichiadis , & kylafis 1996 ) , as discussed in 7 . the failure of the current models to generate the power - law spectral shape characteristic of x - ray pulsars probably stems from the neglect of the critical contribution of the shock in upscattering the soft radiation produced in the thermal mound ( e.g. , burnard , arons , & klein 1991 ) . we demonstrate in this paper that bulk comptonization of the radiation in the accretion shock can produce spectra very similar to the power - law continuum seen in many x - ray pulsars . the main results presented here were summarized by becker & wolff ( 2005 ) . dynamical comptonization has been considered by a number of previous authors in a variety of astrophysical situations , including spherical inflows ( payne & blandford 1981 ; colpi 1988 ; schneider & bogdan 1989 ; titarchuk , mastichiadis , & kylafis 1997 ) , accretion disks around black holes ( laurent & titarchuk 2001 ; titarchuk , cui , & wood 2002 ; titarchuk & shrader 2002 ) , spherical outflows ( becker & begelman 1986 ; titarchuk , kazanas , & becker 2003 ) , and the transport of cosmic rays in the solar wind ( parker 1965 ; stawicki , fichtner , & schlickeiser 2000 ) . the process was also studied in the context of neutron star accretion by titarchuk , mastichiadis , & kylafis ( 1996 ) and mastichiadis & kylafis ( 1992 ) , although their results are not applicable to x - ray pulsar accretion columns due to the assumption of spherical symmetry and the imposition of a power - law velocity profile . lyubarskii & sunyaev ( 1982 ) analyzed dynamical comptonization in the context of plane - parallel pulsar shocks , but the velocity profile they utilized is not consistent with the dynamics of x - ray pulsar accretion , and furthermore they did not account for the escape of radiation through the walls of the column . the relevance of the bulk comptonization process for spectral formation in x - ray pulsars was also recognized by burnard , arons , & klein ( 1991 ) , who pointed out that thermal spectra alone are much too soft to explain the observed emission , and suggested that upscattering in the shock could provide the required hardening . the fundamental character of the green s function describing both thermal and bulk comptonization was studied by titarchuk & zannias ( 1998 ) for the case of accretion onto a black hole . these authors established that the green s function can be approximated using a broken power - law form with a central peak between the high- and low - energy portions of the spectrum , for either type of comptonization . furthermore , they concluded that bulk comptonization dominates over the thermal process if the electron temperature @xmath5k . the theory developed in the present paper represents an extension of the same idea to neutron star accretion , resulting in the exact green s function for the x - ray pulsar spectral formation process . in agreement with titarchuk & zannias ( 1998 ) , we find that the bulk comptonization process is dominant for the temperature range relevant for x - ray pulsars . however , the presence of the event horizon in the black - hole application treated by these authors makes their problem fundamentally different from the neutron star accretion problem studied here , because the neutron star obviously possesses a solid surface . our work therefore represents the first exact , quantitative analysis of the role of bulk comptonization in the x - ray pulsar spectral formation process . radiation pressure governs the dynamical structure of the accretion flows in bright pulsars when the x - ray luminosity satisfies ( becker 1998 ; basko & sunyaev 1976 ) @xmath6 where @xmath7 is the polar cap radius , @xmath8 and @xmath9 denote the stellar mass and radius , respectively , @xmath10 is the thomson cross section , and @xmath11 and @xmath12 represent the electron scattering cross sections for photons propagating parallel or perpendicular to the magnetic field , respectively . when the luminosity of the system is comparable to @xmath13 , the radiation flux in the column is super - eddington and therefore the radiation pressure greatly exceeds the gas pressure ( becker 1998 ) . in this situation the gas passes through a radiation - dominated shock on its way to the stellar surface , and the kinetic energy of the gas is carried away by the high - energy radiation that escapes from the column . the strong gradient of the radiation pressure decelerates the material to rest at the surface of the star . the observation of many x - ray pulsars with @xmath14 implies the presence of radiation - dominated shocks close to the stellar surfaces in these systems ( white et al . 1983 ; white et al . note that radiation - dominated shocks are _ continuous _ velocity transitions , with an overall thickness of a few thomson scattering lengths , unlike traditional ( discontinuous ) gas - dominated shocks ( blandford & payne 1981b ) . in luminous x - ray pulsars , the pressure of the radiation determines the dynamics of the flow , while the dynamics in turn determines the shape of the radiation spectrum . hence the flow dynamics and the radiative transport are intimately connected , which makes this a complex and nonlinear `` photohydrodynamical '' problem . one implication is that the photons scattering through the column can not be regarded as `` test particles '' since it is their pressure that dictates the structure of the flow . we must therefore solve for the radiation spectrum and the velocity profile in a self - consistent manner . radiation pressure may also play an important role in the dynamics of moderate - luminosity pulsars due to the strong dependence of the electron scattering ( cyclotron ) cross section on the magnetic field strength ( langer & rappaport 1982 ) . becker ( 1998 ) and basko & sunyaev ( 1976 ) have considered the dynamics of radiation - dominated pulsar accretion flows , including the effect of the shock and the escape of radiation energy from the column . they find that the gas is decelerated to zero velocity in a manner consistent with accretion onto a solid body , but they do not consider the shape of the escaping radiation spectrum . conversely , the dynamics and the radiation spectrum have been worked out self - consistently by blandford & payne ( 1981b ) for the case of a `` standard '' ( rankine - hugoniot ) radiation - dominated shock , which does not include radiation escape . in the blandford & payne approach , the radiation spectrum everywhere in the flow is determined by solving the transport equation using the incident radiation field as a boundary condition . the solutions have a power - law character at high energies , with a spectral index that depends on the mach number of the upstream flow . since the rankine - hugoniot shocks studied by blandford & payne have a conserved energy flux , the downstream velocity is never less than one - seventh its upstream value . consequently the associated spectral solutions are not applicable to the x - ray pulsar case , since pulsar shocks must be _ radiative _ in nature so that the kinetic energy of the infalling material can be removed and the flow brought to rest at the stellar surface . the analytical approach employed here parallels the work of blandford & payne ( 1981b ) , except that the velocity profile they utilized is replaced with the appropriate solution for an x - ray pulsar accretion column , and the loss of radiation energy from the column is incorporated into the transport equation using an escape - probability formalism . in the spirit of the blandford & payne study of spectral formation in radiation dominated rankine - hugoniot shocks , we are interested here in exploring the direct effect of the radiative accretion shock on the spectrum emerging from an x - ray pulsar . hence it is not our goal in this paper to develop complete models that include additional processes such as free - free emission and absorption , cyclotron features , and iron emission lines . however , even without including any of these processes , we are able to demonstrate qualitative agreement with x - ray pulsar spectra . this suggests that energization in the accretion shock is one of the most important aspects of spectral formation in x - ray pulsars . -1.0truein the remainder of the paper is organized as follows . in 2 we briefly review the conservation equations and discuss the resulting dynamical solution for the velocity profile of the accreting gas . the transport equation governing the scattering of the radiation inside the accretion column is analyzed in 3 , and various constraints are derived in order to ensure that the radiative transfer is modeled in a manner consistent with the flow dynamics . in 4 we obtain the exact analytical solution for the green s function and convolve it with the planck distribution emitted at the thermal mound to compute the radiation spectrum emerging from the accretion column . in 5 we present results obtained for the count - rate spectra using parameters corresponding to specific pulsars , and we compare the theoretical spectra with the observational data . the implications of our results for the production of x - ray spectra in pulsars are discussed in 7 . additional mathematical details are provided in a series of appendices . the model analyzed here is based on the dynamical picture suggested by davidson ( 1973 ) , which is illustrated schematically in figure 1 . the accretion scenario corresponds physically to the flow of a mixture of gas and radiation inside a magnetic `` pipe '' that is sealed with respect to the gas , but is transparent with respect to the radiation . the accretion column incorporates a radiation - dominated , radiative shock located above the stellar surface . although the shock is extended and the velocity profile is continuous , the flow possesses a well - defined sonic surface , as indicated in figure 1 . the model also includes a dense thermal mound located at the base of the flow , where local thermodynamic equilibrium prevails . due to its blackbody nature , the thermal mound acts as both a photon source ( emitting a planck spectrum ) and a photon sink ( absorbing all incident radiation ) . hence the surface of the mound represents the `` photosphere '' for photon creation and absorption , and the opacity is dominated by electron scattering above this point . in our approach to the problem , we assume that the upstream flow is composed of pure , fully ionized hydrogen gas moving at a highly supersonic speed , which is the standard scenario for accretion - powered x - ray pulsars ( basko & sunyaev 1975 , 1976 ) . our transport model employs a cylindrical , plane - parallel geometry , and therefore the velocity , density , and pressure are functions of the distance above the stellar surface , but they are all constant across the column at a given height . in this situation , the vertical variation of the flow velocity is given by the exact analytical solution obtained by becker ( 1998 ) and basko & sunyaev ( 1976 ) , which describes the settling of the transonic flow onto the surface of the star . a detailed consideration of the angular and energy dependences of the electron scattering cross section is beyond the scope of the present paper . we shall therefore follow wang & frank ( 1981 ) and becker ( 1998 ) by treating the directional dependence of the electron scattering in an approximate way in terms of the energy - averaged cross sections @xmath11 and @xmath12 , describing respectively the scattering of photons propagating either parallel or perpendicular to the magnetic field . the gas in the accretion column is radiation - dominated , and therefore it is the pressure of the photons that decelerates the infalling plasma to rest . we will use this fact in 6.1 to evaluate the self - consistency of the coupled radiative transport / dynamical model based on the analytical solution for the radiation pressure profile obtained below . the one - dimensional , time - dependent euler equations governing the flow of a radiation - dominated gas with mass density @xmath15 , flow velocity @xmath16 , and photon energy density @xmath17 inside a cylindrical accretion column are @xmath18 @xmath19 @xmath20 where the mass , momentum , and energy fluxes are given respectively by @xmath21 @xmath22 @xmath23 with @xmath24 denoting the radiation pressure . the spatial coordinate @xmath25 increases along the column axis , in the direction of the flow , and @xmath26 denotes the associated electron - scattering optical depth , which is related to @xmath25 via @xmath27 where @xmath28 is the electron number density , @xmath29 is the proton mass , and @xmath11 is the electron scattering cross section parallel to the magnetic field ( i.e. , the column axis ) . the two coordinates @xmath25 and @xmath26 are calibrated so that they each vanish at the sonic point , as explained below . the quantities @xmath30 , @xmath31 , and @xmath32 appearing in equation ( [ eq2.3 ] ) represent the rates per unit volume at which radiation energy escapes through the column walls , or is absorbed or emitted by the gas , respectively . when the system is radiation - dominated as assumed here , the gas itself has little thermal energy to contribute to the radiation field , although it does possess a large reservoir of _ bulk _ kinetic energy which is transferred to the photons through the first - order fermi energization process in the shock . in this situation , the emission and absorption processes must nearly balance throughout the flow , and therefore we can write @xmath33 in which case equation ( [ eq2.3 ] ) reduces in a steady state to @xmath34 although emission and absorption do not effect the flow dynamics in this situation , these processes profoundly influence the _ shape _ of the radiation spectrum , as discussed in 3 and 4 within the context of our one - dimensional picture , the rate at which radiation energy diffuses through the walls of the column can be approximated using an escape - probability formalism . if the column cross section is optically thick to electron scattering , as expected , then the rate at which radiation energy escapes through the walls is given by @xmath35 where the mean escape time , @xmath36 , is given by @xmath37 with @xmath38 denoting the diffusion velocity perpendicular to the @xmath25-axis , and @xmath39 representing the perpendicular optical thickness of the cylindrical accretion column with radius @xmath40 . note that @xmath39 and @xmath36 are functions of @xmath25 through their dependence on @xmath41 . becker ( 1998 ) confirmed that the diffusion approximation employed in equation ( [ eq2.10 ] ) is valid because @xmath42 for typical x - ray pulsar parameters . equations ( [ eq2.4 ] ) and ( [ eq2.11 ] ) can be combined to express the mean escape time as @xmath43 since the escape timescale is inversely proportional to the flow velocity , the column becomes completely opaque at the surface of the neutron star due to the divergence of the electron number density there . the relationship between the escape - probability approximation and the physical distribution of radiation inside the accretion column is further discussed in 6.3 . in the steady - state situation of interest here , the mass flux @xmath44 and the momentum flux @xmath45 are both conserved , but the energy flux @xmath46 decreases as the gas accretes onto the surface of the neutron star due to the emission of radiation through the column walls . the momentum , energy , and mass conservation equations can be combined in this case to show that the flow velocity @xmath16 satisfies the second - order nonlinear differential equation ( becker 1998 ) @xmath47 where @xmath48 here , @xmath49 represents the flow velocity at the sonic point , where we set @xmath50 and @xmath51 , with the radiation mach number @xmath52 defined by @xmath53 the quantity @xmath54 denotes the speed of sound in the radiation - dominated gas . in the far upstream region , we assume that @xmath55 which is an excellent approximation in pulsar accretion flows ( basko & sunyaev 1975 , 1976 ) . becker ( 1998 ) showed that in order for the flow to come to rest at the stellar surface , the parameters @xmath40 , @xmath44 , @xmath12 , and @xmath11 must satisfy the relation @xmath56 the corresponding exact solution for the velocity @xmath16 as a function of @xmath57 in the steady - state situation of interest here is given by @xmath58\right\ } \ , \label{eq2.17}\ ] ] where @xmath59 note that @xmath60 at the sonic point ( @xmath61 ) as required . the velocity can also be written as an explicit function of the position @xmath25 using ( basko & sunyaev 1976 ) @xmath62 \ , \label{eq2.19}\ ] ] where @xmath63 is the distance between the sonic point and the stellar surface , which can be evaluated using equation ( 4.16 ) from becker ( 1998 ) to obtain @xmath64 note that @xmath25 increases in the direction of the flow , which comes to rest at the surface of the star as expected . the height above the stellar surface @xmath65 is related to the coordinate @xmath25 via @xmath66 the incident ( upstream ) velocity of the infalling material is expected to be close to the free - fall velocity onto the stellar surface , @xmath67 , given by @xmath68 since @xmath69 in the upstream region according to equation ( [ eq2.19 ] ) , it follows that @xmath70 , and therefore the flow velocity at the sonic point is given by @xmath71 with the velocity solution given by equation ( [ eq2.19 ] ) , the corresponding pressure profile can be obtained by combining equations ( [ eq2.4 ] ) and ( [ eq2.5 ] ) , which yields @xmath72 the pressure increases to a maximum value at the surface of the star , where @xmath73 . based on equations ( [ eq2.4 ] ) and ( [ eq2.15 ] ) , we can reexpress the mach number as @xmath74 since the mach number approaches infinity in the far upstream region and @xmath69 , it follows that @xmath75 must vanish asymptotically and consequently the upstream momentum flux is dominated by the ram pressure of the freely falling matter . we therefore obtain @xmath76 by combining equations ( [ eq2.23 ] ) , ( [ eq2.24 ] ) and ( [ eq2.25 ] ) , we find that the pressure profile is related to the velocity variation by @xmath77 \ . \label{eq2.26}\ ] ] using equation ( [ eq2.19 ] ) to eliminate the velocity @xmath78 yields the closed - form result @xmath79 the pressure vanishes in the upstream limit ( @xmath80 ) as expected , and at the surface of the star ( @xmath81 ) , the pressure achieves the stagnation value @xmath82 if the gas is radiation - dominated and fully ionized , then the photons interact with the matter primarily via electron scattering , which controls both the spatial transport and the energization of the radiation . in this situation the opacity is dominated by electron scattering , and therefore absorption is negligible , except at the surface of the thermal mound located inside the cylindrical accretion column at @xmath83 . we are interested in obtaining the photon distribution , @xmath84 , measured at position @xmath25 and energy @xmath85 resulting from the reprocessing of blackbody radiation emitted from the mound . the normalization of @xmath86 is defined so that @xmath87 gives the number density of photons in the energy range between @xmath85 and @xmath88 , and therefore @xmath89 , where @xmath90 is the occupation number . the self - consistency of the solution obtained for the radiation distribution @xmath86 will be confirmed by verifying that the associated pressure distribution agrees with the dynamical result given by equation ( [ eq2.27 ] ) . in the cylindrical geometry employed here , the photon distribution @xmath86 satisfies the transport equation ( e.g. , blandford & payne 1981a ; becker 1992 ; parker 1965 ; becker & wolff 2005 ) @xmath91 where @xmath85 is the photon energy , @xmath25 is the spatial coordinate along the column axis , and @xmath92 is the flow velocity at the top of the thermal mound . the terms on the right - hand side of equation ( [ eq3.1 ] ) represent advection , first - order fermi energization ( `` bulk comptonization '' ) in the converging flow , spatial diffusion parallel to the column axis , the blackbody source , escape of radiation from the column , and the absorption of radiation at the thermal mound , respectively . we are interested here in the steady - state version of equation ( [ eq3.1 ] ) with @xmath93 . the photon number and energy densities associated with the distribution @xmath86 are given respectively by @xmath94 in the present application , we are primarily interested in exploring the effect of the dynamics itself ( i.e. , the radiative shock ) on the observed x - ray emission . the transport equation ( [ eq3.1 ] ) therefore does not include additional processes such as thermal comptonization or cyclotron emission and absorption that are likely to be important in x - ray pulsars . in general , the neglect of thermal comptonization is reasonable because at the temperatures typical of x - ray pulsars ( @xmath95k ) , the kinetic energy associated with the bulk flow far surpasses the thermal energy . however , the failure to include the electron recoil associated with thermal comptonization renders the model unable to reproduce the quasi - exponential cutoffs often observed in x - ray pulsar spectra at high energies . furthermore , in some sources thermal comptonization appears to be necessary in order to flatten the spectrum in the energy range @xmath96kev . we provide further discussion of this issue in 7 . when integrated over the photon energy spectrum , the first - order fermi process considered here corresponds to the @xmath97 work done on the radiation by the compression of the background plasma as it accretes onto the stellar surface . the mean escape time , @xmath36 , is given by equation ( [ eq2.12 ] ) , and the source term @xmath98 is defined so that @xmath99 represents the number of photons injected into the accretion column per second in the energy range between @xmath85 and @xmath100 from the blackbody surface at @xmath83 . the absorption term in equation ( [ eq3.1 ] ) containing the dimensionless constant @xmath101 must be included due to the blackbody nature of the thermal mound , which acts as both a source and a sink of radiation . in the radiation - dominated situation considered here , essentially all of the pressure is provided by the photons , and therefore the rate of absorption of radiation energy at the top of the mound must equal the energy emission rate when integrated with respect to the photon frequency ( see eq . [ [ eq2.8 ] ] ) . we will use this energy balance argument to compute a self - consistent value for @xmath101 below . next we need to specify the appropriate form for the source term @xmath98 in order to treat the injection of the blackbody radiation from the surface of the thermal mound . the flux per unit energy measured at the surface of an isotropically radiating object is equal to @xmath102 times the intensity ( rybicki & lightman 1979 ) . it therefore follows that the amount of energy emitted per second from the upper surface of the thermal mound ( with area @xmath103 ) in the energy range between @xmath85 and @xmath100 is given by @xmath104 where @xmath105 denotes the blackbody intensity and @xmath106 is the gas temperature at the surface of the mound . note that the units for @xmath107 are @xmath108 . hence we obtain for the source term @xmath109 and consequently the total energy injection rate is given by @xmath110 where @xmath111 is the stephan - boltzmann constant . in a radiation - dominated x - ray pulsar accretion column , the emission of fresh blackbody radiation energy at the surface of the thermal mound is almost perfectly balanced by the absorption of energy , as expressed by equation ( [ eq2.8 ] ) . most of the energy appearing in the emergent x - rays is therefore not supplied from the internal energy of the gas , but rather from its bulk kinetic energy , which is transferred directly to the photons via collisions with infalling electrons . we can gain some insight into the energy balance at the thermal mound by rewriting the steady - state version of the transport equation ( [ eq3.1 ] ) in the flux - conservation form ( skilling 1975 ; gleeson & axford 1967 ) @xmath112 where @xmath113 represents the `` specific flux '' ( becker 1992 ) . by operating on equation ( [ eq3.7 ] ) with @xmath114 and utilizing equations ( [ eq3.2 ] ) and ( [ eq3.6 ] ) , we arrive at the photon energy equation @xmath115 where the integrated photon energy flux , @xmath116 , is given by @xmath117 equation ( [ eq3.9 ] ) states that the divergence of the radiation energy flux @xmath116 is equal to the net rate of change of the photon energy density @xmath17 due to the combined influence of compression , escape , absorption , and photon injection . note that @xmath116 is related to the _ total _ energy flux @xmath46 via ( see eq . [ [ eq2.6 ] ] ) @xmath118 since the absorption and emission of radiation energy must balance at the surface of the mound in the radiation - dominated situation of interest here , we can integrate equation ( [ eq3.9 ] ) with respect to @xmath25 across the mound location to obtain @xmath119 where @xmath120 denotes the radiation energy density at the mound surface . the fact that the photon energy flux @xmath116 is continuous at @xmath83 therefore requires that the dimensionless absorption constant @xmath101 in the transport equation ( [ eq3.1 ] ) be given by @xmath121 by utilizing this expression for @xmath101 , we ensure that the results obtained for the radiation spectrum @xmath84 are consistent with the dynamical structure of the accretion column . this is confirmed after the fact by calculating the radiation pressure @xmath122 using equation ( [ eq3.2 ] ) and comparing the result with the dynamical solution given by equation ( [ eq2.27 ] ) . in our approach to solving equation ( [ eq3.1 ] ) for the spectrum @xmath84 inside the accretion column , we shall first obtain the green s function , @xmath123 , which is the radiation distribution at location @xmath25 and energy @xmath85 resulting from the injection of @xmath124 photons per second with energy @xmath125 from a monochromatic source at location @xmath126 . the determination of the green s function is a useful intermediate step in the process because it provides us with fundamental physical insight into the spectral redistribution process , and it also allows us to calculate the particular solution for the spectrum @xmath86 associated with an arbitrary continuum source @xmath98 using the integral convolution ( becker 2003 ) @xmath127 note that the upper bound of integration is set equal to @xmath85 because in our model all of the injected photons gain energy via the first - order fermi process , and none lose energy . the technical procedure used to solve for the green s function involves the derivation of eigenfunctions and associated eigenvalues based on the set of spatial boundary conditions for the problem ( see , e.g. , blandford & payne 1981b ; payne & blandford 1981 ; schneider & kirk 1987 ; colpi 1988 ) . the steady - state transport equation governing the green s function is ( cf . [ [ eq3.1 ] ] ) @xmath128 and the associated radiation number and energy densities are given by ( cf . [ [ eq3.2 ] ] ) @xmath129 further simplification of the mathematical derivation is possible if we work in terms of the new spatial variable @xmath130 , defined by @xmath131 where @xmath63 is given by equation ( [ eq2.20 ] ) . note that @xmath132 in the far upstream region ( @xmath80 ) , and @xmath133 at the surface of the star ( @xmath134 ) . based on equations ( [ eq2.19 ] ) and ( [ eq4.4 ] ) , we find that the variation of the velocity @xmath16 as a function of the new variable @xmath130 is given by the simple expression @xmath135 likewise , we can also combine equations ( [ eq2.27 ] ) and ( [ eq4.4 ] ) to show that the exact dynamical solution for the radiation pressure profile as a function of @xmath130 is given by @xmath136 note that in the limit @xmath133 , the pressure approaches the stagnation value @xmath137 in agreement with equation ( [ eq2.28 ] ) . equations ( [ eq2.27 ] ) and ( [ eq4.6 ] ) will prove useful when we seek to confirm the validity of the results obtained for the photon distribution @xmath84 in 5 . utilizing equations ( [ eq2.4 ] ) , ( [ eq2.12 ] ) , ( [ eq2.16 ] ) , and ( [ eq4.5 ] ) along with the differential relation @xmath138 we find that equation ( [ eq4.2 ] ) can be transformed from @xmath25 to @xmath130 to obtain @xmath139 where we have made the definition @xmath140 so that @xmath141 denotes the value of @xmath130 at the top of the thermal mound ( see eq . [ [ eq4.4 ] ] ) . according to equation ( [ eq4.5 ] ) , the flow velocity at the top of the mound , @xmath142 , is related to @xmath141 via @xmath143 note that we can now write the green s function as either @xmath144 or @xmath145 since @xmath146 and @xmath147 are interchangeable via equations ( [ eq4.4 ] ) and ( [ eq4.9 ] ) . when @xmath148 , equation ( [ eq4.8 ] ) is separable in energy and space using the functions @xmath149 where @xmath150 is the separation constant , and the spatial function @xmath151 satisfies the differential equation @xmath152 in order to avoid an infinite spatial diffusion flux at @xmath153 , the function @xmath151 must be continuous there , and consequently we obtain the condition @xmath154 we can also derive a jump condition for the derivative @xmath155 at the top of the mound by integrating equation ( [ eq4.12 ] ) with respect to @xmath130 in a very small region around @xmath153 . the result obtained is @xmath156 where we have also used equation ( [ eq4.10 ] ) . the spatial eigenfunctions for our problem are denoted by @xmath157 where @xmath158 represents the @xmath159th eigenvalue . these functions satisfy the continuity and derivative jump conditions given by equations ( [ eq4.13 ] ) and ( [ eq4.14 ] ) , respectively , as well as the boundary conditions discussed below . the homogeneous version of equation ( [ eq4.12 ] ) for @xmath151 obtained when @xmath160 has fundamental solutions given by @xmath161 @xmath162 where @xmath163 denotes the hypergeometric function ( abramowitz & stegun 1970 ) , and the parameters @xmath164 , @xmath165 , and @xmath166 are defined by @xmath167 the `` seed '' photons injected into the flow from the thermal mound located near the base of the column are unable to diffuse very far up into the accreting gas due to the extremely high speed of the inflow . most of the photons therefore escape through the walls of the column within a few scattering lengths of the mound , forming a `` fan '' type beam pattern , as expected for accretion - powered x - ray pulsars ( e.g. , harding 1994 , 2003 ) . based on this physical picture , we conclude that the eigenfunction @xmath168 must _ vanish _ in the upstream limit , @xmath169 . conversely , in the downstream limit , @xmath133 , the gas settles onto the surface of the star and the radiation pressure achieves the stagnation value given by equation ( [ eq2.28 ] ) . since this is a finite pressure , we conclude that @xmath168 must approach a constant as @xmath133 . analysis of the asymptotic behaviors of the hypergeometric functions @xmath170 and @xmath171 shows that the first function vanishes and the second diverges as @xmath169 . hence in the region upstream from the thermal mound ( @xmath172 ) , the eigenfunction @xmath168 must be given by @xmath170 . however , in the downstream region ( @xmath173 ) , the situation is more complicated because the two functions @xmath170 and @xmath174 each diverge logarithmically as @xmath133 . we must therefore utilize a suitable linear combination of these functions in order to obtain a convergent solution in the downstream region . consequently we define the new function ( see appendix a for details ) @xmath175 which remains finite in the limit @xmath133 as required . by utilizing the various relations derived above , we find that the global solution for the spatial eigenfunction @xmath176 ( eq . [ [ eq4.15 ] ] ) can now be written as @xmath177 where the constant @xmath178 is determined using the continuity condition ( eq . [ [ eq4.13 ] ] ) , which yields @xmath179 we can combine equations ( [ eq4.14 ] ) , ( [ eq4.20 ] ) , and ( [ eq4.21 ] ) to show that the eigenvalue equation for @xmath150 is given by @xmath180 the left - hand side of this expression can be evaluated using the exact solution for the wronskian ( see appendix b ) , @xmath181 which is applicable for arbitrary values of @xmath150 and @xmath130 . equations ( [ eq4.22 ] ) and ( [ eq4.23 ] ) can be combined to obtain the alternative form for the eigenvalue equation , @xmath182 where @xmath164 , @xmath165 , and @xmath166 are functions of @xmath150 given by equations ( [ eq4.18 ] ) . the roots of this expression are the eigenvalues @xmath183 , and the associated eigenfunctions are evaluated using equation ( [ eq4.20 ] ) . the first eigenvalue , @xmath184 , is especially important because it determines the slope of the high - energy spectrum emerging from the accretion column according to equation ( [ eq4.11 ] ) . the spectral index of the emitted photon count rate distribution , @xmath185 , is related to @xmath184 via @xmath186 . in figure 2 we plot the photon index @xmath185 as a function of the dimensionless parameters @xmath101 and @xmath141 . note that @xmath185 is a double - valued function of @xmath141 for fixed @xmath101 , which is a consequence of the imposed velocity profile ( eq . [ [ eq4.5 ] ] ) . physically , this behavior reflects the fact that it is always possible to achieve a desired amount of compression ( first - order fermi energization ) by placing the source in a specific location in either the upstream or downstream region of the flow . we also observe that if we increase the absorption parameter @xmath101 while holding @xmath141 fixed , then @xmath185 increases monotonically , and therefore the high - energy spectrum becomes progressively steeper . this behavior is expected physically because as the absorption parameter is increased , the injected photons spend less time on average being energized by collisions with electrons before either escaping from the column or being absorbed at the source location . the reduction in the amount of energization naturally leads to a steepening of the radiation spectrum . when @xmath187 , no absorption occurs , and the index @xmath185 achieves it minimum ( limiting ) value of 2 . this limit is unphysical , however , since it yields a divergent result for the total photon energy density @xmath188 according to equation ( [ eq4.3 ] ) . we demonstrate in appendix c that the eigenfunctions @xmath176 form an orthogonal set , as expected since this is a standard sturm - liouville problem . once the eigenvalues and eigenfunctions are determined , the solution for the green s function can therefore be expressed as the infinite series @xmath189 where the expansion coefficients @xmath190 are computed by employing the orthogonality of the eigenfunctions along with the condition @xmath191 which is obtained by integrating the transport equation ( [ eq4.8 ] ) with respect to @xmath85 in a very small range surrounding the injection energy @xmath125 . the result obtained for the @xmath159th expansion coefficient is @xmath192 where the quadratic normalization integrals @xmath193 are defined by @xmath194 in appendix d , we show that the normalization integrals can be evaluated using the closed - form expression @xmath195 where @xmath196 \ , \label{eq4.30}\ ] ] and @xmath197 this provides an extremely efficient alternative to numerical integration for the computation of @xmath193 . the eigenfunction expansion converges fairly rapidly and therefore we are able to obtain an accuracy of at least five significant figures in our calculations of @xmath198 by terminating the series in equation ( [ eq4.25 ] ) after the first 20 terms . equation ( [ eq4.25 ] ) represents the exact solution for the green s function describing the radiation spectrum inside a pulsar accretion column resulting from the injection of @xmath124 seed photons per unit time from a monochromatic source located at @xmath153 ( or , equivalently , at @xmath83 ) . in the escape probability approach employed here , the associated green s function for the _ photon spectrum _ emitted through the walls of the cylindrical column is defined by @xmath199 so that @xmath200 represents the number of photons escaping from the column per unit time between positions @xmath25 and @xmath201 with energy between @xmath85 and @xmath100 . we remind the reader that the quantities @xmath146 and @xmath147 are interchangeable via equations ( [ eq4.4 ] ) and ( [ eq4.9 ] ) , and therefore we are free to work in terms of the more convenient parameters @xmath147 without loss of generality . by substituting for @xmath36 using equation ( [ eq2.12 ] ) we can obtain the alternative expression @xmath202 eliminating @xmath44 using equation ( [ eq2.16 ] ) and substituting for @xmath16 and @xmath198 using equations ( [ eq4.5 ] ) and ( [ eq4.25 ] ) , respectively , yields the equivalent result @xmath203 where the expansion coefficients @xmath204 are defined by @xmath205 the green s function @xmath206 describes the photon distribution escaping from the accretion column as a function of energy @xmath85 and location @xmath130 for the case of monoenergetic photon injection . analysis of @xmath206 therefore reveals some interesting details about the energization of the photons as they are transported through the column via diffusion and advection , and ultimately escape through the column walls . we plot @xmath206 as a function of the energy ratio @xmath207 and the location @xmath130 in figure 3 for the parameter values @xmath208 and @xmath209 . in this case the first eigenvalue is given by @xmath210 , and the corresponding photon index is @xmath211 ( see fig . 2 ) . the selected value of @xmath141 corresponds to a source located near the bottom of the accretion column , just above the stellar surface . at the source location , @xmath212 , the energy spectrum extends down to the injection energy , @xmath125 . however , at all other radii the spectrum displays a steep turnover above that energy because all of the photons at these locations have experienced fermi energization due to collisions with the infalling electrons . the photons with energy @xmath213 at the source location have been injected so recently that they have not yet experienced significant energization . note that for small values of @xmath130 , the escaping spectrum is greatly attenuated due to the inability of the photons to diffuse upstream through the rapidly infalling plasma . the escaping spectrum is also attenuated in the far downstream region ( @xmath214 ) due to the very high opacity of the flow , which inhibits the diffusion of photons through the walls . this latter effect stems from the divergence of the electron density near the stellar surface . the energy of maximum brightness , corresponding to the peak in @xmath206 , achieves its maximum value as @xmath169 because the photons that manage to diffuse far upstream from the source are the ones that have resided in the flow the longest and therefore experienced the most energy amplification . we plot @xmath206 as a function of energy and location in figure 4 for the parameters @xmath215 and @xmath216 , in which case we obtain for the first eigenvalue and the photon index @xmath217 and @xmath218 , respectively . in this scenario the source is located in the upstream region and the absorption is stronger than that in figure 3 . due to the increased absorption resulting from the larger value of @xmath101 , the photons spend less time on average in the flow being energized by collisions with the electrons before they escape through the column walls or are `` recycled '' by absorption . this causes a steepening of the spectrum at high energies , as evidenced by the increase in the photon index @xmath185 . in this situation the energy of maximum brightness ( where @xmath206 displays a peak ) achieves its greatest value in the downstream region . this is the reverse of the behavior displayed in figure 3 because in the present example , the source is located in the upstream region and therefore the photons that diffuse further upstream towards @xmath219 do not experience as much compression as those that are advected downstream . the escaping photon distribution in the far upstream and downstream regions is strongly attenuated due to the same processes operative in figure 3 , and therefore most of the escaping radiation is emitted from the accretion column around the source location . the analytical results for the green s function obtained in this section provide the basis for the consideration of any source distribution since the fundamental differential equation ( [ eq3.1 ] ) is linear . this is further discussed in 5 , where we use equations ( [ eq3.5 ] ) and ( [ eq4.1 ] ) to convolve the green s function with the blackbody spectrum produced by the thermal mound . by integrating over the vertical structure of the accretion column , we can compute the total emitted radiation distribution , which corresponds to the phase - averaged spectrum of the x - ray pulsar . in the cylindrical geometry employed here , the formal integration domain is the region @xmath220 , where @xmath63 is the distance between the sonic point and the stellar surface ( see eq . [ [ eq2.20 ] ] ) . for the case of a monochromatic source , we define the _ column - integrated green s function _ by writing @xmath221 where @xmath222 represents the number of photons escaping from the column per unit time with energy between @xmath85 and @xmath100 . using equation ( [ eq4.7 ] ) , the variable of integration can be transformed from @xmath25 to @xmath130 to obtain the alternative form @xmath223 where @xmath141 is related to @xmath126 via equation ( [ eq4.9 ] ) . despite the appearance of the factor @xmath224 inside the integral in equation ( [ eq4.37 ] ) , the contribution from small values of @xmath130 is actually negligible because the spectrum declines exponentially in the upstream region due to advection . furthermore , the escaping spectrum is also strongly attenuated in the downstream region due to the divergence of the electron density . hence most of the radiation is emitted from the column around @xmath225 , as indicated in figures 3 and 4 . by combining equations ( [ eq4.34 ] ) , ( [ eq4.35 ] ) , and ( [ eq4.37 ] ) , we can reexpress the column - integrated green s function as @xmath226 where the quadratic normalization integrals @xmath193 are evaluated using equation ( [ eq4.29 ] ) and we have made the definition @xmath227 in appendix e , we demonstrate that the integral in the expression for @xmath228 can be carried out analytically to obtain @xmath229 + { \varphi_1^*(\lambda_n , y_0 ) \over ( a-1 ) \ , ( b-1 ) \ , \gamma(1-c ) } \bigg [ f(a , b\,;c-1;y_0 ) \nonumber \\ & + & { ( c-2 ) \over ( a-2 ) \ , ( b-2 ) } \ , f(a , b;c-2;y_0)\bigg ] \bigg\ } + { ( 1-c ) \over ( a-1 ) \ , ( b-1 ) } \left[1 + { ( c-2 ) \over ( a-2 ) \ , ( b-2)}\right ] \ , \qquad \label{eq4.40}\end{aligned}\ ] ] where @xmath230 , @xmath231 , and @xmath232 are given by equations ( [ eq4.16 ] ) , ( [ eq4.17 ] ) , and ( [ eq4.19 ] ) , respectively , and the constants @xmath164 , @xmath165 , and @xmath166 are computed in terms of @xmath158 using equations ( [ eq4.18 ] ) . in figure 5 we plot the dependence of the column - integrated green s function @xmath233 on the radiation energy @xmath85 using the same parameters employed in figures 3 and 4 . note that when @xmath234 and @xmath209 , which corresponds to the thin line in figure 5 , the column - integrated spectrum displays a peak at @xmath235 . the peak forms because the source is located close to the bottom of the accretion column , whereas most of the photons escape from higher altitudes around @xmath225 ( see fig . the escaping radiation has therefore experienced significant compression in the flow and consequently the photon energies are generally boosted well above the injection energy @xmath125 . on the other hand , when @xmath215 and @xmath216 , which corresponds to the thick line in figure 5 , the absorption is so strong that the spectrum is quite steep and therefore the peak is suppressed . consequently the column - integrated spectrum achieves it maximum value at the injection energy in this case . the closed - form solutions for the green s function @xmath206 ( eq . [ [ eq4.34 ] ] ) and for the column - integrated green s function @xmath233 ( eq . [ [ eq4.38 ] ] ) provide a very efficient means for computing the x - ray spectrum escaping through the walls of the accretion column due to the injection of monochromatic seed photons . however , in our physical application to x - ray pulsars , we are primarily interested in computing the emitted spectrum resulting from the reprocessing of _ blackbody radiation _ injected into the accretion column from the thermal mound . in the escape probability approach utilized here , the photon distribution emitted through the walls of the cylindrical column at position @xmath25 , denoted by @xmath236 , is computed using @xmath237 where the particular solution @xmath84 is evaluated using equation ( [ eq4.1 ] ) , and @xmath238 represents the number of photons escaping from the column per unit time between positions @xmath25 and @xmath201 with energy between @xmath85 and @xmath100 . by combining equations ( [ eq4.1 ] ) and ( [ eq4.41 ] ) , we find that the escaping photon spectrum can be expressed as @xmath239 where the blackbody source function @xmath240 is evaluated using equation ( [ eq3.5 ] ) . the spatial variables @xmath146 and @xmath147 are interchangeable by virtue of equations ( [ eq4.4 ] ) and ( [ eq4.9 ] ) , and therefore we can use equations ( [ eq4.32 ] ) and ( [ eq4.42 ] ) to write @xmath241 where @xmath242 is computed using equation ( [ eq4.34 ] ) . by analogy with equation ( [ eq4.36 ] ) for the green s function , we define the particular solution for the _ column - integrated _ photon spectrum escaping from the plasma due to the blackbody source , @xmath243 , by writing @xmath244 so that @xmath245 gives the number of photons escaping from the column per unit time with energy between @xmath85 and @xmath246 . the variable of integration can be transformed from @xmath25 to @xmath130 using equation ( [ eq4.7 ] ) , which yields @xmath247 where @xmath141 is related to @xmath126 by equation ( [ eq4.9 ] ) . by substituting for @xmath248 using equation ( [ eq4.43 ] ) and employing equation ( [ eq4.37 ] ) , we can obtain the alternative form @xmath249 the column - integrated green s function @xmath250 is evaluated using equation ( [ eq4.38 ] ) . in 5 we will utilize equations ( [ eq4.43 ] ) and ( [ eq4.46 ] ) to compute the photon spectrum emitted from the accretion column using parameters corresponding to specific x - ray pulsars . for given values of the stellar mass @xmath251 and the stellar radius @xmath252 , our model has three free parameters , namely the column radius @xmath40 , the temperature at the top of the thermal mound @xmath253 , and the accretion rate @xmath254 . in this section we investigate the relationship between these quantities and the dimensionless parameters @xmath141 and @xmath101 appearing in the theory . although it is not our intention in this paper to develop complete models for x - ray pulsar spectra , it is nonetheless interesting to compare the simplified model developed here with actual data for a few sources . hence in this section we will also compute the spectrum emerging from the accretion column for parameters corresponding to two specific x - ray pulsars . in order to understand how the theory developed here can be related to observables such as the temperature and the luminosity , we first need to determine how the position of the thermal mound , @xmath126 ( or , equivalently , @xmath141 ) , depends on the values of @xmath40 , @xmath253 , and @xmath255 . the blackbody surface of the mound represents the photosphere for photon creation and destruction in this problem , and therefore the opacity is dominated by free - free absorption inside the mound . for a given value of the temperature @xmath253 , the density at the top of the mound , @xmath256 , can be calculated by setting the free - free optical thickness of the column equal to unity , so that @xmath257 where the rosseland mean of the free - free absorption coefficient is evaluated in cgs units using ( rybicki & lightman 1979 ) @xmath258 for pure , fully - ionized hydrogen with the gaunt factor set equal to unity . the density at the top of the thermal mound is therefore given by @xmath259 the velocity at the top of the mound , @xmath142 , is related to @xmath256 via the continuity equation @xmath260 , so that @xmath261 by combining this result with equations ( [ eq2.23 ] ) and ( [ eq4.10 ] ) , we find that the value of @xmath141 is given by @xmath262 or , equivalently , @xmath263 the values obtained for @xmath141 are extremely close to unity , indicating that the top of the thermal mound is just above the stellar surface . once @xmath141 is computed , we can determine @xmath126 by using equation ( [ eq4.9 ] ) to write @xmath264 where @xmath63 is the distance between the sonic point and the stellar surface given by equation ( [ eq2.20 ] ) . the height of the thermal mound above the stellar surface is then given by ( see eq . [ [ eq2.21 ] ] ) @xmath265 or , equivalently , @xmath266 where we have substituted for @xmath63 using equation ( [ eq2.20 ] ) . specific numerical examples are given in 5.4 . for typical x - ray pulsar parameters , we find that the mound is situated very close to the surface of the star , as expected . in order to compute the eigenvalues @xmath158 using equation ( [ eq4.24 ] ) , we must also evaluate the dimensionless absorption parameter @xmath101 appearing in the transport equation ( [ eq3.1 ] ) . based on energy conservation considerations in the radiation - dominated plasma , we have shown in 3.2 that the photon energy flux @xmath116 must be continuous across the location of the thermal mound . this leads to equation ( [ eq3.13 ] ) , which can be rewritten as @xmath267 where @xmath268 is the radiation pressure at the top of the thermal mound . according to equation ( [ eq4.6 ] ) , @xmath269 , and therefore @xmath270 by substituting for @xmath49 and @xmath142 using equations ( [ eq2.23 ] ) and ( [ eq4.10 ] ) , respectively , and setting @xmath271 , we can obtain the alternative form @xmath272 or , equivalently , @xmath273 taken together , equations ( [ eq5.6 ] ) and ( [ eq5.13 ] ) allow the determination of the two dimensionless model parameters @xmath141 and @xmath101 in terms of @xmath252 , @xmath251 , @xmath40 , @xmath253 , and @xmath254 . this closes the system and facilitates the calculation of the emergent photon number spectrum @xmath248 using equation ( [ eq4.43 ] ) , and the calculation of the column - integrated spectrum @xmath274 using equation ( [ eq4.46 ] ) . the values of the parameters @xmath40 , @xmath253 , and @xmath254 can therefore be extracted by comparing the theoretically predicted photon count - rate distributions with x - ray pulsar spectral data . the eigenvalues @xmath158 are functions of the dimensionless parameters @xmath141 and @xmath101 . based on the dependences of these parameters on @xmath40 , @xmath253 , and @xmath254 expressed by equations ( [ eq5.6 ] ) and ( [ eq5.13 ] ) , we note that the eigenspectrum _ remains unchanged _ if @xmath40 , @xmath253 , and @xmath254 are simultaneously varied according to the prescription @xmath275 in this case , the column - integrated green s function ( eq . [ [ eq4.38 ] ] ) also remains unchanged since it depends only on the values of @xmath101 and @xmath141 . these findings suggest that it is useful to introduce the new `` similarity variables '' @xmath276 and @xmath277 , defined by @xmath278 in terms of these parameters , equations ( [ eq5.6 ] ) and ( [ eq5.13 ] ) for @xmath141 and @xmath101 now become @xmath279 and @xmath280 the introduction of the new variables @xmath276 and @xmath277 is convenient because it reduces the dimensionality of the parameter space that determines the eigenvalues @xmath158 from the three - dimensional domain @xmath281 to the two - dimensional space @xmath282 . in figure 6 we plot the spectral index @xmath186 of the photon count - rate spectrum as a function of @xmath276 and @xmath277 , where @xmath184 is the first eigenvalue and we have set @xmath283 km and @xmath284 ( the same quantity was plotted as a function of @xmath141 and @xmath101 in figure 2 ) . note that when @xmath277 is increased for a fixed value of @xmath276 , the values of @xmath141 and @xmath101 both increase according to equations ( [ eq5.16 ] ) and ( [ eq5.17 ] ) . the source therefore moves further downstream , and the absorption at the source location becomes stronger . this causes the spectrum to steepen as we move upwards along a vertical line in figure 6 . however , if @xmath277 becomes very large , the spectrum starts to harden again due to the increasing amount of compression experienced as the source approaches the base of the accretion column . our primary goal in this paper is to explore the effect of bulk comptonization on the shape of the x - ray continuum spectrum emerging from a pulsar accretion column . the development of complete models that include additional effects such as thermal comptonization , cyclotron processes , free - free emission and absorption , and iron line formation will be pursued in future work . here we compare spectra computed using our simplified model with the observations for a few sources in order to focus attention on the role of the accretion shock in the formation of the x - ray continuum . in figures 7 and 8 we plot the position - dependent photon count rate spectra @xmath248 ( eq . [ [ eq4.43 ] ] ) escaping from the accretion column computed using the two sets of model parameters listed in table 1 . these results were obtained using 20 eigenvalues and eigenfunctions , which yields extremely high precision due to the rapid convergence of the expansion . the theoretical spectra describe the upscattering of planckian seed photons injected into the base of the accretion column from the surface of the thermal mound . in both cases we use @xmath285 and @xmath283 km for the stellar mass and radius , respectively . for model 1 ( fig . 7 ) , the parameters adopted are @xmath286 km , @xmath287 , and @xmath288 . in model 2 ( fig . 8) we set @xmath289 km , @xmath290k , and @xmath291 . we use equations ( [ eq5.15 ] ) , ( [ eq5.16 ] ) , and ( [ eq5.17 ] ) to compute the theoretical parameters @xmath276 , @xmath277 , @xmath141 and @xmath101 based on the selected values for @xmath40 , @xmath253 , and @xmath254 . in model 1 we obtain @xmath292 and @xmath293 , and in model 2 we find that @xmath294 and @xmath295 . additional parameters are provided in table 1 . the fact that the thermal mound location @xmath141 is extremely close to unity in both cases indicates that the mound lies just above the surface of the star , as expected . it is interesting to compute the actual physical distance between the thermal mound and the stellar surface , @xmath296 , which is given by equation ( [ eq5.9 ] ) . using the parameters corresponding to models 1 and 2 yields @xmath297 cm and @xmath298 cm , respectively . note that in model 2 the thermal mound is essentially on the surface of the star , whereas in model 1 there is a much larger separation . this reflects the fact that the accretion rate in model 1 is over 800 times larger than in model 2 . the results obtained for the first eigenvalue and for the photon index in model 1 are @xmath299 and @xmath300 , respectively . in model 2 , we find that the spectrum is significantly steeper , with @xmath301 and @xmath302 . the steeper spectrum obtained in model 2 results from the large value obtained for @xmath101 , as indicated in table 1 . in general , the escaping spectrum has a power - law shape at high energies , as expected for a fermi process , and a low - energy turnover due to the planck distribution of the seed photons . note that the emitted radiation is concentrated in a layer just above the thermal mound location , around @xmath225 . this reflects the fact that advection due to collisions with the infalling stream of high - speed electrons tends to keep the photons trapped at low altitudes in the column , and therefore few of them are able to escape through the column walls for @xmath303 . since the spin axis and the column axis are not aligned in pulsars , the portion of the accretion column visible along the line of sight to the earth changes as the pulsar rotates . when combined with the vertical variation of the energy dependence of the escaping photon distribution , this effect will produce a pulse - phase dependence in the observed x - ray spectrum . in figures 9 and 10 we plot the photon count rate flux measured at earth , @xmath304 computed using models 1 and 2 , respectively , where @xmath305 is the distance to the source and the column - integrated spectrum @xmath306 is evaluated using equation ( [ eq4.46 ] ) . also included in figures 9 and 10 for comparison are plots of the unfolded phase - averaged spectra for the x - ray pulsars 4u 1258 - 61 ( gx 304 - 1 ) and 4u 0352 + 30 ( x persei ) , respectively . the 4u 1258 - 61 data was reported by white et al . ( 1983 ) , and the 4u 0352 + 30 data is the result of xspec analysis of an archival rxte observation taken in july 1998 and reported by delgado - mart et al . ( 2001 ) . based on estimates from negueruela ( 1998 ) , the values adopted for the distance @xmath305 are 2.5kpc for 4u 1258 - 61 and 0.35kpc for 4u 0352 + 30 . the observational data represent the de - convolved ( incident ) spectra , which depend weakly on the detector response model . the theoretical spectra were computed using 20 eigenvalues and eigenfunctions for high accuracy , and various amounts of interstellar absorption were included as indicated in the figures . although the results presented here are not fits to the data , we note that the general shape of the pulsar spectrum predicted by the theory agrees fairly well with the observations in each case , including both the turnover at low energies and the power law at higher energies . several other sources yield similar agreement . in our model , the turnover at @xmath307kev is due to planckian photons that escape from the accretion column without experiencing many scatterings . this effect will tend to reduce the amount of absorption required to fit the observational data , compared with the amount required using the standard ad hoc models usually employed in x - ray pulsar spectral analysis ( see the discussion in hickox , narayan , & kallman 2004 ) . the second source , x persei , presents an interesting test for the model due to its relatively low luminosity , @xmath308 . the agreement between the theory and the observations suggests that radiation pressure may still be playing an important role in the dynamics of this source , despite its low luminosity . in fact , langer & rappaport ( 1982 ) pointed out that radiation pressure may have a strong effect on the flow dynamics of low - luminosity pulsars if the photon energies are close to the cyclotron resonance , which greatly increases the electron scattering cross section . furthermore , the dynamical importance of the gas pressure in this luminosity range may lead to the development of a gas - mediated `` subshock '' within the overall gradual compression of the radiation - dominated shock . this type of hybrid shock is analogous to the `` two - fluid '' model of cosmic - ray mediated supernova shocks ( e.g. , drury & vlk 1981 ) . it is know that in the cosmic - ray case , multiple flow solutions can occur for the same set of upstream boundary conditions ( e.g. , becker & kazanas 2001 ) . there is some possibility that the same type of multi - state behavior may occur in x - ray pulsars , which could have interesting observational implications . the analytical approach employed in this paper is based on the conservation of energy and momentum in the radiation - dominated flow . this leads to the self - consistent determination of the thermal mound location , @xmath126 , for given values of the fundamental parameters @xmath40 , @xmath253 , and @xmath254 , as discussed in 5.1 . in this section we verify that the model satisfies the self - consistency conditions required for global energy conservation . we also examine the implications of the one - dimensional escape probability formalism that has been utilized to model the emission of radiation from the column . the exact dynamical solution for the radiation pressure inside the pulsar accretion column is given by equation ( [ eq2.27 ] ) , which states that @xmath309 where @xmath63 is the distance between the sonic point and the stellar surface , given by equation ( [ eq2.20 ] ) . in principle , the radiation pressure can also be computed by integrating the particular solution for the photon distribution @xmath86 inside the accretion column . since @xmath24 , we can use equation ( [ eq3.2 ] ) to write @xmath310 where @xmath86 is computed using the convolution ( see eq . [ [ eq4.1 ] ] ) @xmath311 here , the blackbody source @xmath240 is given by equation ( [ eq3.5 ] ) and the green s function @xmath198 is evaluated using equation ( [ eq4.25 ] ) , bearing in mind that the quantities @xmath146 and @xmath147 are interchangeable via equations ( [ eq4.4 ] ) and ( [ eq4.9 ] ) . we can check the accuracy of the entire solution technique , including the determination of the eigenfunctions and the eigenvalues , by comparing the exact dynamical result for @xmath75 with that obtained by integrating the spectrum . in figure 11 we plot the exact analytical solution for the radiation pressure computed using equation ( [ eq5.19 ] ) along with the results obtained by integrating the model 1 and 2 spectra using equation ( [ eq5.20 ] ) . note that the agreement between the various solutions for @xmath75 is excellent , and in each case the pressure achieves the correct stagnation value @xmath312 ( eq . [ [ eq2.28 ] ] ) as the gas approaches the stellar surface . the agreement between the various results for the pressure confirms that our solution for the radiation spectrum is consistent with the dynamics of the accretion flow . in order to further explore the self - consistency of the formalism developed here , we have also computed the total x - ray luminosity , @xmath313 , emerging from the column by integrating the escaping number spectrum @xmath274 ( eq . [ [ eq4.44 ] ] ) using @xmath314 we have confirmed that the values obtained using this expression agree precisely with the accretion luminosity given by @xmath315 the calculations performed in this section and in 6.1 validate our entire solution procedure for the determination of the radiation spectrum by demonstrating that our results are fully consistent with global energy conservation . within the context of our one - dimensional model , the diffusion of radiation through the walls of the accretion column is treated in an approximate manner using an escape - probability formalism . in this approach , the probability per unit time that a randomly - selected photon at location @xmath25 inside the column will escape through the walls is equal to @xmath316 , where the mean escape time @xmath36 is a function of @xmath25 given by ( see eqs . [ [ eq2.11 ] ] ) @xmath317 with @xmath40 and @xmath12 denoting the radius of the cylindrical column and the electron scattering cross section for photons propagating perpendicular to the column axis , respectively . the escape of radiation through the walls of the accretion column can therefore be modeled approximately using a term of the form @xmath318 which is incorporated into the transport equation ( [ eq3.1 ] ) . from a technical point of view , the utilization of the escape - probability formalism implies that the distribution of the photon number density across the column is proportional to the first eigenfunction of the operator describing the diffusion of radiation perpendicular to the column axis ( e.g. , sunyaev & titarchuk 1980 ) . in a comprehensive , multidimensional calculation , the diffusion of photons perpendicular to the column axis is treated by replacing the escape - probability term @xmath319 in the transport equation ( [ eq3.1 ] ) with the more rigorous expression @xmath320 where @xmath321 is the radial coordinate measured from the central axis of the cylindrical accretion column . if the gas density is constant across the column at a given height , as assumed here , then equation ( [ eq5.26 ] ) can be separated to obtain for the first eigenfunction the solution @xmath322 where @xmath323 is a constant , @xmath324 denotes the first eigenvalue , @xmath325 represents the zeroth - order bessel function , and @xmath326 is the electron scattering mean free path for photons propagating perpendicular to the column axis . the value of @xmath324 is determined by satisfying the flux boundary condition at the surface of the column . by employing the eddington approximation , we can write the boundary condition as @xmath327 where @xmath328 is a positive constant of order unity and we have assumed that no radiation is incident on the column from the outside . the precise value of @xmath328 is somewhat arbitrary , with rybicki & lightman ( 1979 ) setting @xmath329 and sunyaev & titarchuk ( 1980 ) setting @xmath330 . combining equations ( [ eq5.27 ] ) and ( [ eq5.29 ] ) , we find that @xmath324 is the smallest positive root of the equation @xmath331 where @xmath332 denotes the order - unity bessel function . the rate of change of the first eigenfunction ( eq . [ [ eq5.27 ] ] ) is given by @xmath333 comparison of equations ( [ eq5.25 ] ) and ( [ eq5.31 ] ) reveals that the escape - probability approximation and the rigorous diffusion model yield the same result for the logarithmic time derivative @xmath334 if the mean escape time @xmath36 and the first eigenvalue @xmath324 are related via @xmath335 this equation is well satisfied in all of our sample calculations , which suggests that the escape - probability formalism provides a reasonable description of the leakage of photons from the accretion column . we have shown that the simplified model considered here , comprising a radiation - dominated accretion column with a blackbody source / sink at its base and a radiative shock , is able to reproduce the power - law continuum spectra observed in accretion - powered x - ray pulsars with a range of luminosities , as indicated in figures 9 and 10 . our results represent the first ab initio calculations of the x - ray spectrum associated with the physical accretion scenario first suggested by davidson ( 1973 ) . for given values of the stellar mass @xmath251 and the stellar radius @xmath252 , our model has only three free parameters , namely the column radius @xmath40 , the temperature at the top of the thermal mound @xmath253 , and the accretion rate @xmath254 . we have confirmed that the solution obtained for the radiation spectrum is consistent with the dynamics of the flow in the accretion column ( see fig . many pulsars are found to have spectra that are well fitted using a combination of a power - law plus a blackbody component , but until now this form has been adopted in a purely ad hoc manner . the new model described here finally provides a firm theoretical foundation for this empirical result . our analytical solution , based on a rigorous eigenfunction expansion , is relatively straightforward to implement computationally because it avoids numerical integration of the transport equation . the work presented here therefore represents a significant step in the development of a comprehensive theory for the spectral formation process in x - ray pulsars . the photon spectral index @xmath185 is related to the first eigenvalue @xmath184 via @xmath336 , and therefore the leading eigenvalue determines the slope of the high - energy portion of the x - ray spectrum . the value of @xmath185 is quite sensitive to the strength of the absorption at the thermal mound , which is determined by the parameter @xmath101 . according to equation ( [ eq5.13 ] ) , @xmath101 in turn is a strong function of the mound temperature @xmath253 . the higher the temperature , the stronger the emission and absorption at the blackbody surface of the mound . model 2 has a higher temperature than model 1 , and therefore the value of @xmath101 is larger . hence the spectrum is steeper in model 2 because the photons spend less time on average upscattering in the flow before either escaping through the column walls or being `` recycled '' by absorption at the mound location . acceptable values for @xmath185 , in the range between 2 and 4 , are obtained for reasonable source parameters , as indicated in figures 2 and 6 . although a broad range of spectral indices can be produced , we must have @xmath337 in order to avoid an infinite photon energy density since the model considered here does not include a high - energy cutoff . the shape of the green s function spectrum derived here is qualitatively similar to the results obtained using the `` bulk motion comptonization '' model coded into the xspec package using the bmc function ( see shrader & titarchuk 1998 , 1999 ; borozdin et al . 1999 ) . the bmc model includes parameters corresponding to the high - energy spectral index , the temperature of the soft photon source , the fraction of the comptonized component in the resulting spectrum , and the size of the emitting region . however , the bmc model is designed to treat spectral formation in black - hole accretion flows , whereas our focus here is on neutron star accretion . hence the model developed in this paper can be viewed as a parallel formalism that is optimized for the treatment of the spectral formation process in accretion - powered x - ray pulsars . in this sense , our theory translates the free parameters of a generalized bulk comptonization model into specific values for the accretion rate , temperature , and column radius for a particular x - ray pulsar . our goal in this paper is to explore the direct role of the accretion shock in producing the observed continuum spectra in x - ray pulsars via the first - order fermi process . although the model produces spectra that are quite similar to those observed from several x - ray pulsars , it can not explain the spectra from other sources . for example , the observed spectra for several of the brightest pulsars with luminosities in the range @xmath338 ( e.g. , her x-1 ) are very flat power - laws leading up to a sharp exponential cutoff . the photon index in these sources is typically @xmath339 , which is outside the range allowed by our model since it does not include a high - energy cutoff . in order to understand spectral formation in the brightest pulsars , additional effects such as thermal comptonization , cyclotron features , bremsstrahlung , and iron emission must be incorporated into the model . in principle , one should also consider higher - order relativistic effects that can alter the form of the fundamental transport equation ( psaltis & lamb 1997 ) . thermal comptonization is particularly important since this process redistributes energy from the highest frequency photons to lower energies via electron recoil . this naturally leads to the development of a high - energy exponential turnover and a concomitant flattening of the spectrum at moderate energies , as observed . the thermal comptonization process was studied by lyubarskii & sunyaev ( 1982 ) and becker ( 1988 ) in the context of plane - parallel shocks . however , the velocity profiles employed by these authors are not consistent with the structure of an x - ray pulsar accretion column . poutanen & gierliski ( 2003 ) computed x - ray pulsar spectra based on the thermal comptonization of soft radiation in a hot layer above the magnetic pole , but their model did not include a complete treatment of the bulk process , which is of crucial importance in x - ray pulsars . in this paper we have carefully analyzed the effect of bulk comptonization on the emitted spectrum , while neglecting the corresponding thermal process . the inclusion of thermal comptonization is beyond the scope of the present paper because it renders the transport equation insoluble via analytical means . however , in future work we intend to develop a comprehensive numerical model that incorporates both the bulk and thermal processes . the authors would like to thank drs . lev titarchuk , philippe laurent , and nikos kylafis who provided useful comments on the manuscript . the authors are also grateful to the anonymous referee whose careful reading and stimulating suggestions led to significant improvements in the paper . pab would also like to acknowledge generous support provided by the office of naval research . in the following sections we provide further details regarding the various mathematical methods that form the foundation of our model . when @xmath160 , the spatial separation functions @xmath340 satisfy the homogeneous differential equation ( cf . eq . [ [ eq4.12 ] ] ) @xmath341 which has the fundamental solutions @xmath342 @xmath343 where @xmath163 denotes the hypergeometric function ( abramowitz & stegun 1970 ) , and we have made the definitions @xmath344 as explained in the discussion following equation ( [ eq4.18 ] ) , in the downstream limit the radiation pressure approaches the finite stagnation value given by equation ( [ eq2.28 ] ) . the spatial separation functions @xmath340 should mimic this behavior , and therefore we require that @xmath151 approaches a constant in the limit @xmath133 . this does not happen in general , but only if the separation constant @xmath150 is equal to one of the eigenvalues @xmath158 , in which case we obtain the eigenfunction @xmath345 the hypergeometric functions appearing in equations ( [ eqa.2 ] ) and ( [ eqa.3 ] ) can be evaluated at @xmath346 using equation ( 15.1.20 ) from abramowitz & stegun ( 1970 ) , which gives for general values of @xmath164 , @xmath165 , and @xmath166 @xmath347 however , for the values of @xmath164 , @xmath165 , and @xmath166 in the current application , we find that @xmath348 and therefore the hypergeometric functions @xmath349 and @xmath350 each _ diverge _ in the downstream limit @xmath133 . based on the asymptotic behaviors of the hypergeometric functions appearing in the expressions for @xmath170 and @xmath174 , we conclude that in the downstream region ( @xmath173 ) , the eigenfunction @xmath168 must be represented by a suitable linear combination of @xmath170 and @xmath174 that remains finite as @xmath133 . in order to make further progress , we need to employ equation ( 15.3.10 ) from abramowitz & stegun ( 1970 ) , which yields for general @xmath164 , @xmath165 , and @xmath130 @xmath351 ( 1-y)^n \ , \label{eqa.8}\end{aligned}\ ] ] where @xmath352 by making use of this expression , we note that the logarithmic divergences of the two functions @xmath170 and @xmath174 in the limit @xmath133 can be balanced by creating the new function @xmath353 which remains finite as @xmath133 . hence @xmath354 represents the fundamental solution for @xmath168 in the region downstream from the thermal mound located at @xmath355 . we can use the asymptotic behaviors of @xmath170 and @xmath174 as @xmath133 to conclude that @xmath356 \over \gamma(a ) \ , \gamma(1-b ) } \ . \label{eqa.11}\ ] ] since the solutions @xmath170 and @xmath354 are applicable in the upstream and downstream regions , respectively , the global expression for the eigenfunction @xmath168 is therefore @xmath357 where the constant @xmath178 is evaluated using the continuity condition @xmath358 it follows from equations ( [ eqa.11 ] ) , ( [ eqa.12 ] ) , and ( [ eqa.13 ] ) that the downstream value of @xmath168 is given by @xmath359 \over \gamma(a ) \ , \gamma(1-b ) } \ , { \varphi_1(\lambda_n , y_0 ) \over \varphi_2(\lambda_n , y_0 ) } \ , \label{eqa.14}\ ] ] where @xmath164 and @xmath165 are evaluated using equations ( [ eq4.18 ] ) . the derivation of the eigenvalue equation ( [ eq4.24 ] ) in 4.2 makes use of the wronskian of the two functions @xmath170 and @xmath354 , defined by @xmath360 it is useful to derive an analytical expression for @xmath361 based on the differential equation ( [ eq4.12 ] ) governing the two functions @xmath170 and @xmath354 , which can be rewritten in the self - adjoint form @xmath362 + { \lambda \over 4 \ , y^{3/4 } } \ , \varphi - t \ , \varphi = 0 \ , \label{eqb.2}\ ] ] where @xmath363 by applying equation ( [ eqb.2 ] ) to the function @xmath354 and multiplying the result by @xmath170 , and then subtracting from this the same expression with @xmath170 and @xmath354 interchanged , we obtain @xmath364 - \varphi_2 \ , { d \over dy}\left[y^{1/4 } \ , ( 1-y ) \ , { d\varphi_1 \over dy}\right ] = 0 \ , \label{eqb.4}\ ] ] which can be rewritten as @xmath365 = 0 \ , \label{eqb.5}\ ] ] where we have made use of the result @xmath366 equation ( [ eqb.5 ] ) can rearranged in the form @xmath367 \ , \label{eqb.7}\ ] ] which can be integrated to obtain the exact solution @xmath368 where @xmath369 is an integration constant that may depend on @xmath150 but not on @xmath130 . the exact dependence of @xmath305 on @xmath150 can be derived by analyzing the behaviors of the functions @xmath170 and @xmath354 in the limit @xmath169 . for small values of @xmath130 , we have the asymptotic expressions ( abramowitz & stegun 1970 ) @xmath370 we therefore find that asymptotically , @xmath371 comparing this result with equation ( [ eqb.8 ] ) , we conclude that @xmath372 and therefore the exact solution for the wronskian is given by @xmath373 this result is used in 4.2 in the derivation of the eigenvalue equation ( [ eq4.24 ] ) . the evaluation of the green s function @xmath198 using equation ( [ eqc.1 ] ) relies on knowledge of the expansion coefficients @xmath190 , which are given by @xmath384 where the quadratic normalization integrals @xmath193 are defined by @xmath385 the direct computation of the normalization integrals @xmath193 via numerical integration is costly and time consuming , and therefore it is desirable to have an alternative procedure available for their evaluation . in fact , it is possible to derive an analytical expression for the normalization integrals based on manipulation of the fundamental differential equation ( [ eq4.12 ] ) governing the eigenfunctions @xmath176 . let us suppose that @xmath340 is a general solution to equation ( [ eq4.12 ] ) for an arbitrary value of @xmath150 ( i.e. , not necessarily an eigenvalue ) that has the asymptotic upstream behavior @xmath386 we shall also require that @xmath151 be continuous at @xmath153 , and that it satisfy the derivative jump condition ( see eq . [ [ eq4.14 ] ] ) @xmath387 after a bit of algebra , we find that the global solution for @xmath151 can be expressed as @xmath388 where @xmath389 and @xmath390 are given by @xmath391 and the wronskian @xmath361 is evaluated using equation ( [ eqb.12 ] ) . in 4.5 we demonstrated that the column - integrated green s function @xmath411 can be written as ( see eq . [ [ eq4.38 ] ] ) @xmath412 where the quadratic normalization integrals @xmath193 are computed using equation ( [ eq4.29 ] ) , and @xmath413 as an alternative to numerical integration for the evaluation of the integral @xmath228 , here we shall derive an analytical expression for this quantity . first we use equation ( [ eq4.20 ] ) to substitute for @xmath176 in equation ( [ eqe.2 ] ) to obtain @xmath414 substitution for @xmath354 using equation ( [ eq4.19 ] ) and rearranging terms , we can write @xmath415 where @xmath416 to make further progress , we need to evaluate the fundamental indefinite integrals @xmath417 fortunately , these integrals can be carried out analytically . by applying equation ( 15.2.1 ) from abramowitz & stegun ( 1970 ) twice , we obtain for @xmath418 the result @xmath419 where @xmath420 is an integration constant . likewise , we can apply equation ( 15.2.4 ) from abramowitz & stegun ( 1970 ) twice to find that @xmath421 where @xmath422 is another integration constant . combining relations and simplifying , after some algebra we obtain the final result @xmath229 + { \varphi_1^*(\lambda_n , y_0 ) \over ( a-1 ) \ , ( b-1 ) \ , \gamma(1-c ) } \bigg [ f(a , b\,;c-1;y_0 ) \nonumber \\ & + & { ( c-2 ) \over ( a-2 ) \ , ( b-2 ) } \ , f(a , b;c-2;y_0)\bigg ] \bigg\ } + { ( 1-c ) \over ( a-1 ) \ , ( b-1 ) } \left[1 + { ( c-2 ) \over ( a-2 ) \ , ( b-2)}\right ] \ . \qquad \label{eqe.9}\end{aligned}\ ] ] this formula allows the efficient computation of the integrals @xmath228 appearing in equation ( [ eq4.38 ] ) for the column - integrated green s function . cccccccccccccc 1 & 6.0 & @xmath423 & @xmath424 & 0.890 & 1.087 & 0.999770 & @xmath425 & 2.0398 & @xmath426 & @xmath427 + 2 & 1.3 & @xmath428 & @xmath429 & 0.286 & 0.954 & 0.999998 & @xmath430 & 2.6382 & @xmath431 & @xmath432 +
on the whole , measurements of observables in the @xmath3 system agree with the standard model ( sm ) . however , some cracks have started to appear . there are now several quantities whose measured values differ from the predictions of the sm . although these disagreements are not statistically significant they are typically at the level of @xmath4 they are intriguing since there are a number of different @xmath3 decays and effects involved , and they all appear in @xmath5 transitions . because of this , there have been numerous papers examining new - physics ( np ) flavour - changing neutral - current ( fcnc ) contributions to the various @xmath5 processes . these analyses have been performed in the context of specific np models , or model - independently . in general , such np can also contribute to fcnc processes involving the top quark . this has been looked at , though much less so than in @xmath3 decays . however , given that the lhc will produce a large number of top quarks and will be able to measure flavour - changing @xmath6 decays , it is important to explore the possibility of np contributions to fcncs in the top sector . in the past , analyses have focused on rare top decays such as @xmath7 ( @xmath8 ) and @xmath9 @xcite . other top decays where np effects have been examined include @xmath10 @xcite and @xmath11 @xcite . in this paper we examine the decay @xmath12 . in the sm , this decay occurs at tree level , via @xmath13 . on the other hand , because it involves the small element @xmath14 ( @xmath15 ) of the cabibbo - kobayashi - maskawa ( ckm ) quark mixing matrix , the amplitude for this process is also rather small , and is therefore quite sensitive to np . for example , there could be np contributions to this decay in models with a charged higgs boson @xcite , or via @xmath16 , where @xmath17 corresponds to some neutral particle ( such as a @xmath18 or a non - sm higgs boson ) . such processes could interfere with the sm process , leading to observable consequences , even if the intermediate np particle were heavier than the top quark . rather than restricting our attention to any one particular model , we examine np contributions to @xmath19 model - independently ( i.e. , using an effective lagrangian ) . our model - independent treatment of @xmath12 takes into account the effects of the 10 possible four - fermi operators . these operators contribute to both cp - conserving and cp - violating observables . for the cp - even observables , we consider the cp - averaged partial width , a forward - backward - like asymmetry , and an asymmetry that depends on the spin of the top quark . for the cp - odd observables , we note that the decay @xmath19 is dominated by one amplitude in the sm ; i.e. , there is only one weak phase involved . as such , all cp - violating asymmetries are very suppressed in the sm , so the observation of a non - zero asymmetry would be a smoking - gun signal of np . in this paper , we discuss two types of cp - odd asymmetries : the partial - rate asymmetry ( pra ) and a triple - product asymmetry ( tpa ) . pras require a strong phase in order to be non - zero . strong phases can arise due to gluon exchange , but it is expected that such phases will be small since the energies involved are so large . another source of a strong phase is the width of the @xmath20 . in our calculation , we ignore qcd - based strong phases and assume that the required strong phase is due entirely to the width of the @xmath20 . this means that only sm - np interference can lead to a pra . on the other hand , in contrast with pras , tpas do not require a strong phase in order to be non - zero . thus , np - np interference terms can give rise to tpas . as we will see , tpas generated by sm - np interference tend to be small , but np - np tpas can be large . these are particularly interesting . we show that the measurement of the partial width by itself can reveal the presence of np . however , if the np exists , we will want to know its identity , i.e. , which of the 10 operators is responsible , and the partial width measurement does not give us this information . in order to do this , it is necessary to measure the other quantities mentioned above . since the various observables depend differently on the operators , the knowledge of their sizes will give us an idea of which np operators are present . this will allow us in turn to deduce which model(s ) might be responsible for the observed effects . although we confine our attention to @xmath12 in this work , we note that most of our results are easily transferable to @xmath21 by the replacement @xmath22 in feynman diagrams and expressions . since @xmath1 , the branching ratio for @xmath21 is similar to that for @xmath12 , and the two processes would apriori have similar sensitivities to np effects . one difference between @xmath0 and @xmath21 is that the `` cpt '' correction to the pra would be significant for @xmath21 , whereas it is miniscule for @xmath12 ( see the discussion in sec . [ sec : pra ] and appendix [ sec : cpt ] ) . the remainder of this paper is organized as follows . in sec . [ sec : smnp ] we write down the sm contribution to @xmath0 , and also parameterize np contributions to this decay in terms of an effective lagrangian containing ten terms . in sec . [ sec : cpeven ] ( cp - even observables ) we compute the cp - averaged partial width for the decay under consideration , as well as a forward - backward - like asymmetry and an asymmetry based on the spin of the top quark . the latter two asymmetries are both constructed in such a way that they are zero within the context of the sm . we close this section with a brief numerical study , noting that the cp - even asymmetries could reasonably be of order 10 s of percent . section [ sec : cpodd ] contains our analysis of two cp - odd observables the partial rate asymmetry and the triple - product asymmetry . section [ sec : discussion ] concludes with a discussion of our results . appendices [ sec : appa ] , [ sec : cpt ] and [ sec : appb ] contain some technical details . in particular , appendix [ sec : cpt ] contains a discussion of results related to the cpt theorem , namely which vertex - type corrections must be considered in the calculation of pras in order not to violate cpt . in this section we parameterize the np contributions to @xmath23 in terms of an effective lagrangian . we then write down expressions for the sm and np amplitudes . these expressions are used in following sections to determine various cp - even and cp - odd observables . figure [ fig : feynman_diagram1](a ) shows the feynman diagram for the sm contribution to @xmath19 . the resulting amplitude is given by , @xmath24 , ~ \label{eq : wamp}\end{aligned}\ ] ] where @xmath25 is the ckm matrix . we work in the standard representation of the ckm matrix , in which @xmath14 and @xmath26 are both real . note that colour indices have been suppressed . the expression in square parentheses is proportional to the @xmath20 propagator , with @xmath27 , @xmath28^{-1}$ ] and @xmath29 , where @xmath30 . ( note : throughout this paper we neglect the leptons and light quarks masses . however , if this not done , the @xmath20 propagator is modified to i , where @xmath31^{-1}$ ] . @xmath32 and @xmath33 are related to the transverse and longitudinal widths of the @xmath20 @xcite , and they both depend on the quark masses . there has been considerable discussion in the literature regarding the correct form of the @xmath20 propagator . ( see , for example , refs . the above expression has been derived by performing a dyson summation of the absorptive parts of the @xmath20 self - energy diagrams in unitary gauge , with quarks and leptons in the loops . some of the disagreement in the literature has focused on the form of @xmath34 ( see the brief discussion in ref . @xcite , for example ) . still , when all light masses are neglected , none of the observables in the present work depend on @xmath34 . there seems to be broader agreement on the form for @xmath35 in the literature , although many authors drop the @xmath36 dependence in @xmath37 . finally , we should note that the pinch technique may be used to reorganize perturbative calculations even those involving resonances in such a way that results are explicitly gauge - invariant ( see , for example , ref . rigorous application of the pinch technique to the problem at hand is beyond the scope of this work . ) we parameterize new - physics effects via an effective lagrangian @xmath38 , where , @xmath39 in the above expressions , @xmath40 is assumed to be of order @xmath41 , @xmath42 is the np mass scale and the @xmath43 and @xmath44 couplings may include weak ( cp - violating ) phases . for the levi - civita tensor , we adopt the convention @xmath45 . the np contributions to @xmath46 are illustrated in fig . [ fig : feynman_diagram1](b ) . colour indices are not shown in the above expressions , but are assumed to contract in the same manner as those of the sm ( i.e. , @xmath47 with @xmath6 and @xmath48 with @xmath49 ) . in fcnc models those with a flavour - changing neutral particle such as a @xmath50 or a scalar the colour indices would contract in the opposite manner ( i.e. , @xmath48 with @xmath6 and @xmath47 with @xmath49 ) . it is straightforward to incorporate colour - mismatched terms into the effective lagrangian . this topic is discussed further in appendix [ sec : appb ] . it is useful to define @xmath51 and similarly for the other @xmath43 and @xmath44 couplings . in terms of the `` @xmath52 '' parameters , we have the following expression for the np contribution to @xmath12 , @xmath53 to get a sense of the possible order of magnitude of the @xmath52 couplings , note that , if @xmath54 and @xmath55 gev , then @xmath56 . thus , @xmath57 could reasonably be assumed to be of order unity . in other words , the sm and np contributions to @xmath12 can very well be about the same size . when computing the effect of np on a particular observable , it is therefore important to include both the sm - np interference and np@xmath58 pieces . at present , there are no direct constraints on the @xmath52 couplings . the precision measurements of @xmath14 place an indirect constraint via the loop diagram shown in fig . [ fig : vcb ] . it is known that some care must be taken when attempting to incorporate terms from an effective lagrangian into loop calculations @xcite . incorporating the diagram shown in fig . [ fig : vcb ] , we find the following expression for the effective lagrangian for @xmath59 , @xmath60\overline{\ell}_l\gamma^\mu \nu_l + \mbox{h.c . , } \label{eq : bcellnu}\end{aligned}\ ] ] in which we have dropped corrections of order @xmath61 . using the feynman rules for the various vertices , and employing dimensional arguments , we estimate @xmath62 since semileptonic @xmath63 transitions are used to determine @xmath14 , the experimental value of @xmath14 can be used to bound @xmath57 and @xmath64 . let us first consider the @xmath57 term in eq . ( [ eq : bcellnu ] ) , ignoring the @xmath64 term . the @xmath57 term has exactly the same structure as the sm term . its effect is thus simply to multiply any inclusive or exclusive @xmath65 width by a factor of @xmath66 $ ] . the current experimental value of @xmath14 is @xmath67 @xcite , implying a @xmath68 uncertainty on @xmath69 . if we assume that the @xmath57 contribution to @xmath59 is hiding in the experimental uncertainty of @xmath69 , we find the bound , @xmath70 in which we have neglected the quadratic contribution of @xmath71 , since it is small . since @xmath64 is associated with the right - handed quark current in eq . ( [ eq : bcellnu ] ) , its effect is process - dependent . for example , for @xmath72 , the hadronic matrix element is only sensitive to the vector part of the hadronic current , so left - handed and right - handed couplings both have the same effect , and they can be absorbed in with the sm current @xcite . for other modes , such as @xmath73 , the right - handed and left - handed currents must be treated differently @xcite . since our expression in eq . ( [ eq : zetavllr ] ) is somewhat of an approximation in any case , we assume the same upper bound for @xmath64 as for @xmath57 , i.e. , @xmath74 the np operators considered in this work could contribute , via loops , to other observables as well . as an example , consider the decay @xmath75 , which proceeds at tree - level in the sm . in the present context , the np operators contribute to this decay via a diagram similar to fig . [ fig : vcb ] , but with @xmath76 replaced by @xmath77 in the final state . the resulting effective lagrangian would be very similar to eq . ( [ eq : bcellnu ] ) , which could lead to effects in the measurement of @xmath78 @xcite . we do not consider such effects further here . although we do not perform any model calculations in this work , it is worthwhile to consider which types of models could give rise to the various np operators . the terms in eqs . ( [ eq : eff1 ] ) and ( [ eq : eff2 ] ) arise in models that contain new charged vector or scalar bosons . for example , extensions of the sm containing gauge bosons with left- and right - handed charged - current couplings ( such as the left - right model ) would contribute terms such as those appearing in @xmath79 including the @xmath80 and @xmath81 terms if there were some amount of mixing between the left- and right - handed gauge bosons . models containing charged scalars ( such as the charged higgs bosons that appear in many extensions of the sm ) could give rise to the terms in the expression for @xmath82 . alternatively , there are many fcnc models containing a heavy neutral np particle ( such as a @xmath18 or a neutral higgs boson ) with flavour - changing @xmath6-@xmath83 couplings . here fierz rearrangements of the eight operator combinations @xmath84 $ ] and @xmath85 $ ] ( in the notation employed in ref . @xcite ) lead to all ten of the operator combinations in eqs . ( [ eq : eff1])-([eq : eff3 ] ) . in this case there would be mismatched colour indices between the np and sm diagrams . appendix [ sec : appb ] explains how to deal with this situation . in the following sections we compute various observables associated with the decays @xmath86 and @xmath87 . we take as our starting point the expressions for the sm and np amplitudes in eqs . ( [ eq : wamp ] ) and ( [ eq : npamplitude2 ] ) , respectively . we begin by considering three cp - even observables associated with the decays in question . the first is the cp - averaged partial width , normalized to the sm result ; the second is a forward - backward - like asymmetry ; and the third is a cp - even asymmetry that employs the spin of the top quark . in secs . [ sec : cpaveragedwidth]-[sec : singlespin ] we work out expressions for the various observables . section [ sec : discussioncpeven ] contains a numerical analysis and discussion of the results . we first consider the partial width for @xmath12 . the expression for the differential partial width , including the various np terms from the effective lagrangian , may be found in eq . ( [ eq : appdgam ] ) in appendix [ sec : appa ] . performing the integrations over @xmath88 and @xmath36 , we find , @xmath89 \nonumber\\ & & + ~\frac{3g_f m_t^2}{\sqrt{2}\pi^2 \left(1-\zeta_w^2\right)^2\left(1 + 2\zeta_w^2\right ) } \bigg [ \left|x_{ll}^v\right|^2 + \left|x_{lr}^v\right|^2 + \left|x_{rl}^v\right|^2 + \left|x_{rr}^v\right|^2 \nonumber\\ & & + ~ \frac{1}{4}\left(\left|x_{ll}^s\right|^2 + \left|x_{lr}^s\right|^2 + \left|x_{rl}^s\right|^2 + \left|x_{rr}^s\right|^2\right ) + 24\left|x_t\right|^2 + 96\left|x_{te}\right|^2 \bigg ] \bigg\}\ , , \label{eq : gam_tbbbarc}\end{aligned}\ ] ] where @xmath90 and @xmath91 in calculating the expressions in eqs . ( [ eq : gam_tbbbarc ] ) and ( [ eq : gamsm ] ) , we have used the narrow width approximation , in which @xmath92 is replaced by a @xmath93-function in @xmath36 , appropriately normalized . [ one exception is the term proportional to re(@xmath94 ) in eq . ( [ eq : gam_tbbbarc ] ) , which was computed numerically . ] ( note that the term proportional to im(@xmath94 ) in eq . ( [ eq : gam_tbbbarc ] ) , which is involved in the partial rate asymmetry discussed below , is not quite complete . in its present form , it would lead to a violation of cpt . to avoid running into problems with the cpt theorem , certain vertex - type corrections need to be included in the calculation when computing partial rate asymmetries . we discuss these extra terms in appendix [ sec : cpt ] . ) the partial width for the cp - conjugate decay may be obtained from eq . ( [ eq : gam_tbbbarc ] ) by complex conjugating all weak phases . '' multiplying @xmath95 in the np amplitude does _ not _ get complex conjugated when computing the amplitude for the cp - conjugate process . ] this has the effect of changing the sign of the im(@xmath94 ) term , while leaving the other terms unchanged . adding the widths for @xmath12 and @xmath96 , and dividing by twice the sm result yields @xmath97 , \label{eq : rcal}\end{aligned}\ ] ] in which we have inserted the known values for the various physical constants , and used the expression for @xmath98 noted below eq . ( [ eq : wamp ] ) . note that , in practice , the term proportional to re(@xmath94 ) is always small compared to the other terms . above , we noted that the @xmath52 s could well be @xmath99 . thus , from eq . ( [ eq : rcal ] ) , we see that the cp - averaged partial width could be used as a tool to search for np . in particular , an experimental value for @xmath43 that is different from unity would give clear evidence for np . ( depending on the size of the np signal , it may be important to include corrections to eq . ( [ eq : gamsm ] ) @xcite . ) on the other hand , all 10 np operators contribute to @xmath43 in similar ways . thus , even if the measurement of @xmath43 reveals the presence of np , it does not give us any clue as to the type of np . for this reason , it is important to look for signs of np in other quantities . this is most easily done using observables that are strictly zero within the context of the sm . such observables would typically depend upon differing combinations of np parameters , so that the observation of np effects using several different observables would yield insight into the precise nature of the np . in the following , we construct several asymmetries that are zero within the context of the sm and discuss their potential usefulness as tools for searching for np . note that the ratio @xmath43 , defined above , will also appear in the denominators of all asymmetries considered below . since @xmath43 is primarily a sum of positive quantities , and since it will always appear in the denominators of the asymmetries , it will always tend to decrease the values of the asymmetries compared to the analogous expressions employing the approximation @xmath100 . a tool that has historically been useful to experimentalists is the forward - backward ( fb ) asymmetry . the differential width for @xmath23 may be written in terms of @xmath101 and @xmath102 , where @xmath103 is the angle between the momentum of the top quark and that of the charm quark in the @xmath47-@xmath83 rest frame . the fb asymmetry makes use of the following asymmetric integration over @xmath102 , @xmath104 dq^2 \ , . \label{eq : fb1}\end{aligned}\ ] ] we choose instead to work with the kinematical variables @xmath36 and @xmath105 , noting that @xmath106 using eq . ( [ eq : costheta ] ) , we may rewrite eq . ( [ eq : fb1 ] ) as follows , @xmath107 dq^2 \ , . \label{eq : fbdef}\end{aligned}\ ] ] let us first consider the sm contribution to the fb asymmetry . the sm - only contribution to the width is such that @xmath108 ( see eq . ( [ eq : appdgam ] ) in appendix [ sec : appa ] ) . using the integration prescription in eq . ( [ eq : fbdef ] ) , and assuming the narrow - width approximation , we find , @xmath109 thus , we see that the sm contribution to the fb asymmetry is non - zero . as noted above , in order to use a particular asymmetry as a discriminator of np , it is useful if the asymmetry is zero when no np contribution is present . the regular fb asymmetry does not satisfy this requirement , as is evidenced by eq . ( [ eq : afbsm ] ) . it turns out , however , that if we modify the @xmath105 integration prescription somewhat , the sm contribution can be made to disappear . that is , instead of breaking up the integral over @xmath105 at the point @xmath110 , as is done in eq . ( [ eq : fbdef ] ) , we move the boundary to the value @xmath111 , @xmath112 dq^2 \ , , \label{eq : fb - likedgam}\end{aligned}\ ] ] in which @xmath111 is chosen such that [ see eq . ( [ eq : smdqsqdrhosq ] ) ] , @xmath113 then , by construction , @xmath114 dq^2 = 0 . \label{eq : fb - likedgamsmzero2}\end{aligned}\ ] ] the new boundary , @xmath111 , is @xmath36-dependent and can be solved for numerically . is cubic and can also be solved analytically , although the resulting expressions for the roots of the equation are not particularly enlightening . ] figure [ fig : phasespace ] shows the phase space available for @xmath12 and also indicates the two boundary choices described above . the vertical band indicates the location of the @xmath20 resonance for the sm contribution . equation ( [ eq : appdgam ] ) in appendix [ sec : appa ] gives the expression for @xmath115 . the @xmath105 dependence of the sm piece is given by @xmath116 ; the sm - np cross terms and the @xmath117 and @xmath118 terms have this same @xmath105 dependence . since the integration prescription described above is engineered to eliminate the sm term when integrating over @xmath105 , these latter terms also disappear upon integration over @xmath105 in this manner . performing the integration numerically for the other terms yields the following , @xmath119 \nonumber\\ & & + 1.81 \left(\left|x_t\right|^2 + 4\left|x_{te}\right|^2\right ) \bigg ] . \label{eq : gammarhobarsq}\end{aligned}\ ] ] the above expression is cp - even ; i.e. , the analogous expression for @xmath96 is the same . finally , we form an fb - like asymmetry as follows , @xmath120 \nonumber\\ & & + 0.460 \left(\left|x_t\right|^2 + 4\left|x_{te}\right|^2\right ) \bigg ] . \label{eq : arhobarsq}\end{aligned}\ ] ] by construction , this asymmetry is only non - zero if np contributions are present . section [ sec : discussioncpeven ] contains a discussion of the range of sizes that are possible for the fb - like asymmetry . at this point we note only that asymmetries of order tens of percent are possible . in this section we have described three observables that are even under cp and that could be used to detect the presence of np in the decays @xmath19 and @xmath137 . should np be discovered , detailed analysis of such observables could allow experimentalists to map out the nature of the np . the first observable considered in this section was a ratio , @xmath138 , which was defined to be proportional to the cp - averaged partial width . of the observables considered in this work , @xmath43 would probably be the simplest to measure experimentally . decisive experimental deviation from the sm value @xmath139 would be evidence for np . on the other hand , @xmath43 can not be used to distinguish the different np operators . to do this requires the use of other quantities . it is useful to employ observables that give a null signal within the context of the sm . for such observables , a significant departure from zero would be a `` smoking - gun '' signal for new physics . in addition , since they depend differently on the various np operators , the observation of non - zero values for these observables would help in identifying the type of np . the usual forward - backward asymmetry for @xmath12 is expected to be non - zero within the context of the sm . it is possible , however , to alter the kinematical weighting that is used in defining the fb asymmetry in such a way that the resulting `` fb - like '' asymmetry is zero within the context of the sm . equation ( [ eq : arhobarsq ] ) defines the fb - like asymmetry @xmath140 in terms of an asymmetric integration over the kinematical variable @xmath105 . the integration is engineered in such a way that the sm contribution disappears kinematically . in eq . ( [ eq : aparallel ] ) we formed a cp - even asymmetry using the spin of the top quark . this asymmetry was also defined in such a way that it was zero within the context of the sm . a non - zero experimental signal for either of these asymmetries would indicate the presence of np . .some representative values for the fb - like asymmetry @xmath140 and the cp - even spin asymmetry @xmath141 . the value for @xmath43 is also included for each case . [ cols="^,^,^,^,^,^,^,^,^,^,^",options="header " , ] [ tab : acptpnpnp ] finally , we define the tpa as @xmath142 so that @xmath143 } { \left(1-\zeta_w^2\right)^2\left(1 + 2\zeta_w^2\right ) } \nonumber \\ & \simeq & \frac{0.243}{\cal r}\ , \mbox{im}\!\left[x_t\left(x^{s*}_{ll}+x^{s*}_{rr}\right ) -2x_{te}\left(x^{s*}_{ll}-x^{s*}_{rr}\right)\right ] . \label{eq : atpcpnon - suppressed}\end{aligned}\ ] ] table [ tab : acptpnpnp ] contains some numerical results following from the above expression , showing that the tpas can indeed be large of order 10 s of percent if the np coefficients are assumed not to be suppressed . figure [ fig : acpodd ] shows a scatter plot of @xmath144 versus @xmath43 . the points in this plot were obtained by generating combinations of purely real and purely imaginary random numbers for @xmath145 , @xmath146 , @xmath147 and @xmath95 over various ranges . once again , asymmetries were only plotted if @xmath148 . it is evident from the plot that relatively large tpas are possible . in this paper we consider new - physics ( np ) contributions to the decay @xmath12 . in the standard model ( sm ) , this is a tree - level process : @xmath13 . however , the sm amplitude involves the small cabibbo - kobayashi - maskawa element @xmath14 ( @xmath15 ) , and is therefore suppressed . as a result , the decay is quite sensitive to np . rather than working within the context of any one particular extension of the sm , we parameterize the np couplings by an effective lagrangian that includes the 10 possible four - fermi operators . we show that the sm and np contributions to @xmath12 can indeed be about the same size . we first compute the @xmath149 partial width in the presence of np . the ratio @xmath43 , defined in eq . ( [ eq : rcal ] ) , provides a quantitative measure of the deviation of the partial width from its sm expectation ( @xmath150 in the sm ) . this shows clearly that this observable is excellent for showing that np is present significant deviations of @xmath43 from 1 are possible . on the other hand , the partial width is not a good observable to _ identify _ the new physics all 10 np operators contribute to @xmath43 in a similar way . in order to get an idea of the type of np present , it is necessary to consider other quantities . to this end , we construct two cp - conserving and two cp - violating observables : ( i ) cp - even : a forward - backward - like asymmetry [ @xmath140 eq . ( [ eq : arhobarsq ] ) ] and a top - quark - spin - dependent asymmetry [ @xmath141 eq . ( [ eq : aparallel ] ) ] , ( ii ) cp - odd : the partial rate asymmetry [ @xmath151 eq . ( [ eq : pra1 ] ) ] and a triple - product asymmetry [ @xmath152 eq . ( [ eq : atpcpnon - suppressed ] ) ] . in each case , the observable is formulated in such a way that it is zero within the context of the sm . the key point is that these observables depend on differing combinations of the np parameters . this gives them different sensitivities to the various lorentz structures present in the np effective lagrangian . the allowed values of these four observables vary greatly depending on the values of the np parameters , but results in the 10 - 20% range are possible ( see tables [ tab : acpeven ] and [ tab : acptpnpnp ] ) . if np is present , it may well produce measurable values of these observables . taken together , the measurements of these quantities will give a good indication of the type of np present . * acknowledgments * : we would like to thank a. soni for helpful discussions . this work was financially supported by nserc of canada ( d.l . ) the work of k.k . , t.k . and k.w . was supported by the u.s . national science foundation under grant phy0900914 . this work was also partially supported by anpcyt ( argentina ) under grant # pict - prh 2009 - 0054 . the cpt theorem requires the total decay width for the top to be equal to that for the anti - top . an apparent violation of the cpt theorem arises , however , if the np contributions in fig . [ fig : feynman_diagram1 ] ( b ) are the only ones that are kept . that this is the case is straightforward to see , since the diagram in fig . [ fig : feynman_diagram1 ] ( b ) affects the partial widths for @xmath153 and @xmath162 differently [ leading to the pra in eq . ( [ eq : pra1 ] ) ] , but has no effect on the other top or anti - top decay modes . thus , the top and anti - top total widths are not equal if only such contributions are kept , resulting in an apparent violation of the cpt theorem . this phenomenon is well - known ( see , for example , refs . @xcite ) . in this appendix we show that the inclusion of certain vertex - type corrections gives rise to compensating differences in the top and anti - top widths . the sum of the differences is zero , so that the top and anti - top widths no longer differ , in agreement with the cpt theorem . equation ( [ eq : delgamma ] ) gives a correction to the pra for @xmath23 , leading to the following modification of eq . ( [ eq : pra1 ] ) , @xmath172 . \label{eq : pra2}\end{aligned}\ ] ] the correction to the original expression is miniscule , since @xmath173 . an interesting consequence of the cpt theorem is that , if np operators give a pra in a particular decay mode ( such as @xmath174 , as in our case ) , then those same np operators must also contribute to one or more other decay modes in such a way that the total width of the top is the same as that of the anti - top . this means that those other decay modes must also have pras ( barring other accidental cancellations ) . we can use eq . ( [ eq : delgamma ] ) to estimate the pras in other decay modes due to the np operators in eqs . ( [ eq : eff1])-([eq : eff3 ] ) . the resulting expression is given by , @xmath175 thus the contributions of these np operators to pras in other top decay modes are expected to be very small . one could also consider the complementary question : are there np operators , other than those given in eqs . ( [ eq : eff1])-([eq : eff3 ] ) , that could contribute to the pra in @xmath12 ? the answer to this question appears to be yes . for example , the effective operators @xmath176 or @xmath177 could appear in a diagram similar to that in fig . [ fig : cpt_vertex_correction ] , but with the usual sm @xmath178 vertex at the top , and the np - induced one - loop correction to the @xmath179 vertex at the bottom . such operators are constrained by @xmath3 decays . we should note that the np effective operators in eqs . ( [ eq : eff1])-([eq : eff3 ] ) give rise to other loop - level diagrams that could contribute to pras in top decays . the contributions in different decay modes would still complement each other in the sense that the total top and anti - top widths would remain equal . figure [ fig : tczvertex ] shows an example of loop - level corrections to @xmath180 ( @xmath181 ) mediated by the np operators considered in this work . these diagrams could interfere with their corresponding sm diagrams to induce pras . we do not compute such contributions here . the effective lagrangian incorporating np effects given in eqs . ( [ eq : eff1])-([eq : eff3 ] ) assumed that the colour indices contracted in the same manner as those of the sm diagram . this need not be the case , so it is useful to consider the effects of including colour - mismatched terms in the effective lagrangian . to this end , let us generalize the np effective lagrangian in eq . ( [ eq : eff1 ] ) as follows , @xmath182 and similarly for eqs . ( [ eq : eff2 ] ) and ( [ eq : eff3 ] ) . in this expression , the subscripts @xmath183 and @xmath49 are colour indices and the primed coefficients correspond to the new , colour - mismatched terms . the total amplitude for @xmath184 ( with the subscripts @xmath183 , @xmath49 , @xmath83 and @xmath185 representing the colours ) could then be parameterized as @xmath186 in which the sum runs over the sm diagram , plus all np contributions . the factors @xmath187 and @xmath188 are the coefficients for the colour - matched and colour - mismatched terms , respectively , and are assumed to contain all of the weak phases . ( the @xmath189 coefficient for the sm term is assumed to be zero . ) for a given value of @xmath190 , the phases of @xmath187 and @xmath188 could be different . the factors @xmath191 contain all the spinors and @xmath192 matrices and , in the case of the sm diagram , the @xmath20 propagator . summing over the quarks colours and dividing by 3 for the average over the top quark s colours , we find , @xmath193\left|{\cal m}_i\right|^2 \nonumber\\ & + & 6 \sum_{j > i } \left\{\mbox{re}\!\left[\mathbb{r}_i\mathbb{r}_j^ * + \mathbb{r}_i^\prime \mathbb{r}_j^{\prime * } + \frac{1}{3}\left ( \mathbb{r}_i\mathbb{r}_j^{\prime * } + \mathbb{r}_i^\prime\mathbb{r}_j^*\right)\right ] \!\mbox{re}\!\left({\cal m}_i { \cal m}_j^*\right ) \right . \nonumber\\ & - & \left . \mbox{im}\!\left[\mathbb{r}_i\mathbb{r}_j^ * + \mathbb{r}_i^\prime \mathbb{r}_j^{\prime * } + \frac{1}{3}\left ( \mathbb{r}_i\mathbb{r}_j^{\prime * } + \mathbb{r}_i^\prime\mathbb{r}_j^*\right)\right ] \!\mbox{im}\!\left({\cal m}_i { \cal m}_j^*\right ) \right\ } . \label{eq : colourmismatchcorrection}\end{aligned}\ ] ] the @xmath194 terms in the above expression correspond to the `` colour - matched '' terms that we have taken into account in this work . the other terms are new . equation ( [ eq : colourmismatchcorrection ] ) can be used to generalize the expressions in this paper , provided the expressions have already been split cleanly into pieces containing the weak phases ( @xmath187 , etc . ) and those containing the spinors and any strong phases ( @xmath191 ) . expressions containing sm - np cross - terms may safely set @xmath195 and incorporate the entire amplitude into the `` @xmath196 '' part [ in eq . ( [ eq : mabcd ] ) ] , since @xmath26 and @xmath14 have been taken to be real . as an example , the generalized form for eq . ( [ eq : rcal ] ) would be @xmath197\ , , \end{aligned}\ ] ] in which we have used the fact that @xmath198 . similarly , eq . ( [ eq : atpcpnon - suppressed ] ) would become , @xmath199\right\},\end{aligned}\ ] ] and eq . ( [ eq : pra1 ] ) would become , @xmath200 finally , an expression such as `` re@xmath201 '' in eq . ( [ eq : appdgam ] ) , which contains both a strong phase ( in @xmath35 ) and a weak phase ( in @xmath157 ) , first needs to be separated into two pieces using @xmath202 . g. eilam , j. l. hewett and a. soni , phys . d * 44 * , 1473 ( 1991 ) [ erratum - ibid . d * 59 * , 039901 ( 1998 ) ] ; m. e. luke and m. j. savage , phys . b * 307 * , 387 ( 1993 ) [ arxiv : hep - ph/9303249 ] ; t. han , r. d. peccei and x. zhang , nucl . b * 454 * , 527 ( 1995 ) [ arxiv : hep - ph/9506461 ] ; t. han , k. whisnant , b. l. young and x. zhang , phys . rev . d * 55 * , 7241 ( 1997 ) [ arxiv : hep - ph/9603247 ] ; t. han , k. whisnant , b. l. young and x. zhang , phys . lett . b * 385 * , 311 ( 1996 ) [ arxiv : hep - ph/9606231 ] ; g. eilam , a. gemintern , t. han , j. m. yang and x. zhang , phys . b * 510 * , 227 ( 2001 ) [ arxiv : hep - ph/0102037 ] ; j. a. aguilar - saavedra , acta phys . b * 35 * , 2695 ( 2004 ) [ arxiv : hep - ph/0409342 ] ; a. datta and m. duraisamy , phys . d * 81 * , 074008 ( 2010 ) [ arxiv:0912.4785 [ hep - ph ] ] . j. l. diaz - cruz and g. lopez castro , phys . b * 301 * , 405 ( 1993 ) . d. atwood , g. eilam , a. soni , r. r. mendel and r. migneron , phys . lett . * 70 * , 1364 ( 1993 ) ; d. atwood , g. eilam and a. soni , phys . rev . lett . * 71 * , 492 ( 1993 ) [ arxiv : hep - ph/9303268 ] . r. cruz , b. grzadkowski and j. f. gunion , phys . b * 289 * , 440 ( 1992 ) . j. liu , phys . d * 47 * , r1741 ( 1993 ) [ arxiv : hep - ph/9209294 ] . d. atwood , g. eilam , a. soni , r. r. mendel and r. migneron , phys . d * 49 * , 289 ( 1994 ) . t. arens and l. m. sehgal , phys . d * 51 * , 3525 ( 1995 ) [ arxiv : hep - ph/9404259 ] . x. j. bi and y. b. dai , eur . phys . j. c * 12 * , 125 ( 2000 ) [ arxiv : hep - ph/9904228 ] . d. atwood , s. bar - shalom , g. eilam and a. soni , phys . rept . * 347 * , 1 ( 2001 ) [ arxiv : hep - ph/0006032 ] . b. grzadkowski and w. y. keung , phys . b * 319 * , 526 ( 1993 ) [ arxiv : hep - ph/9310286 ] ; e. christova and m. fabbrichesi , phys . b * 320 * , 299 ( 1994 ) [ arxiv : hep - ph/9307298 ] ; s. bar - shalom , d. atwood and a. soni , phys . d * 57 * , 1495 ( 1998 ) [ arxiv : hep - ph/9708357 ] . m. nowakowski and a. pilaftsis , z. phys . c * 60 * , 121 ( 1993 ) [ arxiv : hep - ph/9305321 ] . g. lpez castro , j. l. lucio and j. pestieau , int . j. mod . a * 11 * , 563 ( 1996 ) [ arxiv : hep - ph/9504351 ] . j. papavassiliou and a. pilaftsis , phys . lett . * 75 * , 3060 ( 1995 ) [ arxiv : hep - ph/9506417 ] . j. papavassiliou , in _ `` beyond the standard model iv , '' _ pp . 509 - 513 , eds . j. gunion , t. han and j. ohnemus ( 1995 ) [ arxiv : hep - ph/9504386 ] . d. binosi and j. papavassiliou , phys . rept . * 479 * , 1 ( 2009 ) [ arxiv:0909.2536 [ hep - ph ] ] . c. p. burgess and d. london , phys . d * 48 * , 4337 ( 1993 ) [ arxiv : hep - ph/9203216 ] . k. nakamura _ et al . _ [ particle data group ] , j. phys . g * 37 * , 075021 ( 2010 ) . g. h. wu , k. kiers and j. n. ng , phys . d * 56 * , 5413 ( 1997 ) [ arxiv : hep - ph/9705293 ] . k. kiers , a. soni and g. h. wu , phys . rev . d * 59 * , 096001 ( 1999 ) [ arxiv : hep - ph/9810552 ] . c. c. nishi , am . j. phys . * 73 * , 1160 ( 2005 ) [ arxiv : hep - ph/0412245 ] . for a discussion of some corrections to the tree - level result , see w. bernreuther , j. phys . g * 35 * , 083001 ( 2008 ) [ arxiv:0805.1333 [ hep - ph ] ] , and references therein . the effect of the higher - order corrections is to reduce the value of the top width somewhat . for a discussion of top quark polarization , see m. beneke , i. efthymiopoulos , m. l. mangano , j. womersley , a. ahmadov , g. azuelos , u. baur , a. belyaev _ et al . _ , [ hep - ph/0003033 ] , and references therein . g. eilam , j. l. hewett and a. soni , phys . lett . * 67 * , 1979 ( 1991 ) . k. kiers , a. szynkman and d. london , phys . d * 74 * , 035004 ( 2006 ) [ arxiv : hep - ph/0605123 ] . a. szynkman , k. kiers and d. london , phys . d * 75 * , 075009 ( 2007 ) [ arxiv : hep - ph/0701165 ] . m. nagashima , k. kiers , a. szynkman , d. london , j. hanchey and k. little , phys . d * 80 * , 095012 ( 2009 ) [ arxiv:0907.1063 [ hep - ph ] ] . j. m. gerard and w. s. hou , phys . lett . * 62 * , 855 ( 1989 ) . l. wolfenstein , phys . d * 43 * , 151 ( 1991 ) . j. m. soares , phys . lett . * 68 * , 2102 ( 1992 ) . g. eilam , j. l. hewett and a. soni , phys . lett . * 68 * , 2103 ( 1992 ) . for a similar discussion in the context of @xmath203 , see g. l. kane , g. a. ladinsky and c. p. yuan , phys . rev . d * 45 * , 124 ( 1992 ) .
the kondo necklace ( kn ) type models are useful to discuss the quantum phase transitions between kondo singlet and antiferromagnetically ordered states such as found in heavy fermion compounds @xcite . they have been originally proposed by doniach @xcite for the one - dimensional case as a simplified version of the itinerant kondo lattice ( kl ) models @xcite . thereby the kinetic energy of conduction electrons is replaced by an intersite exchange term . for a pure xy - type intersite exchange this may be obtained by a jordan - wigner transformation . however in higher d the replacement can not be justified easily . the intuitive argument is that at low temperatures the charge fluctuations in the kondo lattice model are frozen out and the remaining spin fluctuation spectrum can be simulated by an antiferromagnetic inter - site interaction term of immobile @xmath1 spins coupled by a kondo interaction to the local noninteracting spins * s*. recent exact diagonalisation studies on finite clusters for both kondo lattice and kondo necklace models have indeed found that the competition between on - site kondo singlet formation and af inter - site correlations are very similar in both models @xcite . a more formal way to get rid of charge fluctuations in the kl model is an inclusion of a hubbard term for conduction electrons ( klu model ) which leads to the isotropic kn model in the large u limit at half filling . nevertheless one should consider the kondo necklace model for d@xmath22 as a model in its own right which is suitable for studying quantum phase transitions between a kondo singlet ( ks ) and an antiferromagnetic ( af ) phase . in its original form the local kondo exchange is isotropic while the intersite exchange is of xy - type . this model has u(1 ) symmetry . later more general models with arbitrary anisotropy of kondo as well as intersite exchange terms have been considered @xcite . indeed compounds which exhibit the ks to af quantum phase transition have mostly uniaxial symmetries . a full account of the influence of uniaxial anisotropies of both terms in the kondo necklace hamiltonian on the quantum critical point has been given in ref . . the d=2 kn model without any anisotropy may also be understood as a special case of a bilayer heisenberg model @xcite where the intersite bonds are cut in one of the layers . a reintroduction of holes in this case leads to the kntj model which is related to the klu model away from half filling @xcite . in the general anisotropic kn model the quantum phase transition is achieved by varying the ratio of hopping t , i.e. , the intersite interactions of @xmath1 spins to the on - site kondo interaction j. in practice this is achieved by varying pressure ( hydrostatic or chemical ) . an alternative way to arrive at the qcp is to apply an external magnetic field which breaks the local kondo singlets and leads to a field dependence of the critical t@xmath0 . starting from a noncritical or above critical t at zero field the system may then be tuned to the qcp by varying the field strength . this is indeed a practical method frequently applied @xcite . to investigate field - induced quantum phase transitions from kondo singlet to af ordered state or vice versa we have extended our previous work to include the effect of the magnetic field within the kondo necklace model . however , this introduces an additional parameter , namely the ratio of effective g- factors for the local kondo spins and the interacting spins . they can be different due to the different strength of spin - orbit coupling and crystalline electric field effects involved in the formation of the @xmath1 and * s * ( pseudo- ) spins . a mainly numerical study of the itinerant isotropic kl model in a magnetic field has been given previously @xcite . in sec . [ sec2 ] we define the anisotropic kn model in a magnetic field and in sec . [ sec2a ] briefly discuss the local state space of the kondo term in an external field . in sec . [ sec3 ] we perform the transformation from spin to bosonic variables and in sec . [ sec4 ] derive the selfconsistent equations for the mean field boson condensate amplitudes in the singlet and antiferromagnetic phases .. the influence of higher order terms in the hamiltonian is discussed in sec . [ sec4a ] . in sec . [ sec5 ] we investigate the numerical solutions and discuss the resulting quantum critical t - h phase diagram . a discussion is provided in sec . [ sec5a ] and sec . [ sec6 ] finally gives the conclusions . to investigate field induced quantum critical behaviour we start from the anisotropic kondo necklace ( kn ) model with u(1 ) symmetry @xcite where the field is applied along the anisotropy ( z ) direction : & = & _ t+_j+_z + & = & t_<n , m > ( _ n^x _ m^x + _ n^y _ m^y + m^z ) + j_n(_n^x s_n^x+_n^y s_n^y + _ n^z s_n^z ) + _ is_n^z+_i_n^z , [ ham ] here , the summation over nearest neighbor ( n.n . ) is indicated by brackets and @xmath3 is the @xmath4-component ( @xmath4 = x , y , z ) of the itinerant electron spin at site @xmath5 whereas @xmath6 is the @xmath4-component of localized spins at position @xmath5 . for the exchange coupling between the itinerant and localized spins we generally use @xmath7 as reference energy scale in all figures except when stated otherwise . the local anisotropy parameter @xmath8 is defined by the relation j@xmath9=@xmath8j@xmath10 between the z - axis and in - plane ( xy ) local exchange . the hopping parameter of the itinerant electrons is proportional to @xmath11 with the anisotropy in the z - direction given by @xmath12 . the present model has three control parameters : @xmath13 and the anisotropy parameters ( @xmath12,@xmath8 ) . in the zeeman term we defined @xmath14 and @xmath15 with h = @xmath16h where h is the strength of the applied field . furthermore @xmath17 and g@xmath18 are the gyromagnetic ratios for localised ( @xmath6 ) and itinerant ( @xmath3 ) spins respectively . they are determined by the combined effect of spin - orbit coupling and crystalline electric fields which depends on the degree of localisation or itineracy of electrons . therefore g@xmath19 and g@xmath18 in general need not be equal . we shall consider two extreme cases , namely ( @xmath20)=(2,0 ) and ( @xmath20)=(2,2 ) . the former is more realistic , since in real heavy fermion compounds most of the magnetic response is due to the localised electrons with pseudo - spin * s*. as in previous work @xcite the present study of field induced quantum phase transitions in the anisotropic kn model is based on the bond operator formulation . its hilbert space is spanned by local singlet - triplet states of ( @xmath21 ) spin dimers represented by bosonic degrees of freedom . in applying this technique to the finite field case we will largely use the same or similar notations as in the previous zero field case @xcite for consistency . before we perform the transformation to bosonic variables it is useful to have a clear understanding of the local singlet-triplet level scheme as function of anisotropy and magnetic field because certain tendencies of the quantum critical behaviour are correlated with the splitting of the ground state and the first excited state of a local bond . therefore we first diagonalise the local hamiltonian @xmath22 . this leads to eigenvalues @xmath23=e@xmath24/@xmath25 ( i=1 - 4 ) given by _ 1,2&=&_+ , + _ 3,4&=&(4+_-^2)^- , + _ & = & -(g_sg _ ) , [ llev ] with @xmath26 and @xmath27 . defining s@xmath28=@xmath29 + s@xmath9 then @xmath30 are the energies of the triplet states @xmath31 with s@xmath32 1 respectively which exhibit the linear zeeman effect . furthermore @xmath33 correspond to the triplet @xmath34 state @xmath35 and singlet state @xmath36 respectively . they show level repulsion in a magnetic field , except for g@xmath19=g@xmath18 when the total s@xmath28-component is conserved and commutes with @xmath37 . it is obvious that the spectrum of eigenstates in eq.([llev ] ) is invariant under the transformation ( g@xmath19,g@xmath18 ) @xmath38 ( g@xmath18,g@xmath19 ) . ) on the local anisotropy @xmath8 ( xy - side ) or @xmath39=1 - 1/@xmath8 ( ising side ) at zero field . numbers in parentheses denote the degeneracy of each level . bottom : field dependence ( h=@xmath16h ) of local energy level for three extreme cases . note that in the xy - case the excitation energy from ground state to first excited state vanishes asymptotically ( h@xmath40j@xmath10 ) whereas in the other cases it reaches a constant ( haf ) or is equal to a constant ( ising ) . here we used ( g@xmath19,g@xmath18)=(2,0).,title="fig:",width=302 ] + ) on the local anisotropy @xmath8 ( xy - side ) or @xmath39=1 - 1/@xmath8 ( ising side ) at zero field . numbers in parentheses denote the degeneracy of each level . bottom : field dependence ( h=@xmath16h ) of local energy level for three extreme cases . note that in the xy - case the excitation energy from ground state to first excited state vanishes asymptotically ( h@xmath40j@xmath10 ) whereas in the other cases it reaches a constant ( haf ) or is equal to a constant ( ising ) . here we used ( g@xmath19,g@xmath18)=(2,0).,title="fig:",width=302 ] the variation of zero - field energy levels with @xmath8 is shown in fig . [ fig1 ] ( top ) from the xy - u(1 ) limit ( @xmath8=0 ) via the heisenberg su(2 ) point(@xmath8=1 ) to the ising z@xmath41 limit ( @xmath8=@xmath42 ) . the field dependence for a few selected @xmath8-values and g - factors ( g@xmath19,g@xmath18)=(2,0 ) are shown in fig . [ fig1 ] ( bottom ) . for ( g@xmath19,g@xmath18)=(2,2 ) ( not shown in fig . [ fig1 ] ) the ground state and first excited state levels cross at h@xmath43/j@xmath10=0.25 , 0.50 for ( @xmath44)=(0,0 ) and ( 1,1 ) respectively . in such a case one may expect that the quantum critical t@xmath0(h@xmath43 ) for the transition from kondo singlet to af phase vanishes . we apply the transformation from spin variables ( @xmath45,@xmath46 ) to bond variables ( s , t@xmath47 ) ( @xmath4 = x , y , z ) to the hamiltonian in eq . ( [ ham ] ) . it is given by @xcite : s_n , = ( s^_n t_n , + t_n , ^ s_n -i _ t_n , ^t_n , ) , + _ n , = ( -s^_n t_n , - t_n , ^ s_n -i _ t_n , ^t_n , ) , [ bond ] where @xmath48 is the fully antisymmetric tensor . by construction the singlet ( s ) and triplet ( @xmath49 ) operators generate the local eigenstates of @xmath37 for zero field and no anisotropy @xcite : s^|0&=&(|-|);t_x^|0=-(|-| ) , + t_z^|0&=&(|+|);t_y^|0=(|+| ) . here the first and second arrows indicate the z - component of @xmath1 and @xmath50 spins respectively . the singlet and triplet operators satisfy the usual bosonic commutation relations according to [ s_n , s_n^]=1 ; [ t_n , , t_n , ^]= _ , ; = 0 . all other commutators vanish . the physical states have to satisfy the local constraint @xmath51 . this transformation leads to an effective hamiltonian in terms of singlet ( s ) and triplet ( t@xmath47 ) bosons . it has been argued in @xcite that one may restrict to the terms which are bilinear in the triplet bosons and we will later in sect . [ sec4a ] discuss to which extent this is justified . for the moment we restrict to the bilinear contribution . it can be written as the sum of a local term @xmath37 and an inter - site interaction term @xmath52 . here _ l=_j+_z&= & _ n ( -(2+)s_n^s_n + ( 2-)t_n , z^t_n , z + ( t_n , x^t_n , x+t_n , y^t_n , y ) ) + & & + _ -_n(s^_nt_nz+t^_nzs_n ) -i_+_n(t^_nxt_ny - t^_nyt_nx ) , [ haml ] contains the on - site kondo interaction ( @xmath53 ) and the zeeman term ( @xmath54 ) . the bilinear interaction term is given by ^(2)_1&=&_<n , m > _ = x , y ( s_n^ t_n , ( s_m^ t_m , + t_m , ^s_m)+ h.c . ) + & & + _ < n , m>(s_n^ t_n , z ( s_m^ t_m , z+t_m , z^s_m)+ h.c . ) . [ ham1 ] the physical constraint on the local hilbert space is enforced by adding a lagrange term at each site with an associated chemical potential @xmath55 , leading to = _ l+_n _ n ( s_n^s_n+ _ = x , y , z t_n , ^t_n , -1)+^(2)_1 _ 0+^(2)_1 . [ hammu ] this hamiltonian is diagonalised within a mean field approximation for the bond - operator singlet and triplet bosons . we assume that in general there are three independent bosonic amplitudes which characterise the phases in the ( t , h)-plane : singlet @xmath56 denoting the strength of local singlet formation , staggered triplet @xmath57 which determines the af order parameter ( polarised along x ) and homogeneous triplet @xmath58 which determines the magnetisation caused by the external field along z - direction . for technical reasons it is of advantage to transform to circular polarised transverse triplet coordinates ( u@xmath59,d@xmath59 ) with respect to the field direction ( z axis ) . the transformation and its inverse are given by u_n&=&-(t_nx - it_ny ) ; t_nx=-(u_n - d_n ) , + d_n&=&(t_nx+it_ny ) ; t_ny=-(u_n+d_n ) . [ circ ] in circular triplet coordinates the hamiltonian may then be written as _ 0&=&_n\ { [ -(2+)+_n]s_n^s_n+ [ ( 2-)+_n]t_nz^t_nz+ + & & [ + _ + + _ n]u_n^u_n+ [ -_++_n]d_n^d_n+ _ -(s_n^t_nz+t_nz^s_n ) -_n } , [ ham0 ] and ^(2)_1&=&-_n , m[(s_n^s_m^(u_nd_m+d_nu_m)+h.c . ) -(s_n^s_m(u_nu_m^+d_nd_m^)+h.c . ) ] + & & + _ n , m[(s_n^s_m^t_nzt_mz+h.c . ) + ( s_n^s_m t_nzt^_mz+h.c . ) ] . [ ham1ci ] separating the mean values and the corresponding fluctuations the fourier components of singlet and triplet operators are then given by s_k&=&|s , + u_k&=&|u_k , q + _ k , + d_k&=&|d_k , q + _ k , + t_k , z&=&|t_0_k,0+_k , z .[separate ] here we assumed that if af order appears it will be of the in - plane type ( @xmath60 ) . therefore the case of af order will only be considered for @xmath61 . instead of @xmath62 it will later be convenient to use @xmath63 which is defined by the relation @xmath64 . in the paramagnetic phase at zero field only @xmath65 will be different from zero with a value close to one @xcite . as the field increases @xmath65 will decrease and simultaneously the triplet amplitude @xmath62 associated with the uniform induced moment increases . if an af transition occurs , @xmath65 will further decrease and u , d will increase accordingly with the field . the constraint @xmath66 is always respected for any field strength . in the limit h / j@xmath10 @xmath401 the amplitudes @xmath65 and @xmath62 become asymptotically equal . on the mean field level this signifies that the ground state of the local level scheme becomes an equal amplitude mixture of the two states with s@xmath67=0 which show the level repulsion in fig . therefore it is possible to use the zero - field singlet - triplet bosons in eq . ( [ bond ] ) as a basis even in the present finite field problem . on the mean field level , the physical constraint automatically takes care of the change in the local ground state wave function . inserting the above expressions into the hamiltonian of eqs . ( [ ham0],[ham1ci ] ) one obtains a bilinear form in the triplet fluctuation operators ( @xmath68 , @xmath69 , @xmath70 ) which may be diagonalised with two separate bogoliubov transformations given for the z - polarisation by a_k=(_k , z)_k , z + ( _ k , z)_-k , z^ , + a_-k^=(_k , z)_k , z + ( _ k , z)_-k , z^ , [ bglz ] and for the two circular polarised triplets by [ bglper ] a_k&=&(_k,)_k + ( _ k,)_-k^ , + b_-k^&=&(_k,)_k + ( _ k,)_-k^ , + b_k&=&(_k,)_k + ( _ k,)_-k^ , + a_-k^&=&(_k,)_k + ( _ k,)_-k^.the transformation angles @xmath71 are obtained from the diagonalisation conditions 2_k , z=;2_k,=- [ angles ] where the longitudinal auxiliaryy functions f@xmath9 and d@xmath9 are defined by f_z(k)&= & ( k ) ; d_z(k)=++ ( k ) . [ fdz ] and for the in - plane case the auxiliary functions f@xmath72 and d@xmath72 are given by f_(k)&= & ( k ) ; d_u , d(k)=++ ( k)_+ , + d_(k)&=&(d_u+d_d)=++ ( k ) . [ fdp ] here @xmath73 with @xmath74 denotes the n.n . structure factor in dimension @xmath75 ( @xmath76 = coordination number of the simple cubic lattice ) . it should not be confused with the zeeman energies @xmath77 defined in sect . [ sec2a ] . the bogoliubov transformations in eq . ( [ bglz ] ) and eq . ( [ bglper ] ) yield the diagonalised bilinear hamiltonian _ mf = e_0+_k [ _ a(k)a^(k)a(k ) + _ b(k)b^(k)b(k)+_z(k ) a_k^a_k ] . [ hmf ] where the triplet mode frequencies @xmath78 ( @xmath4=a , b , z ) in the mean field hamiltonian are given by [ modefreq ] _ a(k)&=&_a(k)+_+ , + _ b(k)&=&_b(k)-_+ , + _ z(k)&=&_z(k ) , + _ a(k)=_b(k)&=&[d_(k)^2 - 4f_(k)^2]^ _ ( k ) , + _ z(k)&=&[d_z(k)^2 - 4f_z(k)^2]^ .the excitation energies @xmath78 depend on the field both explicitly through @xmath79 ( for @xmath4=a , b ) and implicitly via @xmath80 ( k ) which is determined by the field dependent singlet and triplet amplitudes . in the nonmagnetic phase ( @xmath81=0 ) the singlet - triplet excitation gap is given by the minimum excitation energy at the incipient ordering wave vector . for d=2 this is at q=(@xmath82 ) , explicitly @xmath83 . in the approach from the nonmagnetic side the quantum critical line t@xmath0(h ) is then defined by the vanishing of e@xmath84 . the ground state energy e@xmath85 is a function of three control parameters ( t / j@xmath10,@xmath8,@xmath12 ) , four singlet - triplet expectation values ( @xmath86 ) and the chemical potential @xmath87 . writing the transverse mean values explicitly in terms of amplitudes and phases according to |u = u(i_u ) , e_0 ( , , ; , , , ) & = & n[-(2j_x+j_z)^2+^2-+ + & & ( + -zt^2)(u^2+d^2)+_+(u^2-d^2 ) + t^2ud(_u+_d ) + & & + ( ( 2j_x - j_z)++zt^2)^2 + 2_- ] + & & + _ k[_k^-d _ ] .as a first step we determine the triplet condensate phases ( @xmath88 ) by minimization . since only one term depends on the phases we obtain the extremal condition @xmath89)=0 . the minimum must also satisfy @xmath90 . this is achieved for @xmath91 with n an odd integer , i.e. the sum is only determined modulo 2@xmath92 . since there is no condition for the difference of phases one of them is arbitrary . we then choose ( @xmath88)=(0,@xmath92 ) . the remaining continuous degeneracy with respect to the phase difference is a signature of the goldstone mode which is present throughout the af phase . ) and triplet ( @xmath93 ) amplitudes and their associated uniform ( @xmath94 ) and staggered ( @xmath95 ) moments as function of field strength normalised to the critical field in the ks and af region . here at the critical field @xmath96 . this case with t / t@xmath0(@xmath44)=0.6 ( t / j@xmath10 = 0.517 ) corresponds to the the upper curve of fig . [ fig4 ] ( top ) . here ( @xmath97)=(2.0).,width=302 ] the minimisation of the ground state energy e@xmath85 in eq . ( [ gsen ] ) leads to selfconsistent coupled equations for the condensate amplitudes @xmath98 and the chemical potential @xmath87 . their structure is slightly different in the nonmagnetic ( @xmath99=0 ) and magnetic ( @xmath100 ) case , therefore we write them explicitly for both . for a convenient expression of the selfconsistency equations we define the quantity @xmath63 by the relation @xmath101 . furthermore we introduce the brillouin zone integrals f@xmath47 and g@xmath47 ( @xmath4=a , b , z ) given by f_&=&_k , + g_&=&_k_k^. [ fg ] which appear in the extremal conditions when differentiating the last term in e@xmath85 with respect to @xmath87 or @xmath65 respectively . here we have defined @xmath102 and @xmath103 . note that even in the case of finite field ( @xmath104 ) it is the frequency @xmath105 and not the mode frequency @xmath106 which appears in the expressions for f@xmath47 and g@xmath47 . + ( @xmath12,@xmath8 ) for quantum phase transition between kondo singlet ( t@xmath107t@xmath0 ) and antiferromagnetic ( t@xmath108t@xmath0 ) phases at zero field . for ( @xmath12,@xmath8)=(0,@xmath8 ) or ( @xmath12,1 ) t@xmath0 is obtained from a closed analytical expression ( eq.(25 ) in ref . ) ( full lines ) , for the other cases it is calculated from the zero - field version of eq . ( [ para ] ) ( dashed lines ) . top : t@xmath0-dependence on anisotropy @xmath8 of local spins . in the ising case ( @xmath109 1 ) t@xmath0 for @xmath12=0 does not vanish because the af order is due to the mixing with the doublet separated by a finite excitation gap . bottom : t@xmath0-dependence on anisotropy @xmath12 of interacting spins . the value t@xmath0(0,1)/j@xmath10=0.7 agrees with results from mc simulatons for the genuine kondo lattice model in 2d . @xcite , title="fig:",width=302 ] + ( @xmath12,@xmath8 ) for quantum phase transition between kondo singlet ( t@xmath107t@xmath0 ) and antiferromagnetic ( t@xmath108t@xmath0 ) phases at zero field . for ( @xmath12,@xmath8)=(0,@xmath8 ) or ( @xmath12,1 ) t@xmath0 is obtained from a closed analytical expression ( eq.(25 ) in ref . ) ( full lines ) , for the other cases it is calculated from the zero - field version of eq . ( [ para ] ) ( dashed lines ) . top : t@xmath0-dependence on anisotropy @xmath8 of local spins . in the ising case ( @xmath109 1 ) t@xmath0 for @xmath12=0 does not vanish because the af order is due to the mixing with the doublet separated by a finite excitation gap . bottom : t@xmath0-dependence on anisotropy @xmath12 of interacting spins . the value t@xmath0(0,1)/j@xmath10=0.7 agrees with results from mc simulatons for the genuine kondo lattice model in 2d . @xcite , title="fig:",width=302 ] _ nonmagnetic kondo spin singlet phase in external field _ + in this case the transverse triplet amplitudes vanish , i.e. , u = d=0 and the minimization of e@xmath85 with respect to ( @xmath110 ) leads to the selfconsistent set of equations [ para ] ^2&= & , + & = & - , + & = & [ ( j_x+j_z)-_g_-2_--zt^2 ^ 2 ] .here @xmath111 are found by iteration which determines @xmath63 and hence @xmath62 completely via the second equation above . for zero field , i.e. @xmath112 , the induced @xmath63 vanishes and then eq . ( [ para ] ) reduces to the paramagnetic case of ref . + = singlet gap ) and antiferromagnetic af(m@xmath19 = staggered moment ) phases as function of increasing field strength h are shown : ( i ) field induced af with sequence ks -af . ( ii ) destruction of af and field induced gap opening corresponding to the sequence ks - af . ( iii ) reentrant behaviour corresponding to ks - af - ks . top : singlet - triplet excitation gap e@xmath84 and af order parameter m@xmath19 as function of external field for the two cases @xmath113 = ( 0,0 ) ( circles ) and @xmath113 = ( 1,1 ) ( diamonds ) as function of the external field . a subcritical scaled value t/ t@xmath0(@xmath44 ) = 0.6 is used in both cases ( t@xmath0(0,0)/j@xmath10=0.350 , t@xmath0(1,1)/j@xmath10=0.862 ) . here ( g@xmath19,g@xmath18)=(2,0 ) is used . this plot corresponds to the field induced af case ( i ) . bottom : similar plot for @xmath113 = ( 0,1 ) and an above critical t/ t@xmath0(@xmath44 ) = 1.1 ( t / j@xmath10=0.77 ) and subcritical t/ t@xmath0(@xmath44 ) = 0.94 ( t / j@xmath10=0.658 ) . in the former case ( ii ) one has the sequence af - ks of phases ( @xmath114 ) and in the latter ( iii ) a reentrant situation with ks - af - ks sequence is observed ( @xmath115).,title="fig:",width=302 ] + = singlet gap ) and antiferromagnetic af(m@xmath19 = staggered moment ) phases as function of increasing field strength h are shown : ( i ) field induced af with sequence ks -af . ( ii ) destruction of af and field induced gap opening corresponding to the sequence ks - af . ( iii ) reentrant behaviour corresponding to ks - af - ks . top : singlet - triplet excitation gap e@xmath84 and af order parameter m@xmath19 as function of external field for the two cases @xmath113 = ( 0,0 ) ( circles ) and @xmath113 = ( 1,1 ) ( diamonds ) as function of the external field . a subcritical scaled value t/ t@xmath0(@xmath44 ) = 0.6 is used in both cases ( t@xmath0(0,0)/j@xmath10=0.350 , t@xmath0(1,1)/j@xmath10=0.862 ) . here ( g@xmath19,g@xmath18)=(2,0 ) is used . this plot corresponds to the field induced af case ( i ) . bottom : similar plot for @xmath113 = ( 0,1 ) and an above critical t/ t@xmath0(@xmath44 ) = 1.1 ( t / j@xmath10=0.77 ) and subcritical t/ t@xmath0(@xmath44 ) = 0.94 ( t / j@xmath10=0.658 ) . in the former case ( ii ) one has the sequence af - ks of phases ( @xmath114 ) and in the latter ( iii ) a reentrant situation with ks - af - ks sequence is observed ( @xmath115).,title="fig:",width=302 ] _ antiferromagnetic phase in external field _ + the additional minimization with respect to u and d amplitudes leads to a direct relation for the chemical potential @xmath87 in terms of the singlet amplitude @xmath65 which is a generalised version of the zero - field af case @xcite : ( h)=zt^2-+[(zt^2)^2+_+^2]^ . at this stage it is convenient to introduce an auxiliary parameter @xmath116 which controls the effect of the external field on the boson amplitudes ; it is defined by ( h)=[1+()^2]^ + . [ kappa ] obviously @xmath116(0)=1 and @xmath117 since @xmath118 . using the expression for the zero - field chemical potential @xmath119=@xmath120 we may also write ( h)&=&_0+t^2 , + _ 0&=&t^2-j_z . we define a total transverse triplet amplitude @xmath99 by @xmath121 which should not be confused with the hopping matrix element t of the hamiltonian . the minimization with respect to u , d leads to a relation between the two amplitudes from which we obtain u^2+d^2=2 ^ 2 ; u^2-d^2=2 ^ 2 ( ) , [ ud1 ] or equivalently u^2= ; d^2==^2u^2 . [ ud2 ] the remaining set of singlet and triplet amplitudes ( @xmath122 ) is again determined by the solution of three coupled equations : [ af ] ^2&= & , + & = & - -2_-_0 , + ^2&=&[2-(2j_x+j_z)+2_-+ _ g_]+^2 ^ 2 .these amplitudes are related to the spin expectation values by s_z&=&+(u^2-d^2 ) + _ z&=&-+(u^2-d^2 ) [ spinav ] for zero field ( @xmath63=0 ) the system of equations in eq . ( [ af ] ) may be shown to reduce to the one already discussed in ref . [ ] . in this case @xmath116=1 and u = d , i.e. , circular polarised triplet modes with equal amplitudes . finally we discuss the uniform magnetisation m@xmath85 and the staggered magnetisation m@xmath19=m@xmath123=-m@xmath124 associated with af order . by definition m_0&=&g_ss_z+g__z + m_s&=&g_ss^b_x+g_^b_x . [ unsta ] here the staggered moment is assumed to be polarised along x. using the expressions in eq . ( [ spinav ] ) for @xmath125 , @xmath126 and similar ones for @xmath127 , @xmath128 we obtain with the help of eqs . ( [ kappa],[ud1 ] ) m_0&=&2g_-2g_-m_0 , + m_s&=&2g_-2 g _ -m_s . [ magnet ] here we used @xmath27 and defined = . when both staggered and uniform component are nonzero the total magnetic moment is canted . the canting angle @xmath4 , counted from the xy plane is then given by [ canting ] =( ) . ( h ) curves ) for various cases of anisotropies ( @xmath12,@xmath8 ) and g - factors ( g@xmath19,g@xmath18 ) . here circles ( @xmath129 ) correspond to ( @xmath12,@xmath8)=(0,0 ) stars ( @xmath130 ) to ( @xmath12,@xmath8)=(1,1 ) and triangles ( @xmath114 ) to ( @xmath12,@xmath8)=(0,1 ) . full symbols are obtained by m@xmath19=0 from the af side , open symbols by e@xmath84=0 from the kondo singlet side . for unequal g - factors the heisenberg case shows saturation at @xmath131 0.4 at larger fields , in the ising case t@xmath0 continues decreasing . for the mixed case ( @xmath12,@xmath8)=(0,1 ) ( the genuine kn model in 2d ) and unequal g - factors the t@xmath0(h ) is nonmonotonic . in each case the ks phase is below and the af phase above the corresponding t@xmath0(h ) curve . a decreasing t@xmath0(h ) leads to the phase sequence ks - af ( field- induced af ) whereas an increasing t@xmath0(h ) entails the opposite sequence af - ks . in the nonmonotonic region the reentrance behaviour ks - af - ks is observed . for comparison see fig . for equal g - factors in all cases t@xmath0 decreases monotonically and vanishes at rather small fields.,width=302 ] the terms @xmath132 are higher order corrections , e.g. @xmath133 for m@xmath19 . for moderate fields they may be neglected . in sec . [ sec5 ] ( fig . [ fig4 ] ) we only show the main contribution @xmath134 in accordance with ref . . ( [ canting ] ) a similar small field approximation may be employed for the canting angle leading to @xmath135 . for equal g - factors g@xmath19=g@xmath18 ( g@xmath136=0 and @xmath62=0 ) one has a special case : m@xmath19 = 0 due to moment compensation of local and itinerant spins , although this is still a phase with an af order parameter m@xmath19=@xmath137 . the uniform moment m@xmath85 is now given by @xmath138 , oriented parallel to the external field ( @xmath139 ) . + _ the ground state energy : condensation vs. fluctuation _ + it is instructive to rewrite the ground state energy in a different way which renders a clearer interpretation of its individual contributions . the total mean field hamiltonian in eq . ( [ hmf ] ) may be rewritten as _ mf=_0+_=z , a , b_k_(k)[n_+ ] , [ hmftil ] where n@xmath47 is the occupation number of bosons for each mode . the total ground state energy is then given by e_0=_0+_k_(k ) , [ gsentot ] where @xmath140 consists of three parts according to [ gsentil ] _ 0/n&=&-[(2j_x+j_z)(^2+)+(-^2 ) ] + & & + t(-1)^2 ^ 2 ^ 2+((2j_x - j_z)++t^2)^2 + & & + 2_-.here the first part ( @xmath141 ) in eq.([gsentot ] ) is the condensate contribution . the second always positive term is due to triplet quantum fluctuations . according to eq.([gsentil ] ) @xmath141 is composed of three parts : the first ( negative ) contribution is due to the singlet formation . the second ( positive ) part is due to the field induced triplet polarisation . the last term is the zeeman energy contribution . for zero field only the first term in eq.([gsentil ] ) is present @xcite . the ground state with @xmath142 will be determined by the competition of these terms . for example when h = 0 the quantum critical point where af order appears will be determined by the balance of two terms : the negative singlet formation energy and the positive energy of triplet quantum fluctuations . the transformation of the hamiltonian in eq . ( [ ham ] ) to bond operator variables also creates third and fourth order terms in the triplet operators . they have been neglected in the previous analysis based on eqs . ( [ haml],[ham1 ] ) . in fact it was argued in @xcite that third order terms do not contribute to the ground state energy and fourth order terms are quantitatively negligible . the latter was also found for the related bilayer heisenberg model @xcite . hence higher order terms have no influence on the zero - field quantum critical properties . in this section we investigate to what extent this is still justified in the presence of an external field . indeed in mean field approximation used here the higher order terms lead to field induced effective bilinear triplet terms which have to be added to the genuine bilinear term in eq . ( [ ham1ci ] ) which is present already at zero field . the quantitative influence of higher order terms on t@xmath0(h ) or h@xmath0(t ) is controlled by the ratio @xmath143 at the critical field h@xmath0 where @xmath144 is the field induced triplet amplitude . if it is still moderate compared to the singlet amplitude @xmath145 then higher order terms have negligible influence . the field dependence of these amplitudes as obtained from eqs . ( [ para],[af ] ) is shown in fig . [ fig2 ] for a typical case . the contribution of third order in t@xmath146 ( n= site , @xmath4=x , y , z ) to the interaction term is given by @xcite : ^(3)_1&=&_n , m [ ( s_n^t_nx+t_nx^s_n)(t^_my t_mz - t^_mz t_my ) + ( s_n^t_ny+t_ny^s_n)(t^_mz t_mx - t^_mx t_mz ) + & + & ( s_n^t_n , z+t_n , z^s_n)(t^_m , x t_m , y - t^_m , y t_m , x ) ] . [ ham3 ] which has to be added to eq . ( [ hammu ] ) . we are only interested in the influence of this term on the quantum critical lines h@xmath0(t ) or t@xmath0(h ) to be discussed in sect . as explained there this may be achieved both from the paramagnetic ( ks ) and af side of the critical line . here we consider only the former for reasons explained at the end of the section . in this case the circular triplet amplitudes @xmath147 . the mean field approximation to h@xmath148 then contains only terms proportional to @xmath149 and no constant contribution to e@xmath85 appears . transforming these terms to circular triplet coordinates one finally obtains another bilinear contribution with ^(3)_1mf = -z_n(u^_nu_n - d^_nd_n ) + _ nm(u^_nu_m - d^_nd_m ) . [ ham3mf ] this contribution is field induced since for small fields @xmath150 . the first single site term in eq . ( [ ham3mf ] ) has the same structure as the @xmath79 part of the zeeman term and may be accommodated by a simple ( nonlinear ) rescaling of the applied field such that @xmath79 is replaced by @xmath151 according to [ hcscale ] _ + ( h)=-g_+hf_s(h ) ; f_s(h)=1+t here @xmath152 is the rescaling function for the applied field . note that for @xmath153 one has @xmath154 , i.e. no rescaling will occur in this case . the second contribution in eq . ( [ ham3mf ] ) is an interaction to which the same ( transverse ) bogoliubov transformation as before may be applied . these terms then have a simple effect : in eq . ( [ fdp ] ) one has to replace @xmath155 . however only the average @xmath156 of the auxiliary functions , which is unchanged , enters the expression for @xmath157 in eq . ( [ modefreq ] ) . therefore @xmath157 is also unchanged by the third order contribution , as a result the total energy e@xmath85 in eq . ( [ gsen ] ) will be exactly the same as before . also the selfconsistent equations eq . ( [ para ] ) will be unchanged . note that in these equations @xmath158 is not rescaled . the only effect of the third order terms is therefore the above rescaling of the external field in the @xmath79 zeeman term leading to the modified transverse mode frequencies _ a , b(k)=_a , b(k)_+ . therefore the critical field h@xmath0 of the quantum phase transition which is defined as the field where one of the above modes ( @xmath159 for @xmath160 ) vanishes at the af wave vector * q*=@xmath161 will be changed by the scaling factor @xmath162 . if h@xmath163 is the critical field without third order terms then approximately @xmath164 is the critical field with the effect of third order terms included . this is only a quantitative modification which we will discuss in sec . [ sec5 ] in connection with fig . the qualitative topology of the phase diagram will not be changed by the third order term . we note again that in the special case @xmath165=@xmath166 the third order term does not have any effect at all because @xmath167 . furthermore for all cases g@xmath19=g@xmath18 there is no field induced triplet amplitude , i.e. @xmath168 according to eq . ( [ para ] ) leading again to @xmath167 . we conclude that the influence of third order terms is not important for the field induced quantum critical behaviour . + now we discuss the effect of terms which are of fourth order in the triplet operators . they are given by @xcite ^(4)_1&=&_n , m((t^_ny t_nz - t^_nzt_ny ) ( t^_my t_mz- t^_mzt_my ) + ( t^_nx t_nz- t^_nzt_nx)(t^_mx t_mz - t^_mzt_mx ) + & + & ( t^_nx t_ny- t^_nyt_nx ) ( t^_mx t_my- t^_myt_mx ) ) . the mean field approximation to @xmath169 for the nonmagnetic case ( @xmath170 ) does not produce a constant term but leads to an effective bilinear contribution in circular triplet coordinates ^(4)_1mf=^2_nm [ u_n^u_m + d_n^d_m + ( u_nd_n+d_nu_m+u_n^d_m^+d_n^u_m^ ) ] . again this is a field - induced bilinear contribution with @xmath171 for small fields . due to @xmath172 it has no longitudinal part . it may be diagonalised by the same transverse bogoliubov transformation in eq . ( [ bglper ] ) as before . the new mode frequencies including fourth order contributions are simply obtained by the replacement @xmath173 and @xmath174 in eq . ( [ fdp ] ) . this will lead to the modified transverse frequency _ a , b(k)^2=_a , b(k)^2 [ 1+^2 ] , and new mode frequencies @xmath175 at the quantum critical point where @xmath176 this may be approximately written as _ a , b(k)_a , b(k ) [ 1+()^2_c(k ) ] , [ fourthc ] where the index c denotes the values of singlet and triplet amplitudes at the qcp . from fig . [ fig2 ] we can estimate that the prefactor in eq.([fourthc ] ) is of the order @xmath177 since at the qcp the triplet amplitude is still quite small and the singlet amplitude basically unchanged from the zero - field value . therefore the fourth order terms lead to corrections of the order one per cent in the frequencies @xmath157 which determine the last term in the ground state energy of eq . ( [ gsen ] ) . indeed the correction to @xmath178 is even smaller since the momentum summation over @xmath157 in the last term of eq . ( [ gsen ] ) leads to a large amount of cancellation because @xmath179 is positive in one half of the brillouin zone and negative in the other half . therefore the selfconsistent equations for @xmath65 and @xmath62 are nearly unchanged and we can conclude that the field induced quantum critical behaviour is not influenced by the inclusion of fourth order terms . our analysis shows that the quantum critical lines h@xmath0(t ) are only marginally influenced by the presence of higher order terms in one example and have no effect in other cases . however for fields with @xmath180 one would expect that the higher order terms should lead to non - negligible corrections , at least in the case of third order contributions . from fig . [ fig2 ] one can see that for @xmath181 the ratio @xmath143 approaches 0.5 and the bilinear approximation might become inadequate . in fact , by its very construction the bond - operator method is a strong coupling theory which assumes the dominance of the singlet state . it can not be expected to be quantitatively correct for high fields when saturation of moments is approached , i.e. when singlet and triplet amplitudes become equal . in this limit it is more appropriate to start from the polarised canted state and perform a conventional spin wave expansion for the two types of spins . + as a consequence of the above analysis we will only consider the genuine bilinear hamiltonian eq . ( [ ham1ci ] ) as used in the previous sections for the following numerical calculations . this restriction has an additional reason : although the critical @xmath182 or t@xmath0(h ) may be calculated by searching for the vanishing of the excitation gap e@xmath84 from the nonmagnetic singlet side , it is more satisfactory to search also from the af magnetic side for the vanishing of the staggered magnetisation m@xmath19 and confirm the agreement . this can not be done easily when third and fourth order terms are included . on the af side the latter lead to a mixing of the longitudinal and transverse modes and a closed analytic form of the bogoliubov transformation can not be found . therefore it is better to include only the bilinear terms and be aware of the trivial effect of field rescaling by the third order terms and the tiny effect of fourth order terms . this standpoint will be adopted for the following numerical analysis . we shall now discuss the numerical solutions of the selfconsistent eqs . ( [ para],[af ] ) which describes the af - ks quantum phase transitions , i.e. , the field dependence of the kondo singlet gap e@xmath84 and the field dependence of the staggered magnetisation m@xmath19 . the former was defined below eq . ( [ modefreq ] ) and the latter in eq . ( [ magnet ] ) . we shall only discuss the results for d=2 and @xmath61 , i.e. , the easy xy - plane situation with @xmath183 c. in this case @xmath184 . as shown before the qualitative behaviour of d=3 is similar to d=2 @xcite . the main purpose is to study the dependence of the quantum critical point t@xmath0(h ) on the external field , or equivalently the dependence of the quantum critical field h@xmath0(t ) on the hopping strength t. to check consistency the qcp has been obtained both by variation of t and of h and both from the paramagnetic ( e@xmath84=0 ) as well as the antiferromagnetic ( m@xmath19=0 ) side of the qcp . as mentioned in the previous section this requires restriction to the genuine bilinear hamiltonian terms . + of total moment moment ( counted from the xy - plane ) in the af phase of two different cases ( cf . fig . [ fig4 ] ) with ( g@xmath19,g@xmath18)=(2,0 ) . full curve is obtained from the exact expression eq . ( [ canting ] ) and dashed curve from the approximate expression @xmath185 . at the quantum critical field h@xmath0 the staggered moment vanishes and the total moment is aligned with the external field ( @xmath139 ) . here h@xmath0(0,1)/j@xmath10=1.23 and h@xmath0(1,1)/j@xmath10=0.55.,width=302 ] before discussing the field dependent results we show the behaviour of the quantum critical @xmath186 as function of the anisotropies in the zero field case as a starting point of our analysis . the calculation of @xmath186 has been described in ref . [ ] . here we present the results for the full range of anisotropies ( fig . [ fig3 ] ) from the u(1 ) -xy case via heisenberg- su(2 ) to ising - z@xmath41 symmetry . note the two different scales on the left ( j@xmath10 ) and right ( j@xmath9 ) half of each part of fig . the quantum critical t@xmath0 for the ks - af transition reaches a singular maximum for the heisenberg case and generally vanishes for the ising case , except for @xmath12=0 . this may be understood by comparing with fig . [ fig1 ] . for @xmath187 ( i.e. @xmath188 ) the two singlets with s@xmath28=0 become degenerate and an arbitrary small interaction t leads to the af state . however for @xmath12=0 the s@xmath28=0 states are not connected by a matrix element of @xmath189 and hence af can occur only via matrix elements between the singlet ground state and the excited doublet which requires a finite t@xmath0/j@xmath10 = 0.35 for af order to occur . + the effect of the external field for ( g@xmath19,g@xmath18 ) = ( 2,0 ) is shown in fig . [ fig4 ] ( top ) for the two cases ( @xmath190 ) = ( 0,0 ) ( @xmath129 ) and ( 1,1 ) ( @xmath191 ) . their corresponding t@xmath0 values may be taken from fig . [ fig3 ] . in fig . [ fig4 ] ( top ) we choose a subcritical value t / t@xmath192 = 0.6 meaning a nonmagnetic singlet phase with finite e@xmath84 exists for zero field . when the field is increased the gap is gradually reduced until it is closed at t@xmath0(h@xmath0 ) . for the ( normalised ) t value of 0.6 obviously the difference in critical fields h@xmath0 is not significant . however , the qualitative behaviour of the spin gap e@xmath84 in the two cases for h@xmath107h@xmath0 is different with a much smaller slope of e@xmath84 in the xy - case as compared to the heisenberg case . in both cases for h@xmath108h@xmath0 a field induced easy plane af order parameter appears , i.e. , a phase sequence ks - af for increasing field . the opposite behaviour is observed in the genuine kn case ( @xmath190 ) = ( 1,0 ) with ( g@xmath19,g@xmath18 ) = ( 2,0 ) in fig . [ fig4 ] ( bottom ) . for an above- critical value t / t@xmath193 = 1.1 ( t / j@xmath10=0.77 ) the af order parameter is gradually suppressed until it vanishes at h@xmath0/j@xmath10 = 1.23 and the kondo singlet phase appears , i.e. , the opposite phase sequence af - ks is realised . finally , for slightly subcritical t / t@xmath193 = 0.94 ( t / j@xmath10=0.658 ) an interesting reentrance sequence ks - af - ks of phases as function of field is observed . it exists only in a narrow range of subcritical values @xmath194 . as discussed in sect . [ sec4a ] this behaviour is robust because third order triplet contributions are exactly zero for @xmath12=0 . + we may collect the data of quantum critical fields h@xmath0 from numerous calculations such as presented in fig . [ fig4 ] and similar ones for quantum critical t@xmath0 at fixed h in a h - t phase diagram . it is shown in the form of t@xmath0(h ) curves in fig . [ fig5 ] for two choices of the g - factors . for ( g@xmath19,g@xmath18 ) = ( 2,0 ) and the cases ( @xmath8,@xmath12)= ( 0,0 ) , ( 1,1 ) the scaled t@xmath195 is identical close to the zero - field qcp , i.e. , the slope does not depend on the anisotropy ( @xmath8,@xmath12 ) . for larger fields one observes the plateau formation in the case ( @xmath8,@xmath12)=(1,1 ) and the further monotonic decrease for ( @xmath8,@xmath12)= ( 0,0 ) . this behaviour may qualitatively be understood from the field dependence of the local energy levels of @xmath53 shown in fig . [ fig1 ] . for increasing h the gap between the ground state and first excited state decreases and hence a smaller t@xmath0 is necessary to achieve the softening of the triplet excitation at q. for @xmath8=1 this effect eventually levels off because the splitting of the two lowest local levels becomes constant at large field and approaches 2@xmath8j@xmath10 so that t@xmath195 reaches a plateau at large fields for nonzero @xmath8 . on the other hand for @xmath8=0 the two lowest levels become asymptotically degenerate and hence t@xmath0(h ) should approach zero for large fields . as already noted in the discussion of fig . [ fig4 ] the intermediate case ( @xmath8,@xmath12 ) = ( 1,0 ) and ( g@xmath19,g@xmath18 ) = ( 2,0 ) behaves quite differently . after an initial but much weaker decrease of t@xmath0(h ) it reaches a minimum at ( h / j@xmath10,t / j@xmath10)=(0.275,0.638 ) and then starts to increase again . the region of the nonmonotonic behaviour with slightly subcritical t / t@xmath0(0,1 ) corresponds to the region where reentrance behaviour ks - af - ks is observed in fig . [ fig4 ] . for above critical t / t@xmath0(0,1 ) , once t@xmath0(h)@xmath108 t the initial af phase is suppressed and we obtain the sequence af - ks , i.e. a field induced ks phase corresponding to fig .. [ fig4 ] ( bottom ) . in all three cases with ( g@xmath19,g@xmath18 ) = ( 2,0 ) one has a finite t@xmath0(h ) for finite fields . + this is different if we chose equal g - factors ( g@xmath19,g@xmath18 ) = ( 2,2 ) . for this choice the lowest two local levels cross at h=0.25 ( @xmath8=0 ) and at h=0.5 ( @xmath8=1 ) ( not shown in fig . [ fig1 ] ) . this is due to the fact that s@xmath28 commutes with the hamiltonian and hence there is no level repulsion of the two s@xmath67=0 states . consequently at this point the quantum critical t@xmath0(h ) vanishes in all three cases of ( @xmath8,@xmath12 ) . it decreases essentially linearly from its maximum value at h=0 . the phase diagram in fig . [ fig5 ] shows a collection of the critical t@xmath0(h ) curves for the six cases considered . the corresponding gapped ks and af ordered phases are below and above each curve respectively . we comment again on the influence of third order terms discussed in sect . [ sec4a ] on the phase diagram of fig . remembering that all cases with @xmath196 or @xmath197 are unaffected by the third order contributions , only one of the critical field curves in fig . [ fig5 ] ( @xmath130 with ( g@xmath19,g@xmath18 ) = ( 2,0 ) ) may be influenced . in this case third order terms lead to an almost field independent and moderate rescaling of h@xmath0 by a factor @xmath1981.2 according to eq . ( [ hcscale ] ) and fig . [ fig2 ] . + finally in fig . [ fig6 ] we show the dependence of the tilting angle of the total moment out of the xy - plane for two possible cases with af order . it is calculated from eq . ( [ canting ] ) ( full line ) and a simplified ( low field ) expression given in the caption . close to the quantum critical field the staggered moment vanishes and the total moment is then alligned with the external field ( @xmath199 z ) , meaning @xmath139 . as we have mentioned in the introduction the interest for the kondo necklace ( kn ) model stems from two facts : ( i ) it is thought to be a simplified version of the original kondo lattice ( kl ) model with itinerant electrons and ( ii ) it is expected to describe some basic features of the observed quantum phase transitions in real kondo materials . although there is no mapping or exact equivalence of the two models the agreement of their qualitative features by simply identifying the intersite interaction ( kn ) and hopping ( kl ) energies ( both given by t ) is generally quite good . as discussed in @xcite the 1d kondo necklace model does not exhibit quantum critical behaviour , this agrees with earlier numerical work on the kl model @xcite where the disordered spin gapped state always prevails @xcite . + we discuss the more interesting 2d case where quantum phase transitions in the kn model are possible @xcite . one can compare the values of the quantum critical t@xmath0/j ( we set j@xmath200 j@xmath10 in this section to comply with literature conventions ) obtained for the kn model from fig . [ fig3 ] with those for the kl model obtained in mc simulations obtained previously @xcite . in the spirit of doniach s replacement in 1d we should compare with the kn model for xy - type interactions , i.e. with the @xmath201 case in fig . [ fig3 ] which has a t@xmath0/j=0.7 . the mc simulations for the 2d su(2 ) fermionic kl model lead to quantum critical parameters ( average values ) t@xmath0/j= 0.71 @xcite , 0.69 @xcite and 0.68 @xcite , very close to the appropriate value in fig . [ fig3 ] ( upper panel : center of full line ; lower panel : left corner of full line ) . note that the critical t@xmath0 s corresponding to full lines in fig . [ fig3 ] are obtained from a simple analytical formula within the bond - operator mean field approach @xcite . further support for this scenario comes from exact diagonalisation ( ed ) results for the kl model on small clusters @xcite using open boundary conditions . the total average local moment @xmath202 calculated as function of j / t shows a reduction to half of the free moment size due to kondo singlet formation in the lattice for the value t@xmath0/j = 0.67 . if it is interpreted as the critical value in the thermodynamic limit this is again close to the above results . this near equality of analytical kn and numerical kl results seems to suggest that even in 2d the su(2 ) kl model is better described by the u(1 ) xy - type kn model with ( @xmath44)=(0,1 ) since the su(2 ) kn model with ( @xmath44)=(1,1 ) has a considerably larger critical value @xmath203 . + the fermionic kl model is obtained from the periodic anderson model ( pam ) by a schrieffer - wolff transformation which eliminates charge fluctuations of f - electrons . the resulting kondo coupling j is then given in terms of the anderson model parameters by @xmath204 where @xmath205 is the f - level position with respect to the fermi level at zero energy , u is the on - site f - electron repulsion and v their hybridisation strength with conduction electrons . the periodic anderson model may be studied with the dynamical mean field theory ( dmft ) @xcite . this method has been used more recently to investigate its quantum critical properties @xcite which should be related to those of the kondo lattice and hence also kn models . the method is formally a d=@xmath42 approximation but has nevertheless been used for studying electron correlations in finite dimensional lattices such as the 3d pam in @xcite . using the value of j from the schrieffer - wolff transformation the extrapolated t= 0 phase diagram suggests that the critical hopping strength at the qcp is given by @xmath206 . this is again in reasonable agreement with the mean - field bond operator result for the 3d kn model which predicts @xmath207 @xcite . + while the quantum critical behaviour as function of control parameter t / j is well presented , there are few microscopic investigations concerning the field induced quantum critical behaviour of kl / kn type models such as provided here . the fermionic 2d kl model in an external field with equal g - factors corresponding to ( g@xmath19,g@xmath18 ) = ( 2,2 ) was studied in ref . by using variational and mc methods . the numerical results for j / t =3 suggest that the transition between kondo and af ordered regime takes place in a field region ( b@xmath208 in notation of ref . ) between b@xmath209/t = 1.0 and n b@xmath210/t = 2.25 . in this regime both kondo - like features and transverse magnetic order coexist . from the present calculations of the 2d kn model again with ( @xmath44)=(0,1 ) , g - factors as above and using the data shown in figs . [ fig3],[fig5 ] one obtains b@xmath0/t = 1.86 ( b@xmath0/j=0.62 ) which is close to the average 1.63 of the above b@xmath211 values . this value for the critical field was also confirmed by exact diagonalisation ( ed ) results for the kl model on small clusters @xcite where the breaking up of on - site singlet formation was observed above the value b@xmath0/j @xmath212 . we conclude that kl and kn models also seem to exhibit similar field induced quantum critical behaviour . another matter is the scaling behaviour of singlet gap and staggered magnetisation around the critical field . the present mean field type theory does not give sufficient insight into that issue . it has sofar been treated within continuum field theories involving order parameter fluctuations around the critical field @xcite . + the present bond operator mean field results for the kn model are apparently consistent with known numerical results for the fermionic kondo lattice models . it is less clear whether they are useful for the interpretation of experimental results , especially for the field induced quantum phase transitions . since charge degrees of freedoms are eliminated in the kn model it is strictly speaking more appropriate for the kondo - insulator compounds which have a hybridisation gap due to half filled conduction bands . however we ignore this subtlety in the following and also apply it to the magnetism of metallic kondo compounds . the zero field quantum critical behaviour as function of the control parameter t / j can experimentally be mimicked by applying hydrostatic or chemical pressure ( by substitution of elements ) . this changes mostly the hybridisation and hence the kondo coupling j. in the accessible pressure regime one may assume that t / j varies linearly with pressure . in this manner it is feasible to drive af ordered heavy fermion systems to the quantum critical point where they become nonmagnetic heavy fermi liquids . since hydrostatic ( positive ) pressure generally increases j it tends to suppress the af phase while with suitable substitution of elements ( negative ) chemical pressure may decrease j which favors af order . thus the af qcp may be approached from both sides . there are many examples to be found especially among ce- compounds , for a review see @xcite . in most cases of ce - compounds , however , the af qcp may not be reached directly because it is enveloped by a dome of the superconducting phase and the critical pressure has to be obtained from extrapolation to t=0 . while the kl / kn type models discussed above naturally suggest the af qcp there is no direct way of experimental determination of the critical ( t / j)@xmath0 since its relation to the experimentally accessible critical pressure p@xmath0 ( or critical concentration of substituent ) is unknown . in fact most of the interest on pressure induced qcp s focuses more on the scaling exponents of various quantities with respect to the distance ( p - p@xmath0 ) to the qcp . this is extensively discussed in the reviews cited in the introduction . + for the field induced quantum phase transition much less experimental results are available . a review of materials investigated is given in ref . . a recent example of a field induced destruction of af order is the tetragonal ybrh@xmath41si@xmath41 compound @xcite . at ambient pressure and zero field it has an af order which is indeed of the easy plane ( xy)-interaction type ( @xmath12=0 ) . the af order is destroyed at 0.66 t for field along the hard c - axis and a nonmagnetic fermi liquid state appears above the critical field . if one assumes that the local kondo interaction which results from the schrieffer - wolff mechanism is more of the isotropic nature ( @xmath8=1 ) it qualitatively corresponds to the af - ks scenario shown in the lower panel of fig . [ fig4 ] . as mentioned before the reentrant scenario shown also in this figure exists only in a narrow parameter range and it is perhaps understandable that no such example is known . what is really surprising is the following observation : most of the parameter cases in fig . [ fig5 ] would predict the ks - af sequence , i.e. the field - induced af order out of the kondo phase . this is a natural consequence that in most cases a field reduces the singlet - triplet gap and supports the onset of magnetic order as shown in the upper panel of fig . we would like to stress that this is also the only phase sequence found in the fermionic kl model @xcite . however field induced af magnetic order is not easily found in heay fermion compounds . for example ceni@xmath41ge@xmath41 at ambient pressure and zero field is in the kondo singlet phase and rather close to the af quantum phase transition which may be achieved by appropriate substitution of ni . thus it is a complementary case to ybrh@xmath41si@xmath41 . however in an external field there is no field induced transition to transverse af order as one might naively expect from the model discussion , rather it is driven further away from the af qcp as specific heat and resistivity measurements suggest @xcite . there is however one known example of a heavy fermion system which exhibits field induced af order , though complicated by the appearance of superconductivity . the compound cerhin@xmath213 is already an af at ambient pressure @xcite . application of hydrostatic pressure moves it to the nonmagnetic ( ks ) side where it also becomes superconducting . additional application of a magnetic field destroys the superconductivity and leads to the reestablishment of a field induced af order in a wide range of pressure form 1.4 to 2.4 gpa @xcite . this is indeed the ks - af sequence of phases corresponding to fig . [ fig4 ] ( top ) which appears in the majority of cases studied here . experimentally , however , it seems to be the one most rarely encountered . + these examples suggest that while the kondo lattice or necklace type models have some qualitative and instructive properties which are relevant for magnetic quantum phase transitions they may not be able to explain some experimental observations in heavy fermion compounds . it is well possible that the complete elimination of charge degrees of freedom is too radical by suppressing hybridisation effects of moments with conduction electrons and thus favoring the magnetic phases . therefore the more fundamental pam model , including appropriate local and conduction electron zeeman terms may be a more promising starting point for future investigations of field induced quantum phase transitions in heavy fermion compounds . in this work we have investigated the general anisotropic kondo necklace model and its quantum phase transition from paramagnetic singlet phase to af phase and vice versa in an external field . we have derived a bosonic hamiltonian using the bond operator representation of local and interacting spins and diagonalised the bilinear part by a generalised bogoliubov transformation to circular polarised modes to include the effect of the external field . higher order parts of the hamiltonian were found to be insignificant for the quantum critical behaviour . in mean field approximation one obtains a total ground state energy that depends on the singlet amplitude @xmath65 , staggered and uniform triplet amplitudes @xmath214 and the chemical potential . minimization leads to two sets of selfconsistent equations for kondo singlet and af phase , respectively . from their numerical solution we find that in most investigated cases the t@xmath0 value of the quantum critical point decreases monotonously with field strength leading to the kn - af phase sequence for increasing field . in the case of unequal g - factors and ( @xmath8,@xmath12 ) = ( 1,1 ) it levels off at a plateau value t@xmath195/t@xmath215 0.4 while for ( @xmath8,@xmath12 ) = ( 0,0 ) it continues to decrease at larger field . however for the genuine kn case with ( @xmath8,@xmath12 ) = ( 1,0 ) the t@xmath0(h ) curve is nonmonotonic with a minimum at @xmath216 0.275 at small fields and continues to increase for larger fields . depending on the size of t this implies two possibilities : for above critical t a suppression of af and the opening of a spin excitation gap for @xmath217 ( af - kn sequence ) or the reentrance behaviour ( kn - af - kn sequence ) of the gapped singlet phase for slightly subcritical t ( as compared to t@xmath0(0 ) ) . for equal g - factors a crossing of local levels occurs due to a conserved s@xmath67 and the critical t@xmath0 reaches zero already at relatively small fields for all investigated cases of ( @xmath44 ) . since our treatment is of mean - field type we will get mean - field exponents for the spin gap e@xmath84 and magnetisation m@xmath19 at the field - induced qcp . this is apparent from fig.[fig4 ] and has not been discussed further . improvement for the critical exponents will require a selfconsistent renormalisation theory for the triplet excitations close to the quantum critical point . the authors would like to thank v. yushankhai for helpful suggestions . a. l. would like to acknowledge the support of sharif university of technology . l. also would like to acknowledge the max planck institute for the physics of complex systems - dresden where the initial part of this work was started during his visit . 99 a. langari and p. thalmeier , phys . rev . b * 74 * , 024431 ( 2006 ) .
in the last decade , studies about the structure and dynamics of complex networks have blossomed , thanks in particular to the versatility of the network representation , which has turned out to be adequate for systems as diverse as the internet or social networks . a large body of knowledge about the empirical description of networked systems has thus been accumulated , together with a wealth of modeling techniques ; a good level of understanding of how dynamical processes taking place on networks depend on their structure has been as well reached @xcite . many network studies have been concerned with systems of interest in several scientific areas a priori remote from physics ( social sciences , biology , computer science , epidemiology , ... ) , and they have also reached more traditional fields of statistical physics , such as the study of glassy systems , as we now describe . the many puzzles raised by the glass transition , and in particular the slow dynamics displayed by glassy systems at low temperatures , have been the subject of a large interest in the past decades @xcite . one of the approaches which has led to promising insights consists in the description of the dynamics of a glassy system inside its configuration space . the energy landscape of a glassy system is typically rugged , made of many local minima ( metastable states ) , whose huge number makes it difficult to reach equilibrium . in this framework , the energy landscape is seen as a set of basins of attractions of local minima ( `` traps '' ) , and the system evolves through a succession of harmonic vibrations inside traps and jumps between minima @xcite . this picture has stimulated the definition and study of various simplified models of dynamical evolution between traps , in order to reproduce the phenomenology of glassy dynamics @xcite . on the other hand , several studies have focused on obtaining a better understanding of the structure of these local minima . a way to attain this goal is to perform numerical simulations of small systems , at a fixed temperature , and quenching them at regular time intervals in order to make them reach the nearest local minimum . information is then gathered on the various local minima , and on the sizes of their basins of attraction . various studies have investigated , among other issues , the detailed structure of the potential energy landscape , the substructure of minima , and the properties of energy barriers between minima @xcite . several works have also used the information on the energy landscape to study a master equation for the time evolution of the probability to be in each minimum . the considered systems range from clusters of lennard - jones atoms to proteins or heteropolymers @xcite . an interesting property of the modeling of the energy landscape in terms of a set of traps linked by energy barriers , lies in the possibility to define and study its network representation within the context of network theory . in this representation , each local minimum is associated to a node , and a link is drawn between two nodes whenever it is possible for the system to jump between the basins of attraction of the corresponding minima . the links can then be defined as weighted and directed , as jumps between minima are not equiprobable , and may be easier in one direction than in the other . networks of local minima of the energy landscape have thus been built and studied . these networks have been found to exhibit a small - world character @xcite . the number of links of each node ( its degree ) turns out to be strongly heterogeneous , possibly with scale - free degree distributions , which have been linked to scale - free distributions of the areas of the basins of attraction @xcite . complex network analysis tools have also been used to investigate the structure of energy landscapes of various systems of interest , such as lennard - jones atoms , proteins , or spin glasses , among others @xcite . the energy of a minimum and its degree ( i.e. , the number of other minima which can be reached from this minimum ) have been shown to be correlated , as well as the barriers to overcome to escape from a minimum . in particular , a logarithmic dependence of the energy of a minimum on its degree has been exhibited , as well as energy barriers increasing as a ( small ) power of the degree of a node @xcite . no systematic study of these issues has however been performed , and most investigations have been limited to relatively small systems because of computational limitations . most importantly , the investigations cited above have focused on the topology of the network of minima , conceived as a tool to characterize the energy landscape . the structure of a network has however a deep impact on the properties of the dynamical processes which take place on it @xcite . it seems thus adequate to put to use the tools and techniques developed for the analysis of dynamical processes on networks to achieve a better understanding of how the energy landscape structure , represented as a network , affects the system performing a random walk in it , and how the onset of glassy dynamics can be described in this way in a general framework . in a previous paper @xcite , we have made a first step to fill this gap by focusing on the trap model put forward in ref . @xcite . in this paper , we generalize our approach to more involved transition rates between energy minima . we show how the heterogeneous mean - field ( hmf ) theory @xcite can be used in this context to highlight the connection between the topological properties of the network of minima and the dynamical exploration of these minima . we show in particular that the relationship between energy and degree of the minima is a crucial ingredient for the existence of a transition and the subsequent glassy phenomenology . our results shed light on the empirically found relationship between the energy of a local minimum and its degree , and we hope that they will stimulate more systematic investigations on this issue . we have organized our paper as follows : in section [ sec : model ] we define our model of energy landscape dynamics as a random walk on a complex network . different physical transition rates are proposed , and the corresponding numerical implementation is discussed . in section [ sec : formalism ] we present a theoretical analysis based on the heterogeneous mean - field approximation for dynamical processes on complex networks . this formalism is applied in section [ sec : general - formalism ] , where general analytical approximate expression are presented for the main quantities characterizing the glassy transition and dynamics . these expressions are applied to the different physical transition rates considered in section [ sec : appl - phys - trans ] , where checks against numerical simulations are also presented . in section [ sec : barriers ] we discuss the relation between energy basins and energy barriers . finally , in section [ sec : conclusions ] we present our conclusions . we consider a network of @xmath0 nodes , in which each vertex @xmath1 corresponds to a minimum in the energy landscape , and a link is drawn between two minima @xmath1 and @xmath2 if the system can jump directly from @xmath1 to @xmath2 . to each node @xmath1 is associated the energy @xmath3 of the corresponding minimum ( energies are defined from a reference level , in such a way that @xmath4 for all @xmath1 ) . moreover , an energy gap @xmath5 is associated to the edge between vertices @xmath1 and @xmath2 , as depicted in fig . [ landscape ] : @xmath5 is a symmetric function , such that the energy barrier that must be overcome to jump from vertex @xmath1 to vertex @xmath2 can be written as @xmath6 and , analogously , @xmath7 . obviously , we will have in general @xmath8 . are measured positive downwards . energy gaps @xmath5 are defined positive upwards.,width=340 ] the system under investigation is pictured as a walker exploring the network through a biased random walk . the rate ( probability per unit time ) @xmath9 to go from vertex @xmath1 to vertex @xmath2 depends a priori on the energy at vertices @xmath1 and @xmath2 and/or the energy barrier between @xmath1 and @xmath2 that must be overcome . the random walk model is defined in discrete time @xmath10 as follows : * at time @xmath10 , the walker is in vertex @xmath1 . * it chooses at random a neighbor of @xmath1 , namely @xmath2 . * with a probability @xmath9 , that depends on the energy @xmath11 and/or on the energy barrier @xmath12 , the walker hops to vertex @xmath2 . * time is updated @xmath13 . the relationship between the probabilities @xmath9 and the energy and energy barriers can be of different forms . in usual unbiased random walks , @xmath9 is a constant independent of both @xmath1 and @xmath2 @xcite . as a first step to introduce a dependence on the nodes , a possible approach is that in which the energy barriers depend only on the local minima themselves , i.e. we consider @xmath14 . for example , in the bouchaud trap model considered in ref . @xcite , the probability to exit from a trap is just an arrhenius law depending only on the departing trap s depth , namely @xmath15 where @xmath16 is the inverse temperature and @xmath17 is a constant that determines a global timescale . other possible definitions include the metropolis one @xmath18 and the glauber rate @xmath19 we note that the rates considered in the bouchaud trap model are quite different from the metropolis or glauber rates . indeed , while the former depends only on the depth of the originating trap , the latter depend also on the energy of the arriving vertex . this translates in the fact that , in the limit of zero temperature , the dynamics of the bouchaud trap model is frozen for any @xmath11 , while metropolis and glauber dynamics still allow jumps to lower energy minima @xcite . within an even more realistic representation of glassy dynamics , one can also contemplate the case @xmath20 , allowing for the transition rates to depend explicitly on the energy barriers between adjacent minima . as a paradigm of this choice , we propose a rate of the arrhenius form @xmath21 which acts a straightforward generalization of the local transition rate ( [ eq : rate_traps ] ) . the case of rate ( [ eq : rate_traps ] ) ( local trapping ) was studied in a previous publication @xcite . in the following , we will consider in turn non - local rates ( [ eq : rate_metropolis ] ) , ( [ eq : rate_glauber ] ) and ( [ eq : rate_barriers ] ) and discuss the fundamental differences due to the introduction of energy barriers in the model . we emphasize that our model differs both from usual unbiased random walks , as the local energy determines the transition rates , and from mean - field trap models in which jumps between any pair of energy minima are a priori possible : here , the system can jump only between neighboring nodes . the dynamical evolution depends therefore both on the network topology and on the energies associated to the nodes . to implement numerically the random walk , it is convenient to resort to the techniques developed for general diffusion processes on complex weighted networks @xcite . the main advantage of this method is to avoid rejection steps , thus improving dramatically the computational efficiency @xcite . therefore , at each simulation step the random walker sitting at node @xmath1 selects a neighbor @xmath2 with probability @xmath22 , where the sum in the normalizing factor is extended to all of @xmath1 s neighbors . as the walker hops on node @xmath2 , the physical time is incremented by an interval @xmath23 drawn from the exponential distribution @xmath24 , where @xmath25 is the inverse of the average escape rate out of node @xmath1 . in this way the simulation time is disentangled from the physical time and the latter has no impact on the simulation efficiency . no matter how much physical time a walker spends in a node , from the simulation time point of view it is always just a time step . the network substrates on which we will focus are scale - free networks with a degree distribution of the form @xmath26 and @xmath27 . we will generate them with the uncorrelated configuration model ( ucm ) @xcite , that allows us to tune the degree distribution to the desired form and prevents the formation of degree - degree correlations . networks are therefore generated as follows : a number of stubs ( or semi - links ) extracted from the desired final degree distribution is assigned to each node . stubs are then randomly paired to form links between nodes , with the prescription that multiple links as well as self - loops must be avoided . a minimal degree @xmath28 is fixed a priori . to avoid spurious effects due to the possible presence of tree - like structures @xcite it is convenient to adopt @xmath29 . we will choose @xmath30 in all of our simulations . so far the algorithm coincides with the one of the configuration model @xcite , but the ucm introduces moreover a cutoff to the degree distribution , @xmath31 , which avoids the formation of degree correlations by limiting the size of the hubs @xcite . in order to gain analytical understanding of the role of the different transition rates in the corresponding glassy dynamics , we apply a standard heterogeneous mean - field ( hmf ) formalism @xcite . the basic tenet of hmf is the assumption that all the dynamical properties of a vertex depend only on its degree . vertices are thus grouped into classes according to their degree , and vertices with the same degree are treated as equivalent . this approximation is consistent with previous findings that have uncovered the correlations between the energy of a local minimum and the degree of the corresponding node in the network @xcite . we therefore make the assumption that there exists a relationship @xmath32 where the function @xmath33 is a characteristic of the model . this also means that the distributions of energies @xmath34 of the system s landscape , and the degree distribution @xmath35 of the corresponding network are linked through @xmath36 . in the same spirit , we make the further assumption that the energy gap between minima @xmath1 and @xmath2 depends only on the degrees of @xmath1 and @xmath2 , i.e. , that it can be written as @xmath37 , where @xmath38 is a symmetric function of @xmath39 and @xmath40 . under the hmf approximation the dynamics will thus focus on the transitions between different degree classes . the rate to go from a vertex @xmath39 to a vertex @xmath40 can be written as @xmath41 the function @xmath42 , defined as the conditional probability that a vertex of degree @xmath39 is connected with another vertex of degree @xmath40 @xcite , takes into account the topological features of the network , by gauging the probability of selecting a vertex @xmath40 as neighbor of @xmath39 . the function @xmath43 measures the rate of jumping from a vertex of degree @xmath39 to a vertex of degree @xmath40 ( given that they are connected by an edge ) , and depends on @xmath39 and @xmath40 through the rates @xmath44 and the functions @xmath36 and @xmath45 . obviously , the rate @xmath43 is not in general a symmetric function of @xmath39 and @xmath40 . it is worth noting that , apart from a normalization , equation ( [ eq : weighted_propagator ] ) is simply the so - called weighted propagator describing the probability that a node in class @xmath39 interacts with a node in class @xmath40 @xcite . we also note that the rates @xmath43 depend on the inverse temperature @xmath46 through the microscopic rates @xmath9 . in this section , we apply the hmf theory to compute different quantities relevant for the characterization of the dynamics of a random walk in a complex energy landscape represented in terms of a network of minima . the description of a random walk dynamics starts from the occupation probability @xmath47 , defined as the probability for the walker to be in _ any _ node of degree @xmath39 at a time @xmath48 . its time evolution can be easily represented in terms of a master equation of the form @xmath49 upon describing the state at time @xmath48 with the row vector @xmath50 , where @xmath51 is the cutoff or largest degree in the network , equation ( [ rate_scalar ] ) can be rewritten in vector form as @xmath52 where the matrix @xmath53 , with elements @xmath54 is a generalization of a laplacian matrix to the case of directed weighted graphs . the matrix elements satisfy @xmath55 which ensures conservation of probability and states that the columns of @xmath53 are not linearly independent . the real part of every eigenvalue of @xmath53 is non - negative @xcite . as a consequence , all solutions of eq . ( [ rate_vectorial ] ) , which can be formally written as @xmath56 are stable according to lyapunov criteria . in particular , since @xmath57 , @xmath53 always has the eigenvalue @xmath58 , which corresponds to a constant solution of the problem . at this point one can proceed in close analogy with discrete - time regular markov chains @xcite . by making the assumption that the matrix @xmath59 is non - negative and irreducible ( indeed it is for every choice of @xmath60 in the following ) , we can prove that the @xmath58 eigenvalue of @xmath53 has algebraic multiplicity @xmath61 . hence , the stationary solution of eq . ( [ rate_vectorial ] ) is unique . in order to calculate the steady solution @xmath62 in the limit @xmath63 , one can impose @xmath64 . this leads to the condition @xmath65 so that we are left with the task of finding the left nullspace of @xmath53 . ( [ homogeneous_v ] ) is a homogeneous system of algebraic linear equations . it admits non - trivial solutions since @xmath66 . in our case , the solution to ( [ homogeneous_v ] ) can be easily found by imposing the detailed balance condition . namely , writing eq . ( [ homogeneous_v ] ) as @xmath67=0 , \label{eq:1}\ ] ] we can obtain a solution by imposing that the terms inside the summation in eq . ( [ eq:1 ] ) cancel individually , that is @xmath68 substituting the form of @xmath69 , we obtain @xmath70 where in the last step we have used the degree detailed balance condition @xmath71 which simply expresses that the number of edges from a node of degree @xmath39 to a node of degree @xmath40 is equal to the number of edges from a node of degree @xmath40 to a node of degree @xmath39 @xcite . from eq . ( [ eq:2 ] ) , we see that its right - hand - side must be expressible as a simple ratio of a function of @xmath39 over a function of @xmath40 . a general way to obtain this is to impose a coarse - grained rate @xmath60 taking the general form @xmath72 in other words , we assume that the rate @xmath60 can be written as the product of a function of @xmath40 , a function of @xmath39 , and a symmetric function @xmath73 ( where @xmath39 and @xmath40 need not be separable ) . we will see later that all the rates @xmath74 defined in sec . [ sec : definition ] ( traps , glauber , metropolis , and energy barriers ) can be written in such a form . the stationary solution is then given by @xmath75 where @xmath76 is a normalizing constant determined by the condition @xmath77 . such a solution is unique , as proven above . interestingly , the symmetric function @xmath78 does not enter the steady solution , although it will play a role in affecting the transient behavior , as we will see in the next sections . the steady state solution found above for the occupation probability is defined if and only if the normalization constant @xmath79 is finite . when this condition is met , the random walker reaches an equilibrium state with a distribution @xmath80 . on the other hand , whenever such condition is not met , the random walker is unable to reach a steady state , i.e. the steady solution to the rate equation does not correspond to any physical steady state in equilibrium @xmath81 . we identify this region of the phase space with the glass phase for our random walker @xcite . the functions @xmath82 and @xmath83 depend on the temperature , on the precise dynamics chosen ( traps , glauber , metropolis ) , and encode the relationship @xmath36 between energy and degree of the minima . the degree distribution moreover enters explicitly the expression @xmath84 . as the various parameters of the model are changed , it is thus a priori possible to go from one phase in which @xmath84 is finite to one in which @xmath84 diverges . in a physical system in particular , the control parameter is usually the temperature , while the topology of the network of minima and the function @xmath36 are given . it is then clear from eq . ( [ eq:4 ] ) that the presence or absence of a finite glass transition temperature @xmath85 , such that @xmath84 becomes infinite for @xmath86 , depends on the interplay between the topology of the landscape network ( as determined by @xmath35 ) and the functions @xmath82 and @xmath83 . interestingly , at this mean - field level , the existence of a transition does not depend on the network degree correlations , since the conditional probability @xmath42 do not enter eq . ( [ eq:4 ] ) . let us consider for instance a network of minima with a heavy - tailed degree distribution such as @xmath26 . a transition between a finite and an infinite @xmath84 can be observed if and only if @xmath87 behaves at large @xmath39 in the form @xmath88 where the exponent @xmath89 depends on the temperature , and can take values smaller or larger than @xmath90 depending on the temperature . another example is given by a stretched exponential form for @xmath35 , @xmath91 , in which case a transition is observed if and only if @xmath87 is of the form @xmath92 , with @xmath93 a function of the temperature ( the transition is then given by @xmath94 ) . in any finite system , unless the product function @xmath87 exhibits some sort of singularity , the normalization constant eq . ( [ eq:4 ] ) is finite and the steady state distribution @xmath95 exists , the occupation probability @xmath47 converging to it after an equilibration time , i.e. @xmath96 the corresponding thermalization of the occupation probability occurs in a way depending on the function @xmath36 . shallow energy minima are indeed explored first , while deep traps ( large @xmath97 ) are visited at larger times @xcite . if @xmath36 is a growing function of @xmath39 , as indeed found empirically @xcite , small degree nodes correspond to shallow minima , and deeper minima are associated to larger nodes . the evolution of @xmath47 takes then place in a hierarchical fashion : the small degree region equilibrates first , and progressive equilibration of larger degree regions takes place at larger times . in this respect , we obtain a strong difference between the biased random walk that the glassy system experiences and usual diffusion processes corresponding to unbiased random walks , which visit first large degree vertices and then cascade down towards small degree nodes @xcite , in the present case we observe an `` inverse cascade process '' from small vertices to hubs . we have found in @xcite that , in the case of a random walk among traps , this hierarchical thermalization is summarized in a scaling form for @xmath47 , which can be written as @xmath98 where @xmath99 represents the maximum degree of the vertices equilibrated up to time @xmath48 , and @xmath100 interpolates between @xmath101 at small @xmath102 and the short time form of @xmath103 which is proportional to @xmath104 . we will see in the next section that a similar scaling is obeyed for other transition rates . in general , for the glassy dynamics , the functional form of @xmath99 can moreover be obtained through the following argument : the total time @xmath48 can be written as the sum of the trapping times spent in the vertices that have been visited since the beginning of the dynamics . trapping times increase with the depth of the minimum , hence with the degree ( we are still considering the case of an increasing function @xmath33 ) , and , in the glassy phase , the consequence is that the sum of trapping times is dominated by the vertex with the largest degree visited up to that point , namely @xmath105 . moreover , the average trapping time @xmath106 at a given vertex @xmath39 can be estimated as the inverse of the average escape rate @xmath107 from that vertex : @xmath108 we can therefore estimate @xmath99 , the typical degree up to which nodes are `` equilibrated '' at time @xmath48 , by approximating @xmath109 , and solving the equation @xmath110 to obtain @xmath105 as a function of @xmath48 . note that the result depends here on the function @xmath111 and not only on @xmath82 , @xmath83 , @xmath36 and @xmath35 . the properties of the system can be further quantified by measuring the average time @xmath112 required by the random walker to escape from the vertex it occupies at time @xmath48 @xcite . for small waiting times @xmath48 , @xmath113 increases as a result of the transient equilibration of @xmath114 . for large @xmath48 , such that @xmath115 is close enough to the equilibrium @xmath116 , @xmath113 can be calculated instead as the average @xmath117 \tau_k\ ] ] where @xmath118 is the inverse of the equilibrium escape rate , cf eq . ( [ eq:7 ] ) , yielding @xmath119 most interestingly , the explicit form of the average escape time @xmath113 depends explicitely on the symmetric function @xmath111 as well as on the network degree correlations , as expressed by the conditional probability @xmath42 . let us go back to the issue of the existence of a glass transition in the model . we first recall the phenomenology of the fully connected trap model , with transition rates @xmath120 for any @xmath1 and @xmath2 , where the energies @xmath11 are random numbers extracted from a distribution @xmath34 @xcite . as all traps are connected with each other , all traps are equiprobable after a jump , so that the probability for the system to be in a trap of depth @xmath97 is simply @xmath34 , and the average rest time spent in a trap is @xmath121 . a transition between a high temperature phase and a glassy one is thus obtained if and only if , when @xmath46 increases , @xmath122 is finite at small @xmath46 and diverges at a finite @xmath85 . such a phenomenology is obtained if and only if @xmath34 is of the form @xmath123 at large @xmath97 ( else the transition temperature is either @xmath58 or @xmath124 ) , and the transition temperature is then @xmath125 @xcite . in the present case of a nework of minima , the average rest time that the walker spends in a minimum is @xmath126 where the symbol @xmath127 refers to the average performed over the measure @xmath128 , which represents the probability that the walker is in any vertex of degree @xmath39 after a hop . note that we disregard here the _ physical _ time , which is the sum of times spent in the various minima , but consider only the number of hops between minima . in the case of the traps model , @xmath129 is simply given by the probability to be in a node of degree @xmath39 after a hop in a random walk , i.e. by @xmath130 @xcite , since the transition rates do not depend on the arrival node . in a non - local trapping model instead , we need to write a master equation of the form @xmath131 where the matrix @xmath132 is now stochastic and the derivative is intended with respect to the number of hops . in the long time limit , we impose @xmath133 and calculate @xmath128 as we did for @xmath95 , imposing the detailed balance condition , and obtaining @xmath134 r_k \ , \ ] ] where @xmath135 is a normalization factor , given by @xmath136 we finally obtain for the average @xmath137 : @xmath138 where @xmath84 is the normalization factor of @xmath139 defined in eq . ( [ eq:4 ] ) . as for the average escape time , the average rest time @xmath137 thus depends on all the parameters of the system , including the network s degree correlations and the symmetric function @xmath140 . in this section , we apply the general hmf results obtained in the previous section to physical transition probabilities between local minima given by the trap model , glauber , metropolis and barrier - mediated rates . we will focus for definiteness on scale - free networks characterized by a power - law degree distribution @xmath141 with @xmath142 , which turns out to be the interval reported in the literature @xcite . let us first consider the explicit form of the transition rates in each case , to show that they can be cast in the form given by eq . ( [ eq:3 ] ) . in the case of the trap model , the rate to jump from a vertex @xmath39 to a vertex @xmath40 is simply @xmath143 , where we recall that @xmath33 gives the depth of a node of degree @xmath39 : it depends only on the degree of the starting node , and not on the node reached after the jump . we can therefore use @xmath144 the glauber rate can be written as @xmath145 leading to @xmath146 the metropolis transition , on its turn , reads @xmath147.\ ] ] since , for positive @xmath89 , @xmath148 , we can choose @xmath149 finally , in the presence of energy barriers , the transition rate reads @xmath150 where @xmath151 is a symmetric function of its arguments , so that we can use @xmath152 interestingly , for all the transition rates considered above , the product of the functions @xmath153 and @xmath83 , which controls the existence of the steady state solution of the occupation probability , takes the form @xmath154 the normalization constant @xmath84 defined in eq . ( [ eq:4 ] ) can thus be written as @xmath155 for a power - law degree distribution @xmath26 , a finite glass transition temperature is then obtained if and only if @xmath36 is of the form @xmath156 which is precisely what has been found , in conjunction with a scale - free degree distribution , in ref . @xmath84 is then indeed given by a sum of terms of the form @xmath157 , which converges if and only if @xmath158 in other words , a transition between a high temperature phase in which @xmath95 exists and a low temperature glassy phase is obtained at the critical temperature @xcite @xmath159 quite noticeably , the existence of a transition at a finite temperature , as well as the value of this temperature , does not depend on the form of the transition rates between the local minima , but only on the existence of a particular interplay between the topology of the network of minima and the relationship between energy and degree in this network , as determined by the function @xmath36 . we emphasize that this result is also independent of the network degree correlations @xmath42 , as already noted in the previous section . let us focus on the case of a scale - free network of minima , with @xmath160 and @xmath161 . for any of the rates discussed above , the steady state measure , when it exists , is given by @xmath162 where @xmath163 is the riemann @xmath163 function . a plot of @xmath164 as a function of @xmath39 and @xmath165 is given in fig . [ q_k ] , while data from simulations are reported in fig . [ q_k_sim ] for the evolution of @xmath115 under glauber dynamics . above the transition ( @xmath166 ) , low-@xmath39 states ( i.e. , shallow minima ) are more probable . as the temperature decreases , @xmath95 becomes less and less peaked at low values of @xmath39 , and large-@xmath39 states , which correspond to lower energies , become more and more probable . for the random walker to be in any node of degree @xmath39 . for @xmath167 the system undergoes a transition to a glassy state.,width=340 ] in any finite system , the sum defining @xmath84 , is finite at any temperature as the degree distribution has a cut - off at a finite @xmath51 : @xmath168 for instance , for @xmath169 , and with @xmath170 , the sum @xmath171 is analytic in @xmath172 for any finite @xmath51 . here @xmath173 is the harmonic number of order @xmath174 , which tends to @xmath175 for @xmath176 . the probability @xmath95 is thus well defined for every @xmath165 and for any finite system . in particular , performing a continuous degree approximation in eq . ( [ eq:19 ] ) , we can obtain an estimate of the network size dependence of @xmath84 as @xmath177 for @xmath178 , @xmath84 tends to a constant as the network size ( and thus @xmath51 ) increases . on the other hand , for @xmath179 , @xmath84 diverges as @xmath180 , i.e. , as @xmath181 in uncorrelated scale - free networks , which obey @xmath182 . for glauber dynamics . the distribution measured after a small waiting time @xmath48 is determined by a usual unbiased random walk behavior , i.e @xmath183 , while at larger times it relaxes to the equilibrium @xmath95 . the relaxation towards equilibrium starts from small degree nodes . data refer to ucm networks with @xmath184 and @xmath185 for @xmath186 ( @xmath187 ) : with these parameter values , for small times @xmath188 , while at large enough times @xmath189.,width=340 ] at low temperatures , even for a finite system , the evolution of @xmath190 towards @xmath95 is slow , as displayed in fig . [ q_k_sim ] , and an ageing regime takes place , in which the function @xmath115 obeys the scaling form @xmath191 where the characteristic degree @xmath105 can be estimated from eq . ( [ eq:8 ] ) . in order to simplify its computation , we will consider an uncorrelated network of minima , such that @xmath192 , and we will work in the continuous degree approximation , using the normalized form @xmath193 , where @xmath28 is the minimum degree present in the network . in the case of the glauber dynamics , the escape rate can be expressed , within the above approximations , as @xmath194 , \label{eq:21}\end{aligned}\ ] ] where we have used the relation @xmath195 and where @xmath196 $ ] is the gauss hypergeometric function . using the asymptotic expansion for @xmath197 @xcite , we obtain that the leading behavior for large @xmath39 yields @xmath198 which leads to @xmath199 from here , using the relation @xmath109 , we obtain @xmath200 at different temperatures ( glauber dynamics ) . data refer to ucm networks with @xmath184 and @xmath201 , so that @xmath202 ( where we have taken @xmath187 ) . the top panel presents data for @xmath203 while both middle and bottom panels concern the low temperature case @xmath204 . accordingly , for the top panel we use @xmath205 for the rescaling , while for the central and the bottom ones it holds @xmath206 ( see eq . ( [ eq:24 ] ) ) . the curves corresponding to different @xmath48 collapse well under this rescaling . note that we use rather small values of @xmath48 , because the equilibration time , defined by @xmath207 , is @xmath208 for @xmath204 and @xmath209 for @xmath210 . each curve is obtained by averaging over @xmath211 simulation runs.,width=340 ] in fig . [ collapsep ] we check the validity of eqs . ( [ eq:25 ] ) and ( [ eq:24 ] ) by performing a data collapse analysis for different values of @xmath48 . the curves obtained for different @xmath48 collapse indeed as predicted . in the case of the metropolis transition rates , a similar analysis yields @xmath212 leading to the same asymptotic behavior as in eq . ( [ eq:22 ] ) , and therefore to the same scaling picture as for the glauber rate . as pointed out for the case of the traps model @xcite , however , in finite systems the scaling relations in eq . ( [ eq:24 ] ) hold only as far as @xmath213 , i.e. it exists an equilibration time @xmath214 , obtained by inverting eq . ( [ eq:24 ] ) , above which the system has completely relaxed and eq . ( [ eq:25 ] ) is no more valid . finally , it is worth stressing that , while for large temperatures the scaling exponent relating @xmath105 and @xmath48 depends on the temperature , in the low temperature phase it becomes independent , being just proportional to the transition temperature . we note that this saturation of the exponent at @xmath215 is very different from the phenomenology obtained in the trap model @xcite , for which @xmath216 . an immediate consequence is that the equilibration time strongly depends on @xmath46 , as @xmath217 , for a system described by traps , but is given by @xmath218 for any @xmath204 for glauber and metropolis rates . contrarily to results for the glass transition temperature @xmath219 and the steady state , the glassy dynamics for barrier - mediated rates does not yield the same results as for glauber and metropolis rates , since @xmath107 does depend on the symmetric function @xmath220 . in particular , we need here to choose a functional form for @xmath45 . we propose to use @xmath221 which will be justified in section [ sec : barriers ] . in this case we obtain @xmath222 in order to keep an interesting phenomenology , the constant @xmath223 can not be chosen arbitrarily . if @xmath223 were independent of the system size , the escape rate would be dominated by the exponential at all temperatures . this behavior would reflect the fact that in this case the rate is suppressed in transitions involving nodes of large @xmath39 , eventually generating unphysically large barriers at large @xmath51 . to prevent the system from building up infinite barriers , one can impose @xmath224 such that the maximum barrier is always comparable to the lowest energy minimum and neither term dominates the other . as a consequence , we take @xmath225 where @xmath226 is now constant and size independent . contrarily to the previous cases , @xmath99 is hard to determine , as no explicit inversion of eqs . ( [ eq:8 ] ) and ( [ eq:29 ] ) can be provided for the range of parameters of interest in our study . a numerical evaluation of @xmath99 , is reported in figure [ f : kw_barriers ] . the maximum degree of equilibrated vertices @xmath105 has an initial power law increase in time , which is reminiscent of the local trapping model . however , as larger degree nodes are equilibrated , exponential barriers come into play and the hierarchical thermalization becomes logarithmic in time . for barrier - mediated dynamics . curves are obtained from the numerical inversion of eq . ( [ eq:8 ] ) . values of parameters are : @xmath201 , @xmath227 , @xmath228 , @xmath229 , @xmath230 . , width=340 ] , so that @xmath231 ( @xmath187 ) . the scaling forms of eq . ( [ eq : escape ] ) produce a collapse of the curves concerning different systems sizes in the three regimes of high , intermediate and low temperature ( top , center and bottom panels , respectively ) . the slightly worse collapse obtained for @xmath232 may be due to logarithmic corrections as @xmath46 is close to both @xmath85 and @xmath233 . each point is averaged over @xmath234 simulation runs ( @xmath235 runs on each of @xmath235 network realizations ) . , width=340 ] the average escape time , defined as the average time required by the system to escape from the vertex it occupies , can be computed in the long time limit from eq . ( [ eq : t_esc ] ) , as a function of the average trapping time @xmath106 in vertices of degree @xmath39 . from the asymptotic expansions of @xmath106 in eq . ( [ eq:23 ] ) , valid for glauber and metropolis dynamics , evaluation of eq . ( [ eq : t_esc ] ) allows us to observe that , whenever a finite @xmath219 exists , @xmath236 diverges at @xmath237 in an infinite system , as was already observed in the case of local trapping @xcite . it is noticeable that the same divergence temperature is obtained , as eq . ( [ eq : t_esc ] ) a priori involves the network s degree correlations and the function @xmath140 . within the continuous degree approximation , the divergence of the escape time with the system size follows the scaling laws @xmath238 as noted in the previous paragraph for the equilibration time , we note that the scaling for @xmath239 differs from the form @xmath240 encountered for local trapping @xcite . figure [ f : escape ] reports simulation data that confirm the validity of eq . ( [ eq : escape ] ) . as the temperature is lowered , the initial transient becomes longer , but for large enough times @xmath48 the asymptotic behavior predicted in eq . ( [ eq : escape ] ) is reached , as made clear from the collapse of curves concerning different system sizes . in the case of barrier - mediated dynamics , @xmath106 depends on the symmetric function @xmath151 , as expressed in eq . ( [ eq:29 ] ) , namely @xmath241 proceeding as above we obtain @xmath242e_0 } & \beta_c/(2+\epsilon ) < \beta < \beta_c \\ \mbox{const . } & \beta<\beta_c/(2+\epsilon ) \\ \end{array } \right . , \end{aligned}\ ] ] where we recall that @xmath226 does not depend on the system size . , as a function of the inverse temperature , for different system sizes . data are obtained by numerical computation of eq . ( [ tau_zi ] ) ( with @xmath187 ) . note the exponential increase with @xmath46 for @xmath243 , which saturates for @xmath244 as predicted by eq . ( [ e : rest]).,width=340 ] , as a function of the degree distribution cut - off and for different temperatures . data are obtained by numerical computation of eq . ( [ tau_zi ] ) . @xmath245 grows as a power - law of @xmath51 , with an exponent that grows as @xmath46 increases ( going from bottom to top in the figure ) . the thick - gray line corresponds to @xmath246 , while the thick - black line corresponds to @xmath247 . for larger values of @xmath46 , the power - law behavior corresponds to the predicted @xmath248 , which no more depends on @xmath46 . the @xmath249 curve is reported as a dashed line for reference . values of @xmath46 represented here are comprised between @xmath250 and @xmath251.,width=340 ] the hmf expression for the asymptotic average rest time , defined as the average time spent by the system in a minimum , is given by eq . ( [ tau_zi ] ) , namely @xmath252 , where the quantities @xmath84 and @xmath135 , for uncorrelated scale - free networks and a degree - energy relation @xmath253 , take the form , in the continuous degree approximation , @xmath254 let us first recall the case of the local trap model . both @xmath255 and @xmath111 are then constants , so that @xmath256 . thus , the average rest time behaves as @xmath84 : it is finite for @xmath210 , and diverges with the system size as @xmath257 for @xmath204 , signaling the emergence of the glassy regime at low temperatures . in the case of glauber and metropolis dynamics ( both leading to the same results ) , the situation is more involved , since the product @xmath258 entering @xmath135 is not constant . in fact , @xmath135 diverges with @xmath51 for @xmath259 , that is , at a lower temperature given by @xmath260 . the interplay of these two temperatures determines the behavior of the system for finite sizes within the low temperature phase . in particular , for the glauber dynamics with @xmath261 and @xmath262 $ ] , we have @xmath263 upon considering lower values of @xmath46 , @xmath137 first encounters the divergence of @xmath84 at @xmath85 , which is then partially regularized by the divergence of @xmath135 at @xmath233 . from these results , we obtain the emergence of three scaling regimes for the behavior of @xmath137 as a function of the system size : @xmath264 for different transition rates , for a random walker on ucm networks with @xmath265 ( @xmath187 ) and @xmath184 . discrete points : simulation results ; continuous lines : theoretical predictions from @xmath266 , based on simulation parameters . the glauber and metropolis transition rates induce two changes of behavior at @xmath85 and at @xmath233 , the first being a steep increase of the average rest time @xmath137 and the second a smoothing / saturation of this increase . no saturation of @xmath137 is instead observed when barriers are present . the agreement with theoretical predictions is remarkable , thus corroborating the validity of the hmf assumptions . moderate deviations are found only in the case of barriers , where exponential growth is expected to add greater fluctuations . in the inset , the difference between the two behaviors is more evident thanks to a different scale of the plot . note that , since in the simulations @xmath187 , the high temperature limit of the rest time is different for the glauber and metropolis dynamics , being @xmath267 and @xmath268 respectively . for the case of barriers we have chosen @xmath269 . each point is obtained by averaging the rest times corresponding to the first @xmath270 hops of the random walker in each of @xmath271 network realizations . , width=340 ] the direct numerical computation of eq . ( [ tau_zi ] ) is shown in figs . [ 2temperatures ] and [ kcscaling ] , showing the validity of this analysis . in particular , the exponential increase of @xmath272 with @xmath46 in the intermediate temperature range @xmath273 is clearly apparent in fig . [ 2temperatures ] , and fig . [ kcscaling ] confirms that , for @xmath274 , the exponent in the scaling law for the system size @xmath51 does not depend on the temperature . while the temperature @xmath85 signals the onset of the low temperature phase with glassy dynamics for all considered transition rates , for glauber / metropolis dynamics the low temperature phase can be further divided into two regions that correspond to different behaviors of the timescales with the system size . ( @xmath187 , @xmath231 ) . data for different system sizes collapse well when rescaled according to the theoretical values of eq . ( [ e : rest ] ) . while the agreement is excellent both for high and low temperature ( top and bottom panels , respectively ) , logarithmic corrections are probably present for the regime of intermediate temperatures @xmath275 ( central panel ) . in each simulation run the rest interval starting before @xmath48 and ending after @xmath48 is considered , and each point in the figure is averaged over @xmath234 simulation runs ( @xmath235 runs on each of @xmath235 network instances ) . , width=340 ] figures [ simul_2temperatures ] and [ f : rest ] moreover show the result of numerical simulations of random walkers on scale - free networks for glauber and metropolis dynamics as well as in the case of barriers , globally confirming the above discussed picture . dynamics in the presence of barriers do not yield the same phenomenology as glauber and metropolis rate . in this case , we have @xmath276 and @xmath277 . selecting @xmath278 , as in section [ sec : glassy - dynamics-1 ] , we are led to @xmath279 ^ 2.\ ] ] as for the escape time @xmath113 , upon choosing @xmath45 size independent , the rest time @xmath122 will be diverging exponentially with @xmath51 at every temperature . by introducing the size dependence as in eq . ( [ eq : size_depen ] ) , instead , one can see that the @xmath135 integral converges to a constant for large @xmath51 so that one is left with @xmath280 we therefore obtain the same picture as in the case of traps , with an exponential increase of @xmath122 as @xmath46 increases , as confirmed by numerical simulations in fig . [ simul_2temperatures ] . inspired by analogies with systems governed by the arrhenius law , we have introduced a transition rate that takes into account the energy barriers between states . within the heterogeneous mean - field approximation , in which all variables depend only on the degree of the vertices , and choosing @xmath281 , the transition rate we have considered becomes @xmath282 where @xmath151 is a symmetric function of the degrees of the two nodes . this model represents in essence an extension of the local trap model , where non - locality enters only in the form of symmetric energy gaps @xmath5 ( see fig . [ landscape ] ) . the steady state has exactly the same form as the ones discussed so far , which incidentally is the same as for the local trap model . as shown in the previous sections , the presence of barriers affect transient relaxation phenomena , but not the steady state . a different question is whether one can be more specific about the realistic functional form of the coarse - grained function @xmath151 . in the previous section we have already introduced a definition of @xmath151 . here we provide the rationale behind that choice . numerical simulations of the energy - landscape network of lennard - jones clusters show that the average barrier to escape from state @xmath39 follow the power law @xmath283 , with @xmath284 @xcite . in our model , such average can be computed as @xmath285.\ ] ] for simplicity we focus on uncorrelated networks , as simulations indeed show weak degree correlations . under this assumption , the first term of the sum on the right - hand side of eq . ( [ avg_barrier ] ) will contribute as a logarithm of @xmath39 and the power - law behavior of @xmath286 is possible whenever @xmath287 , which leads to consider the form proposed in previous sections , @xmath288 where @xmath289 has the dimensions of an energy ( a discussion about the possible values of @xmath223 is given in section [ sec : glassy - dynamics-1 ] ) . more complicated functional forms can also be proposed , for example accounting for barriers of different signs , as long as they retain the same power - law behavior of eq . ( [ sigma_linear ] ) in the large @xmath39 limit . it is interesting to notice that @xmath283 implies that the average escape rate @xmath290 has the form of a stretched exponential @xmath291 , if we neglect the logarithmic correction . in this paper , we have presented a simple mathematical framework for the description of the dynamics of glassy systems in terms of a random walk in a complex energy landscape . we have shown how to incorporate into this picture the network representation of this landscape , put forward and studied by several authors @xcite , in order to go beyond simple mean - field models of random walks between traps that are all connected to each other . while our previous work had focused on the case of a landscape consisting of traps connected by a network @xcite , we have here generalized our study to more involved and realistic transition rates between minima , including glauber or metropolis rates , and the possibility of energy barriers between minima . we have shown how the interplay between the topology of the network of minima and the relationship between the energy and the degree of a minimum may determine a rich phenomenology , with the existence of two phases and of glassy dynamics at low temperature . interestingly , the existence of these phases , and the transition temperature , do not depend on the network s degree correlations nor on the precise form of the transition rates , but other more detailed properties do . in the case of glauber and metropols dynamics , the low temperature phase can be further divided into two regions with different scaling properties of the average trapping time as a function of the temperature . overall , our results rationalize and link the empirical findings about correlations between the energy of the minima and their degree , and should stimulate further investigations on this issue . our work has also interesting applications in terms of diffusion phenomena on complex networks , and shows that non trivial transition rates can lead to a very interesting phenomenology . usual random walks lead to a higher probability for the random walker to be in a large degree node ( @xmath292 ) , with respect to the random choice of a node ( @xmath293 ) ; here , the models we have studied can lead to various stationary probabilities , such as for instance a uniform coverage which does not depend anymore on the degree . interestingly , the biased random walks among traps that we have studied can even display a phase transition phenomenon , as either a temperature parameter or the network s properties are changed , with the possible presence of a glassy phase with slow dynamics . p. m. , r.p .- s . , and a. baronchelli acknowledge financial support from the spanish mec , under project fis2010 - 21781-c02 - 01 , and the junta de andaluca , under project no . p09-fqm4682 .. r.p .- s . acknowledges additional support through icrea academia , funded by the generalitat de catalunya . a. baronchelli acknowledges support of spanish mci through the juan de la cierva program funded by the european social fund .
low mass x - ray binaries ( lmxbs ) are systems in which a low mass ( @xmath1 m@xmath2 ) star transfers matter to either a low magnetic field ( typically @xmath3 g ) neutron star or a black hole . in both cases plasma flows down to a few stellar radii and produces observable properties in x - rays ( spectra and timing ) . over the last several years comprehensive studies of these properties have provided crucial information on fundamental properties of compact objects , e.g. , the equation of state of neutron stars ( see van der klis 2000 for a review ) , or general relativistic effects near black hole horizons ( see tanaka & lewin 1995 , esin et al . however , there exist several sources whose x - ray properties indicate they may also be lmxbs but whose x - ray emission was so faint that no instruments were sensitive enough to study them until the launch of _ this paper describes our _ chandra _ observations of eight such objects listed as candidate lmxbs in @xcite ; the accurate locations we derived with _ chandra _ will enable follow up observations in the infrared / optical . below we briefly summarize prior observations for each of these objects . table [ tab : fluxsum ] provides a convenient comparison of the observed fluxes discussed below with our _ chandra _ observations . [ [ u-170840 ] ] 4u 170840 + + + + + + + + + + is a persistent x - ray source that was observed by most of the early x - ray instruments , e.g. _ uhuru _ , _ oso-7 _ , _ ariel v _ , and _ heao-1 _ * and references therein ) . rosat _ source , 1rxs j17141224.8405034 , at r.a . @xmath4 , decl . = 405034.5 ( j2000 , 1-@xmath5 error radius = 10 ) , detected in the _ rosat _ all - sky survey , lies in the error box for 4u 170840 @xcite . since 1996 , 4u 170840 has been a persistent source in the _ rossi x - ray timing explorer ( rxte ) _ all - sky monitor ( asm ) . its 2 - 10 kev flux ergs @xmath6 s@xmath7 ( 2 - 10 kev ) . ] slowly declined from about @xmath8 ergs @xmath6 s@xmath7 in 1996 - 1997 to a minimum of about @xmath9 ergs @xmath6 s@xmath7 in mid-1998 , followed by a slow rise to about @xmath10 ergs @xmath6 s@xmath7 in mid-2001 , and a second slow decline that continued in 2002 - 2003 . at the time of our _ chandra _ observations , the 2 - 10 kev flux level in the _ rxte _ asm was about @xmath11 ergs @xmath6 s@xmath7 . @xcite reported the discovery of x - ray bursts in a _ bepposax _ observation of 4u 1708408 in 1999 august . during this observation , the spectrum of the persistent flux ( before and after the bursts ) was well fitted with an absorbed power - law with a column density @xmath12 @xmath6 , a photon index of @xmath13 , and a gaussian iron line . the unabsorbed 2 - 10 kev flux was @xmath14 ergs @xmath6 s@xmath7 . in addition , @xcite also reported results for several _ rxte _ observations in 1997 and 2000 along with an additional _ bepposax _ observation in 2001 during which no bursts were seen ; however , steady persistent emission was observed . the energy spectra in these observations required more complicated models , an absorbed power law with photon index @xmath15 for an assumed @xmath16 @xmath6 , and a blackbody component with @xmath17 kev . the unabsorbed 2 - 10 kev fluxes were @xmath18 ergs @xmath6 s@xmath7 and @xmath19 ergs @xmath6 s@xmath7 in the 2000 june 18 _ rxte _ and 2001 august _ bepposax _ observations , respectively . [ [ s-1711339 ] ] 2s 1711339 + + + + + + + + + + + is an x - ray burster originally detected with _ ariel v_@xcite and more precisely located with _ a radio source was detected in the _ sas-3 _ error circle at r.a . = @xmath20 , decl . = 340036(b1950 , 90% confidence error radius 2.2 ) in 1978 july @xcite . rosat _ source , rxs j171419.334023 @xcite , also lies within the _ sas-3 _ error circle , but just outside the radio source error circle . more recently , @xcite reported on observations of 2s 1711339 with _ bepposax _ , the _ rxte _ asm , and _ chandra_. the _ rxte _ asm detected an x - ray outburst of 2s 1711339 from 1998 july until 1999 may at a 2 - 10 kev flux level of @xmath21 ergs @xmath6 s@xmath7 . during that time period , _ bepposax _ detected 10 short x - ray bursts , with a persistent flux level ( before and after the bursts ) of @xmath22 ergs @xmath6 s@xmath7 ( 2 - 28 kev ) . the persistent emission was fitted with a cutoff power - law with photon index 0.7 , high energy cutoff 2.8 kev , and an assumed @xmath23 of @xmath24 @xmath6 , derived from _ bepposax _ narrow field instrument ( nfi ) observations on 2000 february 29 , when the 2 - 6 kev flux was @xmath25 ergs @xmath6 s@xmath7 . on 2000 march 22 , one burst was detected with _ bepposax _ , but the 3-@xmath5 upper limit on the persistent flux ( before and after the burst ) was @xmath26 ergs @xmath6 s@xmath7 ( 2 - 28 kev ) . chandra _ observation , also described in this paper , on 2000 june 9 , yielded a location of r.a . = @xmath27 , decl . = 340247 ( j2000 , 90% confidence error radius 1 ) , which was consistent with the wide field camera @xcite , _ ariel v _ @xcite , and _ rosat _ @xcite positions . chandra _ position , however , lies 0.2 outside the 90% confidence error circle of the radio source . [ [ slx-1735269 ] ] slx 1735269 + + + + + + + + + + + + is an x - ray burster first reported in 1985 from spacelab-2 x - ray telescope ( xrt ) observations @xcite . it was present in the _ einstein _ slew survey 5 years earlier @xcite . rosat _ source , 1rxs j173817.0265940 , lies within the _ einstein _ error circle @xcite . _ granat _ sigma observations showed it to be a persistent hard x - ray source @xcite . type i x - ray bursts were discovered with the _ bepposax _ wide field camera ( * ? ? ? _ asca _ observations showed the 0.6 - 10 kev spectrum to be well fitted with a power - law with photon index 2.15 and @xmath28 @xmath6 @xcite , which was consistent with the values derived from _ rosat _ and _ granat _ art - p observations @xcite . using _ rxte _ observations from 1997 february - may , @xcite showed that the power spectrum was characterized by a strong band limited noise component which was approximately flat below a break frequency of 0.1 - 2.3 hz . above the break frequency the power spectrum declines as a power law of index 0.9 . at the highest count rates , a broad bump was observed around 0.9 hz . the strength of the noise ( 2 - 60 kev , integrated over 0.01 - 100 hz ) was @xmath29% and @xmath30% rms in the high and low count rate data , respectively . fits to energy spectra yielded power law indices of 2.2 and 2.4 ( with an assumed @xmath23 of @xmath31 @xmath6 ) and 3 - 25 kev fluxes of @xmath32 ergs @xmath6 s@xmath7 and @xmath33 ergs @xmath6 s@xmath7 , respectively . power spectra , from extended _ rxte _ observations in 1997 october @xcite at a 1 - 20 kev flux level of @xmath34 ergs @xmath6 s@xmath7 , had a noise strength of 27.6% ( 2 - 40 kev , integrated over 0.005 - 300 hz ) and a break frequency of 0.08 - 0.15 hz that appeared to increase with intensity . @xcite fitted this power spectrum with a four - lorentzian model and found a remarkable resemblance to observations of the low - luminosity x - ray burster 1e 17243045 . slx 1735269 has been detected as a persistent source with the _ rxte _ asm since 1996 at an average 2 - 10 kev flux level of about @xmath35 ergs @xmath6 s@xmath7 . [ [ grs-1736297 ] ] grs 1736297 + + + + + + + + + + + + was discovered with the _ granat _ art - p telescope in september 1990 and was subsequently observed about a month later @xcite at a similar flux level . its spectrum was characterized by a power law with photon index of 1.8 and a 3 - 12 kev flux of @xmath36 ergs @xmath6 s@xmath7 . @xcite identified rx j1739.52942 with grs 1736297 , since it lies well within the 90% confidence 90 radius _ granat _ error circle @xcite . further , @xcite identified a be star in the _ rosat _ error circle , suggesting that , although grs 1736297 has been previously classified as a candidate lmxb @xcite , it may instead be a transient be / x - ray binary system . [ [ ks-1739304 ] ] ks 1739304 + + + + + + + + + + + is a transient x - ray source discovered with _ mir_/kvant in 1989 at a 2 - 30 kev flux level of about 9 mcrab and located to within a 1.6 error circle @xcite . subsequent observations with _ granat _ art - p in autumn 1990 did not detect the source , with a 3-@xmath5 upper limit of 2 mcrab ( 8 - 20 kev ) @xcite . very little is known about this object . [ [ slx-1746331 ] ] slx 1746331 + + + + + + + + + + + + was discovered with the spacelab-2 xrt in 1985 august @xcite . it had a very soft spectrum , best fitted with a thermal bremsstrahlung with @xmath37 kev and a 2 - 10 kev flux of @xmath38 ergs @xmath6 s@xmath7 . based on the very soft spectrum , @xcite suggested slx 1746331 as a potential black hole candidate . the position of slx 1746331 fell between fields in the _ einstein _ galactic plane survey of @xcite and it was not detected in the _ exosat _ galactic plane survey @xcite . during the _ rosat _ survey observation in 1990 september , 1rxs j174948.4331215 , which lies in the error circle of slx 1746331 , was detected in outburst , but it was not detected in subsequent _ rosat _ high resolution imager ( hri ) observations in 1994 october @xcite . hence the _ exosat _ and _ rosat _ non - detections suggest this object is a transient system . [ [ e-1746.73224 ] ] 1e 1746.73224 + + + + + + + + + + + + + + was discovered during the _ einstein _ galactic plane survey @xcite , at a 2 - 10 kev flux level of @xmath39 ergs @xmath6 s@xmath7 . spacelab-2 xrt observations in 1985 august gave an upper limit on the 2 - 10 kev flux of @xmath40 ergs @xmath6 s@xmath7 @xcite . during the _ rosat _ all - sky survey , 1rxs j175003.8 - 322622 was detected in the _ einstein _ error circle of 1e 1746.73224 @xcite . very little is known about this object . [ [ u-181212 ] ] 4u 181212 + + + + + + + + + + was first detected with _ uhuru_@xcite and was later observed by several satellites , including _ oso-7 _ , _ ariel v _ , _ heao-1 _ , and _ exosat _ with a typical 2 - 10 kev flux of @xmath41 ergs @xmath6 s@xmath7 ( * ? ? ? * and references therein ) . 4u 181212 was detected in the _ rosat _ all - sky survey as a bright source denoted 1rxs j181506.1120545 @xcite . x - ray bursts were first detected from this source in 1982 with hakucho @xcite and later with _ bepposax _ @xcite . 4u 181212 is classified as an atoll source and shows band limited noise with a break frequency of 0.095 hz and a qpo at 0.85 hz @xcite . 4u 181212 has been detected as a persistent source with the _ rxte _ asm since 1996 at an average 2 - 10 kev flux level of about @xmath42 ergs @xmath6 s@xmath7 . recent observations in 2000 april with _ rxte _ and _ bepposax _ show 4u 181212 in a hard state with a hard tail extending above 100 kev . the power spectrum was characterized by a @xmath43 hz qpo and three broad noise components , extending above @xmath44 hz @xcite . each of the eight objects was observed using the advanced ccd imaging spectrometer ( acis ) on the _ chandra _ x - ray observatory . table [ tab : obs ] lists the details of the observations . for some sources , data were collected in two different observing modes , timed exposure ( te ) mode and continuous clocking ( cc ) mode . data obtained in te mode allow for two - dimensional imaging . accurate spectroscopy of bright targets , however , is limited due to pulse pile - up . timing studies are limited by the total number of photons collected and the 3.24 s time resolution . in cc mode , the amount of pile - up is negligible owing to the 2.85 ms time resolution , allowing for accurate spectroscopy and timing . unfortunately only one of the sources , slx 1735269 , was detected in cc mode . the best known location for each object was positioned on the nominal target position of acis - s3 , a back illuminated ccd on the spectroscopic array ( acis - s ) with good charge transfer efficiency and spectral resolution , with the exception of the cc mode observation of slx 1735269 which fell on acis - s2 , a front illuminated ccd on the spectroscopic array ( acis - s ) . standard processing was performed by the _ chandra _ data center . the data were filtered to exclude events with grades 1 , 5 , 7 , hot pixels , bad columns , and events on ccd node boundaries . analysis in this paper was done using _ chandra _ interactive analysis of observations ( ciao ) version 2.2.1 and _ chandra _ calibration database ( caldb ) version 2.10 . for piled - up sources , the tool _ acis_detect_afterglow _ in the standard processing pipeline can reject valid source photons in addition to afterglow photons resulting from residual charge from cosmic ray events in the ccd pixels . we examined the spatial distribution of afterglow events for each observation using methods described in the ciao documentation . for te mode observations of 2s 1711339 and 1e 1746.73224 we found that @xmath45% of events consistent with the point source had been rejected ; hence we reprocessed the data to retain the afterglow flagged events . for all other observations , @xmath46% of events were rejected , so the standard processing was retained . using te mode data , we extracted source locations for each of the detected sources . these locations are listed in table [ tab : loc ] . sources were located by one of two methods : ( 1 ) 2s 1711 - 339 and 1e 1746.73224 were located using the ciao tool _ wavdetect_. the latter source was observed twice in te mode and locations from both observations were consistent . ( 2 ) 4u 170840 , slx 1735269 , and 4u 181212 were brighter and suffered from considerable pile - up of photons in the image core , resulting in a source that looked ring shaped with a hole in the center . the ciao detect tools were inadequate to provide a good location of the centroid , therefore we modeled the data using a gaussian function multiplied by a hyperbolic tangent in radius , scaled to approach zero at r = 0.0 @xcite . the uncertainty of these locations is limited by systematic effects to a circle with @xmath47 radius @xcite . no other known sources were found in the images , hence we were unable to use astrometry to further improve the locations . three objects , grs 1736297 , slx 1746331 , and ks 1739304 went undetected in all observations . to derive upper limits for these sources , we used the ciao tool _ dmextract _ to extract count rates for a 6 radius source region and a background annulus ( inner radius = 6 , outer radius = 30 ) centered on the best known position . the 99% confidence upper limits were count rates of @xmath48 cts s@xmath7 , @xmath49 cts s@xmath7 , and @xmath50 cts s@xmath7 ( total counts = 6.0 cts , 5.8 cts , and 4.6 cts ) for grs 1736297 , slx 1746331 , and ks 1739304 , respectively . using webpimms , we estimated an upper limit 1 - 10 kev absorbed flux of @xmath51 ergs @xmath6 s@xmath7 for grs 1736297 , assuming @xmath52 @xmath6 , derived using colden , a web tool based on @xcite , and a power law photon index = 1.8 @xcite . similarly , for slx 1746331 , we estimated an upper limit 1 - 10 kev absorbed flux of @xmath53 ergs @xmath6 s@xmath7 , assuming @xmath54 @xmath6 , using colden , and a thermal bremsstrahlung spectrum with kt = 1.5 kev @xcite . for ks 1739304 , we estimated an upper limit 1 - 10 kev absorbed flux of @xmath55 ergs @xmath6 s@xmath7 , assuming @xmath56 @xmath6 , using colden , and a power law photon index = 1.5 . a fourth object , 2s 1711339 , that was detected in a te mode observation in 2000 , went undetected in the 2002 cc mode observation . to derive an upper limit for it , we again used dmextract , with a 6 radius source region and a background annulus ( inner radius = 6 , outer radius = 30 ) centered on the _ chandra _ position ( shifted to correspond to the 1-dimensional cc mode image ) from the 2000 observation , to obtain 99% confidence upper limits of @xmath57 cts s@xmath7 or a total 30.3 counts for the observation . using webpimms with our spectral parameters from the 2000 observation table [ tab : spec ] , we estimate a 1 - 10 kev upper limit absorbed flux of @xmath58 ergs @xmath6 s@xmath7 , a factor of more than 100 fainter than in the 2000 observation . in the 5 may 2002 observation of ks 1739304 , an object was present at r.a . = @xmath59 , decl.= @xmath60 , approximately @xmath61 from the expected location of ks 1739304 . we measured a total of approximately 45 counts from this object in the 745 s observation . given its large separation from ks 1739304 , and the fact that it is well outside the 1.6 radius _ mir_/kvant error circle @xcite , the two objects are most likely unrelated . energy spectra were difficult to study for most of these sources because in all but one of the observations where a source was detected , we had only imaging ( te ) mode data and pile - up was a problem . for these sources we used two approaches to extract and fit the energy spectra : ( 1 ) direct extraction from the te mode data , fitting the data using the pile - up model and ( 2 ) extraction of trailed image spectra . to directly extract spectra and associated response files for each source , we used a source region consisting of a 6 radius circle and a background region consisting of an annulus with inner and outer radii of 6 and 30 , respectively , in the ciao tool _ psextract_. these spectra were then fitted using xspec 11.2 . for all sources , we fitted data for energies from 0.3 - 10.0 kev using an absorbed power law model with pile - up ( phabs*powerlaw*pile - up ) . for most sources we found that the psf fraction treated for pile - up ( parameter 5 ) in the pile - up model , needed to be reduced from its default value of 0.95 to 0.85 to allow the grade morphing parameter ( parameter 4 ) to not be pegged at 1.0 and to allow the model to fit the data at higher energies . for some sources , the pile - up model had no effect on the spectral fits , even though the source was obviously piled - up ; hence we used trailed images to characterize those spectra . results of our fits with the pile - up model are listed in table [ tab : spec ] . figure [ fig : pile - up ] shows examples of spectra before and after the pile - up model is applied . trailed images were extracted using the ciao tool _ acisreadcorr_. during each row transfer , the detector is exposed to the source for 40@xmath62s , but the photons are recorded in the wrong y position , resulting in a streak along the x - axis . we calculated the total transfer streak exposure time using the following expression @xmath63 where @xmath64s , the time to transfer charge from one row to the next ; 1024 is the number of pixels in a row ; @xmath65 is the subarray fraction , @xmath66 is the number of pixels excluded near the source ; @xmath67 is the total exposure time in seconds ; and @xmath68 s , the time for one readout of the subarray . spectra were extracted using the ciao tool _ psextract _ with the source region defined as two boxes along the y - axis , each 6 pixels wide in the x - direction . the two boxes included the entire y - axis , except a region @xmath69 pixels from the y - coordinate of the piled - up source . the exposure and backscal keywords in the spectra were corrected for using the results of the ciao tool _ acisreadcorr _ and the calculation above . the resulting exposure times and total counts measured from the source in each trailed image are listed in table [ tab : trspec ] . for five of the observations , trailed image spectra could be extracted . we grouped the source spectra into bins containing 15 counts . then these spectra were fitted in xspec 11.2 with an absorbed power - law model ( phabs*powerlaw ) . for two of the sources , 2s 1711339 and 1e 1746.73224 , very few counts remained in the trailed image spectra . the results of fitting trailed image spectra are listed in table [ tab : spec ] and examples are shown in figure [ fig : trailed ] . for one of the sources , slx 1735269 , energy spectra were extracted from cc mode data taken on 2000 april 4 . a source region consisting of a 6 radius circle and a background region consisting of an annulus with inner and outer radii of 6 and 30 , respectively , was used in the ciao tool _ psextract _ to extract source and background spectra for the cc mode data and to create associated detector response files . we then grouped the data in the source spectrum so that each bin contained 25 counts . this observation fell on the acis - s2 chip , a front illuminated chip , so the low - energy response was reduced . in xspec 11.2 @xcite , we fitted data for energies of 1.0 - 1.8 kev and 2.2 - 9.0 kev with an absorbed power - law model ( phabs*powerlaw ) . fit results are listed in table [ tab : spec ] and are shown in figure [ fig : ccspec ] . the cc mode observations of slx 1735269 resulted in the largest number of source counts of any of our sources . hence we used these observations to investigate if more complicated spectral models are warranted . for example , @xcite reported an iron line with energy @xmath70 kev , fixed width 0.9 kev , and equivalent width 344 ev in addition to an absorbed power - law in observations of 4u 1708408 . to search for evidence of a similar line in slx 1735269 , we included a gaussian with a fixed centroid at 6.5 kev and fixed width of 0.9 kev . our fit resulted in a 3-sigma upper limit on the equivalent width of 975 ev , indicating that we had insufficient statistics to detect such a line . next , since a soft thermal component has been observed in several lmxbs , we fitted a blackbody in addition to the absorbed power - law . the fit gave @xmath71 with 253 degrees of freedom . an f - test comparing this fit to the absorbed power law had a value of 6.02 and a probability of @xmath72 , suggesting that a softer component may be present . the resulting fit parameters were : @xmath73 @xmath6 , blackbody temperature @xmath74 kev , blackbody normalization @xmath75 , power law index @xmath76 , and power law normalization @xmath77 photons kev@xmath7 @xmath6 s@xmath7 at 1 kev . since this observation with the largest number of source counts resulted in only a suggestion of a softer component , we concluded that we had insufficient statistics to distinguish spectral models in the rest of our observations , which had significantly fewer source counts . rxte _ observations of slx 1735269 with 3 - 25 kev fluxes @xmath78 ergs @xmath6 s@xmath7 show band limited noise levels with fractional amplitudes of 24 - 28% and break frequencies of 0.1 - 0.2 hz @xcite and a qpo near 0.9 hz with a fractional rms of @xmath79% and fwhm = 0.3 - 0.8 hz . @xcite also reported a very different power spectrum from low count rate " 1997 _ rxte _ observations of slx 1735269 with 3 - 25 kev fluxes @xmath80 ergs @xmath6 s@xmath7 , that had a break frequency of 2.3 hz and a fractional rms of 17% . @xcite fitted _ bepposax _ power spectra for 4u 181212 with a 4 lorentzian model after @xcite . the low - frequency lorentzian , fitting the low - frequency end of the band limited noise , had a fractional amplitude of 9.8% rms and a break frequency of 0.15 hz . a low - frequency qpo was also detected with _ bepposax _ at 0.73 hz with a fractional amplitude of 3.9% rms . we computed the sensitivity of our chandra observations to features in the power spectrum using @xmath81 where @xmath82 is the fractional root - mean - squared ( rms ) of the signal ; @xmath83 is the significance of the expected detection ; @xmath84 is the total observed count rate ; @xmath85 is the source count rate ; @xmath86 is the full - width half - max ( fwhm ) of the signal , which is approximately equal to twice the break frequency obtained with a broken power law model for band limited noise ( see belloni , psaltis , & van der klis 2002 ) ; and @xmath87 is the exposure time . for te - mode observations of slx 1735269 , 4u 1708408 , 2s 1711339 , 4u 181212 , and the 2000 aug 30 observation of 1e 1746.73224 , @xmath84 = 0.2 - 0.7 cts s@xmath7 , @xmath85 = 0.2 - 0.7 cts s@xmath7 , and @xmath87 = 950 - 6800 s , resulting in 3-@xmath5 sensitivities of 30 - 50% rms for features with @xmath88 hz . the 2002 july 16 observation of 1e 1746.73224 was longer , with @xmath87 = 8500 s , and had slightly higher count rates of @xmath89 cts s@xmath7 and @xmath90 cts s@xmath7 , resulting in a 3-@xmath5 sensitivity of 23% rms . our _ chandra _ te mode observations were not sensitive enough to detect broadband features at previously detected levels . however , the cc mode observation of slx 1735269 , with @xmath91 cts s@xmath7 , @xmath92 cts s@xmath7 , and @xmath93 s , had a much better 3-@xmath5 sensitivity of 9.3% rms for @xmath94 hz . hence band limited noise at levels observed in the brighter state of slx 1735269 @xcite should have been detectable in the cc mode observations ; however , the observed qpo was not detectable . we estimated a 3-@xmath5 sensitivity of 18% rms for @xmath95 hz , corresponding to the fainter state observed by @xcite , for our cc mode observation of slx 1735269 ; hence we were unable to confirm this state with our _ chandra _ observations . for the cc mode observation of slx 1735269 , we created a light curve using ciao tools that account for the fact that event times recorded in cc mode data were the times the events were read , not the times when the charge was deposited on the detector @xcite . we generated a power spectrum for the energy range 0.3 - 10 kev from the 2.85 ms cc mode light curve . no obvious variability was present in either the light curve or the power spectrum . integrating from 0.01 - 1 hz , we estimated a noise strength of @xmath96% rms for this observation ; since this noise level was near our detection threshold , we can not easily confirm that this was entirely source related . however , the lack of strong band limited noise at levels @xmath97% rms , combined with our 1 - 10 kev flux measurement of @xmath98 ergs @xmath6 s@xmath7 suggests that slx 1735269 was most likely in a state similar to the lowest count rate " state reported in @xcite . we obtained images of slx 1735269 , 1e 1746.73224 , and 4u 181212 using the united kingdom infrared telescope ( ukirt ) to search for near - infrared counterparts . for slx 1735269 , the seeing was approximately 0.7 , the airmass was 1.46 , the upper limit on the presence of a star in the _ chandra _ error circle was @xmath99 . the astrometric solution was performed using five stars from the 2mass in j catalog . the fit was good with an rms of 0.26 . the astrometric accuracy of the 2mass catalogue is @xmath100 . for 1e 1746.73224 the seeing was approximately 0.6 , the airmass was 1.64 , and the upper limit for a source in the _ chandra _ error circle was @xmath101 . the astrometric solution was performed using eight stars from the 2mass in j catalog . the fit was good with an rms of 0.23 . there was no 2mass image available for 4u 181212 ; hence , the astrometric solution of the ukirt image of the field containing 4u 181212 ( seeing @xmath102 , airmass 1.18 ) was obtained by identifying four usno - a1.0 stars in an optical digital sky survey ( dss ) image with a near infrared star . however , with an rms of 0.8 the astrometric solution was poor . a star appears to be present near the error circle , but its j band magnitude is difficult to assess since we have no j band magnitudes for other stars in the field for comparison . chandra _ we observed eight faint , little studied likely lmxbs . of the eight systems , we detected five in at least one observation : 4u 1708408 , 2s 1711339 , slx 1735269 , 1e 1746.73224 , and 4u 181212 . chandra _ observations reported in this paper resulted in precise locations for all five of these objects , allowing for future deep optical and infrared observations . ukirt images of slx 1735269 , 1e 1746.73224 , and 4u 181212 did not reveal counterparts in the j band . energy spectra for these five systems were generally consistent with hard power laws ( with photon index 1.5 - 2 ) typical for faint lmxbs ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . absorbed fluxes ( 1 - 10 kev ) for the detected sources ranged from @xmath103 ergs @xmath6 s@xmath7 . the relatively narrow range of power law indices and observed fluxes suggest that all five of the detected systems may be in a similar state . energy spectra and fluxes for 4u 1708408 , slx 1735269 , and 4u 181212 were consistent with previous measurements in similar energy bands , supporting earlier findings that these systems are persistent . we observed slx 1735269 at a 1 - 10 kev flux level of @xmath104 ergs @xmath6 s@xmath7 in both observations . this combined with the lack of detection of strong band limited noise in the cc mode observation suggests that slx 1735269 was in a similar state to the low count rate " state reported in @xcite . the energy spectrum and flux measured using the pile - up model with _ chandra _ on 2000 june 9 for 2s 1711339 were consistent with that measured with the _ bepposax _ nfi on 2000 feb 29 and with _ bepposax _ wfc upper limits on 2000 march 22 , suggesting that we were observing either an extended tail of the bright 1998 - 1999 outburst observed with _ bepposax _ and _ rxte _ or a faint outburst . an estimate of the 2 - 6 kev flux reported in @xcite from the 2000 june 9 _ chandra _ observation was a factor of @xmath105 fainter than the flux we measured with _ chandra _ for the same observation . we believe that @xcite did not properly account for the effects of pulse pile - up . in our second _ chandra _ observation of 2s 1711339 , in 2002 march , it had faded below detectability in cc mode . 1e 1746.73224 , the least studied of our five detected sources , and the only one of the five not to have previously shown x - ray bursts , was observed with _ chandra _ at absorbed 1 - 10 kev fluxes of @xmath106 ergs @xmath6 s@xmath7 and @xmath107 ergs @xmath6 s@xmath7 in 2000 and 2002 , respectively . in the _ rosat _ all - sky survey @xcite , 1e 1746.73224 was measured at a count rate of 0.08 counts s@xmath7 in the position sensitive proportional counter ( pspc ) . this corresponds to an absorbed 1 - 10 kev flux of @xmath108 ergs @xmath6 s@xmath7 , if the spectral shape we measured with _ chandra _ is assumed . these fluxes are a factor of @xmath96 higher than that observed in the _ einstein _ galactic plane survey @xcite . the similarity of the _ chandra _ and _ rosat _ fluxes suggests this source is persistent , and the _ einstein _ measurement suggests that it , like most persistent sources , is also variable . the three undetected systems , grs 1736297 , ks 1739304 , and slx 1746331 , had shown previous transient behavior . these non - detections were not due to positional errors . upper limits on their fluxes were @xmath109 ergs @xmath6 s@xmath7 , two to three orders of magnitude fainter than our faintest detection . if these sources were located at the galactic center , at a distance of 8 kpc , then these fluxes would correspond to luminosities of @xmath110 ergs s@xmath7 . the united kingdom infrared telescope is operated by the joint astronomy centre on behalf of the u.k . particle physics and astronomy research council , and some of the data reported here were obtained as part of the ukirt service programme . ck and sp are party supported by sao grants go0 - 1054a and go2 - 3046b . whgl is grateful for support from nasa . aldcroft , t.l . , karovska , m. , cresitello - ditmar , m.l . , cameron , r.a . , & markevitch , m.l . 2000 , proc . spie , 4012 , 650 arnaud , k.a . 1996 , astronomical data analysis software and systems v , eds . g. jacoby & j. barnes , asp conf . series , 101 , 17 barret , d. et al . 2000 , apj 533 , 329 barret , d. , mcclintock , j.e . , & grindlay , j.e . 1996 , apj , 473 , 963 barret , d. , olive , j.f . , & oosterbroek , t. 2003 , in press , astro - ph/0301130 bazzano , a. et al . 1997 , iau circ . , 6668 , 2 belloni , t. , psaltis , d. , & van der klis , m. 2002 , apj , 572 , 392 carpenter , g.f . , eyles , c.j . , skinner , g. k. , wilson , a. m. , & willmore , a.p . 1977 , mnras , 179 , 297 cherepashchuk , a.m. et al . 1994 , a&a , 289 , 419 cocchi , m. et al . 2000 , , 357 , 527 cornelisse , r. et al . 2002 , , 392 , 885 david , p. et al . 1997 , , 322 , 229 dickey , j.m & lockman , f.j . 1990 , ara&a , 28 , 215 elvis , m. et al . 1992 , apjs , 80 , 257 esin , a.a . 2001 , apj , 555 , 483 forman , w. , jones , c. , tananbaum , h. 1976 , apj , 206 , l29 forman , w. et al . 1978 , apjs , 38 , 357 goldwurm , a. et al . 1996 , , 310 , 857 grebenev , s.a . , pavilinsky , m.n . , & sunyaev , r.a . 1996 , in the 2nd integral workshop : the transparent universe , esa sp-382 ( noordwijk : esa ) , 183 greenhill , j.g . thomas , r.m . & duldig , m.l . 1979 , nature , 279,620 hertz , p. & grindlay , j.e . 1984 , apj , 278 , 137 hulleman , f. , tennant , a.f . , van kerkwijk , m.h . , kulkarni , s.r . , kouveliotou , c. & patel , s.k . 2001 , apj , 563 , l49 migliari , s. et al . 2003 , mnras , in press , astro - ph/0303136 motch , c. et al . 1998 , a&as , 132 , 341 murakami , t. et al . 1983 , pasj , 35 , 531 pavilinsky , m.n . , grebenev , s.a . , & sunyaev , r.a . 1992 , sval , 18 , 88 pavilinsky , m.n . , grebenev , s.a . , & sunyaev , r.a . 1994 , apj , 425 , 110 skinner , g. k. et al . 1987 , nature , 330 , 544 skinner , g.k . , foster , a.j . , willmore , a.p . , & eyles , c.j . 1990 , mnras , 243 , 72 sunyaev , r. et al . 1991 , adv . space res . , 11(8 ) , 177 tanaka , y. & lewin , w.h.g 1995 , in x - ray binaries , eds . lewin , j. van paradijs , & e.p.j . van den heuvel , ( new york : cambridge university press ) , 126 van der klis , m. 1994 , apjs , 92 , 511 van der klis , m. 2000 , ann . astrophys . , 38 , 717 van paradijs , j. 1995 , in x - ray binaries , eds . lewin , j. van paradijs , & e.p.j . van den heuvel , ( new york : cambridge university press ) , 536 van paradijs , j. & van der klis , m. 1994 , , 281 , l17 voges , w. et al . 1999 , , 349 , 389 warwick , r.s . , norton , a.j . , turner , m.j.l . , watson , m.g . & willingdale , r. 1988 , mnras , 232 , 551 wijnands , r. & van der klis , m. 1999a , apj , 514 , 939 wijnands , r. & van der klis , m. 1999b , , 345 , l35 zavlin , v.e . , pavlov , g.g , sanwal , d. , & trmper , j. 2000 , apj , 540 , l25 lcccc 4u 170840 & 1996 - 1997 & _ rxte _ asm & 2 - 10 kev & @xmath8 + & mid-1998 & _ rxte _ asm & 2 - 10 kev & @xmath9 + & 1999 august & _ bepposax _ & 2 - 10 kev & @xmath14 + & 2000 may 15 & _ chandra _ & 1 - 10 kev & @xmath111 + & 2000 jun 18 & _ rxte _ pca & 2 - 10 kev & @xmath18 + & mid-2001 & _ rxte _ asm & 2 - 10 kev & @xmath10 + & 2001 august & _ bepposax _ & 2 - 10 kev & @xmath112 + 2s 1711339 & 1998 july - 1999 may & _ rxte _ asm & 2 - 10 kev & @xmath113 + & 1998 july - 1999 may & _ bepposax _ wfc & 2 - 28 kev & @xmath22 + & 2000 february 29 & _ bepposax _ nfi & 2 - 6 kev & @xmath25 + & 2000 march 22 & _ bepposax _ wfc & 2 - 28 kev & @xmath114 + & 2000 june 9 & _ chandra _ & 1 - 10 kev & @xmath115 + & 2002 march 12 & _ chandra _ & 1 - 10 kev & @xmath116 + slx 1735269 & 1996 - 2003 & _ rxte _ asm & 2 - 10 kev & @xmath35 + & 1997 february - may & _ rxte _ pca & 3 - 25 kev & @xmath117 + & 1997 october & _ rxte _ pca & 1 - 20 kev & @xmath118 + & 2000 april 4 & _ chandra _ & 1 - 10 kev & @xmath98 + & 2000 may 23 & _ chandra _ & 1 - 10 kev & @xmath119 + grs 1736297 & 1990 september - october & _ granat _ art - p & 3 - 12 kev & @xmath120 + & 2000 may 31 & _ chandra _ & 1 - 10 kev & @xmath121 + ks 1739304 & 1989 & _ mir_/kvant & 2 - 30 kev & @xmath104 + & 1990 & _ granat _ art - p & 8 - 20 kev & @xmath122 + & 2002 may 5 & _ chandra _ & 1 - 10 kev & @xmath116 + slx 1746331 & 1985 august & spacelab-2 xrt & 2 - 10 kev & @xmath123 + & 2000 june 9 & _ chandra _ & 1 - 10 kev & @xmath124 + 1e 1746.73224 & 1978 - 1981 & _ einstein _ & 2 - 10 kev & @xmath39 + & 1985 august & spacelab-2 xrt & 2 - 10 kev & @xmath125 + & 1990 - 1991 & _ rosat _ pspc & 1 - 10 kev & @xmath108 + & 2000 august 30 & _ chandra _ & 1 - 10 kev & @xmath106 + & 2002 july 15 - 16 & _ chandra _ & 1 - 10 kev & @xmath107 + 4u 181212 & & _ oso-7 _ , _ ariel v _ , _ heao-1 _ , _ exosat _ & 2 - 10 kev & @xmath126 + & 1996 - 2003 & _ rxte _ asm & 2 - 10 kev & @xmath42 + & 2000 june 14 & _ chandra _ & 1 - 10 kev & @xmath127 + [ tab : fluxsum ] lclcccc 4u 170840 & 661 & 2000 may 15 & te & 0.841 s & 1.2 & y + 2s 1711339 & 662 & 2000 jun 9 & te & 0.841 s & 1.0 & y + & 2695 & 2002 mar 12 & cc & 2.85 ms & 4.7 & n + slx 1735269 & 664 & 2000 apr 4 & cc & 2.85 ms & 1.4 & y + & 663 & 2000 may 23 & te & 0.841 s & 1.7 & y + grs 1736297 & 665 & 2000 may 31 & te & 0.841 s & 1.3 & n + & 666 & 2000 jul 7 & cc & 2.85 ms & 12.2 & n + ks 1739304 & 2696 & 2002 apr 27 & cc & 2.85 ms & 3.4 & n + & 2697 & 2002 may 5 & te & 3.24 s & 0.8 & n + slx 1746331 & 667 & 2000 jun 9 & te & 0.841 s & 1.7 & n + & 668 & 2000 jul 24 & cc & 2.85 ms & 4.1 & n + 1e 1746.73224 & 669 & 2000 aug 30 & te & 3.24 s & 6.8 & y + & 2698 & 2002 jul 15 - 16 & te & 0.841 s & 8.5 & y + 4u 181212 & 670 & 2000 jun 14 & te & 3.24 s & 1.0 & y + [ tab : obs ] lll 4u 170840 & @xmath128 & @xmath129 + 2s 1711339 & @xmath130 & @xmath131 + slx 1735269 & @xmath132 & @xmath133 + 1e 1746.73224 & @xmath134 & @xmath135 + & @xmath136 & @xmath137 + 4u 181212 & @xmath138 & @xmath139 + [ tab : loc ] lllllllc 4u 170840 & 2000 may 15 & tr & @xmath140 & @xmath141 & & @xmath142 & 87 + 2s 1711339 & 2000 jun 9 & te & @xmath143 & @xmath144 & @xmath145 & @xmath146 & 4.4 + & & tr & @xmath147 & @xmath148 & & @xmath149 & 11 + & 2002 mar 12 & cc & 1.4 & 1.9 & & & @xmath150 + slx 1735269 & 2000 apr 4 & cc & @xmath151 & @xmath152 & & @xmath153 & 19 + & 2000 may 23 & tr & @xmath154 & @xmath155 & & @xmath156 & 21 + grs 1736297 & 2000 may 31 & te & 1.1 & 1.8 & & & @xmath157 + ks 1739304 & 2002 may 5 & te & 1.4 & 1.5 & & & @xmath150 + slx 1746331 & 2000 jun 9 & te & 0.4 & 1.5 & & & @xmath158 + 1e 1746.73224 & 2000 aug 30 & te & @xmath159 & @xmath160 & @xmath161 & @xmath162 & 2.1 + & 2002 jul 15 - 16 & te & @xmath163 & @xmath164 & @xmath165 & @xmath166 & 3.2 + & & tr & @xmath167 & @xmath168 & & @xmath169 & 3.3 + 4u 181212 & 2000 jun 14 & tr & @xmath170 & @xmath171 & & @xmath172 & 44 + [ tab : spec ] llcc 4u 170840 & 2000 may 15 & 11.2 & 412 + 2s 1711339 & 2000 jun 9 & 8.84 & 86 + slx 1735269 & 2000 may 23 & 16.0 & 228 + 1e 1746.73224 & 2002 jul 15 - 16 & 79.4 & 117 + 4u 181212 & 2000 jun 14 & 11.9 & 312 + [ tab : trspec ]
* spatial gsa allowed the production of sobol index maps , enhancing the relative weight of each uncertain parameter on the variability of calculated output parameter of interest . * the sobol index maps illustrate the major influence of the modeller choices , when using the hr topographic data in 2d hydraulic models with respect to the influence of hr dataset accuracy . * added value is for modeller to better understand limits of his model . * requirements and limits for this approach are related to subjectivity of choices and to computational cost . in hydraulics , deterministic numerical modelling tools based on approximating solutions of the 2d shallow water equations ( swe ) system are commonly used for flood hazard assessment @xcite . this category of tools describes water free surface behavior ( mainly elevation and discharge ) according to an engineering conceptualization , aiming to provide to decision makers information that often consists in a flood map of maximal water depths . as underlined in @xcite , good practice in hydraulic numerical modelling is for modellers to know in detail the chain of concepts in the modelling process and to supply to decision makers possible doubts and deviation between what has been simulated and the reality . indeed , in considered swe based models , sources of uncertainties come from ( @xmath2 ) hypothesis in the mathematical description of the natural phenomena , ( @xmath3 ) numerical aspects when solving the model , ( @xmath4 ) lack of knowledge in input parameters and ( @xmath5 ) natural phenomena inherent randomness . errors arising from @xmath2 , @xmath3 and @xmath4 may be considered as belonging to the category of epistemic uncertainties ( that can be reduced _ e.g. _ by improvement of description , measurement ) . errors of type @xmath5 are seen as stochastic errors ( where randomness is considered as a part of the natural process , _ e.g. _ in climatic born data ) @xcite . at the same time , the combination of the increasing availability of high resolution ( hr ) topographic data and of high performance computing ( hpc ) structures , leads to a growing production of hr flood models @xcite . for non - practitioner , the level of accuracy of hr topographic data might be erroneously interpreted as the level of accuracy of the hr flood models , disregarding uncertainty inherent to this type of data use , not without standing the fact that other types of above mentioned errors occur in hydraulic modelling . topographic data is a major input for flood models , especially for complex environment such as urban and industrial areas , where a detailed topography helps for a better description of the physical properties of the modelled system @xcite . in the case of an urban or industrial environment , a topographic dataset is considered to be of hr when it allows to include in the topographic information the elevation of infra - metric elements @xcite . these infra - metric elements ( such as sidewalks , road - curbs , walls , _ etc . _ ) are features that influence flow path and overland flow free surface properties . at megacities scale , hr topographic datasets are getting commonly available at an infra - metric resolution using modern gathering technologies ( such as lidar , photogrammetry ) through the use of aerial vectors like unmanned aerial vehicle or specific flight campaign @xcite . moreover , modern urban reconstruction methods based on features classification carried out by photo - interpretation process , allow to have high accuracy and highly detailed topographic information @xcite . photo - interpreted hr datasets allow to generate hr dems including classes of impervious above ground features @xcite . therefore generated hr dems can include above ground features elevation information depending on modeller selection among classes . based on hr classified topographic datasets , produced hr digital elevation model ( dem ) can have a vertical and horizontal accuracy up to @xmath6 @xcite . even though being of high accuracy , produced hr dems are assorted with the same types of errors as coarser dems . errors are due to limitations in measurement techniques and to operational restrictions . these errors can be categorized as : ( @xmath2 ) systematic , due to bias in measurement and processing ; ( @xmath3 ) nuggets ( or blunder ) , which are local abnormal value resulting from equipment or user failure , or to occurrence of abnormal phenomena in the gathering process ( _ e.g. _ birds passing between the ground and the measurement device ) or ( @xmath4 ) random variations , due to measurement / operation inherent limits ( see @xcite ) . moreover , the amount of data that composes a hr classified topographic dataset is massive . consequently , to handle the hr dataset and to avoid prohibitive computational time , hydraulic modellers make choices to integrate this type of data in the hydraulic model , possibly decreasing hr dem quality and introducing uncertainty @xcite . as recalled in the literature @xcite , in hr flood models , effects of uncertainties related to hr topographic data use on simulated flow is not yet quantitatively understood . to evaluate uncertainty in deterministic models , uncertainty analysis ( ua ) and sensitivity analysis ( sa ) have started to be used @xcite and @xcite and become broadly applied for a wild range of environmental modelling problems @xcite . ua consists in the propagation of uncertainty sources through model , and then focuses on the quantification of uncertainties in model output allowing robustness to be checked @xcite . sa aims to study how uncertainty in a model output can be linked and allocated proportionally to the contribution of each input uncertainties . both ua and sa are essential to analyze complex systems @xcite , as study of uncertainties related to input parameters is of prime interest for applied practitioners willing to decrease uncertainties in their models results @xcite . in 1d and 2d flood modelling studies , approaches based on sampling based methods are becoming used in practical applications for ua . for sa , depending on applications and objectives , different categories of variance based approaches have been recently applied in flood modelling studies ( mainly in 1d ) such as local sensitivity analysis ( lsa ) @xcite or more recently , a global sensitivity analysis ( gsa ) based on a screening method has been implemented in 2d flood modelling application @xcite . lsa focuses on fixed point in the space of the input and aims to address model behavior near parameters nominal value to safely assume local linear dependences on the parameter . lsa can use either a differentiation or a continuous approach @xcite . lsa based on differentiation approach performs simulations with slight differences in a given input parameter and computes the difference in the results variation , with respect to the parameter variation . lsa based on continuous approach differentiates directly the equations of the model , creating sensitivity equation @xcite . the advantages of lsa approaches are that they are not resource demanding in terms of computational cost , drawback being that the space of input is locally explored assuming linear effects only . linear effects means that given change in an input parameter introduces a proportional change in model output , in opposition to nonlinear effects . lsa approaches perform reasonably well with swe system even if nonlinear effects occur punctually ( see @xcite ) . nonetheless , important nonlinear effects in model output might arise when parameters are interacting and when solution becomes discontinuous . lsa consequently becomes not suited @xcite in such a context , which is likely to occur in case of 2d swe based simulation of overland flow . gsa approaches rely on sampling based methods for uncertainty propagation , willing to fully map the space of possible model predictions from the various model uncertain input parameters and then , allow to rank the significance of the input parameter uncertainty contribution to the model output variability @xcite . gsa approaches are well suited to be applied with models having nonlinear behavior and when interaction among parameters occurs @xcite . these approaches going through an intensive sampling are computationally demanding , as they most often rely on monte - carlo ( mc ) approach , even though some more parsimonious sampling method such as latin hypercube or pseudo - monte carlo are sometimes applied ( see @xcite for a review ) . most commonly , gsa approaches rely on : * screening methods , such as morris method @xcite ; * sobol indices computation , that considers the output hyperspace ( @xmath7 ) as a function ( @xmath8 ) and performs a functional decomposition @xcite or a fourier decomposition ( fast method ) of the variance . as fully detailed in @xcite , screening techniques ( _ e.g. _ morris method ) allow to classify uncertain input parameters in three categories : those that have negligible effect ; those that have linear effect ; and those that have nonlinear effects or effects in interaction with other input parameters . sobol indices ( or variance - based sensitivity indices ) will explain the share of the total variance in the space of output due to each uncertain input parameter and/or input interaction . gsa has started to be applied in 1d hydraulic modelling in practical applications for hierarchical ranking of uncertain input parameters @xcite . as for 1d , applying a gsa to flooding issues in 2d modelling requires method awareness among the community , practical tools development and computational resources availability . moreover an analysis on spatialization of input uncertain parameters and on output variable is specifically needed in 2d @xcite . recently , gsa using a screening method has been implemented in 2d flood modelling application @xcite tackling ranking of uncertain input parameters using points and zonal approaches . computation of sensitivity maps such as maps of sobol index is a promising outcome that has been achieved for other types of water related issues @xcite . to date , ua and sa have not yet been performed to specifically study uncertainty in 2d urban flood simulations related to hr classified topographic data integration . indeed , due to the curse of dimensionality , sa methods have seldom been applied to environmental models with both spatially distributed inputs and outputs @xcite . such a problematic raises needs of specific tools , computational resource and methods application . among sa methods , a global sensitivity analysis ( gsa ) is implemented in this study . gsa approach is selected over lsa as 2d overland flow process simulation through swe system of partial differential equations , is viewed as being largely nonlinear , with discontinuous solution and interactions between parameters . this paper aims to study uncertainty related to hr topographic data integration in 2d flood modelling approach . the objective of the study is to perform an ua and a sa on two categories of uncertain parameters ( measurement errors and uncertainties related to operator choices ) relative to the use of hr classified topographic data in a 2d urban flood model having spatial inputs and outputs . among sa methods , a global sensitivity analysis ( gsa ) is implemented to produce sensitivity maps based on sobol index computation . carrying out these objectives will demonstrate the feasibility , the added values and limitations of ua and sa implementation in 2d hydraulic modelling , in a context where spatial variability and interaction are likely to occur . moreover , modeller knowledge about challenges and expectations related to hr classified data use in hr urban flood modelling will be enhanced . the study case is the low var river valley ( nice , france ) where flooding events occurred in the last decades in the highly urbanized downstream part of the valley @xcite . the output of interest is the overland flow water level ( @xmath8 ) . the used hr dems are based on classified 3d dataset created from photointerpretation procedure . a proof of concept of gsa application to 2d hydraulic modelling voluntary choosing a resource requiring problem has been developed and the method applied over an innovative concern related to the use of hr topographic data . following this introduction ( part [ sec : intro ] ) , the next part of the paper ( part [ sec : method ] ) introduces the test case context for sa methods uses , then enhances description of used hr topographic dataset , gives overview of implemented methodology for the spatial gsa and introduces developed tools . the third part ( [ sec : results ] ) of the paper presents results of ua and gsa , first at punctual then at spatial levels . the last parts ( part [ sec : discussion ] and [ sec : summary - and - conclusions ] ) discuss outcomes and limits of our approach , providing concluding remarks . the study area is a @xmath9 domain that represents the last five downstream kilometers of the low var valley , located in nice , france ( figure [ fig1 ] ) . in the test basin , two major river flood events occurred in last decades ( @xmath10 of november @xmath11 ; @xmath12 of november @xmath13 ) . a hr topographic data gathering campaign fully covered the domain in @xmath14 . the characteristics of the river basin and of the @xmath11 flood event are described in @xcite . between @xmath11 and @xmath14 ( date of event used for simulation and the date of the hr topographic data gathering campaign ) , the studied area has considerably changed . indeed , levees , dikes and urban structures have been implemented , changing physical properties of the river / urban flood plain system . thus , the objective is not to reproduce the event , but simply to use the framework of this event as a case study to carry out the ua and the sa . as mentioned in the introduction section , a gsa approach using sobol index is suitable to compute sensitivity maps @xcite . the method implemented to carry out the spatial gsa is presented in detail in @xcite , and the upcoming description of the implemented gsa gives to the reader a summary of key elements for understanding . the design and the quality of a photo - interpreted dataset are highly dependent on photogrammetry dataset quality , on classes definition , and on method used for digitalization of vectors @xcite . reader can find details regarding principle of modern aerial photogrammetry technology in @xcite . the photogrammetric campaign carried out over the lower var valley , at a low flight elevation , allowed a pixel resolution at the ground level of @xmath6 and had a high overlapping ratio ( @xmath15 ) among aerial pictures . consequently , these characteristics allowed the production of a high quality photogrammetric dataset . using the photogrammetric dataset , the photo - interpretation process has been carried out , to create a classified vectorial dataset through digitalization of classes of polylines , polygons and points . this photo - interpreted dataset has been designed with a total number of @xmath16 different classes representing large and thin above ground features ( _ e.g. _ buildings , concrete walls , road - gutters , stairs , _ etc . _ ) . the specificities of the given photo - interpretation process can be found in @xcite . for hydraulic modelling purpose , @xmath17 classes over the @xmath16 are considered to represent above ground features impacting overland flow path , as shown in figure [ fig1 ] ( a , b ) . classified data mean horizontal and vertical accuracy is @xmath18 . errors in photo - interpretation which results from feature misinterpretation , addition or omission are estimated to represent @xmath19 of the total number of elements . to control average level of accuracy and level of errors in photo - interpretation , the municipality has carried out a terrestrial control of data accuracy over @xmath20 of the domain covered by the photogrammetric campaign . for our application , the 3d classified data of the low var river valley is used to generate specific dem adapted to surface hydraulic modelling . for the gsa approach ( see next section ) , only the input dem changes from one simulation to another and the hydraulic parameters of the model are set identically for the simulations . hydraulic conditions of the study case implemented in the models can be summarized as follow : a constant discharge of @xmath21 is applied as input boundary condition to reach a steady flow state condition almost completely filling the var river bed . this steady condition is the initial condition for the gsa and a @xmath22 hours estimated hydrograph from the @xmath11 flood event is simulated @xcite . the manning s friction coefficient ( @xmath23 ) is spatially uniform on overland flow areas with a standard value of @xmath24 , which corresponds to a concrete surface . no energy losses properties have been included in the 2d hydraulic model to represent the bridges , piers or weirs . downstream boundary condition is an open sea level ( neumann boundary condition ) to let water flows out . falling within category of gsa approaches , screening methods allow in a computationally parsimonious way , to discriminate among numerous uncertain input parameters those that have little effect from those having linear , nonlinear or combined effects in output variance @xcite . screening methods principle consists in fixing an input parameters set and performing an initial run . then , for one parameter at a time , a new value of the parameter is randomly chosen and a new run is performed . variation in the run output is checked . this operation is completed for all the parameters , @xmath23 times with @xmath23 equals to the total number of input parameters . screening methods perform well to discriminate influencing parameters on output variability with 2d flood modelling studies @xcite . gsa approaches relying on sobol index computation go one step further , allowing to quantify the contribution to the output variance of the main effect of each input parameters @xcite . sobol index is based on functional decomposition of variance ( anova ) , considering @xmath25 the model output of interest as follow : @xmath26 ; where @xmath27 is the model function , @xmath28 are @xmath2 independent input uncertain parameters with known distribution . sobol indices ( @xmath29 ) of parameter @xmath30 are defined as : @xmath31/var(y),\ ] ] where @xmath32 is the expectation operator . @xmath33 being the variance of conditional expectation of @xmath25 for @xmath30 over the total variance of @xmath25 , @xmath33 value will range between @xmath34 $ ] . @xmath29 computations are computationally costly as it requires to explore the full space of inputs and therefore an intensive sampling is necessary @xcite . objective being to quantify impacts of input parameters , gsa approach using sobol index is best suited for sensitivity maps production . in this study , the implemented gsa follows standard steps used for such type of approach as summarized in @xcite or in @xcite . the steps of the method are presented in the figure [ fig2 ] : specification of the problem notably by choosing uncertain parameters and output of interest ( step a ) ; assessing probability density function ( pdf ) of uncertain parameters ( step b ) ; propagating uncertainty , using a random sampling approach in our case ( step c ) ; ranking the contribution of each input parameters regarding the output variance ( step d ) . first steps of the approach ( a and b ) are the most subjective ones . for the study purpose , steps a and b are treated as follow . three input parameters related to uncertainties when willing to use hr 3d classified data in 2d hydraulic models are ( @xmath2 ) one parameter related to the topographic input error ( called var . @xmath35 ) and ( @xmath3 ) two parameters related to modeller choices , when including hr data in 2d hydraulic code ( called var . @xmath36 and var . @xmath37 ) are considered in this gsa practical case . these three parameters are considered as independent . first , the uncertainties related to measurement errors in hr topographic dataset are considered through var . this parameter is an error randomly introduced for every point of the highest resolution dem ( @xmath38 ) following a draw according to a normal distribution pdf , where the standard deviation is equal to the rmse value @xmath39 : @xmath40 . as from one point to the next one , the normal pdf is drawn independently , it results in a spatialization that follows a uniform distribution . hundred maps of var . @xmath35 are generated and combined with the ( @xmath41 ) resolution dems . then , for uncertainties related to modeller choices when including hr data in hydraulic code , two variables are considered : var . @xmath36 and var . @xmath37 . @xmath36 is a categorical ordinal parameter having values representing the level of above ground features details impacting flow direction included in dsm . @xmath42 is a dtm ( digital terrain model ) only , @xmath43 is @xmath42 combined with buildings elevation inclusion , @xmath44 is @xmath43 completed with walls , and @xmath45 is @xmath44 plus thin concrete structures ( sidewalks , roar - curbs , _ etc . _ ) . @xmath37 represents choices made by the modeller concerning the computational grid cells resolution in the model . in the hydraulic code used for this study ( fullswof_2d described in next subsection ) , the grid cells are regular . this parameter var . @xmath37 can have five discrete values from @xmath38 to @xmath46 . at 1 m resolution , number of computational points of the grid is above @xmath47 million and at the @xmath46 resolution grid size is @xmath48 computational points . the bounding of this parameter is justified as on one hand , a grid resolution lower than @xmath38 would result in prohibitive computational time . on the other hand , resolutions coarser than @xmath46 do not sound to be a relevant choice for a modeller willing to create a hr model , as up - scaling effects would make the use of the hr topographic data that are used as input irrelevant . a total of @xmath0 dems are generated and used in the implementation of the gsa . the dems generation process ( step b , figure [ fig2 ] ) , explains as follow . four dems at the finest resolution ( var . @xmath49 ) are generated to implement all of the four var . @xmath36 possible scenarios . then each of these four dems is combined with the hundred var . @xmath35 grids producing @xmath50 dems . eventually , the @xmath50 dems combining all of the var . @xmath36 and var @xmath35 possibilities combinations are resampled to resolution @xmath51 creating a database of @xmath0 dems where all the defined input parameters can possibly be combined . the propagation of uncertainty ( step c ) is carried out using a mc approach to randomly sample in the database . non - parsimonious approach consisting in computing a maximum number of simulations among the @xmath0 possible cases has been carried out to generate a database of results . in total , @xmath52 simulations out of the @xmath0 possible were computed to feed the result database using the available @xmath53 cpu hours on a cluster ( cluster described in next sub - section ) . therefore , to make sure that the input space would be extensively explored , for all of the @xmath54 possible var . @xmath37/ var . @xmath36 combinations , at least @xmath16 over the @xmath55 possible var . @xmath35 drawn were performed . as the exploration of the space of input is restricted to @xmath52 simulations over @xmath0 possible cases , an evaluation of the convergence is performed to assess if the convergence of the mc method is reached . figure [ fig3 ] illustrates the evolution of the convergence of the mean of the hyperspace of the output of interest @xmath8 for three points ( points located in figure [ fig4 ] ) , increasing @xmath56 through a random sampling in the result database . it is reminded here that the output of interest is the simulated maximal overland flow water depth . an asymptotic convergence of the mc method is observed for the three points , respectively when the sample size ( @xmath56 ) is larger to @xmath57 simulations . globally , over the 20 selected points , when n reaches a threshold value between @xmath58 , the stabilization of the convergence is observed . points of interest ( points @xmath59 , @xmath60 and @xmath61 located on figure [ fig4]).,scaledwidth=98.0% ] step d consists in the computation of si using the output database . sobol index of var . @xmath37 , var . @xmath36 and var . @xmath37 , respectively @xmath62 , @xmath63 and @xmath64 are computed following eq . ( 1 ) at points of interests . spatialization of gsa approach is based on discrete realization of spatially distributed input variables as described in @xcite , and discrete computation of output to produce sensitivity maps ( as described in @xcite ) . promthe ( a parametric modelling environment ) , has been coupled with fullswof_2d ( a 2d free surface modelling code ) over a high performance - computing ( hpc ) structure @xcite . promthe is an environment for parametric computation that allows carrying out uncertainties propagation study , when coupled ( or warped ) to a code . this software is freely distributed by irsn ( http://promethee.irsn.org/doku.php ) . promthe allows the parameterization with any numerical code and is optimized for intensive computing . moreover , statistical post - treatment , such as ua and sa can be performed using promthe as it integrates r statistical computing environment @xcite . fullswof_2d ( for full shallow water equation for overland flow in 2 dimensions ) is a code developed as free software based on 2d swe @xcite . two parallel versions of the code have been developed allowing to run calculations under hpc structures @xcite . in fullswof_2d , the 2d swe are solved using a well - balanced finite volume scheme based on the hydrostatic reconstruction ( see @xcite ) . the finite volume scheme is applied on a structured spatial discretization using regular cartesian meshing . for the temporal discretization , based on the cfl criterion , a variable time step is used . the hydrostatic reconstruction ( which is a well - balanced numerical strategy ) allows to ensure that the numerical treatment of the system preserves water depth positivity and does not create numerical oscillation in case of a steady states , where pressures in the flux are balanced with the source term here ( topography ) . different solvers can be used hll , rusanov , kinetic @xcite , vfroe - ncv combined with first order or second order ( muscl or eno ) reconstruction . the hll solver has been used in this study with a first order muscl reconstruction method . on the hpc structure ( interactive computation centre of nice sophia antipolis university ) , up to @xmath65 cpus are available and up to @xmath66 simulations can be launched simultaneously using promthe - fulswof_2d . a database of flood maps results has been produced using a total of @xmath53 cpu hours . the required unitary computation time is two hours over @xmath67 cpus , for simulations using the finest resolution grid size ( @xmath38 ) , which has @xmath68 millions of computational points . at the coarsest resolution ( @xmath46 ) , the grids size is decreased to @xmath69 computational points and using @xmath67 cpus , the computational time decreases to few minutes . this section presents the results of the ua and the gsa . a subarea is selected in the flooded area of the domain to carry out the spatial analysis . this subarea is @xmath70 , representing one quarter of the total spatial extent of the model . @xmath54 points of interest are defined in the selected flooded area of the subarea ( figure [ fig4 ] ) . points @xmath71 to @xmath72 are spread in and around the main streets . these streets are densely urbanized . points @xmath73 to @xmath74 are located in less urbanized areas ( stadium , parking , small agricultural field , _ etc . _ ) . moreover , from points @xmath75 to @xmath54 , points are located in areas which are at the edge of the flood extent , either in open area ( points @xmath75 and @xmath74 ) or where above ground features are densely present ( points @xmath76 to @xmath54 ) . mean and variance of computed maximal water depth ( @xmath8 ) at the different points of interest are presented in figure [ fig4 ] . means and standard deviations of @xmath8 values are computed using the full size database ( @xmath77 ) . over the @xmath54 points of interest , importance of the variability introduced by uncertain input parameter is significant ( @xmath78 in average ) . moreover , variability in @xmath8 variance can be important as the minimal variance is @xmath79 ( point @xmath76 ) , and the maximal variance is @xmath80 ( point @xmath81 ) . further interpretation , such as the analysis of the trend in the magnitude of variance changes from one point to another , is not accessible for generalization using punctual observation only . nevertheless , studying the distribution of @xmath8 at points of interest gives another insight to carry out the uncertainty analysis . figure [ fig5 ] illustrates @xmath8 distributions using the complete set of available model runs in the database for three points . @xmath8 follows a normal distribution , as observed for point @xmath60 or distribution can be bi - modal as observed for point @xmath61 . the difference between the normal and bi - modal distribution of @xmath8 is not always clearly observed ( point @xmath59 ) . most of the clearly observed bimodal distribution ( ten out of the twenty points ) occurs for the points located in the central part of the highly urbanized area ( points @xmath71 to @xmath72 ) . this area is largely flooded , and seven points have here a clearly marked bimodal distribution . in largely flooded but relatively less urbanized areas , the trend is reversed as five out of six points have a normal distribution . lastly for the points located at the edge of the flooded areas , two over four points have a bimodal distribution , whereas the two others have a normal one . bimodal distributions lead to larger amplitude in @xmath8 distribution . the bimodal distribution illustrates the non linearity between the input and output . explanations to link these observations with physical properties of phenomena and of uncertain input parameter properties are given , combining these observations with sa results , in the discussion section . moreover , it is noticeable that points are sometimes not flooded at all , when @xmath8 is equal to zero . reasons for these zero values are that in seldom cases , var . @xmath35 value gives at the point of interest a high ground elevation value ( above @xmath8 value ) or that var . @xmath36 produces critical threshold effects diverting flow direction . at three points of interest ( points , located on figure [ fig4]).,scaledwidth=98.0% ] the comparison of maps of @xmath8 mean and variance ( figure [ fig6 ] ) puts to the light the fact that areas densely urbanized and having a high water depths , have a high variance in @xmath8 . this maps comparison also underlines the fact that areas having a high mean water depths in less densely urbanized , and in areas close to the edge of flood spatial extent ( having a smaller mean water depth ) have a lower variance value of @xmath8 . this confirms the local observations at points of interests . moreover , a high variance of @xmath8 is observed in the map for places that have steep slope such as river bank , access roads , highway ramps or dikes . intuition would lead to incriminate here resolution of discretization effects ( var . @xmath37 ) as it will be confirmed by the sa ( see next section and discussion part ) . over the river , variance is locally important . the spatial changes in variance in the river bed ranges from @xmath82 to @xmath38 . amplitudes in variance in the river bed are most likely due to above ground features additions when var . @xmath36 changes ( features such as walls , dikes levees , and roads elements in the main riverbed ) , that does change the width of the river bed itself . consequently , these local important variance values are not surprising . our study focuses on overland flow areas . a gsa over riverbed itself would be out of the range of the spatial gsa defined for this study . .,scaledwidth=98.0% ] @xmath83 order sobol index ( @xmath29 ) of var . @xmath36 ( @xmath63 ) , var . @xmath37 ( @xmath62 ) and var . @xmath35 ( @xmath64 ) are computed for the @xmath54 points of interest . figure [ fig7 ] ( a ) shows the evolution of computed @xmath29 increasing @xmath56 through a random sampling in the results database for the same three points used in the figure [ fig5 ] . stabilization of the computed @xmath29 values is observed when @xmath56 is approximately @xmath84 , confirming that convergence of the random sampling is reached around this @xmath56 value . it has to be noticed that below a value of @xmath85 , the samples are too small to compute @xmath64 with our algorithm ( draws of var . @xmath35 are too scarcely distributed in the matrix to compute conditional expectation of var . @xmath35 ) . a bootstrap is performed , to check confidence interval of the computed @xmath29 as can be seen in figure [ fig7 ] ( b ) . for each point , independent samples of size @xmath86 are randomly drawn @xmath87 times in the results data base to compute @xmath87 times @xmath29 . then the @xmath88 confidence interval is computed . + over the @xmath54 selected points , the average @xmath63 value is @xmath89 , the average @xmath62 value is @xmath90 and the average @xmath64 value is @xmath91 . @xmath63 is ranked as the highest among the three @xmath29 for @xmath92 out of the @xmath54 points . for the seven other points , @xmath62 is ranked as the highest @xmath29 . the results show that var . @xmath35 is never the variable which influences the most @xmath8 variance and @xmath64 is ranked as the second highest @xmath29 only for points @xmath75 and @xmath74 . these points are located at the edge of the flood extent area where the @xmath8 values are in average below @xmath38 . + for the @xmath54 points , the difference between the highest ranked @xmath29 and the second one is often clear ( around @xmath93 ) , but the difference between the @xmath29 ranked as @xmath94 and @xmath95 is often not important ( around @xmath96 ) and can be smaller than the @xmath97 confidence interval calculated from the bootstrap . + the main outcome from the punctual gsa is that var . @xmath36 and var . @xmath37 , which are two modeller choices when including hr topographic data in the model , are always the parameters contributing the most to @xmath8 variance . the analysis also highlights that var . @xmath35 does not introduce much variance on @xmath8 . for the @xmath54 points , @xmath29 ranking varies from one point to another one , enhancing the spatial variability of uncertain parameters influence on @xmath8 variance and strengthening the interest of sensitivity maps production . over the selected subarea , @xmath29 are computed every @xmath46 to produce sensitivity maps . with this level of discretization , it represents a total of @xmath98 points where @xmath29 are calculated . a test has been carried out at a finer resolution ( @xmath38 ) over a @xmath99 per @xmath99 area for @xmath29 mapping . results in @xmath29 maps at @xmath38 and @xmath46 are similar over this small area . therefore , the @xmath29 maps are computed at a resolution @xmath46 as the number of points to compute is @xmath100 times less important than for a @xmath38 resolution . the @xmath29 are computed at every points using @xmath56 equal to the full size of available simulations in the results database ( @xmath52 ) . + a first analysis of the distribution of computed @xmath29 is illustrated in figure [ fig8 ] ( a ) . non flooded areas are removed for this analysis as well as areas covered by buildings . indeed , inside the buildings which are represented as impervious blocks in the model , @xmath63 is equal to one . therefore , var . @xmath36 explains the entire variance of @xmath8 in building areas . moreover , at the edges of buildings , @xmath62 is equal to one as well , due to buildings resolution effects . the number of points where @xmath29 have been calculated and that are plotted in figure [ fig8 ] ( a ) is around @xmath101 . the results show that : * @xmath63 is highly distributed around @xmath96 and has two peaks in distribution around @xmath102 and @xmath103 that have a flatter shape ; * @xmath62 is highly distributed around a value of @xmath104 . a second minor distribution peak around @xmath105 ; @xmath64 distribution is a single peak centered in @xmath106 , which is a value lower than both @xmath62 and @xmath63 peaks . analysis of these multi - modal distributions , confirms punctual gsa results regarding the non - spatially homogeneous ranking of @xmath29 . the analysis of @xmath29 maps will help to understand the spatial distribution and the ranking of uncertain input parameters according to their influence over the output variance . + figure [ fig8 ] ( c ) presents the sobol index maps . analyzing in the first place the maximal @xmath29 spatial distribution , it appears that , @xmath62 and @xmath63 are always ranked with the highest value . @xmath62 is ranked as the highest over @xmath107 of the subarea whereas @xmath63 is ranked as the highest over @xmath108 of the subarea . @xmath35 is rarely the most impacting parameter . this confirms the punctual gsa results and the @xmath29 distribution analysis . in the second place , using the spatial repartition of @xmath29 values presented as sensitivity maps ( figure [ fig8 ] ( b ) ) , the following remarks arise : * @xmath109 is ranked as the highest index in locations where @xmath8 has a high variance . in the areas with a high @xmath8 variance , @xmath109 values range between @xmath110 and @xmath111 . those high @xmath109 areas are characterized by a highly urbanized environment where above ground features strongly impact @xmath8 . * @xmath63 is ranked as the highest @xmath29 , where a given above ground element strongly impact locally hydrodynamic and consequently @xmath8 . * @xmath62 happens to be the most impacting parameters in areas less densely urbanized . * moreover , high ranking of @xmath109 also occurs when a given aboveground structure impacts upstream or downstream calculation of @xmath8 whatever is the urban configuration / density of affected upstream or downstream areas . * @xmath62 is ranked as the highest @xmath29 when @xmath8 is low ( below @xmath38 ) , and when in the meantime , variance of @xmath8 is low as well . it corresponds to areas close to the edge of the flood extent . * @xmath62 is ranked as the highest @xmath29 in areas which are less densely urbanized and where no above ground features , at the given area , neither upstream nor downstream , have any important effects on @xmath8 . * @xmath62 is ranked as the highest @xmath29 in areas where the ground slope is steep . indeed the level representation of a sloping area is highly affected locally by the degree of resolution of the discretization . * @xmath64 low and almost homogeneous over the subarea . ( a ) , details of @xmath29 maps ( b ) and map of highest ranked @xmath29 ( c).,scaledwidth=98.0% ] the implemented approach is a proof of concept of applicability of spatially distributed gsa to 2d hydraulic problems . ua and spatial ranking of influent uncertain input parameters over the 2d hr flood modelling study case have been achieved . nevertheless , being a first attempt , the approach can be improved . outcomes , limits and perspectives are underlined in this section and compared with other research fields in geomatics , sa and hydraulic modelling . a basic ua leads to the following conclusions on : output variability quantification , nonlinear behavior of the model and spatial heterogeneity . within established framework for the ua , the considered uncertain parameters related to the hr topographic data accuracy and to the inclusion in hydraulic models influence the variability of @xmath8 in a range that can be up to @xmath80 . this stresses out the point that even though hydraulic parameters were set - up as constant , the uncertainty related to hr topographic data use can not be omitted and needs to be assessed and understood . these warnings were already raised up in @xcite and @xcite , and are strengthened in this study by @xmath8 variance quantification . the quantification is not easily transposable in other contexts and it is not an easy process to give general trend for practical applications given the fact that ( @xmath2 ) spatial heterogeneity of @xmath8 variance is observed and ( @xmath3 ) specificities of different hr classified dataset is highly variable . nevertheless , this quantification of uncertainty goes in the direction of improvement of state of the art as common practice is still to quantify uncertainty using expert opinion only ( see @xcite ) . investigations on the ua can lead to deeper understanding of mechanisms leading to @xmath8 variability . the punctual analyses of the @xmath8 distributions ( either unimodal or multimodal ) illustrate the nonlinearity of uncertain parameters effects over the output . this nonlinearity in the output distributions is most likely due to var . @xmath36 which represents the level of details of above ground features incorporated in hr dems . + punctual sa highlighted that depending on location of considered point of interest , maximal first order @xmath29 are different . this goes in the direction of a need of spatial representation of @xmath29 under the form of sensitivity maps for consistent analyses . this spatial distribution of @xmath29 showed the major influence of the modeller choices when using the hr topographic data in 2d hydraulic models ( var . @xmath36 and var . @xmath37 ) with respect to the influence of hr dataset accuracy ( var . @xmath35 ) . hence as underlined in @xcite , if one wants to reduce variability of @xmath8 at a given point of interest , the use of sensitivity maps helps to determine the most influential input at this point . moreover , sensitivity maps give possibility to link the spatial distribution of @xmath29 to the properties of the model , especially with the physical properties of represented urban sector topography . the fact that var . @xmath36 is the most contributing parameter in densely urbanized areas is not surprising as it introduces a change in the representation in the model of physical properties of the urban environment . @xmath37 indirectly impacts quality of small scale elements representation well . gsa allowed to compute sensitivity maps , but simplifications and choices , especially regarding the way step a ( setting up of the spatial gsa framework by choosing uncertain parameters and choosing a way to spatialize them ) and step b ( assigning pdf to input parameters ) , lead to simplifications which are interesting to enhance . + for the uncertainties related to errors in hr topographic data ( var . @xmath35 ) , the followed normal pdf having properties of the rmse is randomly introduced , for every points of the highest resolution dem ( @xmath38 ) . nevertheless , as from one point to the next one , the normal pdf is drawn independently , it results in a uniform spatial distribution . in practice a uniform repartition should increase entropy and maximize errors / uncertainties effects . in the present case , this consideration is not valid . indeed , the used parameter is a rmse which is already averaged over the space . in fact as reminded in @xcite , the rmse is calculated based on assumption of normality which is often violated . for instance , over open and flat areas ( _ e.g. _ parking , roads ) , relative accuracy from one point to another should increase . in the study case , a comparison with ground topographic data measurement revealed that accuracy of hr dem rmse increases to @xmath112 . hence , over flat areas where the var . @xmath35 appears to be ranked as the second most contributing parameter to @xmath8 variability , not without standing the fact that the @xmath29 confidence interval of ranked second and third parameters overlaps , it sounds reasonable to think that var . @xmath35 is overestimated . opposite effect is observable over sloping areas ( _ e.g. _ dikes ) , where in our cases , after a regional control of the measurement quality , it is found that rmse value is about @xmath38 . therefore , especially over steep slope areas such as dikes where var . @xmath37 has been found to be the most important parameter contributing to @xmath8 variability , our @xmath29 ranking has to be taken with caution as var . @xmath35 has probably been locally underestimated . for further work , it would be interesting to improve the approach , by using spatialized value of rmse in function of topographic properties . this regionalization of characteristics of pdf might not be easy to implement by practitioners as regional information of accuracy might not be available . in that case , basic assumption to attribute regionally different characteristics to pdf could be relevant . for var . @xmath35 , a component related to photointerpretation errors should have been taken into consideration . moreover , in order to improve our study , it would be relevant to include a new variable that would reflect errors in photointerpretation . basically , this should consist in a random error in classified data for @xmath19 of the number of elements used for dem generation . from a technically point of view , implementation of such process is not straight forward particularly , recalling that this study is a first proof of concept on the topic . therefore , it has not been included in the sa . nevertheless , errors in photo - interpretation , which are uncertainties inherent to the hr dataset would have locally a considerable impact on @xmath8 variability and would require further research . + for modeller choices , in terms of level of details in classified features to be integrated in the hydraulic models ( var . @xmath36 ) , it is reasonable to consider this parameter as a categorical ordinal parameter having a uniform pdf . indeed , depending on availability of information of features influencing overland flow defined as classes and depending on model objective , modeller will select one of the available options in increasing complexity of dem . the choice of a row hr dtm ( without buildings , var . @xmath42 ) is mostly responsible of the observed binomial distribution in the ua , leading to an under estimation of maximal water depth @xmath8 compare to other cases . nevertheless it appears as well that punctually , at @xmath38 and @xmath113 resolutions , var . @xmath45 leads to low @xmath8 value as well due to local effects over flow paths . + for modeller choices in terms of level of discretization ( var . @xmath37 ) , hr dem were used , we constrained ourselves to resolution levels which are realistic with the use of such type of data considering that a resolution higher than @xmath46 is not compatible with the idea of producing hr models . nonlinear effects of resolution are long time known by practitioners in the sense that the grid resolution will impact the level of details included in the model @xcite . implemented approach is a proof of concept of applicability of spatially distributed variance - based global sensitivity analysis ( gsa ) to 2d flood modelling , allowing to quantify and to rank the defined uncertainties sources related to topography measurement errors and to operator choices when including high resolution ( hr ) classified dataset in hydraulic models . interest focuses on ( @xmath2 ) applying an uncertainty analysis ( ua ) and spatial gsa approaches in a 2d hr flood model having spatial inputs and outputs and ( @xmath3 ) producing sensitivity maps . summary of outcomes and remarks are put to the front concerning these aspects . * using @xmath53 cpu hours on the hpc architecture of the centre de calcul interactif , a database of @xmath52 simulations of a river flood event scenario over a densely urbanized area described based on a hr classified topographic dataset has been built . a random sampling on the produced result database was performed to follow a monte - carlo approach . after convergence check , a ua and a variance based functional decomposition gsa have been performed over the output of interest . output of interest being the maximal overland flow water depth ( @xmath8 ) reached at every point of the computational grids . * feasibility of spatial gsa approach for hr 2d flood modelling was achieved by this proof of concept test study . * important requirements are involved when implementing ua and gsa as expertise and efforts are required ( @xmath2 ) for method establishment ( specification of the problem ) and ( @xmath3 ) for characterization of input parameters as complexity of this step increases to consider spatial variability of the input parameters and can involve an important pretreatment phase ( _ e.g. _ for dems generation ) . eventually spatial information of hr topographic dataset accuracy might not be available . in that case , basic assumption attributing regionally different characteristics to pdf could be relevant . not only this part of the process is subject to subjectivity , but it can be time consuming and his application in dedicated tools ( such as promthe - fullswof_2d ) might not be straight forward . * for practical application , restrictive computational resources requirement is raised for this specific case ( in terms of cpu and in terms of hard drive storage ) due to the use of big data combined with a monte carlo approach . more parsimonious strategies like pseudo monte carlo sampling could be used or , depending on objective other gsa method than used sobol functional variance decomposition can be carried out : see @xcite for a review on optimization of gsa strategy in function of objectives and complexity of models . * the ua has allowed to quantify uncertain parameters impacts on output variability and to describe the spatial pattern of this variability . the spatial gsa has allowed the production of sobol index ( @xmath29 ) maps over the area of interest , enhancing the relative weight of each uncertain parameter on the variability of calculated overland flow . * within established framework , the considered uncertain parameters related to the hr topographic data accuracy and to the inclusion of hr topographic data in hydraulic models influence the variability of @xmath8 , in a range that can be up to 1 m. this enhances the fact that the uncertainty related to hr topographic data use is considerable and deserves to be assessed and understood before qualifying a 2d flood model of being hr or of high accuracy . moreover , ua reveals non linear effects and spatial heterogeneity of @xmath8 variance . nonlinearity in the output distributions is most likely due to var . @xmath36 which represents the level of details of above ground features incorporated in dems . * quantification of uncertainty through ua goes in the direction of improvement of state of the art , compared to quantification of uncertainty based on expert opinion only . investigations on the ua can lead to deeper understanding of mechanisms leading to @xmath8 variability . moreover , an analysis of @xmath8 extreme quantiles distribution could have been performed to find the combination of penalizing parameters . * the spatial distribution of @xmath29 illustrates the major influence of the modeller choices , when using the hr topographic data in 2d hydraulic models ( var . @xmath36 and var . @xmath37 ) with respect to the influence of hr dataset accuracy ( var . @xmath35 ) . as underlined in @xcite , if one wants to reduce variability of @xmath8 at a given point of interest , the use of sensitivity maps helps to determine the most influential input at this point . moreover , possibility to link the spatial distribution of the @xmath29 to the properties of the model , especially with the physical properties of represented urban sector topography . the fact that var . @xmath36 is the most contributing parameter in densely urbanized areas is not surprising . indeed , in that case , a change in var . @xmath36 highly influences the representation in the model of physical properties of the urban environment , therefore impacting model results . @xmath37 indirectly impacts quality of small scale elements representation as well . nevertheless var . @xmath35 assumes a spatially uniform rmse and does not take into consideration errors in photo - interpretation . therefore , errors related to hr measurement are probably underestimated locally in this study . * gsa use to spatially rank uncertain parameters effects gives a valuable insight to modeller . moreover , it can help to reduce variability in the output putting effort on improving knowledge about a given parameter or helps for optimization ( _ e.g. _ to define relevant areas where spatial discretization is important prior to non structured mesh use ) . * quantification and ranking helps modeller to have a better knowledge of limits of what has been modelled . nevertheless , as reminded in @xcite depending in the method gsa might produce different results . photogrammetric and photo - interpreted dataset used for this study have been kindly provided by nice cte dazur metropolis for research purpose . the nice cte dazur metropolis direction of geographic information , particularly g. tacet , f. largeron and l. andres gave comments of great value regarding geomatic aspects . authors are thankful to cemracs 2013 organizers , y. richet and b. ioss , which provided valuable comments for gsa aspects . this work was granted access to the hpc and visualization resources of the " centre de calcul interactif hosted by university nice sophia antipolis . abily , m. , delestre , o. , amoss , l. , bertrand , n. , laguerre , c. , duluc , c .- m . , and gourbesville , p. ( 2014 ) . use of 3d classified topographic data with fullswof for high resolution simulations of river flood event over dense urban area . in _ @xmath95 iahr europe congress : book of proceedings , 14 - 15 april 2014 , porto - portugal _ , number isbn 978 - 989 - 96479 - 2 - 3 . abily , m. , delestre , o. , amoss , l. , bertrand , n. , richet , y. , duluc , c .- m . , gourbesville , p. , and navaro , p. ( 2015 ) . uncertainty related to high resolution topographic data use for flood event modeling over urban areas : toward a sensitivity analysis approach . , 48:385399 . alliau , d. , de saint seine , j. , lang , m. , sauquet , e. , and renard , b. ( 2015 ) . etude du risque dinondation dun site industriel par des crues extrmes : de lvaluation des valeurs extrmes aux incertitudes hydrologiques et hydrauliques . , 2:6774 . , coullon , h. , delestre , o. , laguerre , c. , le , m. h. , pierre , d. , and sadaka , g. ( 2013 ) . fullswof paral : comparison of two parallelization strategies ( mpi and skelgis ) on a software designed for hydrology applications . , 43:5979 . cunge , j. a. ( 2014 ) . what do we model ? what results do we get ? an anatomy of modelling systems foundations . in gourbesville , p. , cunge , j. , and caignaert , g. , editors , _ advances in hydroinformatics _ , springer hydrogeology , pages 518 . springer singapore . delestre , o. , cordier , s. , darboux , f. , and james , f. ( 2012 ) . a limitation of the hydrostatic reconstruction technique for shallow water equations / une limitation de la reconstruction hydrostatique pour la rsolution du systme de saint - venant . , 350:677681 . fewtrell , t. j. , duncan , a. , sampson , c. c. , neal , j. c. , and bates , p. d. ( 2011 ) . benchmarking urban flood models of varying complexity and scale using high resolution terrestrial lidar data . , 36(78):281291 . hunter , n. , bates , p. , neelz , s. , pender , g. , villanueva , i. , wright , n. , liang , d. , falconer , r. , lin , b. , waller , s. , crossley , a. , and mason , d. ( 2008 ) . benchmarking 2d hydraulic models for urban flooding . in _ proceedings of the ice water management _ , volume 161 , pages 1330 . lafarge , f. and mallet , c. ( 2011 ) . building large urban environments from unstructured point data . in _ computer vision ( iccv ) , 2011 ieee international conference on _ , volume 0 , pages 10681075 , los alamitos , ca , usa . ieee computer society . le bris , a. , chehata , n. , briottet , x. , and paparoditis , n. ( 2013 ) . very high resolution land cover extraction in urban areas : very high resolution urban land cover extraction using airborne hyperspectral images . in _ proceedings of the @xmath114 earsel imaging spectrometry workshop , april 2013_. mastin , a. , kepner , j. , and fisher , j. i. ( 2009 ) . automatic registration of lidar and optical images of urban scenes . in _ computer vision and pattern recognition , 2009 . cvpr 2009 . ieee conference _ , pages 26392646 . remondino , f. , barazzetti , l. , nex , f. , scaioni , m. , and sarazzi , d. ( 2011 ) . uav photogrammetry for mapping and 3d modeling current status and future perspectives . in _ archives of photogrammetry , remote sensing and spatial information sciences _ , volume 38(1/c22 ) . isprs conference uav - g , zurich , switzerland . walker , w. , harremos , rotmans , j. , van der sluijs , j. , van asselt , m. , janssen , p. , and krayer von krauss , m. ( 2003 ) . defining uncertainty a conceptual basis for uncertainty management in model - based decision support . , 4(1):517 .
gauge theories constitute the building blocks of our present understanding of natural phenomena . the standard model ( sm ) of high energy particle interactions is , in fact , entirely based on a semi - simple gauge group @xmath0 . the sm accounts for roughly four percent of the known universe . however , several puzzles remain still unexplained . for example , why do we observe , at least , three generations of elementary particles . besides , the remaining @xmath1 of the universe , of which @xmath2 dark matter , and @xmath3 dark energy , remains largely unknown . it is therefore natural to expect that new gauge theories , or extensions of the present ones play a fundamental role in explaining the unknown side of the universe . the space of four dimensional gauge theories at our disposal , without theoretical prejudice , is very large and moreover a large fraction of it is still _ terra incognita _ due to our limited methods to tackle nonperturbative dynamics . a fascinating possibility is that different gauge theories display the same physical dynamics . this is possible since the physical quantities are determined by the gauge singlet operators of a generic gauge theory . therefore it is perfectly legal to imagine to have two different gauge theories leading to the same physics , at least , in certain regimes . quantum chromodynamics ( qcd ) itself is a famous example . in fact , at low energy the spectrum can be represented by hadrons interacting via yukawa interactions and its electric dual description is in terms of the @xmath4 gauge theory featuring quarks and gluons . it is therefore natural to imagine that generic asymptotically free gauge theories have magnetic duals . in fact , in the late nineties , in a series of ground breaking papers seiberg @xcite provided strong support for the existence of a consistent picture of such a duality within a supersymmetric framework . supersymmetry is , however , quite special and the existence of such a duality does not automatically imply the existence of nonsupersymmetric duals . one of the most relevant results put forward by seiberg has been the identification of the boundary of the conformal window for supersymmetric qcd as a function of the number of flavors and colors . recently several analytic predictions have been provided for the conformal window of nonsupersymmetric gauge theories using different approaches @xcite . we initiated in @xcite the exploration of the possible existence of a qcd nonsupersymmetric gauge dual providing a consistent picture of the phase diagram as a function of number of colors and flavors . arguably the existence of a possible dual of a generic nonsupersymmetric asymptotically free gauge theory able to reproduce its infrared dynamics must match the t hooft anomaly conditions @xcite . we have exhibited several solutions of these conditions for qcd in @xcite . an earlier exploration already appeared in @xcite . in @xcite theories with fermions transforming according to higher dimensional representations were analyzed . in @xcite we have put the idea of nonsupersymmetric gauge duality on a firmer ground by showing that for certain scalarless gauge theories with a spectrum similar to the one of qcd the gauge dual passes a large number of consistency checks . the theory studied in @xcite is qcd with @xmath5 dirac flavors and one adjoint weyl fermion . an important feature of this theory is that it possesses the same global symmetry of super qcd despite the fact that squarks are absent . this means that there are four extra anomaly constraints not present in the case of ordinary qcd , moreover we have shown that the potential dual can be constructed for any number of colors greater than two . the magnetic dual is a new gauge theory featuring magnetic quarks and a weyl adjoint fermion , new gauge singlet fermions which can be identified as states composite of the electric variables , as well as scalar states needed to mediate the interactions between the magnetic quarks and the gauge singlet fermions . the new scalars allow for a consistent flavor decoupling which was an important consistency check in the case of supersymmetry . in @xcite it was also shown that the candidate dual allows to bound the anomalous dimension of the dirac fermion mass operator to be less than one in the conformal window , and there we also estimated the critical number of flavors below which large distance conformality is lost in the electric variables . in these previous studies we implicitly assumed that there exist nontrivial infrared fixed points in the magnetic description matching the electric ones . however , given that we do not start in the ultraviolet with a supersymmetric spectrum nor we impose supersymmetric coupling relations it is legitimate to ask whether these fixed point exist at all in the magnetic description , and if they do what are their universal properties . the goal of this work is exactly to establish the existence of these fixed points assuming the most general coupling structure . we will investigate the regime where the computations are trustable , i.e. when the magnetic theory features fixed points accessible in perturbation theory . we discover , for the first time , a large number of calculable fixed points and classify them by analyzing their basin of attraction . an extremely interesting result , according to us , is that among these very many fixed points we discover supersymmetric fixed points to emerge in the infrared along nonsupersymmetric flows . our results have an impact on the ( non)supersymmetric gauge - gauge duality and even recent ideas on why the sm can be seen as the magnetic dual description of an electric strongly coupled theory featuring no scalars @xcite . the consequences of this interpretation of the sm leads to a novel way to tackle the hierarchy problem and even to shed light on the mystery of the observed number of matter generations @xcite . the phenomenological interest of our studies relies also on the fact that theories similar to the ones investigated here and featuring infrared fixed points have been used to construct sensible extensions of the standard model of particle interactions of technicolor type passing precision data and known as minimal walking technicolor models @xcite . duality has also been used to infer relevant insights on the constraints on these models from precision electroweak tests @xcite . in @xcite we investigated the possible existence of a magnetic dual for the electric theory constituted by a scalarless @xmath6 gauge theory with @xmath5 dirac fermions and @xmath7 larger than two , as in qcd , but with an extra weyl fermion transforming according to the adjoint representation of the gauge group . the quantum global symmetry of the electric theory is therefore : @xmath8 at the classical level there is one more @xmath9 symmetry destroyed by quantum corrections due to the adler - bell - jackiw anomaly . of the three independent @xmath10 symmetries only two survive , a vector like @xmath11 and an axial - like anomaly free ( af ) one indicated with @xmath12 . the spectrum of the theory and the global transformations are summarized in table [ electric ] . @xmath13 & su_l(n_f ) & su_r(n_f ) & u_v(1)&u_{af}(1 ) \\ \hline \hline \lambda & { \rm adj } & 1 & 1 & ~~0 & ~~1 \\ q & { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}}&{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}}}&1&~~1 & -\frac{n}{n_f } \\ \widetilde{q } & \overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}}}&1 & \overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } } & -1 & -\frac{n}{n_f } \\ g_{\mu}&{\rm adj } & 1&1 & ~~0 & ~~0\\ \hline \end{array}\ ] ] we then postulated the existence of a magnetic dual featuring the minimal spectrum of composite states and gauge group structure needed to : * classify all of the t hooft anomaly conditions and show how to match them , for the first time , for _ any _ number of colors and flavors ; * allow for consistent flavor decoupling both in the electric and in the magnetic theory ; * ensure _ duality involution_. meaning that if we dualize once more the magnetic theory one recovers the gauge structure of the electric theory . the proposed nonsupersymmetric magnetic gauge theory @xcite is summarized in table [ qcdadual ] . @xmath14 & su_l(n_f ) & su_r(n_f ) & u_v(1 ) & u_{af}(1 ) \\ \hline \hline \lambda_m & { \rm adj } & 1 & 1 & 0 & 1 \\ q & { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}}&\overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}}}&1&~~\frac{n_f - x}{x } & - \frac{x}{n_f } \\ \widetilde{q } & \overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}}}&1 & { { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } } & -\frac{n_f - x}{x } & - \frac{x}{n_f } \\ m & 1 & { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } & \overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } } & 0 & -\frac{n_f-2x}{n_f } \\ \hline { \widetilde}{\phi } & \overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } } & 1 & { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } & -\frac{n_f - x}{x } & \frac{n_f - x}{n_f}\\ \phi & { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } & \overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } } & 1 & \frac{n_f - x}{x } & \frac{n_f - x}{n_f}\\ g_\mu & \text{adj } & 1 & 1 & 0 & 0 \\ \hline \end{array}\ ] ] we have also shown that one has to have @xmath15 . differently from the dual of super qcd we did not impose a supersymmetric spectrum or supersymmetric coupling relations . the spectrum above is nonsupersymmetric as it is clear from the fact that there is no complex scalar partner of @xmath16 . moreover we showed that we could build the gauge singlet states using electric variables which do not contain squarks . subsequently in @xcite it was argued that one could also add the complex scalar @xmath17 ( appearing now in table [ qcdadualm ] ) . this is possible since it does not affect the anomaly conditions and can be built naturally out of the electric fermionic variable as follows @xcite : @xmath18 it is crucial to be able to construct all these states directly from the electric fermionic variables . this demonstrates that supersymmetry is not a fundamental ingredient in order to construct these states . @xmath17 in @xcite plays the phenomenologically relevant role of the sm - like higgs , elementary in terms of the magnetic variables . although when adding the new complex scalar field the spectrum of the dual theory looks supersymmetric , the full theory is not since the couplings , in the ultraviolet , are not taken to respect supersymmetric relations . assuming the duality to exist we were able to make a number of relevant predictions for the nonperturbative gauge dynamics of the strongly coupled qcd gauge theory with one fermion in the adjoint representation . the most relevant ones being : * the anomalous dimension of the mass of the electric fermions at the lower boundary of the conformal window can not exceed unity . * estimate the size of the conformal window , i.e. the critical number of flavors below which the theory looses large distance conformality . * when the magnetic theory is used as possible dual of a minimal extension of the sm one can argue that this extension can be re - constructed in terms of a strongly coupled electric theory featuring only fermionic matter , and furthermore argue that the mathematical consistency of the electric dual requires , at least , three generations of ordinary matter @xcite . what was still missing is the actual existence of the fixed points in the magnetic theory . this is what this paper aims to accomplish using perturbation theory . this regime is achieved by choosing the number of flavors @xmath5 , number of colors @xmath19 , and yukawa couplings where perturbation theory holds . @xmath14 & su_l(n_f ) & su_r(n_f ) & u_v(1 ) & u_{af}(1 ) \\ \hline \hline \lambda_m & { \rm adj } & 1 & 1 & 0 & 1 \\ q & { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}}&\overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}}}&1&~~\frac{n_f - x}{x } & - \frac{x}{n_f } \\ \widetilde{q } & \overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}}}&1 & { { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } } & -\frac{n_f - x}{x } & - \frac{x}{n_f } \\ m & 1 & { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } & \overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } } & 0 & -\frac{n_f-2x}{n_f } \\ \hline h & 1 & { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } & \overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } } & 0 & \frac{2x}{n_f}\\ { \widetilde}{\phi } & \overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } } & 1 & { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } & -\frac{n_f - x}{x } & \frac{n_f - x}{n_f}\\ \phi & { \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } & \overline{{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } } & 1 & \frac{n_f - x}{x } & \frac{n_f - x}{n_f}\\ g_\mu & \text{adj } & 1 & 1 & 0 & 0 \\ \hline \end{array}\ ] ] we start with listing the yukawa operators:@xmath20 @xmath21-interactions do not affect our results to the perturbative order we are considering and therefore we will not include them here . this yukawa - sector contributes to the two - loop beta function of the gauge coupling in the following manner : @xmath22 \beta_0 & = \frac{11}{3 } c_2(g ) - \frac{2}{3 } \sum_r t(r ) n_f(r)- \frac{1}{6}\sum_s t(s ) n_f(s)\\ \beta_1 & = \frac{34}{3 } c_2(g)^2- \sum_r \left [ \frac{10}{3 } c_2(g ) + 2 c_2 ( r ) \right ] t(r ) n_f(r ) -\sum_s \left [ \frac{1}{3 } c_2(g ) + 2 c_2 ( r ) \right ] t(s ) n_f(s)\\ \beta_y & = \frac{1}{d(g ) } \sum_r\text{tr}\left [ c_2(r ) y^j { y_j}^\dagger\right]\label{betay}\end{aligned}\ ] ] where @xmath23 denotes the representation of fermions and @xmath24 denotes the representation of the _ real _ scalars . @xmath25 is the trace normalization of the group generators , @xmath26 is the quadratic casimir of these and @xmath27 is the dimension of the gauge group . we refer to appendix [ app : rg ] for a careful derivation of the yukawa contribution @xmath28 to the running of the gauge coupling . the result is : @xmath29 , \end{aligned}\ ] ] with @xmath30 and where we used the notation @xmath31 we now consider the running of the yukawa couplings . the one loop beta function for the yukawa couplings is given by @xcite : @xmath32 & + 2 y^k y^\dagger_j y^k + \frac{1}{2}y^k \text{tr}\left [ y^\dagger_k y^j+y^\dagger_j y^k\right ] - 3 g^2 \ { c_2(r ) , y^j \}.\end{aligned}\ ] ] here @xmath33 is the yukawa coupling matrix defined by the particular interaction : @xmath34 where roman indices contract over the scalar gauge - flavor overall index and the greek indices @xmath35 are again gauge - flavor indices but reserved for the weyl fermions @xmath36 and @xmath37 . @xmath38 is the group invariant : @xmath39 for each scalar contraction we multiply , row by column , the yukawa matrices over the fermion indices . we again report the derivation of the beta function for each yukawa coupling in appendix [ app : rg ] . to the second order in the couplings the set of beta function equations reads : @xmath40\\ \label{rgsystem2 } \beta(\alpha_{\lambda } ) = 2 \alpha_{\lambda } & \left [ \frac{3 ( x^2 - 1)}{4x } \alpha_\lambda+ ( \alpha_\lambda + \alpha_{{\widetilde}{\lambda}})\frac { n_f } { 4 } + n_f \left ( \frac{\alpha_{{\widetilde}{m}}+\alpha_h}{2 } + \alpha_{m } \right ) - 3\alpha_g \frac{3x^2 - 1}{2x } \right ] \nonumber\\ & -4 n_f\frac { y_m y_{\widetilde m } y_\lambda y_{\widetilde \lambda}}{(4\pi)^2 } \\ \label{rgsystem3 } \beta(\alpha_{m } ) = 2 \alpha_{m } & \left [ \frac{3 n_f + x}{2 } \alpha_m+ \frac { x } { 2}\alpha_{{\widetilde}{m } } + \frac{n_f}{2}\alpha_h + \frac{x^2 - 1}{2x } \left ( \frac{\alpha_{{\widetilde}{\lambda}}}{2 } + \alpha_\lambda \right ) - 3\alpha_g \frac{x^2 - 1}{2x}\right ] \nonumber\\ & -4\frac{x^2 - 1}{2x } \frac { y_m y_{\widetilde m } y_\lambda y_{\widetilde \lambda}}{(4\pi)^2 } \\ \beta(\alpha_h ) = 2 \alpha_h & \left [ \frac{\alpha_m + \alpha_{{\widetilde}{m } } + 2\alpha_h}{2}n_f + \left ( \frac{\alpha_{\lambda } + \alpha_{{\widetilde}{\lambda}}}{2}-6\alpha_g\right)c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) + x \alpha_h \right]\\ \beta(\alpha_{{\widetilde}{\lambda } } ) = \beta_{\alpha_{\lambda } } & \left ( y_{\lambda } \leftrightarrow y_{\widetilde \lambda } , y_{\widetilde m } \leftrightarrow y_{m } \right ) , \\ \beta(\alpha_{\widetilde{m } } ) = \beta_{\alpha_{m}}&\left ( { y}_{m } \leftrightarrow y_{\widetilde m } , y_{\widetilde \lambda } \leftrightarrow { y}_{\lambda } \right ) , \label{rgsystem4}\end{aligned}\ ] ] where : @xmath41 we impose @xmath42 invariance and therefore set all the yukawa phases to zero . we discuss the full set of equations , including the phases , in the appendix [ app : rg ] . in the supersymmetric limit @xmath43 , @xmath44 and the beta function system simplifies to : @xmath45\\ \beta(\alpha_{\lambda } ) & = - 2 \alpha_{\lambda } \left [ \beta_0 \alpha_g \right]\\ \beta(\alpha_{m } ) & = 2 \alpha_{m } \left [ ( 2n_f + x ) \alpha_m - 4\alpha_gc_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) \right]\\ \beta(\alpha_h ) & = 2 \alpha_h \left [ ( 2n_f + x ) \alpha_h - 4\alpha_gc_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}})\right],\label{susylimit4}\end{aligned}\ ] ] where @xmath46 is the well - known two - loop term of pure super qcd , i.e. _ without _ the gauge - singlet chiral superfield transforming as bilinear with respect to the non - abelian global symmetries and , if present , leading to the introduction of a superpotential term . the theory with this chiral superfield included is seiberg s magnetic dual . this result shows , as expected , that supersymmetry stays unbroken along the renormalization flow . also note how the running of the gaugino yukawa coupling has collapsed to the one - loop result for the running of the gauge coupling . these results are in agreement with the known result from supersymmetry . the all - orders supersymmetric beta functions for seiberg s magnetic dual are : @xmath47}{1 - 2x\alpha_g}\nonumber\\ \beta^s(\alpha_m ) & = \alpha_m \left [ \gamma_m + 2\gamma_0 \right],\end{aligned}\ ] ] where the one - loop expressions for the anomalous dimensions of the chiral superfields read : @xmath48 by which it is readily seen that the one - loop expansion of @xmath49 and @xmath50 are in agreement with the expressions in equations - . just below the critical number of flavors where the theory looses asymptotic freedom ( @xmath51 ) a banks - zaks perturbatively stable infrared fixed point ( irfp ) emerges once the yukawa interactions are set to zero . the well known expression for the value of the gauge coupling at this fixed point is : @xmath52 in this phase the theory shows large distance conformality . we now investigate whether the perturbatively stable irfp persists when we turn on the yukawa interactions . from eq . , the perturbative fixed point value for the gauge coupling reads : @xmath53 perturbative consistency is ensured by the smallness of the yukawa couplings at the fixed point . there is , clearly , a symmetry between the tilded and untilded yukawa coupling constants at the beta functions level . this symmetry persists at the fixed point effectively reducing the space of solutions . the fixed points associated to zeros of @xmath54 are : @xmath55 a perturbative nontrivial infrared fixed point for all the couplings is achieved when the number of flavors are such that asymptotic freedom is almost lost . to express this point mathematically we introduce the small parameter @xmath56 as follows : @xmath57 we report in table [ solutions0 ] and table [ solutions1 ] the explicit solutions for the fixed points as a function of @xmath56 . there are 18 physical solutions for the system - , where @xmath58 of these correspond to the class @xmath59 and the remaining @xmath60 to @xmath61 . the solutions obtained exchanging tilded couplings with untilted ones @xmath62 are indicated in the tables [ solutions0 ] and [ solutions1 ] by an asterisk . @xmath63 \hline 2^ { ( * ) } & \alpha_g^ * = \frac{2 ( 3 x - n_f ) ( x ( n_f + 3 x)-3)x } { 2 n_f^2 x ( 7 x^2 - 3 ) + 3 n_f ( 5 - 14 x^2 + x^4)-36 x^3 ( x^2 - 1 ) } & = \frac{2 x ( 2 x^2 - 1 ) \epsilon}{5 - 16 x^2 + 11 x^4 } \\ & \alpha_{\lambda}^ * = \frac{12 ( 3 x - n_f ) ( 3x^2 - 1)x } { 2 n_f^2 x ( 7 x^2 - 3 ) + 3 n_f ( 5 - 14 x^2 + x^4)-36 x^3 ( x^2 - 1 ) } & = \frac{4 x ( 3 x^2 - 1 ) \epsilon}{5 - 16 x^2 + 11 x^4}\\ & \alpha_{{\widetilde}{\lambda } , m , { \widetilde}{m}}^ * = 0 & \\[2 mm ] \hline 3^ { ( * ) } & \alpha_g^ * = \frac{2 ( 3 x - n_f ) ( 3n_f + x ) x } { 3 n_f^2 ( 13 x^2 - 5)-12 x^4 - 2 n_f x ( 3 + 11 x^2 ) } & = \frac{20x \epsilon}{91 x^2 - 51 } \\ & \alpha_{m}^ * = \frac{6 ( 3 x - n_f ) ( x^2 - 1)x } { 3 n_f^2 ( 13 x^2 - 5)-12 x^4 - 2 n_f x ( 3 + 11 x^2 ) } & = \frac{6 ( x^2 - 1 ) \epsilon}{x(91x^2 - 51)}\\ & \alpha_{\lambda,{\widetilde}{\lambda } , { \widetilde}{m}}^ * = 0 & \\[2 mm ] \hline 4 & \alpha_g^ * = \frac { ( 3 x - n_f ) ( 3n_f + 2x ) x}{2 ( n_f^2 ( 9 x^2 - 3 ) -6 x^4- n_f x ( 3 + 2 x^2 ) ) } & = \frac{11x \epsilon}{2(23 x^2 - 12 ) } \\ & \alpha_{m}^ * = \alpha_{{\widetilde}{m}}^*= \frac{3 ( 3 x - n_f ) ( x^2 - 1 ) } { 2 ( n_f^2 ( 9 x^2 - 3 ) -6 x^4- n_f x ( 3 + 2 x^2 ) ) } & = \frac{3 ( x^2 - 1 ) \epsilon}{2x(23x^2 - 12)}\\ & \alpha_{\lambda,{\widetilde}{\lambda}}^ * = 0 & \\[2 mm ] \hline 5 & \alpha_g^ * = \frac { ( 3 x - n_f ) ( 3x^2 + 2n_f x -3 ) x}{2 ( 9 x^3 - 9 x^5 + n_f^2 x ( 7 x^2 - 3 ) + n_f ( 3 - 6 x^2 - 9 x^4 ) ) } & = \frac{x \epsilon}{2(x^2 - 1 ) } \\ & \alpha_{\lambda}^ * = \alpha_{{\widetilde}{\lambda}}^*= \frac{3 ( 3 x - n_f ) ( 3x^2 - 1)x } { 2 ( 9 x^3 - 9 x^5 + n_f^2 x ( 7 x^2 - 3 ) + n_f ( 3 - 6 x^2 - 9 x^4 ) ) } & = \frac{x\epsilon}{x^2 - 1 } = 2\alpha_g^ * + \mathcal{o}(\epsilon^2)\\ & \alpha_{m,{\widetilde}{m}}^ * = 0 & \\[2 mm ] \hline 6^ { ( * ) } & \alpha_g^ * = \frac{2 ( 3 x - n_f ) ( 3 ( n_f^2 - 1 ) x -8n_f+ 9 n_f x^2 + 3 x^3)x } { 3 n_f^3 x ( 13 x^2 - 5)-36 x^4 ( x^2 - 1 ) + 3 n_f x ( 5 + 18 x^2 - 31 x^4 ) + 2 n_f^2 ( 18 - 59 x^2 + 9 x^4 ) ) } & = \frac{2 x ( 19 x^2 - 9 ) \epsilon}{41 - 141 x^2 + 100 x^4}\\ & \alpha_{\lambda}^ * = \frac{12 ( 3 x - n_f ) ( 3x^3 - 2n_f - x+8n_fx^2 ) } { 3 n_f^3 x ( 13 x^2 - 5)-36 x^4 ( x^2 - 1 ) + 3 n_f x ( 5 + 18 x^2 - 31 x^4 ) + 2 n_f^2 ( 18 - 59 x^2 + 9 x^4 ) ) } & = \frac{4x ( 27 x^2 - 7 ) ) \epsilon}{41 - 141 x^2 + 100 x^4}\\ & \alpha_{{\widetilde}{m}}^ * = \frac{6 ( 3 x - n_f ) ( n_f x-2)(x^2 - 1 ) } { 3 n_f^3 x ( 13 x^2 - 5)-36 x^4 ( x^2 - 1 ) + 3 n_f x ( 5 + 18 x^2 - 31 x^4 ) + 2 n_f^2 ( 18 - 59 x^2 + 9 x^4 ) ) } & = \frac { ( 6 x^2 - 4 ) \epsilon}{100x^3 - 41x}\\ & \alpha_{{\widetilde}{\lambda},m}^ * = 0 & \\[2 mm ] \hline 7 & \alpha_g^ * = \frac { ( 3 x - n_f ) ( 3 ( n_f^2 - 1 ) x -4 n_f + 6 n_f x^2 + 3 x^3 ) } { 2(9 x^3 - 9 x^5 + n_f^3 ( 9 x^2 - 3 ) + n_f^2 ( x - 15 x^3 ) + n_f(3 + 6 x^2 - 21 x^4 ) ) } & = \frac{x ( 16 x^2 - 5 ) \epsilon}{2(4x^4 - 5x^2 + 1)}\\ & \alpha_{\lambda}^ * = \alpha_{{\widetilde}{\lambda}}^*= \frac{3 ( 3 x - n_f ) ( 5 n_f x^2 -2 n_f - x + 3 x^3 ) } { 9 x^3 - 9 x^5 + n_f^3 ( 9 x^2 - 3 ) + n_f^2 ( x - 15 x^3 ) + n_f(3 + 6 x^2 - 21 x^4 ) } & = \frac{x ( 18 x^2 - 7 ) ) \epsilon}{4x^4 -5x^2 + 1}\\ & \alpha_m^*=\alpha_{{\widetilde}{m}}^ * = \frac{3 ( 3 x - n_f ) ( n_f x+3x^2 - 2)(x^2 - 1 ) } { 2x(9 x^3 - 9 x^5 + n_f^3 ( 9 x^2 - 3 ) + n_f^2 ( x - 15 x^3 ) + n_f(3 + 6 x^2 - 21 x^4 ) ) } & = \frac { ( 3 x^2 - 1 ) \epsilon}{x(4x^2 - 1)}\\[2 mm ] \hline \end{array}\ ] ] [ solutions0 ] @xmath64 \hline 2^ { ( * ) } & \alpha_g^ * = \frac{(3 x - n_f ) ( 6 x ( x^2 - 1 ) + n_f ( 7 x^2 + 2 x n_f-5))}{2 x n_f^2 ( 5 x^2 + 4 x n_f-13 ) -36 x^3 ( x^2 - 1 ) -3n_f ( 4 x^2 + 9 x^4 - 5 ) } & = \frac{x ( 15 x^2 - 7 ) \epsilon}{5 - 26 x^2 + 21 x^4 } \\ & \alpha_h^ * = \frac{3 ( 3 x - n_f ) ( x^2 - 1 ) ( 3 x^2 + 2 x n_f-5)}{x(2 x n_f^2 ( 5 x^2 + 4 x n_f-13 ) -36 x^3 ( x^2 - 1 ) -3n_f ( 4 x^2 + 9 x^4 - 5 ) ) } & = \frac{(9x^2 - 5)\epsilon}{x(21x^2 - 5 ) } \\ & \alpha_{\lambda}^ * = \frac{12 x ( 3 x - n_f ) ( 3 x^2 + 2 x n_f-1 ) } { 2 x n_f^2 ( 5 x^2 + 4 x n_f-13 ) -36 x^3 ( x^2 - 1 ) -3n_f ( 4 x^2 + 9 x^4 - 5 ) } & = \frac{4 x ( 9 x^2 - 1 ) \epsilon}{5- 26 x^2 + 21 x^4}\\ & \alpha_{{\widetilde}{\lambda } , m , { \widetilde}{m}}^ * = 0 & \\[2 mm ] \hline 3^ { ( * ) } & \alpha_g^ * = \frac{(3 x - n_f ) ( 2 x^2 + 8 x n_f + 5 n_f^2 ) } { 20n_f^3x - 12 x^4 + n_f^2(17x^2 - 15 ) - 2 n_f x ( 17x^2 + 3 ) } & = \frac{71x \epsilon}{193x^2 - 51 } \\ & \alpha_h^ * = \frac{3 ( 3 x - n_f ) ( 2 x + 5 n_f ) ( x^2 - 1 ) } { x(20n_f^3x - 12 x^4 + n_f^2(17x^2 - 15 ) - 2 n_f x ( 17x^2 + 3 ) ) } & = \frac{51(x^2 - 1)\epsilon}{x(193x^2 - 51 ) } \\ & \alpha_{m}^ * = \frac{6 ( 3 x - n_f ) ( x^2 - 1 ) } { 20n_f^3x - 12 x^4 + n_f^2(17x^2 - 15 ) - 2 n_f x ( 17x^2 + 3 ) } & = \frac{6 ( x^2 - 1 ) \epsilon}{x(193x^2 - 51)}\\ & \alpha_{\lambda,{\widetilde}{\lambda } , { \widetilde}{m}}^ * = 0 & \\[2 mm ] \hline 4 & \alpha_g^ * = \frac { ( 3 x - n_f ) ( 2n_f + x ) ( n_f+2x)}{2 ( 4 n_f^3 x - 6 x^4 + n_f^2 ( 7 x^2 - 3 ) - n_f x ( 3 + 8 x^2 ) ) } & = \frac{35x \epsilon}{2(47x^2 - 12 ) ) } \\ & \alpha_h^ * = \frac{3 ( 3 x - n_f ) ( n_f+x ) ( x^2 - 1 ) } { x(4 n_f^3 x - 6 x^4 + n_f^2 ( 7 x^2 - 3 ) - n_f x ( 3 + 8 x^2 ) ) } & = \frac{12(x^2 - 1)\epsilon}{x(47x^2 - 12 ) } \\ & \alpha_{m}^ * = \alpha_{{\widetilde}{m}}^*= \frac{3 ( 3 x - n_f ) ( x^2 - 1 ) } { 2 ( 4 n_f^3 x - 6 x^4 + n_f^2 ( 7 x^2 - 3 ) - n_f x ( 3 + 8 x^2 ) ) } & = \frac{3 ( x^2 - 1 ) \epsilon}{2x(47x^2 - 12)}\\ & \alpha_{\lambda,{\widetilde}{\lambda}}^ * = 0 & \\[2 mm ] \hline 5 & \alpha_g^ * = \frac{(3 x - n_f ) ( 3 x ( x^2 - 1 ) + 2 n_f ( 2 x^2 + x n_f-1))}{2 ( 4 n_f x + 3 x^2 - 3 ) ( n_f^2 x - 3 x^3 - n_f ( x^2 + 1 ) ) } & = \frac{x(11x^2 - 3 ) \epsilon}{2(5x^2 - 1)(x^2 - 1 ) } \\ & \alpha_h^ * = \frac{3 ( 3 x - n_f ) ( n_fx-1 ) ( x^2 - 1 ) } { x ( 4 n_f x + 3 x^2 - 3 ) ( n_f^2 x - 3 x^3 - n_f ( x^2 + 1 ) } & = \frac{(3x^2 - 1)\epsilon}{x(5x^2 - 1 ) } \\ & \alpha_{\lambda}^ * = \alpha_{{\widetilde}{\lambda}}^*= \frac{3x ( 3 x - n_f ) ( 3x^2 + 2xn_f-1 ) } { ( 4 n_f x + 3 x^2 - 3 ) ( n_f^2 x - 3 x^3 - n_f ( x^2 + 1 ) ) } & = \frac{x(9x^2 - 1)\epsilon}{(5x^2 - 1)(x^2 - 1 ) } \\ & \alpha_{m,{\widetilde}{m}}^ * = 0 & \\[2 mm ] \hline 6 & \alpha_g^ * = \frac { ( 3 x - n_f ) ( 2 n_f + x ) ( n_f^2 + 2 n_f x + 3 x^2 - 3 ) } { 2(4 n_f^4 x + 9 x^3 - 9 x^5 + n_f^3 ( x^2 - 3 ) - n_f^2 x ( 3 + 23 x^2 ) + 3 n_f ( 1 + 4 x^2 - 9 x^4 ) ) } & = \frac{7x \epsilon}{2(x^2 - 1)}\\ & \alpha_h^ * = \frac{3 ( 3 x - n_f ) ( n_fx - n_f^2 + 1 ) ( x^2 - 1 ) } { x(4 n_f^4 x + 9 x^3 - 9 x^5 + n_f^3 ( x^2 - 3 ) - n_f^2 x ( 3 + 23 x^2 ) + 3 n_f ( 1 + 4 x^2 - 9 x^4 ) ) } & = \frac{\epsilon}{x } \\ & \alpha_{\lambda}^ * = \alpha_{{\widetilde}{\lambda}}^ * \frac{3 ( 3 x - n_f ) ( x + 2 n_f ) ( 3 x^2 + x n_f-1 ) } { 4 n_f^4 x + 9 x^3 - 9 x^5 + n_f^3 ( x^2 - 3 ) - n_f^2 x ( 3 + 23 x^2 ) + 3 n_f ( 1 + 4 x^2 - 9 x^4 ) ) } & = \frac{7x \epsilon}{x^2 - 1 } = 2\alpha_g^ * + \mathcal{o}(\epsilon^2)\\ & \alpha_m^*=\alpha_{{\widetilde}{m}}^ * = \frac{3 ( 3 x - n_f ) ( 3n_f x+3x^2 - 2)(x^2 - 1 ) } { 2x(4 n_f^4 x + 9 x^3 - 9 x^5 + n_f^3 ( x^2 - 3 ) - n_f^2 x ( 3 + 23 x^2 ) + 3 n_f ( 1 + 4 x^2 - 9 x^4 ) ) } & = \frac { \epsilon}{x } = \alpha_h^ * + \mathcal{o}(\epsilon^2)\\ \hline \end{array}\ ] ] [ solutions1 ] solution ( 5 ) in table [ solutions0 ] corresponds to the perturbative infrared fixed point for super qcd while the last solution ( 6 ) in table [ solutions1 ] corresponds to the perturbative fixed point for seiberg s magnetic dual and it is the only one with all nonvanishing couplings . interestingly , when setting to zero @xmath59 ( solution @xmath65 in table [ solutions0 ] ) we find yet another fixed point corresponding to the infrared stable fixed point of the non - supersymmetric magnetic dual theory proposed in @xcite and shown in table [ qcdadual ] . this lends further support to the proposed non - supersymmetric duality . the stability of the fixed points reported in tables [ solutions0 ] and [ solutions1 ] is determined by first evaluating the following matrix @xmath66 at the fixed point of interest denoted in brief by @xmath67 . if it has only real and positive eigenvalues the fixed point is said to be stable . in the tables [ exponents0 ] and [ exponents1 ] we present the eigenvalues corresponding to the fixed points of tables [ solutions0 ] and [ solutions1 ] respectively for @xmath68 and @xmath69 . for convenience , we have marked the cells with negative eigenvalues in gray . @xmath70{0.8}}{\cellcolor{white}}~~1~~ & -0.049 & -0.049&-0.031&-0.016&-0.016 & { \cellcolor{white}}0.0034\\ [ 2 mm ] \hline 2 & 0.20&{\cellcolor[gray]{0.8}}-0.048 & { \cellcolor[gray]{0.8}}-0.045&{\cellcolor[gray]{0.8}}-0.015&0.0063&0.0011\\ [ 2 mm ] \hline 3 & { \cellcolor[gray]{0.8}}-0.049&{\cellcolor[gray]{0.8}}-0.044 & 0.035&{\cellcolor[gray]{0.8}}-0.029&{\cellcolor[gray]{0.8}}-0.015&0.0036\\ [ 2 mm ] \hline 4 & { \cellcolor[gray]{0.8}}-0.064 & 0.038&0.030&{\cellcolor[gray]{0.8}}-0.027&{\cellcolor[gray]{0.8}}-0.023&0.0039\\ [ 2 mm ] \hline 5 & 0.30 & 0.092 & 0.063 & { \cellcolor[gray]{0.8}}-0.060&{\cellcolor[gray]{0.8}}-0.058 & 0.0089\\ [ 2 mm ] \hline 6 & 0.20 & { \cellcolor[gray]{0.8}}-0.062&{\cellcolor[gray]{0.8}}-0.043&0.031 & 0.021 & 0.0065\\ [ 2 mm ] \hline 7 & 3.0 & 1.8 & 1.4 & 1.1 & { \cellcolor[gray]{0.8}}-0.14 & 0.033\\ [ 2 mm ] \hline \end{array}\ ] ] [ exponents0 ] @xmath71{0.8}}-0.067&{\cellcolor[gray]{0.8}}-0.067&{\cellcolor[gray]{0.8}}-0.0073&{\cellcolor[gray]{0.8}}-0.0073&0.0058\\ [ 2 mm ] \hline 2 & 0.37&0.17 & { \cellcolor[gray]{0.8}}-0.078 & 0.026 & 0.012 & 0.00038\\ [ 2 mm ] \hline 3 & 0.12&{\cellcolor[gray]{0.8}}-0.067&{\cellcolor[gray]{0.8}}-0.064&0.015&{\cellcolor[gray]{0.8}}-0.0075&0.0058\\ [ 2 mm ] \hline 4 & 0.12&{\cellcolor[gray]{0.8}}-0.074&{\cellcolor[gray]{0.8}}-0.053&0.015&0.014&0.058\\ [ 2 mm ] \hline 5 & 0.73 & 0.26 & 0.18 & 0.16&{\cellcolor[gray]{0.8}}-0.081 & 0.018\\ [ 2 mm ] \hline 6 & 21 & 5.4 & 4.0 & 4.0 & 0.77 & 0.052\\ [ 2 mm ] \hline \end{array}\ ] ] [ exponents1 ] there is a rich structure of fixed points classified according to the relevant directions clearly visible from the list of associated eigenvectors reported in the appendix [ tables ] . the dimensions of the critical surfaces are dictated by the number of positive eigenvalues . once the unstable directions , in the coupling space , are removed the remaining subset of couplings generate a critical @xmath72-dimensional surface with @xmath72 given by the number of positive eigenvalues containing the nontrivial fixed points . as an example consider the first fixed point of table [ exponents0 ] . here we have a critical line with an infrared stable fixed point and five unstable directions . the critical _ surface _ in this case is just a line . another interesting example is constituted by the fixed point solution 7 of table [ solutions0 ] . in this case we find a five - dimensional critical surface and one unstable sixth direction parallel to the @xmath73 axis . we plot in figure [ fp7 ] the projections of the renormalization group flow around this solution for a subset of two couplings at the time . the other filled small circles in the figure correspond to the projections on the chosen planes of fixed points such as the seiberg s magnetic dual , indicated by the susy label on the plot , or the super qcd banks - zaks fixed point labelled by susybz . and @xmath69 in the planes ( from left to right and top to bottom ) : ( @xmath74 ) , ( @xmath75 , @xmath73 ) , ( @xmath76,@xmath73 ) and ( @xmath77 , @xmath78 ) . the other couplings are kept fixed at the fixed point value of solution 7 given in table [ solutions0 ] . besides solution 7 clearly indicated in the plot we also plot the projections of the seiberg s magnetic dual indicated by the susy label on the plot , and the super qcd perturbative fixed point indicated by susybz . , title="fig:",scaledwidth=45.0% ] and @xmath69 in the planes ( from left to right and top to bottom ) : ( @xmath74 ) , ( @xmath75 , @xmath73 ) , ( @xmath76,@xmath73 ) and ( @xmath77 , @xmath78 ) . the other couplings are kept fixed at the fixed point value of solution 7 given in table [ solutions0 ] . besides solution 7 clearly indicated in the plot we also plot the projections of the seiberg s magnetic dual indicated by the susy label on the plot , and the super qcd perturbative fixed point indicated by susybz . , title="fig:",scaledwidth=45.0% ] + and @xmath69 in the planes ( from left to right and top to bottom ) : ( @xmath74 ) , ( @xmath75 , @xmath73 ) , ( @xmath76,@xmath73 ) and ( @xmath77 , @xmath78 ) . the other couplings are kept fixed at the fixed point value of solution 7 given in table [ solutions0 ] . besides solution 7 clearly indicated in the plot we also plot the projections of the seiberg s magnetic dual indicated by the susy label on the plot , and the super qcd perturbative fixed point indicated by susybz . , title="fig:",scaledwidth=45.0% ] and @xmath69 in the planes ( from left to right and top to bottom ) : ( @xmath74 ) , ( @xmath75 , @xmath73 ) , ( @xmath76,@xmath73 ) and ( @xmath77 , @xmath78 ) . the other couplings are kept fixed at the fixed point value of solution 7 given in table [ solutions0 ] . besides solution 7 clearly indicated in the plot we also plot the projections of the seiberg s magnetic dual indicated by the susy label on the plot , and the super qcd perturbative fixed point indicated by susybz . , title="fig:",scaledwidth=45.0% ] the fixed point featuring all positive eigenvalues is the one corresponding to seiberg s magnetic dual . we uncovered the full spectrum of perturbative fixed points associated to an @xmath79 nonsupersymmetric gauge theory featuring scalars and fermions , of the type summarized in table [ qcdadualm ] . although the noninteracting field theory has a supersymmetric looking spectrum we did not assume the bare couplings to respect supersymmetry . we then analyzed the fixed points by studying the beta functions to second order in the couplings . to this order , which is renormalization scheme independent , we discovered supersymmetric and nonsupersymmetric fixed points . we further analyzed their stability and discovered that : * when all bare couplings are nonzero the theory flows to the seiberg s magnetic dual fixed point . this occurs on supersymmetric and nonsupersymmetric renormalization flow directions . * when the yukawa couplings @xmath77 , @xmath80 and @xmath73 are all set to zero the theory flows to the super qcd fixed point , as above , independently if the renormalization flow is supersymmetric . * when all the yukawas are set to zero we achieve the perturbative nonsupersymmetric fixed point in the gauge coupling . * when @xmath77 and @xmath80 are nonzero and @xmath81 we discover a new nonsupersymmetric fixed point . this can be identified with the magnetic dual fixed point for nonsupersymmetric gauge theories proposed in @xcite . another relevant fixed point emerges when @xmath82 and @xmath83 . * all the other fixed points emerge similarly by setting to zero different yukawas . the generic conclusions we draw from our results are : * supersymmetry naturally emerges as a fixed point theory from a nonsupersymmetric lagrangian , meaning without fine - tuning the bare couplings . this relevant result demonstrates that supersymmetry can be viewed as an emergent phenomenon in field theory . our results , among other things can be used to argue that one does not need to fine - tune the bare couplings when performing lattice simulations aimed to study supersymmetry on the lattice @xcite . * duality is not a prerogative of supersymmetry given that the theory features many nonsupersymmetric fixed points , depending on the details of the interactions . therefore these new fixed points can describe the nonperturbative physics of electric dual gauge theories like the one envisioned in @xcite . the potential physical applications of our results range from ( supersymmetric ) extensions of the sm to providing new ideas on how to investigate nonperturbative supersymmetric dynamics via first principle lattice simulations . we thank simon catterall , stefano di chiara and marco nardecchia for enlightening discussions and comments . we follow the notation of machacek and vaughn@xcite , who derived the general expressions to two - loop order for the running of the gauge coupling and yukawa couplings for a general gauge theory with real scalars and majorana fermions . in ref . @xcite is showed how to use the expressions of machacek and vaughn for complex scalars and weyl fermions . we need only to do a slight rewriting of @xmath84 given in eq . to follow their notation : @xmath85 where @xmath86 is the gauge index reserved for the adjoint majorana fermion with @xmath27 the dimension of its representation , greek gauge indices @xmath87 are reserved for the weyl fermions and roman gauge indices @xmath88 are reserved for the complex scalars . flavor indices have been suppressed . the yukawa matrices are defines as follows , once we take flavor indices into account with @xmath89 denoting the indices of @xmath90 and @xmath91 indices of the @xmath92 global symmetries : @xmath93 where @xmath94 are the generators of the gauge group . the minus sign of @xmath95-term has been chosen to facilitate the connection with supersymmetry , i.e. for susy @xmath96 . we will first derive the yukawa contribution to the 2-loop beta function , given in eq . - , in particular : @xmath97 \ , \end{aligned}\ ] ] where @xmath98 is the quadratic casimir of the spinor representations @xmath23 and @xmath27 is the dimension of the gauge group . it is instructive to consider the contribution from the yukawa sector to the two - loop beta function of the gauge coupling diagrammatically . there is only one type of diagram that has non - vanishing contribution , which is : where the solid line represents spinor fields , and the dashed line represents the scalar(s ) coupled via the yukawa interactions . note that the scalar - gauge interactions do not contribute . the fermion @xmath16 is not coupled to the gauge fields . thus we get : @xmath99 \nonumber\\ & = ( \alpha_\lambda + \alpha_{{\widetilde}{\lambda } } ) t({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) n_f \left [ c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) + c_2(g ) \right ] + ( \alpha_m + \alpha_{\widetilde{m } } + 2 \alpha_h ) n_f^2 \frac{c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) x}{d(g ) } \nonumber \\ & = ( \alpha_\lambda + \alpha_{{\widetilde}{\lambda } } ) \frac { n_f}{2 } \frac{3x^2 - 1}{2x } + \left ( \frac{\alpha_m + \alpha_{\widetilde{m}}}{2 } + \alpha_h \right ) n_f^2,\end{aligned}\ ] ] where we used the notation @xmath31 we now consider the running of the yukawa couplings . the one loop beta function for the yukawa couplings is given in @xcite : @xmath32 & + 2 y^k y^\dagger_j y^k + \frac{1}{2}y^k \text{tr}\left [ y^\dagger_k y^j+y^\dagger_j y^k\right ] - 3 g^2 \ { c_2(r ) , y^j \}.\end{aligned}\ ] ] where @xmath38 is the group invariant : @xmath100 it is again instructive to consider each term in the beta function in terms of the diagrams they originate from . the first two terms in the bracket corresponds to the fermion leg self - energy contribution from the yukawa sector , and the expression tells us that they each contribute with a factor of @xmath101 to the beta function : cccc c @xmath102 + + + + + & & c @xmath103 + + + + + the next term in the beta function is the vertex correction from the yukawa sector : the fourth term is the scalar leg self energy contribution from the yukawa sector : ccc & c @xmath104 + + + + note the different directions of the arrow in the fermion loops . for dirac fermions the above contribution comes naturally with a factor of two . finally the gauge sector has a non - vanishing contribution only to the fermion leg self energy , and each term contributes with a factor of -3 : cccc c @xmath105 + + + + + & & c @xmath106 + + + + + we thus find the one - loop coefficient of the beta function for @xmath33 to be : @xmath107 + 2 { \widetilde}{u}^i u^\dagger_j { \widetilde}{y}^i \nonumber\\[2 mm ] & + y^k \text{tr}\left [ y^\dagger_k y^j\right ] + y^k \text{tr}\left [ u^k u^\dagger_j\right ] - 3 g^2 \left\ { c_2(q ) y^j + y^j c_2(\lambda ) \right\}.\end{aligned}\ ] ] inserting the definitions of the yukawa - matrices , we find : @xmath108{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}})\mid y_\lambda \mid^2 + n_f \mid \tilde{y}_m \mid^2 + n_f \mid y_h \mid^2 + t({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}})n_f(\mid y_\lambda \mid^2 + \mid y_{\widetilde \lambda } \mid^2 ) \right\ } \right . \nonumber \\ & \quad \left . + c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}})\mid y_\lambda \mid^2 + n_f\mid y_m \mid^2 - 3 g^2 \left \ { c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}})+ c_2(g ) \right \ } \right ] y^j -2 n_f y_m^ * y_{\widetilde m } y_{\widetilde \lambda } { \left(t^a\right)_\alpha}^j \delta_l^{l ' } \ ] ] note that @xmath109 in the last term is the matrix part of @xmath33 , which cancels on both sides of the equation once we do the rewriting @xmath110 . the beta function for @xmath111 is obtained via : @xmath112 @xmath113 and @xmath114 are related by a similar transformation . we compute the former : @xmath115 + 2 { \widetilde}{y}^i y^\dagger_j { \widetilde}{u}^i \nonumber\\[2 mm ] & + u^k \text{tr}\left [ u^\dagger_k u^j\right ] + u^k \text{tr}\left [ y^\dagger_k y^j\right ] - 3 g^2 u^j c_2({\widetilde}{q}).\end{aligned}\ ] ] using the expression for the matrices we deduce : @xmath116{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) \alpha_{{\widetilde}{\lambda } } + n_f \alpha_h \right\ } \right.\nonumber \\ & \quad \left . + n_f \alpha_m + c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) \alpha_\lambda - 3 \alpha_g c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}})\right ] u^j - \frac{2 c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) } { ( 4\pi)^2 } y_{\widetilde \lambda } y_\lambda^ * y_{\widetilde m } \delta_\alpha^j \delta_l^{l ' } \delta_{r}^{r ' } , \ ] ] where we used @xmath117 . as before @xmath118 is the matrix part of @xmath119 , which cancels on both sides of the equation . the renormalization of the @xmath73 coupling reads : @xmath120\nonumber \\ & + v \text{tr}\left [ v^\dagger v \right ] - 3g^2 \left \ { c_2(q ) v + v c_2({\widetilde}{q } ) \right \}\nonumber\\[2 mm ] \beta(v)= & \left [ \frac{\alpha_m + \alpha_{{\widetilde}{m } } + 2\alpha_h}{2}n_f + \left ( \frac{\alpha_\lambda + \alpha_{{\widetilde}{\lambda}}}{2}-6\alpha_g\right)c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) + x \alpha_h \right ] v\end{aligned}\ ] ] we summarize now the results for beta functions , after having cancelled the matrices : @xmath121{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) \left [ c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) + c_2(g ) \right ] ( \alpha_\lambda + \alpha_{{\widetilde}{\lambda } } ) + \frac{c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}})x n_f^2}{d(g ) } ( \alpha_m + \alpha_{\widetilde{m } } + 2 \alpha_h ) \right]\\ & \beta(y_\lambda ) = \mid y_\lambda \mid e^{i\theta_\lambda } \left [ \frac{3}{2 } c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) \alpha_\lambda + \frac{t({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) n_f}{2 } ( \alpha_\lambda + \alpha_{{\widetilde}{\lambda } } ) + n_f \left ( \frac{\alpha_{{\widetilde}{m}}+\alpha_h}{2 } + \alpha_{m } \right ) - 3\alpha_g \left \ { c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}})+ c_2(\text{g } ) \right \}\right ] \nonumber\\ & \qquad \quad -2 n_f\frac{\mid y_m \mid \mid y_{\widetilde m}\mid \mid y_{\widetilde \lambda}\mid}{(4\pi)^2}e^{i(\theta_{{\widetilde}{\lambda}}+ \delta \theta_m ) } \\[2 mm ] & \beta ( y_m ) = \mid y_m \mid e^{i\theta_m } \left [ \frac{3}{2}n_f \left ( \alpha_m + \frac{\alpha_h}{3}\right ) + \frac{x}{2 } ( \alpha_m + \alpha_{{\widetilde}{m}})+ c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) \left ( \frac{\alpha_{{\widetilde}{\lambda}}}{2 } + \alpha_\lambda \right ) - 3\alpha_g c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) \right ] \nonumber\\ & \qquad \quad - 2c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) \frac{\mid y_{\widetilde m}\mid \mid y_\lambda\mid \mid y_{\widetilde \lambda}\mid}{(4\pi)^2}e^{i(\theta_{{\widetilde}{m}}+ \delta \theta_\lambda ) } \\[2 mm ] & \beta(\alpha_h ) = 2 \alpha_h \left [ \frac{n_f}{2}(\alpha_m + \alpha_{{\widetilde}{m } } + 2\alpha_h ) + c_2({\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt } } } } ) \left ( \frac{\alpha_{\lambda } + \alpha_{{\widetilde}{\lambda}}}{2}-6\alpha_g\right)+ x \alpha_h \right]\\ & \beta(\alpha_{{\widetilde}{\lambda } } ) = \beta_{\alpha_{\lambda } } \left ( y_{\lambda } \leftrightarrow y_{\widetilde \lambda } , y_{\widetilde m } \leftrightarrow y_{m } \right ) , \quad \beta(\alpha_{\widetilde{m } } ) = \beta_{\alpha_{m}}\left ( { y}_{m } \leftrightarrow y_{\widetilde m } , y_{\widetilde \lambda } \leftrightarrow { y}_{\lambda } \right),\end{aligned}\ ] ] where we have made explicit the phases of the yukawa s couplings : @xmath122 note also that we define @xmath123 where @xmath124 is the complex conjugate of @xmath125 . we have considered the case of real couplings in the main text . we provide here the eigenvectors associated to the eigenvalues given in tables [ exponents0 ] and [ exponents1 ] . shaded cells indicate unstable directions , i.e. corresponding to negative eigenvalues . f. sannino and k. tuominen , phys . d * 71 * , 051901 ( 2005 ) [ arxiv : hep - ph/0405209 ] . d. k. hong , s. d. h. hsu and f. sannino , phys . b * 597 * , 89 ( 2004 ) [ arxiv : hep - ph/0406200 ] . h. gies and j. jaeckel , eur . j. c * 46 * , 433 ( 2006 ) [ arxiv : hep - ph/0507171 ] . j. braun and h. gies , phys . b * 645 * , 53 ( 2007 ) [ arxiv : hep - ph/0512085 ] . j. braun and h. gies , jhep * 0606 * , 024 ( 2006 ) [ arxiv : hep - ph/0602226 ] . d. d. dietrich and f. sannino , phys . rev . d * 75 * , 085018 ( 2007 ) [ arxiv : hep - ph/0611341 ] . t. a. ryttov and f. sannino , phys . rev . d * 76 * , 105004 ( 2007 ) [ arxiv:0707.3166 [ hep - th ] ] . t. a. ryttov and f. sannino , phys . d * 78 * , 065001 ( 2008 ) [ arxiv:0711.3745 [ hep - th ] ] . j. braun and h. gies , jhep * 1005 * , 060 ( 2010 ) [ arxiv:0912.4168 [ hep - ph ] ] . o. antipin and k. tuominen , phys . d * 81 * , 076011 ( 2010 ) [ arxiv:0909.4879 [ hep - ph ] ] . o. antipin , k. tuominen , [ arxiv:0912.0674 [ hep - ph ] ] . j. alanen , k. kajantie and k. tuominen , phys . d * 82 * , 055024 ( 2010 ) [ arxiv:1003.5499 [ hep - ph ] ] . h. s. fukano and f. sannino , phys . d * 82 * , 035021 ( 2010 ) [ arxiv:1005.3340 [ hep - ph ] ] . c. pica and f. sannino , arxiv:1011.3832 [ hep - ph ] . c. pica and f. sannino , phys . d * 83 * , 035013 ( 2011 ) [ arxiv:1011.5917 [ hep - ph ] ] . m. t. frandsen , t. pickup , m. teper , phys . lett . * b695 * , 231 - 237 ( 2011 ) . [ arxiv:1007.1614 [ hep - ph ] ] j. braun , c. s. fischer and h. gies , arxiv:1012.4279 [ hep - ph ] . m. jarvinen and f. sannino , jhep * 1102 * , 081 ( 2011 ) [ arxiv:1009.5380 [ hep - ph ] ] . f. sannino , phys . d * 79 * , 096007 ( 2009 ) [ arxiv:0902.3494 [ hep - ph ] ] . t. a. ryttov and f. sannino , int . j. mod . a * 25 * , 4603 ( 2010 ) [ arxiv:0906.0307 [ hep - ph ] ] . f. sannino , phys . d * 80 * , 065011 ( 2009 ) [ arxiv:0907.1364 [ hep - th ] ] . g. t hooft , _ recent developments in gauge theories _ , plenum press , 1980 , 135 ; reprinted in _ unity of forces in the universe vol . , a. zee ed . , world scientific 1982 , 1004 . j. terning , phys . lett . * 80 * , 2517 ( 1998 ) [ arxiv : hep - th/9706074 ] . f. sannino , nucl . b * 830 * , 179 ( 2010 ) [ arxiv:0909.4584 [ hep - th ] ] . m. mojaza , m. nardecchia , c. pica and f. sannino , phys . d * 83 * , 065022 ( 2011 ) [ arxiv:1101.1522 [ hep - th ] ] . f. sannino , arxiv:1102.5100 [ hep - ph ] . f. sannino , phys . d * 82 * , 081701 ( 2010 ) [ arxiv:1006.0207 [ hep - lat ] ] . f. sannino , phys . lett . * 105 * , 232002 ( 2010 ) [ arxiv:1007.0254 [ hep - ph ] ] . s. di chiara , c. pica and f. sannino , arxiv:1008.1267 [ hep - ph ] . m. e. machacek and m. t. vaughn , nucl . b * 222 * , 83 ( 1983 ) . m. e. machacek and m. t. vaughn , nucl . b * 236 * , 221 ( 1984 ) . m. x. luo , h. w. wang and y. xiao , phys . d * 67 * , 065019 ( 2003 ) [ arxiv : hep - ph/0211440 ] . s. catterall , e. dzienkowski , j. giedt , a. joseph and r. wells , jhep * 1104 * , 074 ( 2011 ) [ arxiv:1102.1725 [ hep - th ] ] . s. catterall , arxiv:1005.5346 [ hep - lat ] .
a fundamental purpose of molecular simulation is the computation of macroscopic quantities , typically through averages of functions of the variables of the system with respect to a given probability measure @xmath0 , which defines the macroscopic state of the system . in the most common setting , the probability measure @xmath0 with respect to which averages are computed corresponds to the canonical ensemble ( see for instance @xcite ) . its distribution is defined by the boltzmann - gibbs density , which models the configurations of a conservative system in contact with a heat bath at fixed temperature . numerically , high - dimensional averages with respect to @xmath0 are often approximated as ergodic averages over realizations of appropriate stochastic differential equations ( sdes ) : @xmath1 a typical dynamics to this end is the langevin dynamics @xmath2 where @xmath3 is a standard wiener process , @xmath4 the potential energy function , @xmath5 a friction coefficient , @xmath6 the mass matrix of the system , and @xmath7 is proportional to the inverse temperature ( see section [ section modified langevin dynamics ] for more precise definitions ) . for references on the ergodicity of langevin dynamics , we refer the reader to @xcite and @xcite , for instance . there are two main sources of error in the computation of average properties such as @xmath8 through time averages as in : ( i ) a systematic bias ( or _ perfect sampling bias _ ) related to the use of a discretization method for the sdes ( and usually proportional to a power of the integration step size @xmath9 ) , and ( ii ) statistical errors , due to the finite lengths of the sampling paths involved and the underlying variance of the random variables . the first point was studied in @xcite for standard langevin dynamics . our focus in this work is on the statistical error . statistical errors may be large when the dynamics is metastable , _ i.e. _ when the system remains trapped for a very long time in some region of the configuration space ( called a metastable region ) before hopping to another metastable region . metastability implies that the convergence of averages over trajectories is very slow , and that transitions between metastable regions ( which are typically the events of interest at the macroscopic level ) are very rare . in fact , metastability arises from the multi - modality of the probability measure sampled by the dynamics . we refer for instance to @xcite for a review on ways to quantify the metastability of sampling dynamics . there are various strategies to reduce the variance of time averages by reducing the metastability . the most famous one is importance sampling : the potential energy function @xmath4 is modified by an additional term @xmath10 so that the langevin dynamics associated with @xmath11 is less metastable . an automatic way of doing so is to consider a so - called reaction coordinate , and define @xmath10 as the opposite of the associated free energy ( see @xcite for further precisions ) . we explore here an alternative path , which consists in modifying the kinetic energy rather than the potential energy . indeed , recall that the difficult part in sampling the canonical measure is in sampling positions ( see section [ section modified langevin dynamics ] for a more precise discussion of this point ) . there is therefore some freedom in the choice of the kinetic energy if the goal is to compute average properties . previous works in this direction focused on changing the mass matrix in order to increase the time steps used in the simulation ( see _ e.g. _ @xcite ) . the mathematical analysis we provide is inspired by a recent work by @xcite where the kinetic energy of each particle is more drastically modified : it is set to 0 when the particle s momenta are small , while it remains unchanged for larger momenta . in such _ adaptively restrained _ ( ar ) simulations , particles may become temporarily _ frozen _ , while their momenta may continue to evolve . since , in many cases , inter - particle forces only depend on relative particle positions , and hence do not have to be updated when particles are frozen , adaptively restrained particle simulations may yield a significant algorithmic speed - up @xmath12 when a sufficiently large number of particles are frozen at each time step ( or , more generally , when inter - particle distances remain constant and particle forces are expressed in local reference frames ) . this has been demonstrated in several contexts , _ e.g. _ for modeling hydrocarbon systems ( @xcite ) , proteins ( @xcite ) , and for electronic structure calculations ( @xcite ) . unfortunately , freezing particles even temporarily may make iterates more correlated , which may translate into an increase of the statistical error @xmath13 observed for modified langevin dynamics , compared to the statistical error @xmath14 observed for standard langevin dynamics . the actual speed - up of the method , in terms of the total wall - clock time needed to achieve a given precision in the estimation of an observable , should therefore be expressed as : @xmath15 our aim here is thus to quantify the increase in the variance as a function of the parameters of the modified kinetic energy . in fact , a first task is to prove that the langevin dynamics with modified kinetic energy is indeed ergodic , and that the variance is well defined . this is unclear at first sight since the modified dynamics fails to be hypoelliptic ( see the discussion in section [ sec : cv_ergo_avg ] ) . this article is organized as follows . in section [ section modified langevin dynamics ] , we introduce the modified langevin dynamics we consider , and present the particular case of the ar - langevin dynamic . the ergodicity of these dynamics is proved in section [ section ergodicity ] , both in terms of almost - sure convergence of time averages along a single realization , and in terms of the law of the process . we also provide a result on the regularity of the evolution semi - group , adapted from similar estimates for standard langevin dynamics in @xcite . such estimates allow us to analyze the statistical error in section [ section analysis of statistical error ] . we state in particular a central limit theorem for @xmath16 , and perform a perturbative study of the asymptotic variance of the ar - langevin dynamics in some limiting regime . our theoretical findings are illustrated by numerical simulations in section [ sec : numerics ] , both in a simple one - dimensional case where the variance can be accurately computed using an appropriate galerkin approximation , as well as for a more realistic system for which we resort to monte - carlo simulations . the proofs of our results are gathered in section [ section proofs ] . we consider a system of @xmath17 particles in spatial dimension @xmath18 , so that the total dimension of the system is @xmath19 . the vectors of positions and momenta are denoted respectively by @xmath20 and @xmath21 . periodic boundary conditions are used for positions , so that the phase - space of admissible configurations is @xmath22 with @xmath23 , @xmath24 being the one - dimensional unit torus and @xmath25 the size of the simulation box . in order to possibly increase the rate of convergence of the ergodic averages , we modify the langevin dynamics by changing the kinetic energy . more precisely , instead of the standard quadratic kinetic energy @xmath26 we introduce a general kinetic energy function @xmath27 . the total energy of the system is then characterized by the hamiltonian @xmath28 in order to ensure that the measure @xmath29 can be normalized , and in order to simplify the mathematical analysis , we make in the sequel the following assumption . [ assumption potential ] the potential energy function @xmath4 belongs to @xmath30 , and @xmath31 grows sufficiently fast at infinity in order to ensure that @xmath32 . the langevin dynamics associated with a general hamiltonian reads @xmath33 where @xmath3 is a standard @xmath34-dimensional wiener process and @xmath35 is the friction constant . for the separable hamiltonian , the general langevin dynamics simplifies as @xmath36 the generator of the process ( [ modified langevin ] ) reads @xmath37 a simple computation shows that the canonical distribution @xmath38 is invariant under the dynamics , _ i.e. _ for all @xmath39 functions @xmath40 with compact support , @xmath41 note that , in view of the separability of the hamiltonian , the marginal of the distribution @xmath0 in the position variables is , for any kinetic energy @xmath42 , @xmath43 in particular , this marginal distribution therefore coincides with the one of the standard langevin dynamics . this allows to straightforwardly estimate canonical averages of observables depending only on the positions with the modified langevin dynamics . in fact , there is no restriction in generality in considering observables depending only on the positions , since general observables @xmath44 depending both on momenta and positions can be reduced to functions of the positions only by a partial integration in the momenta variables . this partial integration is often very easy to perform since momenta are independent gaussian random variables under the canonical measure associated with the standard kinetic energy . a concrete example for the choice of the kinetic energy function @xmath42 in ( [ eq : modified hamiltonian ] ) is the one proposed for the adaptively restrained langevin dynamics in @xcite . it is parameterized by two constants @xmath45 . in this model , the kinetic energy is a sum of individual contributions @xmath46 for large values of momenta , the modified individual kinetic energies are equal to the standard kinetic energy of one particle , but they vanish for small momenta : @xmath47 an appropriate function allows to smoothly interpolate between these two limiting regimes ( see definition [ definition arps langevin ] for the precise expression ) . a possible choice of an individual kinetic energy @xmath48 , as well as the associated canonical distribution of momenta @xmath49 are depicted in figure [ figure 1 left ] and figure [ figure 1 right ] when @xmath50 . ( solid lines ) , and an example of an ar kinetic energy function @xmath48 with parameters @xmath51 and @xmath52 ( dashed line).,scaledwidth=80.0% ] .,scaledwidth=80.0% ] the interest of ar - langevin dynamics is that , when their individual kinetic energies are sufficiently small , particles do not move . when two particles are frozen in this way , their pairwise interactions need not be updated . this allows decreasing the computational complexity of the force computation , which is typically the most time - consuming part of a molecular dynamics solver . note that this can be generalized to higher - order interactions ( such as three - body interactions based on bending angles for instance ) . [ remark individual parameters ] note that , due to the additive structure of the kinetic energy , the momenta @xmath53 are independent and identically distributed ( i.i.d . ) under the canonical measure . it is however possible to choose different parameters @xmath54 and @xmath55 for different particles , for example to focus calculations on a specific part of the particle system , in which case the momenta are still independent but not longer identically distributed . such a situation is considered in the numerical example presented in section [ subsection a more realistic system : second order discretization ] . there are several notions of ergodicity for stochastic processes . we focus here on two of them : the convergence of ergodic averages over a single trajectory , and the convergence of the law of the process . the convergence of ergodic averages over one trajectory is automatically ensured by the existence of an invariant probability measure and the irreducibility of the dynamics ( see for instance @xcite for early results on such convergences for possibly degenerate diffusions ) . since , by construction , an invariant probability measure is known ( namely the canonical measure ) , it suffices to show that the process generated by the modified langevin equation is irreducible to conclude to the convergence of ergodic averages . as reviewed in @xcite , the most standard argument to prove the irreducibility of degenerate diffusions is to prove the controllability of the dynamics relying on the stroock - varadhan support theorem , and the regularity of the transition kernel thanks to some hypoellipticity property . these conditions are satisfied for standard langevin dynamics ( see for instance @xcite ) , but not for the modified langevin dynamics we consider , since the hessian of the kinetic energy function may not be invertible on an open set . this is the case for the ar kinetic energy function presented in section [ sec : arps_presentation ] . to illustrate this point , let us show for instance how the standard way of proving hypoellipticity fails ( the proof of the controllability faces similar issues ) . the first task is to rewrite the generator of the process as @xmath56 where @xmath57 and @xmath58 is the adjoint of @xmath59 on the flat space @xmath60 . we next compute , for @xmath61 , the commutators @xmath62 = x_0x_j - x_j x_0 = \sqrt{\frac{\gamma}{\beta}}\,{\nabla}\left({\partial}_{p_j } u\right)\cdot \left({\nabla}_q-\gamma { \nabla}_p\right ) . \ ] ] when @xmath63 is invertible , it is possible to recover the full algebra of derivatives by an appropriate combination of @xmath64 and @xmath65,\dots,[x_0,x_d]$ ] . here , we consider a situation when this is not the case and , even more dramatically , where the hessian may vanish on an open set . in this situation , @xmath66=0 $ ] on the same open set , and in fact all iterated commutators @xmath67]$ ] also vanish . we solve this problem by a direct constructive approach , where we see the modified dynamics as a perturbation of the standard langevin dynamics . we rely on the following assumption : the kinetic energy function @xmath68 of the modified langevin dynamics is such that @xmath69 for some constant @xmath70 . [ hyp mod langevin ergodicity ] under this assumption , we can prove that the modified langevin dynamics is irreducible by proving an appropriate minorization condition , which crucially relies on the compactness of the position space @xmath71 ( see section [ proof of the minorization condition ] for the proof ) . [ lemma minorization condition ] suppose assumption [ hyp mod langevin ergodicity ] holds . then for any fixed @xmath72 and @xmath73 , there exists a probability measure @xmath74 on @xmath75 and a constant @xmath76 such that , for every borel set @xmath77 , @xmath78 with @xmath79 when @xmath80 . the minorization condition implies the irreducibility of the dynamics , so that the following convergence result readily follows . [ thm convergence of averages ] when assumption [ hyp mod langevin ergodicity ] holds , ergodic averages over trajectories almost surely converge to the canonical average : @xmath81 there are various functional frameworks to measure the convergence of the law of the process . we consider here weighted @xmath82 estimates on the semi - group @xmath83 . more precisely , we introduce a scale of lyapunov functions @xmath84 for @xmath85 . recall indeed that only momenta need to be controlled since positions remain in a compact space . the associated weighted @xmath82 spaces are @xmath86 in order to prove the exponential convergence of the law , we rely on the result of @xcite , which states that if a lyapunov condition and a minorization condition hold true , then the sampled chain converges exponentially fast to its steady state in the following sense . [ thm uniform convergence ] suppose that assumption [ hyp mod langevin ergodicity ] holds . then the invariant measure @xmath0 is unique , and for any @xmath87 , there exist constants @xmath88 such that @xmath89 as mentioned above , the proof of this result directly follows from the arguments of @xcite . the minorization condition is already stated in lemma [ lemma minorization condition ] , while the appropriate lyapunov condition reads as follows ( see section [ proof of the lyapunov condition ] for the proof , which uses the same strategy as @xcite and @xcite ) . [ lemma lyapunov ] suppose that assumption [ hyp mod langevin ergodicity ] holds . then , for any @xmath90 and @xmath73 , there exist @xmath91 and @xmath92 such that @xmath93 we provide in this section decay estimates for the spatial derivatives of @xmath94 , following the approach pioneered in @xcite and further refined in @xcite . such estimates were obtained for the standard langevin dynamics , but can in fact straightforwardly be extended to modified langevin dynamics with hessians bounded from below by a positive constant . our aim in this section is to provide decay estimates for the spatial derivatives of @xmath94 in the situation when @xmath63 fails to be strictly convex , for instance because @xmath63 vanishes on an open set as is the case for ar particle simulations . in order to state our results , we first need to define the weighted sobolev spaces @xmath95 for @xmath96 : @xmath97 these spaces gather all functions which grow at most like @xmath98 , and whose derivatives of order at most @xmath99 all grow at most like @xmath98 . we also introduce the space of smooth functions @xmath100 , the vector space of functions @xmath101 such that , for any @xmath102 , there exists @xmath103 for which @xmath104 . we also make the following assumption on the kinetic energy function , which can be understood as a condition of `` almost strict convexity '' of the hessian @xmath105 . [ hyp thm ] the kinetic energy @xmath106 has bounded second - order derivatives : @xmath107 and there exist a function @xmath108 and constants @xmath109 and @xmath110 such that @xmath111 and @xmath112 a natural choice for the function @xmath113 in assumption [ hyp thm ] is @xmath114 . the condition then holds with @xmath115 . moreover , holds true as soon as @xmath42 is a local perturbation of @xmath116 . the most demanding condition therefore is , especially if @xmath117 has to be small . by following the same strategy as in ( * proposition a.1 . ) ( which refines the results already obtained in @xcite ) , and appropriately taking care of the lack of strict positivity of the hessian @xmath63 by assuming that @xmath117 is sufficiently small , we prove the following result in section [ proof proposition kopec ] . [ main theorem kopec ] suppose that assumptions [ hyp mod langevin ergodicity ] and [ hyp thm ] hold , and fix @xmath118 . for any @xmath119 , there exist @xmath120 and @xmath121 such that , for @xmath122 and @xmath123 with @xmath124 sufficiently small ( depending on @xmath125 but not on @xmath99 ) , there is @xmath103 and @xmath126 for which @xmath127 the parameter @xmath128 can in fact be made explicit , see below . the decay estimate shows that the derivatives of the evolution operator can be controlled in appropriate weighted hilbert spaces . note however that the lyapunov functions entering in the estimates are not the same a priori on both sides of the inequality . let us emphasize , though , that we can obtain a control in all spaces @xmath129 for @xmath125 sufficiently large ( depending on the order of derivation ) . the asymptotic variance characterizes the statistical error . in section [ sec : asymptotic_variance ] , we show that the asymptotic variance is well defined for the modified langevin dynamics . we can in fact prove a stronger result , namely that a central limit theorem ( clt ) holds true for ergodic averages over one trajectory . in a second step , we more carefully analyze in section [ sec : perturbative_variance ] the properties of the variance of the ar - langevin dynamics by proving a linear response result in the limit of a vanishing lower bound on the kinetic energies . to obtain the latter results , we rely on the estimates provided by lemma [ main theorem kopec ] . let us first write the asymptotic variance in terms of the generator of the dynamics . to simplify the notation , we introduce the orthogonal projection @xmath130 onto the orthogonal of the kernel of the operator @xmath131 ( with respect to the @xmath132 scalar product ) : for any @xmath133 , @xmath134 since @xmath135 , we can define @xmath136 . the ergodicity result allows us to conclude that the operator @xmath131 is invertible on @xmath137 since the following operator equality holds on @xmath138 , the banach space of bounded operators on @xmath137 : @xmath139 this leads to the following resolvent bounds ( the second part being a direct corollary of lemma [ main theorem kopec ] ) . [ corollary convergence thm ] suppose that assumption [ hyp mod langevin ergodicity ] holds . then , for any @xmath140 , @xmath141 where @xmath142 are the constants introduced in theorem [ thm uniform convergence ] . suppose in addition that assumption [ hyp thm ] holds , and fix @xmath118 . for any @xmath119 , there exist @xmath120 and @xmath121 such that , for @xmath122 and @xmath123 with @xmath124 sufficiently small ( depending on @xmath125 but not on @xmath99 ) , there is @xmath103 and @xmath126 for which @xmath143 this already allows us to conclude that the asymptotic variance of the time average @xmath16 defined in is well defined for any observable @xmath144 since @xmath145 \\ & = \lim_{t \rightarrow \infty}2\int_0^t\int_{{\mathcal{e}}}\left(1-\frac{s}{t}\right)\left(\rm{e}^{s{\mathcal{l}}}\pi_{\mu}a\right)\left(\pi_{\mu}a \right ) d\mu\\ & = 2\int_0^{\infty}\int_{{\mathcal{e}}}\left({\mathrm}{e}^{s{\mathcal{l}}}\pi_{\mu}a\right)\left(\pi_{\mu}a\right ) d\mu \end{aligned}\ ] ] by the dominated convergence theorem . therefore , @xmath146 in fact , a central limit theorem can be shown to hold for @xmath16 using standard results ( see _ e.g. _ @xcite ) . our aim in this section is to better understand , from a quantitative viewpoint , the behavior of the asymptotic variance for the ar - langevin dynamics defined in section [ sec : arps_presentation ] , at least in some limiting regime where the parameter @xmath54 is small . for intermediate values , we need to rely on numerical simulations ( see section [ sec : numerics ] ) . the regime where both @xmath54 and @xmath55 go to 0 is somewhat singular since the transition from @xmath147 to @xmath148 becomes quite abrupt , which prevents a rigorous theoretical analysis . the regimes where either @xmath54 or @xmath55 go to infinity are also of dubious interest since the dynamics strongly perturbs the standard langevin dynamics . therefore , we restrict ourselves to the situation where @xmath149 with @xmath55 fixed . in order to highlight the dependence of the ar kinetic energy function on the restraining parameters @xmath45 , we denote it by @xmath150 in the remainder of this section . let us however first give a more precise definition of this function , having in mind that @xmath55 is fixed while @xmath54 eventually goes to 0 . we introduce to this end an interpolation function @xmath151 such that @xmath152 and @xmath153 we next define an interpolation function @xmath154 obtained from the function @xmath155 by an appropriate shift of the lower bound and a rescaling . more precisely , @xmath156 with @xmath157 a plot of @xmath158 is provided in figure [ figure arps redefinition shift functions left ] . and @xmath154 for @xmath51 and @xmath52.,scaledwidth=80.0% ] , as well as two ar kinetic energy functions @xmath150 with @xmath51 and @xmath159 or @xmath160.,scaledwidth=80.0% ] [ definition arps langevin ] for two parameters @xmath45 , the ar kinetic energy function @xmath161 is defined as @xmath162 where the individual kinetic energy functions are @xmath163\frac{p_i^2}{2m_i } , & \quad \text{for } \frac{p_i^2}{2m_i } \in [ { k_{\mathrm{min}}},{k_{\mathrm{max } } } ] , \\ \frac{p_i^2}{2m_i } , & \quad \text{for } \frac{p_i^2}{2m_i}{\geqslant}{k_{\mathrm{max}}}. \\ \end{aligned } \right.\ ] ] of course , @xmath164 converges to @xmath165 as @xmath149 . the limiting kinetic energy function @xmath166 corresponds to what we call the zero-@xmath55-ar - langevin dynamics ( see figure [ figure arps redefinition shift functions right ] for an illustration ) . let us emphasize that the limiting dynamics is not the standard langevin dynamics , so that the expansion in powers of @xmath54 of the variance we provide is with respect to the limiting variance of the dynamics corresponding to @xmath166 . to simplify the notation , we denote by @xmath167 the variance associated with the kinetic energy @xmath150 . [ proposition variance perturbation ] there exists @xmath168 such that , for any @xmath169 , there is a constant @xmath170 for which @xmath171 the proof can be read in section [ proof of the proposition variance perturbation ] . the assumption that @xmath55 is sufficiently small ensures that assumption [ hyp thm ] holds ( see section [ proof of lemma arps condition ] ) . the result is formally clear . the difficulty in proving it is that the kinetic energy is not a smooth function of @xmath54 because the shift function is only piecewise smooth . [ rmk : lr_var ] an inspection of the proof of proposition [ proposition variance perturbation ] shows that the linear response result can be generalized to non - zero values of @xmath54 and in fact to linear responses in the parameter @xmath55 as well . for the latter case , we consider @xmath172 . denoting now by @xmath173 the variance associated with the kinetic energy @xmath150 , it can be proved that , for @xmath174 not too large , there are @xmath175 such that , for @xmath176 sufficiently small , @xmath177 the aim of this section is to quantify the evolution of the variance of ar - langevin dynamics as the parameters of the kinetic energy function are modified . we first consider in section [ sec : galerkin ] a simple system in spatial dimension 1 , for which the variance can be very precisely computed using a galerkin - type approximation . we next consider more realistic particle systems in section [ subsection a more realistic system : second order discretization ] , relying on molecular dynamics simulations to estimate the variance . in this section , the function @xmath178 is chosen to be of the form @xmath179 , with @xmath180 a fifth - order spline function . we first consider a single particle in spatial dimension @xmath181 , in the periodic domain @xmath182 and at inverse temperature @xmath183 . in this case , it is possible to directly approximate the asymptotic variance using some galerkin discretization , as in @xcite or @xcite . we denote by @xmath184 the generator of the modified langevin dynamics associated with the ar kinetic energy function @xmath150 defined in , by @xmath185 the associated canonical measure , and by @xmath186 the projector onto functions of @xmath187 with average 0 with respect to @xmath185 . for a given observable @xmath188 , we first approximate the solution of the following poisson equation : @xmath189 and then compute the variance as given by : @xmath190 to achieve this , we introduce the basis functions @xmath191 , where @xmath192 ( for @xmath193 ) and @xmath194 are the hermite polynomials : @xmath195 the choice of @xmath196 is natural in view of the spatial periodicity of the functions under consideration , while hermite polynomials are eigenfunctions of the generator associated with the ornstein - uhlenbeck process on the momenta for the standard quadratic kinetic energy @xmath197 . note however that , when the kinetic energy is modified as @xmath150 , the hermite polynomials are no longer orthogonal for the @xmath187 scalar product . we approximate the poisson equation on the basis @xmath198 for given integers @xmath199 , and we look for approximate solutions of the form @xmath200 with @xmath201_{nk } \psi_{nk},\ ] ] where @xmath202 is a vector of size @xmath203 . restricting to @xmath204 leads to @xmath205 where @xmath206 is a matrix of size @xmath207 and @xmath208 a vector of size @xmath203 , whose entries respectively read @xmath209_{nk , ml } & = \left\langle \psi_{ml},-{\mathcal{l}}_{{k_{\mathrm{min}}},{k_{\mathrm{max } } } } \psi_{nk}\right\rangle_{l^2(\mu_{{k_{\mathrm{min}}},{k_{\mathrm{max } } } } ) } , \\ \left[a_{n_g , n_h}\right]_{ml } & = \left\langle \psi_{ml},\pi_{{k_{\mathrm{min}}},{k_{\mathrm{max } } } } a \right\rangle_{l^2(\mu_{{k_{\mathrm{min}}},{k_{\mathrm{max}}}})}. \\ \end{aligned}\ ] ] the approximated solution @xmath210 of the poisson equation can therefore be computed by solving . note however that some care is needed at this stage since @xmath184 is not invertible on @xmath204 , because the basis functions @xmath211 are not of integral 0 with respect to @xmath185 . we correct this by performing a singular value decomposition of @xmath206 , removing the component of @xmath208 associated with the singular value 0 , and computing the inverse of @xmath206 on the subspace generated by the eigenvectors associated with non - zero eigenvalues . in practice , we compute the entries of @xmath208 and @xmath206 by numerical quadrature . since the hermite polynomials are no longer orthogonal for the @xmath187 scalar product , quadratures are required both in position and momentum variables . the variance is finally approximated as @xmath212 in the simulations presented in this section , the potential is @xmath213 , the observable under study is @xmath214 , and we always set @xmath215 . figure [ fig : convergence galerkin ] presents the convergence of the variance with respect to the basis size , for the standard langevin dynamics and the ar langevin dynamics with @xmath51 and various values of @xmath54 . the results show that the choice @xmath216 is sufficient in all cases to approximate the asymptotic value . we checked in addition in one case , namely for the standard dynamics , that the values we obtain are very close to a reference value obtained with @xmath217 : the relative variation is of order @xmath218 for @xmath219 , @xmath220 for @xmath216 and @xmath221 for @xmath222 . we therefore set @xmath216 in the remainder of this section . and @xmath223 : approximation of the variance of observable @xmath224 for the standard dynamics and the ar dynamics with fixed parameter @xmath51 and various values of @xmath54.,scaledwidth=80.0% ] the variation of the computed variance for @xmath224 is plotted in figure [ fig : variancegalerkinallkmaxkmin ] for various parameters @xmath45 of the ar - langevin dynamics . note that , as expected , the variance increases with increasing values of @xmath54 for fixed @xmath55 , but also with increasing values of @xmath55 for fixed @xmath54 . we next illustrate the linear response results of proposition [ proposition variance perturbation ] and remark [ rmk : lr_var ] in figures [ fig : variance_kmax5_overkmin left ] and [ fig : variance_kmax5_overkmin right ] : in both situations , the variance increases linearly with the parameter under consideration is varied in a sufficiently small neighborhood of its initial value . after that initial regime , nonlinear variations appear . note also that the relative increase of the variance is more pronounced as a function of @xmath55 than @xmath54 . , approximated by the galerkin method , as a function of @xmath54 and for several values of @xmath55.,scaledwidth=80.0% ] and its initial value computed for reference parameters . ( left ) fixed upper bound @xmath225 , and reference value @xmath226.,scaledwidth=80.0% ] . fixed lower bound @xmath159 , and reference value @xmath227.,scaledwidth=80.0% ] [ remark smoothness stability ] in practice , the idea usually is to set the lower bound @xmath54 sufficiently large when performing monte carlo simulations , in order to decrease as much as possible the computational cost . the gap @xmath228 should however not be too small in order to have a sufficiently smooth transition from a vanishing kinetic energy to a quadratic one . this requires therefore @xmath55 to be quite large if @xmath54 is large . the results presented in figure [ fig : variance_kmax5_overkmin right ] suggest that this may not be the optimal choice , unless the algorithmic speed - up is quite large . in order to study the variation of the variance as a function of @xmath54 and @xmath55 in systems of higher dimensions , we resort to monte carlo simulations . this requires discretizing the ar - langevin dynamics , and we resort to a scheme of weak order 2 , obtained by a splitting strategy where the generator of the modified langevin dynamics is decomposed into three parts : @xmath229 the transition kernel obtained by a strang splitting reads @xmath230 . contrarily to the standard kinetic energy functions , the elementary evolution associated with @xmath231 can not be integrated analytically . to preserve the order of the scheme , we approximate @xmath232 by a midpoint rule , encoded by a transition kernel @xmath233 satisfying @xmath234 for smooth test functions @xmath235 . this gives the following discretization scheme : @xmath236 where @xmath237 are i.i.d . standard @xmath34-dimensional gaussian random variables . the first and the last line are obtained by implicit schemes , solved in practice by a fixed point strategy ( the termination criterion being that the distance between successive iterates is smaller than @xmath220 , and the initial iterate being obtained by a euler - maruyama step ) . by following the same approach as in @xcite , it can indeed be proved that this scheme is of weak order 2 ; see @xcite for further precisions . the ergodicity of some second - order schemes was proved for the standard langevin dynamics in @xcite . since the ar - langevin dynamics can be seen as a perturbation of the standard langevin dynamics , it can be proved by combining the proofs from @xcite and the proof of theorem [ thm uniform convergence ] that , when @xmath45 are sufficiently small , the corresponding discretization of the ar - langevin dynamics remains ergodic ( see @xcite ) . the corresponding invariant measure is denoted by @xmath238 . it also follows by the results of @xcite that the error on averages of smooth observables @xmath239 with respect to @xmath238 is of order 2 , i.e. there exists @xmath240 such that @xmath241 as already mentioned in remark [ remark smoothness stability ] , the reduction of the gap between the parameters @xmath54 and @xmath55 reduces the smoothness of the transition between the restrained dynamics and the full dynamics . this raises issues in the stability of the scheme , which can be partly cured by resorting to a metropolis - hastings correction ( @xcite and @xcite ) . the system we consider is composed of @xmath242 particles in dimension 2 , so that @xmath243 and @xmath244 . the masses are set to 1 for all particles . among these particles , two particles ( numbered 1 and 2 in the following ) are designated to form a dimer while the others are solvent particles . all particles , except the two particles forming the dimer , interact through the purely repulsive wca pair potential , which is a truncated lennard - jones potential @xcite : @xmath245 + \varepsilon & \quad { \rm if \ } r { \leqslant}r_0 , \\ 0 & \quad { \rm if \ } r > r_0 , \end{array } \right.\ ] ] where @xmath246 denotes the distance between two particles , @xmath247 and @xmath248 are two positive parameters and @xmath249 . the interaction potential between the two particles of the dimer is a double - well potential @xmath250 ^ 2,\ ] ] where @xmath251 and @xmath252 are two positive parameters . the potential @xmath253 has two energy minima . the first one , at @xmath254 , corresponds to the compact state . the second one , at @xmath255 , corresponds to the stretched state . the total energy of the system is therefore , for @xmath256 with @xmath257 , @xmath258 where the solvent - solvent and dimer - solvent potential energies respectively read @xmath259 we choose @xmath183 , @xmath260 , @xmath261 , @xmath262 , @xmath263 , and set the particle density @xmath264 to 0.56 in the numerical results presented in this section , sufficiently high to ensure that the solvent markedly modifies the distribution of configurations of the dimer compared to the gas phase . the source of metastability in the system is the double - well potential on the dimer . in such a system , it makes sense to restrain only solvent particles ( since they account for most of the computational cost ) , and keep the standard kinetic energy for the particles forming the dimer ( since the observable depends on their positions ) . as noted in remark [ remark individual parameters ] , the method allows us to choose different individual kinetic energies for different particles . since the solvent interacts with the dimer , we study how the variance of time averages of observables related to the configuration of the dimer , such as the dimer potential energy @xmath265 , depend on the restraining parameters chosen for the solvent particles . we also estimate the variance of time averages based on observables depending only on the solvent degrees of freedom , such as the solvent - solvent potential energy @xmath266 . the asymptotic variance of time averages for a given observable @xmath188 is estimated by approximating the integrated auto - correlation function @xmath267dt,\ ] ] where the expectation is with respect to initial conditions @xmath268 and all realizations of the ar langevin dynamics . this is done by first truncating the upper bound in the integral by a sufficiently large time @xmath269 , and using a trapezoidal rule : @xmath270 where @xmath271 , and the empirical averages over @xmath6 realizations of trajectories of @xmath272 steps are defined as @xmath273 with @xmath274 the initial condition @xmath275 for the @xmath276th trajectory is obtained from the last configuration of the @xmath277th configuration , namely @xmath278 . figure [ fig : acf ] presents the auto - correlation function obtained for @xmath265 . the results show that the choice @xmath279 is reasonable . $ ] for @xmath280 as a function of time.,scaledwidth=80.0% ] the results of @xcite show that the errors on the approximation of the variance should be of order @xmath281 when @xmath282 . this is illustrated in figures [ fig : variance zero coeff left ] , [ fig : variance zero coeff right ] , [ fig : variance max coeff left ] and [ fig : variance max coeff right ] , which present the convergence of @xmath283 as a function of @xmath284 for @xmath285 . it is possible to extrapolate the value of the variance at @xmath286 by fitting @xmath283 as @xmath287 . note that the errors on the variance are bigger in the case @xmath288 , which is expected due to the smaller gap between the parameters @xmath289 . in the sequel , all the reported approximations of the variance are obtained by computing @xmath283 for 6 values of the time step @xmath284 , and extrapolating to the limit @xmath290 as in figures [ fig : variance zero coeff left ] , [ fig : variance zero coeff right ] , [ fig : variance max coeff left ] and [ fig : variance max coeff right ] . more precisely , the time steps are chosen as @xmath291 for @xmath292 when @xmath226 , and @xmath293 for @xmath294 . for intermediate values of @xmath54 , the time steps @xmath295 are obtained by a linear interpolation between @xmath296 and @xmath297 . for @xmath298 , as a function of the step size @xmath284 , for @xmath159 and @xmath299.,scaledwidth=80.0% ] for @xmath300 , as a function of the step size @xmath284 , for @xmath159 and @xmath299.,scaledwidth=80.0% ] , except that @xmath288.,scaledwidth=80.0% ] , except that @xmath288.,scaledwidth=80.0% ] the variations as a function of @xmath54 of the approximations of the variances @xmath301 for the solvent - solvent potential energy @xmath302 and the dimer potential energy @xmath253 are reported in figures [ fig : variance over kmin left ] and [ fig : variance over kmin right ] . surprisingly , even though the solvent particles are restrained , the variance of the solvent - solvent potential decreases linearly for moderately small values of @xmath54 ; whereas , as expected , the variance of the dimer potential , which is only implicitly influenced by the restraining parameters , increases linearly for these values of @xmath54 . in order to more easily compare the impacts of the restraining procedure , we plot in figure [ fig : comparisonreldifferencevariance_ss_dd left ] the relative differences of the variance @xmath167 and the variance of zero-@xmath55-ar dynamics @xmath303 as a function of @xmath54 . for the two observables under consideration , the impact of an increase of the parameter @xmath54 on the variance associated with the dimer potential is much weaker than on the variance related to the solvent potential . we also provide in figure [ fig : comparisonreldifferencevariance_ss_dd right ] the percentage of restrained particles , which directly depends on the restraining parameter @xmath54 and dictates the algorithmic speed - up . this supports the idea that the use of the ar - langevin method for heterogeneous systems can be beneficial when the ar parameters are set to non - zero values for the part of the system which is not directly of interest ( e.g. the solvent ) , while the standard kinetic energy should be kept for the degrees of freedom that are directly involved in the observable ( e.g. the dimer ) . for @xmath298 as a function of @xmath304 $ ] for @xmath305.,scaledwidth=80.0% ] for @xmath265 as a function of @xmath304 $ ] for @xmath305.,scaledwidth=80.0% ] with respect to the reference variances @xmath303.,scaledwidth=80.0% ] the modified langevin equation can be written as a perturbation of the langevin equation , namely @xmath306 where @xmath307 is uniformly bounded as @xmath308 in view of assumption [ hyp mod langevin ergodicity ] . by a direct integration in time of the momenta dynamics , @xmath309 where @xmath310 is a gaussian random variable with mean zero and covariance @xmath311 . note also that @xmath312 is uniformly bounded ; more precisely , @xmath313 . let us first consider the case @xmath314 . we introduce @xmath315 for a given time @xmath73 . with this notation , @xmath316 where we used young s inequality to obtain the last line , with a constant @xmath317 sufficiently small so that @xmath318 . we next take the expectation of the previous inequality , conditionally to the filtration of events up to time @xmath319 . since @xmath320=0 $ ] , it follows @xmath321{\leqslant}\alpha^2(1+{\varepsilon}){\mathcal{k}}_1(q_t , p_t ) + r,\ ] ] for some constant @xmath322 . this shows the lyapunov condition for @xmath323 . the higher order conditions ( @xmath324 ) can be proved as in ( * ? ? ? * section 5.1.5 ) , by noting that @xmath325 is equal to @xmath326 plus some lower order polynomial in @xmath327 . the main idea is , as in ( * ? ? ? * section 5.1.5 ) , to compare the modified langevin dynamics to the standard langevin dynamics with zero forces , for which a minorizing measure @xmath74 can be explicitly constructed . from the rewriting , we deduce , in view of the momenta evolution , @xmath328 where periodic boundary conditions are considered , and @xmath329 note that @xmath330 is bounded as @xmath331 whereas @xmath332 is a gaussian random variable , which is correlated to @xmath333 . a simple computation shows that @xmath334= \left(\begin{matrix } \mathcal{v}_{qq } & \mathcal{v}_{qp } \\ \mathcal{v}_{pq } & \mathcal{v}_{pp } \end{matrix } \right ) , \quad \qquad \left\ { \begin{aligned } \mathcal{v}_{qq}&=\frac{1}{\beta\gamma}\left(2t-\frac{1}{\gamma}\left(3 - 4\alpha_t+\alpha_t^2\right)\right),\\ \mathcal{v}_{qp } & = \frac{1}{\beta\gamma}\left(1-\alpha_t\right)^2 , \\ \mathcal{v}_{pp}&= \frac{1}{\beta}\left(1-\alpha_t^2\right ) , \end{aligned } \right.\ ] ] where @xmath335 is the same constant as in section [ proof of the lyapunov condition ] . therefore , for a given measurable set @xmath336 , @xmath337 where @xmath338 are both bounded by some constant @xmath339 ( depending on @xmath340 and @xmath319 ) when @xmath341 . note that there is an inequality in since we neglect in fact the periodic images of @xmath342 when writing it as @xmath343 , the latter two quantities being interpreted as elements of @xmath344 . since the matrix @xmath345 is definite positive , we can finally consider the following minorizing measure : @xmath346 where @xmath347 is a normalization constant . the proof is concluded by defining @xmath348 . the proof follows the strategy of ( * ? ? ? * proposition a.1 ) . we recall in this section the general outline of this proof , and highlight the required extensions . the proofs of these extensions are then provided in section [ proof of lemma in sketch of the proof ] . without restriction of generality , and in order to simplify the notation , we assume that @xmath349 . we introduce weight functions @xmath350 where the lyapunov functions @xmath98 are defined in . we also define @xmath351.\]]the following result , central in this proof , gives estimates on derivatives of @xmath352 in the weighted spaces @xmath353 ( see section [ proof of lemma in sketch of the proof ] for the proof ) . [ proposition kopec 1 ] suppose that assumptions [ hyp mod langevin ergodicity ] and [ hyp thm ] hold . for any @xmath119 , there exists @xmath121 and @xmath354 such that , for @xmath122 and @xmath355 with @xmath128 sufficiently small , there is @xmath356 and @xmath126 for which @xmath357 assume in the sequel that @xmath355 for @xmath125 sufficiently large . in view of the estimates , and using the fact that @xmath358 with @xmath359 as @xmath360 , we obtain that , for any @xmath119 , there exist @xmath354 such that , for @xmath361 , it is possible to find @xmath103 and @xmath126 for which @xmath362 by the sobolev embedding theorem , we can conclude that , for any @xmath119 , there exist @xmath363 such that , for @xmath361 and provided @xmath364 , it is possible to find @xmath103 and @xmath365 for which @xmath366 this concludes the proof of lemma [ main theorem kopec ] . [ lemma a.6 ] let @xmath367 be a linear operator . assume that @xmath106 and @xmath368 . there exists an integer @xmath369 such that , for all @xmath370 , there is a constant @xmath371 for which the following inequality holds true for any @xmath372 : @xmath373u(t ) \ \mathcal{a}u(t ) \pi_s \ , dq \ , dp \right)dt \ . \label{eq : lemma a.6 } \end{aligned}\ ] ] in fact , a careful inspection of the proof shows that , since @xmath374 , it is possible to avoid the assumption @xmath375 by appropriately increasing the lyapunov index @xmath125 . since @xmath375 for ar - langevin dynamics , we however keep this assumption . a simple computation shows that @xmath376 the formal adjoint of the operator @xmath131 in @xmath60 is given by @xmath377 in view of assumption [ hyp mod langevin ergodicity ] , there exists therefore @xmath378 such that @xmath379\pi_s \nonumber\\ & { \leqslant}\left(\omega_s + \gamma\left\|{\delta}u\right\|_{l^{\infty}}\right)\pi_{s}. \label{eq : estimation adjoint}\end{aligned}\ ] ] with this estimation , we can follow exactly the proof of ( * ? ? ? * lemma a.6 ) , _ i.e. _ write the expression for @xmath380 $ ] , use , integrate the resulting expression in time and with respect to @xmath381 ( for @xmath125 sufficiently large ) , and finally use to deduce . let us now prove lemma [ proposition kopec 1 ] . the complete proof is done by induction on @xmath99 . we provide here the complete proofs for @xmath382 and @xmath323 , and only sketch the extension to higher orders of derivation since the proof follows the same lines as in ( * ? ? ? * appendix a ) . recall first that , in view of assumption [ hyp mod langevin ergodicity ] , the exponential convergence of the law provided by theorem [ thm uniform convergence ] holds . denote by @xmath383 the corresponding exponential rate of decay for a given @xmath384 . for any @xmath103 , we directly obtain the following decay estimates in @xmath385 when @xmath386 : there exists @xmath387 such that @xmath388 note that this corresponds to the case @xmath382 in lemma [ proposition kopec 1 ] . we now prove the estimates in the case @xmath323 . we first apply lemma [ lemma a.6 ] with @xmath389 : there exists @xmath390 such that , for all @xmath391 and @xmath392 , there is @xmath126 and @xmath356 for which @xmath393dt{\leqslant}c \left\|a\right\|_{l^{\infty } _ { { \mathcal{k}}_{r}}}^2 . \label{eq : estimation dp}\ ] ] in order to control derivatives in @xmath394 , the key idea , going back to @xcite , is to use mixed derivatives @xmath395 ( for some parameter @xmath396 ) . this allows indeed to retrieve some dissipation in the @xmath394 direction when @xmath63 is positive definite . the next lemma is the most important part of our proof since we show how to extend the use of mixed derivatives to the case when @xmath63 is not positive definite . [ lemma estimation operator a1 ] consider the operator @xmath397 for some parameter @xmath398 . there exists @xmath390 such that , for @xmath391 and provided @xmath364 ( for some constant @xmath399 defined in below ) , there is @xmath400 , @xmath401 , @xmath396 and @xmath126 for which @xmath402dt { \leqslant}c \left\|a\right\|_{w^{1,\infty } _ { { \mathcal{k}}_{r}}}^2 . \label{eq : estimation operator l}\ ] ] define @xmath403 for @xmath404 . the commutator of @xmath405 and @xmath131 is @xmath406&=-\alpha \left({\nabla}_p { \partial}_{p_i}{u}\right)\cdot \left(\gamma { \nabla}_p-{\nabla}_q\right ) + \left({\nabla}_q { \partial}_{q_i}v\right)\cdot { \nabla}_p \\ & = -\alpha \left({\nabla}_p { \partial}_{p_i}{u}\right)\cdot l_\alpha -\alpha \left(\gamma-\alpha\right)\left({\nabla}_p { \partial}_{p_i}{u}\right ) \cdot { \nabla}_p + \left({\nabla}_q { \partial}_{q_i}v\right)\cdot{\nabla}_p \ . \label{eq : [ l_i , lop ] } \end{aligned}\ ] ] introducing @xmath407 , a simple computation shows that @xmath408u(t ) = \\ & \quad= -2\alpha \sum_{i=1}^d \sum_{j=1}^{d } l_iu(t)\left({\partial}_{p_j } { \partial}_{p_i}u\right ) l_{\alpha , j } u(t ) \label{eq : computation [ l , l ] new } \\ & \qquad + 2\alpha(\alpha-\gamma ) \sum_{i=1}^d l_{\alpha , i } u(t ) \left({\nabla}{\partial}_{p_i } u\right)\cdot { \nabla}_p u(t ) + 2\sum_{i=1}^dl_{\alpha , i } u(t ) { \nabla}\left ( { \partial}_{q_i } v\right)\cdot { \nabla}_p u(t ) \\ & \quad{\leqslant}\left ( { \varepsilon}_1+{\varepsilon}_2 - 2\nu\alpha\right ) \left|l_\alpha u(t)\right|^2 - 2\alpha \sum_{i=1}^d\sum_{j=1}^{d } l_{\alpha , i}u(t ) { \partial}_{p_j } { \partial}_{p_i}(u - u_{\nu } ) l_{\alpha , j } u(t ) \\ & \qquad + \frac{\alpha^2 ( \gamma-\alpha)^2}{{\varepsilon}_1}\left ( \sup_{\left|j\right|=2}\left\|{\partial}^j u\right\|_{l^{\infty}}^2 \right ) \left|{\nabla}_p u(t)\right|^2 + \frac{c_v^2}{{\varepsilon}_2}\left|{\nabla}_p u(t)\right|^2 \end{aligned}\ ] ] for any @xmath409 . with this preliminary computation , we can now choose @xmath410 in lemma [ lemma a.6 ] and sum over @xmath411 : for @xmath391 with @xmath369 sufficiently large , @xmath412 l_\alpha u(t ) \ , \pi_s \ , dq \ , dp \right)dt \label{eq : ipp 3}\\ & \quad + \frac{\alpha^2 ( \gamma-\alpha)^2}{{\varepsilon}_1 } \left(\sup_{\left|j\right|=2}\left\|{\partial}^ju\right\|_{l^{\infty}}^2\right)\int_{0}^t\exp(\zeta t ) \left ( \int_{{\mathcal{e } } } \left|{\nabla}_pu(t)\right|^2 \pi_s \ , dq \ , dp \right ) dt \nonumber \\ & \quad + \frac{c_v^2}{{\varepsilon}_2}\int_{0}^t\exp(\zeta t ) \left ( \int_{{\mathcal{e}}}\left|{\nabla}_pu(t)\right|^2 \pi_s \ , dq \ , dp \right ) dt \nonumber \ , .\end{aligned}\ ] ] since @xmath413 for some integer @xmath414 ( upon increasing @xmath369 ) , and in view of , the first and the two last terms of the right hand side of the above inequality can be controlled uniformly in time for @xmath415 . it remains to take care of the terms and . our strategy is to prove that they are negative when @xmath416 , and can hence be transfered to the lef - hand side of the inequality . to simplify the notation , we denote @xmath417 . recall that , by assumption [ condition dnu - dnunu ] , it holds @xmath418 . an integration by parts shows that @xmath419 with this expression we now estimate the term ( [ eq : ipp 3 ] ) by @xmath420 where we have used young s inequality and introduced a constant @xmath421 such that @xmath422 . the following conditions are therefore sufficient to ensure that and are non - positive when @xmath416 : there exists @xmath396 such that @xmath423 and @xmath424 these conditions can be restated as @xmath425 since @xmath426 can be chosen arbitrarily small ( while still being positive ) , the latter condition holds provided @xmath427 : @xmath428 after optimization with respect to @xmath429 , this leads to the final condition @xmath430 in conclusion , defining @xmath431 we see that the estimate holds when the constant @xmath432 from assumption [ hyp thm ] satisfies @xmath364 . the remainder of the proof of lemma [ proposition kopec 1 ] is very similar to the corresponding proof in @xcite . we first combine and lemma [ lemma estimation operator a1 ] : there exists @xmath390 such that for @xmath391 there exists an integer @xmath246 , a sufficiently small @xmath401 and @xmath433 such that if @xmath434 , then there is a constant @xmath365 for which @xmath435 we can now again apply lemma [ lemma a.6 ] , and sum the estimates obtained with @xmath436 . before stating the result , we bound the integrand of the term involving the commutator @xmath437 $ ] ( for @xmath438 ) as : @xmath439u(t){\partial}_{p_i}u(t ) \right| & = \left| \sum_{i=1}^d \nabla_p({\partial}_{p_i}u ) \cdot ( \nabla_q -\gamma\nabla_p)u(t ) \ , { \partial}_{p_i}u(t ) \right| \\ & { \leqslant}d \left ( \sup_{\left|j\right|=2}\left\|{\partial}^j u\right\|_{l^{\infty}}^2 \right ) \left|(\nabla_q -\gamma\nabla_p)u(t)\right| \ , |\nabla_p u(t)| \\ & { \leqslant}d \left ( \sup_{\left|j\right|=2}\left\|{\partial}^j u\right\|_{l^{\infty}}^2 \right ) \left[\frac12 |\nabla_q u(t)|^2 + \left(\gamma+\frac12\right ) |\nabla_p u(t)|^2\right ] . \end{aligned}\ ] ] then , for @xmath391 ( with @xmath369 sufficiently large ) and for all @xmath440 , @xmath441 in view of and and since @xmath442 for some integer @xmath443 , we see that there exists @xmath444 sufficiently large such that , for any @xmath391 and @xmath445 sufficiently small , and provided @xmath364 , there is a constant @xmath365 and an integer @xmath246 for which @xmath446 to conclude to lemma [ proposition kopec 1 ] for @xmath323 , it remains to apply lemma [ lemma a.6 ] with @xmath447 in order to obtain an estimate similar to the one above , but for @xmath448 . this is possible in view of the following bounds on the commutator : for all @xmath438 , @xmath449u(t){\partial}_{q_i}u(t ) \big| = \left| { \nabla}\left({\partial}_{q_i}v\right)\cdot { \nabla}_p u(t){\partial}_{q_i}u(t ) \right| { \leqslant}\left|{\nabla}_{p}u(t)\right|^2+c_v^2\left|{\partial}_{q_i}u(t)\right|^2.\ ] ] the remainder of the proof is done by induction of @xmath99 and relies on the control of the commutators @xmath450 $ ] with @xmath451 , which are independent of @xmath452 , as well as @xmath453\psi\right|{\leqslant}\sum_{\substack{i\in \mathbb{n}^{2d}\\ \left|i\right|{\leqslant}n}}\mathcal{p}_i\left|{\partial}^i\psi\right| , \ ] ] where @xmath454 are positive polynomial functions that depend on the polynomial growth of @xmath452 and its derivatives . these polynomial functions can be controlled with lyapunov weights for sufficiently large indices . in addition , the same approach as in the proof of lemma [ lemma estimation operator a1 ] is used to estimate the extra term arising from missing positivity of @xmath63 , namely @xmath455l_{\alpha , j}\left({\partial}^n_qu(t)\right)\pi_s \ , dq \ , dp \right ) dt \\ & \qquad{\leqslant}2(1+\mathscr{g}_s)\alpha{\varepsilon}_3\left\| { \nabla}\widetilde{u}\right\|_{l^{\infty}}\int_{0}^t\mathrm{exp}(\zeta t)\left(\int_{\mathcal{e}}\left|l_\alpha\left({\partial}^n_qu(t)\right)\right|^2\pi_s \ , dq \ , dp \right ) dt \\ & \qquad \ \ \ + \frac{2\alpha}{{\varepsilon}_3}\left\| { \nabla}\widetilde{u}\right\|_{l^{\infty}}\int_{0}^t\mathrm{exp}(\zeta t ) \left(\int_{\mathcal{e}}\left|{\nabla}_p l_\alpha\left({\partial}^n_qu(t)\right)\right|^2\pi_s \ , dq \ , dp \right ) dt . \end{aligned}\ ] ] therefore , the result is obtained when the same condition on @xmath432 is satisfied . note however that this condition depends on @xmath125 , hence on @xmath99 since @xmath125 has to be larger than some index @xmath456 . we define the ar perturbation function as @xmath457 this allows to write the generator @xmath458 of the ar - langevin dynamics as a perturbation of the generator @xmath459 : @xmath460 for notational convenience we omit the subscript @xmath55 and simply write @xmath461 . we also denote by @xmath462 the invariant measure associated with @xmath463 , and by @xmath464 the projection @xmath465 for a given observable @xmath118 , the asymptotic variance associated with the corresponding time averages reads , in view of : @xmath466 where @xmath467 is the unique solution in @xmath129 ( @xmath125 being such that @xmath468 ) of the following poisson equation : @xmath469 similarly , the limiting variance for @xmath226 can be rewritten as @xmath470 in order to prove the convergence of to and to identify the linear term in @xmath54 , the idea is to expand @xmath462 and @xmath471 in powers of @xmath54 . to this end , we rewrite the poisson equation as @xmath472 the operator @xmath473 is not bounded ( since @xmath474 contains derivatives in @xmath394 , which can not be controlled by @xmath475 ) , so that it is not possible to write the inverse of @xmath476 as some neumann series . it is however possible to consider a pseudo - inverse operator by truncating the neumann series at order @xmath99 . this motivates the introduction of the following approximation of the solution of : @xmath477 the corresponding approximation of the variance reads @xmath478 the connection with the exact variance is given by the following lemma , which is proved in section [ proof of theorem truncated variance ] . we introduce a critical value @xmath479 such that assumption [ hyp thm ] is satisfied for @xmath480 and @xmath481 ( see section [ proof of lemma arps condition ] ) . this allows to resort to lemma [ main theorem kopec ] . proposition [ proposition variance perturbation ] now straightforwardly follows by combining lemma [ theorem truncated variance ] and the following expansion in powers of @xmath54 of the truncated variance ( whose proof can be read in section [ proof of variance perturbation main theorem ] ) . recall that the function @xmath155 ( with @xmath155 defined in ) belongs to @xmath493)$ ] . the next result shows that the same is true for @xmath494 with @xmath495 defined in . this is not obvious a priori since @xmath495 is only piecewise @xmath39 , with singularities on the first order derivative at @xmath54 and @xmath55 . in fact , it can even be proved that @xmath496 and all its derivatives are small when @xmath54 is small . for any @xmath45 , the function @xmath154 belongs to @xmath497)$ ] . moreover , its derivatives have a compact support in @xmath498 $ ] . finally , for any @xmath499 and @xmath500 , there exists a constant @xmath501 such that @xmath502 , \qquad \left\|f_{{k_{\mathrm{min}}},{k_{\mathrm{max}}}}^{(n ) } - f_{0,{k_{\mathrm{max}}}}^{(n)}\right\|_{l^{\infty}}{\leqslant}c_{n_0,{k_{\mathrm{max}}},\delta}{k_{\mathrm{min}}}. \label{eq : corollary ftheta minus f 1}\ ] ] [ lemma cinfty in p f ] the function @xmath495 is defined piecewise on three intervals @xmath503 , @xmath504 and @xmath505 . in the interior of each interval , both @xmath155 and @xmath495 are @xmath39 , and so is therefore their composition . in addition , @xmath154 is constant on @xmath505 , hence all derivatives vanish on this interval . to prove that @xmath154 is @xmath39 with derivatives of compact support , it therefore suffices to prove that all derivatives can be extended by continuity at the points @xmath54 and @xmath55 . since @xmath155 is constant outside the interval @xmath506 $ ] , a simple computation shows that , for @xmath507 , @xmath508 it is therefore obvious to check the continuity at @xmath54 and @xmath55 since all derivatives of @xmath155 vanish at 0 and @xmath55 , and @xmath509 while @xmath510 . moreover , it is easy to check that @xmath511 , so that the estimate already follows in the case @xmath382 since @xmath155 is lipschitz continuous . to obtain the same result for higher order derivatives , we note that the @xmath99-th order derivative can be rewritten as @xmath512{\mathds{1}}_{\left[{k_{\mathrm{min}}},{k_{\mathrm{max}}}\right]}\right)\ , . \ ] ] therefore , @xmath513 is the sum of ( i ) @xmath514 , which is of order @xmath54 in @xmath82 norm by the same argument as before since @xmath515 is lipschitz continuous ; and ( ii ) a remainder term of order @xmath54 since @xmath516 is uniformly bounded ; while for any @xmath517 there exists @xmath518 such that @xmath519 , \qquad \left|\left(\frac{{k_{\mathrm{max}}}}{{k_{\mathrm{max}}}-{k_{\mathrm{min}}}}\right)^n-1\right|{\leqslant}r_{n,\delta } { k_{\mathrm{min}}}\,.\ ] ] this allows to obtain the desired result . in view of the definition - of @xmath520 , we can deduce the following estimates on @xmath521 and its derivatives , which allow in particular to control @xmath522 . to state the result , we introduce @xmath523 [ lemma ueps minus uzero intro ] for any @xmath45 , the function @xmath161 belongs to @xmath524 . for any @xmath525 and @xmath526 , the function @xmath527 has a compact support in @xmath528 . moreover , for any @xmath529 and @xmath500 , there exists a constant @xmath530 such that @xmath531 , \quad \left\|{\partial}^{\alpha } u_{{k_{\mathrm{min } } } , { k_{\mathrm{max}}}}-{\partial}^{\alpha } u_{0,{k_{\mathrm{max}}}}\right\|_{l^{\infty}}{\leqslant}c_{n_0,\delta,{k_{\mathrm{max } } } } { k_{\mathrm{min}}}\ , . \label{eq : ueps minus uzero intro}\ ] ] [ corollary ukmin cinfty ] in order to obtain more precise statements about the behavior of the functions @xmath532 for small values of @xmath54 , a natural idea would be to perform taylor expansions with respect to this parameter . the difficulty is however that the derivatives with respect to @xmath54 of the shift function @xmath533 are not continuous in @xmath534 . this prevents to write directly remainders of order @xmath535 . before stating the precise result in lemma [ lemma dkimkim conv ] , we need another technical ingredient . [ lemma kminkmax tau ] fix @xmath536 and define @xmath537 . then , for any @xmath525 and @xmath500 , there exists @xmath538 such that @xmath539 , \qquad \left\|\frac{f_{0,{k_{\mathrm{max}}}}^{(n ) } \circ \theta_{{k_{\mathrm{min } } } } - f_{0,{k_{\mathrm{max}}}}^{(n ) } } { { k_{\mathrm{min}}}}-f_{0,{k_{\mathrm{max}}}}^{(n+1 ) } \circ \tilde{\tau}\right\|_{l^{\infty}}{\leqslant}c_{n,\delta } \ , { k_{\mathrm{min}}}\,.\ ] ] note that , formally , @xmath540 is the derivative of @xmath495 on @xmath541 $ ] with respect to @xmath54 , evaluated at @xmath226 . recall also @xmath542 . simple computations show that there exists @xmath543 such that @xmath544 since @xmath545 , there exists @xmath546 $ ] such that @xmath547 therefore , for @xmath548 $ ] , @xmath549\right ) } + c_\delta \left\| f^{(n+1)}_{0,{k_{\mathrm{max}}}}\right\|_{l^{\infty}\left(\left[0,{k_{\mathrm{max}}}\right]\right ) } \right){k_{\mathrm{min } } } , \end{aligned}\ ] ] where we have used the following equality : there exists @xmath550 $ ] such that @xmath551 together with the bound @xmath552 . . there exist functions @xmath554 ( for @xmath555 ) , with compact support in @xmath528 , such that , for @xmath556 and @xmath103 , there is @xmath557 such that @xmath558,\quad\left\|d_{{k_{\mathrm{min}}},i}-{k_{\mathrm{min}}}\mathscr{d}_i\right\|_{w^{r,\infty}}{\leqslant}c_{r,\delta } { k_{\mathrm{min}}}^{2}. \label{eq : dkimkim esti kmin2}\ ] ] [ lemma dkimkim conv ] recall that the functions @xmath559 are defined , for @xmath560 , as @xmath561 \frac{p_i}{m_i}\\ & \qquad + \frac{\left|p_i\right|^2}{2m_i } \left[f^{\prime}_{{k_{\mathrm{min}}},{k_{\mathrm{max}}}}\left(\frac{\left|p_i\right|^2}{2m_i}\right)-f_{0,{k_{\mathrm{max}}}}^{\prime}\left(\frac{\left|p_i\right|^2}{2m_i}\right)\right]\frac{p_i}{m_i}\ , . \end{aligned}\ ] ] we next define , for @xmath555 , the function @xmath562\tilde{\tau}\left(\frac{\left|p_i\right|^2}{2m_i}\right)\frac{p_i}{m_i } , & \qquad \mathrm{for } \ \frac{\left|p_i\right|^2}{2m_i}\in \left[0 , { k_{\mathrm{max}}}\right ] , \\ 0 , & \qquad \mathrm{for } \ \frac{\left|p_i\right|^2}{2m_i } { \geqslant}{k_{\mathrm{max } } } , \end{aligned } \right.\ ] ] where @xmath540 is defined in lemma [ lemma kminkmax tau ] . recall that @xmath563 and @xmath515 have compact support on @xmath498 $ ] for @xmath119 . therefore , @xmath564 also has compact support in @xmath528 . the case @xmath565 of follows directly from lemma [ lemma kminkmax tau ] with @xmath382 and @xmath323 . let us now consider the case @xmath566 more carefully . to simplify the presentation , we consider separately the two terms in the sums defining the functions @xmath567 and @xmath568 , _ i.e. _ @xmath569 and @xmath570 with @xmath571 \frac{p_i}{m_i},\ ] ] and @xmath572 , \\ 0 , & \qquad \mathrm{for } \ \frac{\left|p_i\right|^2}{2m_i } { \geqslant}{k_{\mathrm{max}}}. \end{aligned } \right.\ ] ] we present the estimates only for the difference @xmath573 since similar computations allows to control the difference @xmath574 . for @xmath575 , we denote by @xmath576 the @xmath577th component of the momentum of the @xmath578th particle and by @xmath579 and @xmath580 the @xmath581th components of @xmath582 and @xmath583 . then , for @xmath584 , @xmath585\frac{{\displaystyle}\theta_{{k_{\mathrm{min}}}}^{\prime}\left(\frac{\left|p_i\right|^2}{2m_i}\right)-1}{{k_{\mathrm{min}}}}\right| \\ & \quad + \frac{|p_{i,\alpha}p_{i,\alpha'}|}{m_i^2}\left|f_{0,{k_{\mathrm{max}}}}^{\prime}\left(\frac{\left|p_i\right|^2}{2m_i}\right)\left[\frac{{\displaystyle}\theta_{{k_{\mathrm{min}}}}^{\prime}\left(\frac{\left|p_i\right|^2}{2m_i}\right)-1}{{k_{\mathrm{min}}}}-\frac{1}{{k_{\mathrm{max}}}}\right ] \right| , \end{aligned}\ ] ] where we used @xmath586 for @xmath587 $ ] . the first two terms in the last inequality can be bounded by @xmath588 for some constant @xmath589 in view of lemma [ lemma kminkmax tau ] . for the last two terms , distinguish the cases @xmath590 and @xmath591 , where for @xmath592 we define @xmath593 when @xmath590 , the third term disappears since @xmath594 on @xmath595 $ ] . in addition , @xmath596 so that @xmath597\right)}\right){k_{\mathrm{min}}}\ , . \end{aligned}\ ] ] when @xmath598 , we use @xmath599 for @xmath600 $ ] , so that there exists @xmath601 such that @xmath602}\left |\frac{\theta_{{k_{\mathrm{min}}}}^{\prime}(x)-1}{{k_{\mathrm{min } } } } - \frac{1}{{k_{\mathrm{max } } } } \right|{\leqslant}c_{\delta } { k_{\mathrm{min } } } , \quad \sup_{x\in \left[{k_{\mathrm{min}}},{k_{\mathrm{max}}}\right]}\left|\theta_{{k_{\mathrm{min}}}}^{\prime}(x)-1\right|{\leqslant}\frac{1}{{k_{\mathrm{max}}}-{k_{\mathrm{min}}}}{\leqslant}\frac{1}{\delta}\ , . \label{du theta minus x minus tau kminkmax}\ ] ] using these bounds as well as the inequality @xmath542 and for @xmath323 , it follows @xmath603 this concludes the proof of for @xmath566 . [ lemma admukim new ] fix @xmath536 . then for any @xmath608 , there exist @xmath609 such that , for @xmath610 , @xmath611 , \qquad \int_{\mathbb{r}^d } f\left(p\right)\mathrm{e}^{-\beta u_{{k_{\mathrm{min}}}}(p)}dp=\int_{\mathbb{r}^d } f\left(p\right)\mathrm{e}^{-\beta u_{0}(p)}dp+ a_f { k_{\mathrm{min}}}+{\mathrm}{o}\left({k_{\mathrm{min}}}^2\right)\ , . \label{eq : admukim new}\ ] ] recall that @xmath612 . note that @xmath613\,.\ ] ] manipulations similar to the ones used to prove allow to show that there exists a function @xmath614 with compact support in @xmath528 such that , for @xmath610 and @xmath356 , there is @xmath557 for which @xmath558 , \quad \left\|u_{{k_{\mathrm{min}}}}-u_0-{k_{\mathrm{min}}}\mathcal{u}\right\|_{w^{r,\infty } } { \leqslant}c_{r,\delta } { k_{\mathrm{min}}}^2\ , . \label{eq : ukmin minus u w r infty}\ ] ] this allows to write @xmath615 with @xmath616 uniformly bounded in @xmath617 . moreover since @xmath618 also has a compact support , we easily obtain @xmath619 with @xmath620 uniformly bounded in @xmath617 . therefore , there exists a constant @xmath621 such that @xmath622 so that follows with @xmath623 . in order to use lemma [ main theorem kopec ] , we need to check that assumption [ hyp thm ] holds with @xmath432 as small as wanted for appropriate values of @xmath605 . the first condition is easy to check , so we concentrate on the last two conditions . the reference kinetic energy function @xmath624 in lemma [ main theorem kopec ] is chosen as the standard kinetic energy @xmath625 , so that @xmath626 . it therefore remains to check the last condition . an inspection of the proof of lemma [ main theorem kopec ] reveals that it holds provided @xmath605 are such that holds . straightforward computations show that @xmath627 - \frac{p_i |p_i|^2}{m^2_i } f'_{{k_{\mathrm{min}}},{k_{\mathrm{max}}}}\left(\frac{p_i^2}{m_i}\right),\ ] ] so that , using the fact that @xmath628 has compact support in @xmath528 ( hence @xmath629 ) and in view of the expression of @xmath630 , the following bound holds : @xmath631 similarly , there exists a constant @xmath365 ( depending on @xmath632 and @xmath633 ) such that @xmath634 . \ ] ] it is then easy to see that holds upon choosing @xmath635 with @xmath536 sufficiently small . denote by @xmath367 the operator @xmath636 . by corollary [ lemma ueps minus uzero intro ] , for any @xmath525 and @xmath491 , there exists a constant @xmath637 such that @xmath638 by the resolvent estimate , there exists for any @xmath639 a constant @xmath640 such that @xmath641 therefore , choosing an integer @xmath125 for which @xmath642 , there exists a constant @xmath126 such that @xmath643 by the same principle , using the fact that , by , there is for any @xmath644 a constant @xmath645 such that @xmath646 and in view of , there exists , for any @xmath647 , integers @xmath648 and @xmath649 such that , for all @xmath650 , there is a constant @xmath126 and an integer @xmath356 for which @xmath651 by recurrence , there exist , for any @xmath507 , integers @xmath652 such that , for all @xmath361 , there is @xmath653 and @xmath654 for which @xmath655 this gives the claimed result . we start by writing the difference between the variance and the truncated one : @xmath656 a simple computation gives @xmath657 we first use lemma [ corollary : estimate lopd n ] : there exists @xmath658 such that , for @xmath361 , there is @xmath659 and @xmath126 such that @xmath660 therefore , using and , there exists some constant @xmath637 such that @xmath661 we finally apply @xmath662 to both sides of : in view of , it follows @xmath663 the result is then a direct consequence of the equality . we first expand @xmath665 in terms of @xmath666 as @xmath667 estimates on @xmath666 and its derivatives in terms of @xmath188 can be obtained with . lemma [ corollary : estimate lopd n ] then allows to estimate the higher order terms in the above equality : there exists @xmath668 and @xmath365 such that @xmath669 by combining these estimates with , we obtain @xmath670 where @xmath671 is uniformly bounded in @xmath672 due to for @xmath54 small enough ( upon possibly increasing @xmath125 ) . with the notation of lemma [ lemma admukim new ] , for any @xmath608 , @xmath673 with @xmath674 uniformly bounded for @xmath54 small enough . finally , by combining and , we see that there exists @xmath490 such that @xmath675 stephane redon and zofia trstanova gratefully acknowledge funding from the european research council through the erc starting grant n. 307629 . this work was funded by the agence nationale de la recherche , under grant anr-14-ce23 - 0012 ( cosmos ) . gabriel stoltz benefited from the scientific environment of the laboratoire international associ between the centre national de la recherche scientifique and the university of illinois at urbana - champaign . , , , , , , , , adaptive torsion - angle quasi - statics : a general simulation method with applications to protein structure analysis and design . bioinformatics * 23*(13 ) ( 2007 ) .
it is by now well - known that quantum field theory permits violations of all of the local energy conditions used in classical general relativity . these conditions are employed in a variety of ways , such as in the proofs of theorems on the occurrence of singularities in gravitational collapse and cosmology , or of cosmic censorship . two of the weakest such local conditions are the `` weak energy condition '' and the `` null energy condition '' which state that : @xmath0 for all timelike vectors @xmath1 , and @xmath2 for all null vectors @xmath3 , respectively @xcite . note that the null energy condition follows by continuity if the weak energy condition holds . these two local conditions are satisfied by known forms of classical matter , but there are a variety of states of quantum fields which violate them @xcite , the most well - known of which is arguably the casimir vacuum @xcite . the extent to which quantum field theory allows violations of the local energy conditions is not yet completely clear , although progress has been made in recent years to answer this question . two principal approaches for determining the degree of violation have been `` averaged energy conditions '' and `` quantum inequalities . '' the first , originally due to tipler @xcite , involves a suitable averaging of the local conditions over timelike or null geodesics . it can be shown that many of the standard results of classical general relativity obtained via global techniques can be proved using only the averaged , rather than the local , energy conditions @xcite-@xcite . here we take the `` averaged weak energy condition '' ( awec ) to be @xmath4 where the integral is taken over a timelike geodesic with tangent vector @xmath1 and parameterized by the proper time @xmath5 . similarly , we take the `` averaged null energy condition '' ( anec ) to be @xmath6 where the integral is taken over an affinely parameterized null geodesic with tangent vector @xmath3 and affine parameter @xmath7 @xcite . there has been a great deal of effort in the last several years to determine whether quantum field theory enforces averaged energy conditions . most of this activity has been concentrated primarily on anec @xcite-@xcite . this is in part due to the discovery that violations of anec are required to maintain traversable wormholes @xcite . it appears that although anec holds for a wide class of quantum states in a variety of spacetimes , it does not hold in an arbitrary four - dimensional curved spacetime ( see refs . @xcite , @xcite , and @xcite for more detailed discussions ) . however , it is quite plausible that a suitable generalization of anec may hold in more general spacetimes @xcite . the second approach involves uncertainty principle - type inequalities , derived from quantum field theory , which restrict the magnitude and duration of negative energy fluxes or densities @xcite . for example , one such `` quantum inequality '' ( qi ) for negative energy fluxes seen by inertial observers in two - dimensional flat spacetime has the form : @xmath8 where @xmath9 is the magnitude of the negative energy flux and @xmath10 is its duration . this inequality implies that @xmath11 , the amount of negative energy transmitted in time @xmath10 , is limited by @xmath12 therefore , @xmath11 is less than the quantum uncertainty in the energy , @xmath13 , on the timescale @xmath10 . recently , more precise versions of these inequalities have been derived @xcite . these involve an integral of the energy flux multiplied by a `` sampling function '' , i.e. , a peaked function of time with a time integral of unity and characteristic width @xmath14 . a convenient choice @xcite of such a function is @xmath15 $ ] . if the integrated flux , @xmath16 , is defined by @xmath17 then these inequalities may be written as @xmath18 and @xmath19 for all @xmath14 , in two- and four - dimensions , respectively . these inequalities have the form required to prevent macroscopic violations of the second law of thermodynamics @xcite . it was also discovered that similar inequalities hold for a quantized massless , minimally - coupled scalar field propagating on two- and four - dimensional extreme reissner - nordstrm black hole backgrounds . these inequalities were shown to foil attempts to produce an unambiguous violation of cosmic censorship by injecting a negative energy flux into an extreme charged black hole @xcite . the latter results provide perhaps one of the strongest reasons for the belief that the production of large - scale effects via manipulation of negative energy may be forbidden by quantum field theory . classically , _ any _ amount of negative energy , no matter how small , injected into an extreme black hole is enough to produce a naked singularity . therefore , if one had any chance of producing gross effects with negative energy , the scenario discussed above would seem to have offered the best possibility of success . it should also be emphasized that the energy - time uncertainty principle was _ not _ used as input in the derivation of any of the qi restrictions . they arise directly from quantum field theory . in a recent paper , we have shown that there exist deep connections between averaged energy conditions and quantum inequalities , at least in flat spacetime @xcite . for a quantized massless , minimally - coupled scalar field in two- and four - dimensional minkowski spacetime , we have derived analogous qi s to eqs . ( [ eq : fqi/2d ] ) and ( [ eq : fqi/4d ] ) for energy density . in particular , for timelike geodesics in two - dimensional flat spacetime , @xmath20 for all @xmath14 . here @xmath21 is the renormalized expectation value taken in an arbitrary quantum state @xmath22 . in the limit as @xmath23 , we `` sample '' the entire geodesic , and our inequality eq . ( [ eq : tlqi2duncp ] ) reduces to awec . note that it is possible that the integral in eq . ( [ eq : tlqi2duncp ] ) converges , even though the associated awec integral in eq . ( [ eq : awec ] ) diverges . analogous results hold in this case for null geodesics . for a quantized massless scalar field in a two - dimensional minkowski spacetime with compactified spatial dimension , it was also discovered that the _ difference _ between the expectation values of @xmath24 , or @xmath25 , in an arbitrary quantum state and in the casimir vacuum state obey awec and anec - type inequalities @xcite . this is surprising since it is known @xcite that awec and anec are violated for @xmath26 and @xmath27 , respectively , in the ( renormalized ) casimir vacuum state by itself . such `` difference inequalities '' might provide new measures of the degree of energy condition violation in cases where the usual averaged energy conditions fail . this approach has recently been generalized to arbitrary two - dimensional curved spacetimes by yurtsever @xcite . for a quantized massless , minimally - coupled scalar field in four - dimensional minkowski spacetime , we also derived the following inequality for timelike geodesics @xmath28 for all @xmath14 . from this inequality , it was shown that one can derive awec and anec @xcite . ( the application of qi bounds to constrain the dimensions of traversable wormholes is discussed in a separate manuscript@xcite . ) in almost all studies of anec for quantum fields , the bounds on the anec integral have been taken from @xmath29 to @xmath30 @xcite . however , a question of some interest is whether anec is satisfied in the spacetime of an object collapsing to form a black hole . the answer to this question might determine , for example , whether penrose s singularity theorem @xcite will still hold in the presence of local violations of the energy conditions , such as the hawking evaporation process @xcite . in this case , one is usually concerned with the focusing of null geodesics which generate the boundary of the future of a trapped surface . to prove penrose s theorem , one would want anec to hold over these half - complete geodesics @xcite . this is a stronger condition to impose than to demand that anec hold over full ( i.e. , past and future - complete ) geodesics . however , even if this version of anec fails in some regions of a given spacetime , it may hold in enough other regions so that the conclusions of penrose s theorem will still be valid . in addition , it should be noted that anec is a sufficient condition to insure continued focusing of null geodesics . one can also guarantee focusing with other conditions , for example by allowing the right - hand - side of the anec integral to be only periodically non - negative @xcite , or even negative but bounded ( see lemma 3 of ref . @xcite , eq . ( 5 ) of ref . @xcite , and ref . @xcite ) . in the current paper , we examine quantized massless scalar , and electromagnetic fields in the unruh vacuum state in two- and four - dimensional schwarzschild black hole backgrounds . we evaluate awec and anec integrals along half - complete timelike and null geodesics , respectively . our goal is to determine the regions of these spacetimes in which the averaged energy conditions fail , and whether there exist any bounds on the extent of the violation . the paper is organized as follows . the analysis of 2d evaporating black holes is presented in sec . ( [ sec:2devap ] ) ; the results for 4d black holes are given in sec . ( [ sec:4dbh ] ) . the latter section also includes a discussion of anec along orbiting null geodesics . concluding remarks are given in sec . ( [ sec : conc ] ) . our units are taken to be : @xmath31 . in this section , we study awec and anec for a quantized , massless scalar field in the unruh and boulware vacuum states , along geodesic segments in a 2d schwarzschild spacetime . the discussion here is a natural extension of the analysis in ref . @xcite . let @xmath32 be the two - velocity of an inertial observer . the energy density in this observer s frame is given by @xmath33 in the following discussion , we understand @xmath34 to denote a quantum expectation value in a specified vacuum state . the metric is @xmath35 where @xmath36 . an outgoing geodesic observer s two - velocity is @xmath37 the constant @xmath38 is the energy per unit rest mass . in our two - dimensional discussion , we will consider observers moving in both the unruh and boulware vacua . the stress tensor components in the unruh vacuum are @xcite : @xmath39 @xmath40 and @xmath41 the corresponding components in the boulware vacuum are given by : @xmath42 @xmath43 and @xmath44 for an outgoing timelike observer in the unruh vacuum , @xmath45\ , + \,{{cm^2 } \over { r^4 } } \,- \ , { c\over { 32 m^2}}\ , - \ , { { k \sqrt{k^2-c } } \over { 16 m^2 } } \biggr\}. \label{eq : etl / out}\ ] ] let us consider , for simplicity , the case of an observer who starts out initially very close to the horizon and is shot outward at large velocity , i.e. , we are interested in the limits @xmath46 where @xmath47 is the observer s initial position . in this limit @xmath48 , so from the geodesic equation , eq . ( [ eq : umu ] ) , we have that @xmath49 we now wish to multiply @xmath50 by a sampling function , i.e. , a peaked function of time whose time integral is unity . in refs . @xcite and @xcite , this function was chosen to be @xmath51 where the integral was taken over complete geodesics . however , unlike in ref . @xcite , here we are integrating over half - infinite geodesics . since the function given by eq . ( [ eq : origsamp ] ) is symmetric about @xmath52 , in our case we may choose @xmath53 where the proper time @xmath5 is initialized when the observer starts at @xmath54 . we multiply @xmath50 by this sampling function , use eq . ( [ eq : tauexp ] ) , and expand the numerator in inverse powers of @xmath38 . the integral is performed using macsyma and the result is expanded in powers of @xmath55 , yielding @xmath56 \nonumber \\ & + & \frac{m}{3 \pi^2 \,{\tau_0}^3 \,k } \ , \ln \biggl ( \frac{k \tau_0}{2 m } \biggr ) \ , + o(k^{-1 } ) \ , . \label{eq : intit}\end{aligned}\ ] ] therefore , to leading order in @xmath38 and @xmath55 , we have @xmath57 where @xmath58 note that in these limits @xmath59 and @xmath60 @xcite . what is the physical significance of @xmath61 ? the negative energy density drops off very rapidly with increasing @xmath62 . in the limit @xmath63 and @xmath64 , if we consider the proper time spent by the observer in the region of appreciable negative energy to be @xmath65 then typically @xmath66 , so @xmath67 to see this a little more explicitly , note that near the horizon , eq . ( [ eq : etl / out ] ) becomes @xmath68 if the observer starts at @xmath69 , then at @xmath70 , @xmath71 so the energy density drops off to @xmath72 of its initial value in a distance @xmath73 , corresponding to a proper time @xmath74 . therefore , we should consider the time interval @xmath61 not to be the entire proper time over which the energy density is negative , but rather a time scale which characterizes a change in the energy density . for an outgoing timelike observer in the boulware vacuum : @xmath75\ , + \,{{cm^2 } \over { r^4 } } \biggr\}. \label{eq : betl / out}\ ] ] note that this quantity is negative everywhere for @xmath76 . if we perform the same set of operations as discussed above , we find that eq . ( [ eq : intt ] ) holds for the boulware vacuum as well . in the @xmath77 limit of eq . ( [ eq : intt ] ) , we `` sample '' the entire ( i.e. , half - infinite ) geodesic and find @xmath78 and hence @xmath79 the dominant contribution to the integral in eq . ( [ eq : bawec/2d ] ) is from the lower limit , since @xmath50 in the boulware vacuum state also has the form given in eq . ( [ eq : tasyp ] ) near the horizon , but drops off rapidly with increasing @xmath62 . by contrast , the @xmath77 limit of eq . ( [ eq : intt ] ) for the unruh vacuum state results in an integral which diverges positively at large @xmath62 . this is due to the fact that in 2d the hawking radiation does not drop off with distance , but remains constant at large @xmath62 . therefore , awec is satisfied in the unruh vacuum state in 2d . however , as discussed in sec . ( [ sec:4dbh ] ) , the situation is different in 4d . we could perform the integral on the left - hand - side of eq . ( [ eq : bawec/2d ] ) for the unruh vacuum state , but instead truncate the integration at some finite value of @xmath80 , with @xmath81 . for fixed @xmath82 , by making @xmath55 small enough we can always arrange it so that the negative energy contribution from the lower limit , given by eq . ( [ eq : tasyp ] ) , dominates the integral . in that case , using eq . ( [ eq : tasyp ] ) , we have that @xmath83 the inequalities , eqs . ( [ eq : awec / qi ] ) and ( [ eq : uawec / qi ] ) , represent bounds on the degree of awec violation seen by timelike geodesic observers who start out very close to the horizon and are shot outward at high speed . in this limit , the longer the ( proper ) timescale over which the observer sees a significant change in the energy density , the smaller is the magnitude of the integrated negative energy density seen by that observer . since the negative energy density drops off rapidly with increasing @xmath62 , to remain in the negative energy density the observer must stay close to the horizon . the closer the observer is to the horizon , the larger is the magnitude of the negative energy density . however , in order to remain close to the horizon for a long time as seen by a distant observer , the observer s trajectory must be nearly lightlike . therefore , although the observer spends a long time in the negative energy region as seen by the distant observer , the _ proper _ time spent in the region of appreciable negative energy ( as measured by @xmath61 ) decreases with the observer s proximity to the horizon . these inequalities are similar in form to eq . ( [ eq : eti/2d ] ) . we now wish to examine anec for null geodesics in 2d schwarzschild spacetime . from eq . ( [ eq:2dmetric ] ) , and from the equation for null geodesics , we have @xmath84 where @xmath85 is an arbitrary positive constant , whose value fixes the scale of the affine parameter , @xmath7 . we want to examine anec along both ingoing and outgoing future - directed ( i.e. , @xmath86 ) null geodesics , inside and outside the horizon . there are four cases ( see fig . 1 ) : @xmath87 note that @xmath88 changes sign inside the horizon , while @xmath89 remains positive for future - directed null rays . if we use eqs . ( [ eq : out / rg2 m ] ) - ( [ eq : in / rl2 m ] ) , and eqs . ( [ eq : uttt ] ) - ( [ eq : utrr ] ) , we obtain for the unruh vacuum state @xmath90 \ , { \rm ( outgoing)}\ , , \label{eq : enull / o}\ ] ] and @xmath91 { \rm ( ingoing ) } \ , . \label{eq : enull / i}\ ] ] to obtain eq . ( [ eq : enull / i ] ) , we have used the fact that the sign of @xmath92 changes inside the horizon , as can be seen from eq . ( [ eq : uttr ] ) , together with eqs . ( [ eq : in / rg2 m ] ) and ( [ eq : in / rl2 m ] ) . note that for eq . ( [ eq : enull / i ] ) , the term in square brackets vanishes as @xmath93 . a taylor expansion shows that , to leading order , it vanishes as @xmath94 , so it will cancel the @xmath94 divergence in the denominator . therefore eq . ( [ eq : enull / i ] ) is finite on the horizon , as it should be . however , there appears to be a discontinuity in eq . ( [ eq : enull / o ] ) for an outgoing null ray on the horizon , @xmath95 . the situation for null rays on the horizon is sufficiently subtle as to warrant a separate discussion . this is provided in the next subsection . from eq . ( [ eq : enull / o ] ) , @xmath96 is negative for outgoing null geodesics when@xmath97 . therefore the local null energy condition , eq . ( [ eq : nec ] ) is violated along outgoing null vectors slightly inside the horizon , as well as along all outgoing rays outside ( and on ) the horizon . for @xmath76 , consider an outgoing null geodesic starting at @xmath54 , with @xmath64 . examine anec along this ray to obtain @xmath98\,dr\ , , \label{eq : anec/2d out / rg2m}\ ] ] where we have used eq . ( [ eq : out / rg2 m ] ) . here @xmath99 is the value of the affine parameter at @xmath54 . equations ( [ eq : enull / o ] ) and ( [ eq : anec/2d out / rg2 m ] ) apply for the boulware vacuum as well as for the unruh vacuum . this is due to the fact that in the latter case , the outgoing null geodesics are parallel to the hawking radiation . hence the terms which would cause the integral in eq . ( [ eq : uawec / qi ] ) to diverge as @xmath100 are absent in the null geodesic case . for @xmath101 , along an outgoing null geodesic starting just inside the horizon at @xmath102 : @xmath103\,dr\ , , \label{eq : anec/2d out / rl2m}\ ] ] where we have used eq . ( [ eq : out / rl2 m ] ) , and @xmath104 is the value of the affine parameter at @xmath105 , the minimum value of @xmath62 attained by the null geodesics . notice that the integrand in eq . ( [ eq : anec/2d out / rl2 m ] ) will be dominated by the positive @xmath106 term which diverges for small @xmath62 . from eq . ( [ eq : enull / i ] ) , we see that @xmath96 is positive everywhere for ingoing null geodesics . thus we immediately see that anec is satisfied for these two sets of null geodesics . in eq . ( [ eq : anec/2d out / rg2 m ] ) we perform the integration and then expand the result in the limit of small @xmath55 to find @xmath107 note that @xmath108 is the characteristic affine parameter distance over which the negative energy is decreasing along the outgoing geodesic . we can rewrite eq . ( [ eq : lim out / rg2 m ] ) as @xmath109 this is the null version of eq . ( [ eq : uawec / qi ] ) . note that it is invariant under rescaling of the affine parameter . similarly , for eq . ( [ eq : anec/2d out / rl2 m ] ) , we may perform the integration and expand the result in the small @xmath110 limit . for fixed @xmath55 , the result is @xmath111 however , for fixed @xmath110 and @xmath112 , @xmath113 has the same behavior as @xmath114 : @xmath115 thus we see that anec is violated for outgoing null rays just outside the horizon , but it is satisfied for outgoing null rays inside the horizon in the limit @xmath116 . ( the divergence in eq . ( [ eq : lim out / rg2 m ] ) as @xmath112 may be circumvented by an appropriate choice of scaling of the affine parameter , i.e. , by an appropriate choice of @xmath85 . this and related issues will be discussed in the next sub - section . ) outgoing rays which originate in the region @xmath117 initially encounter negative energy , and later encounter positive energy near the singularity , @xmath118 . this is analogous to the case of outgoing timelike geodesics in the unruh vacuum , which encounter negative energy near the horizon and positive energy at large distances due to the hawking radiation . we also see that anec holds for ingoing null geodesics , whether they originate outside or inside the horizon . one might at first sight conclude that the reasoning here is a bit circular . it could be argued that it is no surprise that evaluation of the anec integral for null geodesics inside the horizon yields a divergent positive result . after all , we placed our quantized field on a background which already had a singularity in it . thus one could argue that divergent values of our anec integrals for these geodesics are not unexpected , because the behavior of the fields inside the horizon is dominated by the singularity . however , it is not obvious , a priori , what the _ sign _ of these integrals should be . since these are test fields on a given background , as opposed to fields which generate the background spacetime ( which we can not have in any case in 2d ) , it is possible that the integrals might have turned out to diverge with either sign . similar anec integrals for quantized massless fermion fields in 2d have the same signs as for the massless scalar field . this follows from the fact that the renormalized fermion stress tensor is identical to that in the scalar case @xcite , even though the formally divergent tensors have opposite signs in the two cases . recall that our results for outgoing null geodesics outside the horizon are formulated in terms of the null tangent vector @xmath119 where the components are given in terms of schwarzschild @xmath120 coordinates . since these coordinates are badly - behaved on the horizon , let us switch to kruskal null coordinates , given by @xmath121 where @xmath122 , @xmath123 , with @xmath124 the usual tortoise coordinate and @xmath125 . in these coordinates , our null vector has the form @xmath126 if @xmath85 has the same value for all outgoing null geodesics , then @xmath127 would not be defined in the limit where the geodesics approach the horizon , since @xmath128 . however , @xmath85 need only be constant along each null geodesic , but not the same constant for different null geodesics . consider a sequence of outgoing null geodesics which start at different values of @xmath62 , e.g. , the histories of a sequence of photons emitted by an infalling observer . let @xmath129 be the value of @xmath62 at which the null geodesic begins ( which will be different for each null geodesic ) . we might choose the various values of @xmath85 for different null geodesics by the following prescription . let the observer emit photons of fixed frequency @xmath130 in his rest frame . the time - component of the observer s two - velocity ( in schwarzschild coordinates ) is @xmath131 where @xmath132 is the observer s energy per unit rest mass , measured at infinity . this is just the rate of a clock at infinity as compared to the rate of the clock carried by the infalling observer . the frequency of a photon at infinity is @xmath133 , which is related to @xmath130 by @xmath134 note that the expression on the right - hand - side includes both the doppler effect and the gravitational redshift . the factor of @xmath135 reflects the fact that the faster the observer is shot toward the black hole ( i.e. , the larger @xmath132 is ) , the greater will be the doppler shift . if we choose a scaling for the affine parameter such that @xmath136 , for fixed @xmath130 , then @xmath137 now @xmath138 is finite in the limit @xmath139 , so that the components of @xmath3 are finite on the horizon , in either schwarzschild or kruskal coordinates . however , @xmath127 is still not the same as @xmath140 , the _ affinely _ parameterized null tangent vector on the horizon . in kruskal null coordinates the vector @xmath140 has components @xmath141 ( see the discussion on p. 331 of ref . @xcite . ) on the horizon , @xmath142 so these vectors still differ by a factor of @xmath143 . thus , a null vector such as @xmath3 , which is an affinely parameterized null tangent vector off the horizon , does not necessarily remain affinely parameterized _ on _ the horizon . with these lessons in hand , let us now consider the integral @xmath144 along a portion of the future event horizon . here @xmath145 is the vacuum expectation value of the stress - tensor in the unruh vacuum state expressed in kruskal null coordinates , and @xmath140 is an affinely parameterized null tangent vector on the horizon . the kruskal advanced time coordinate @xmath146 , is an affine parameter on the horizon ( see for example , p. 331 of ref . @xcite or p. 122 of ref . @xcite ) . from eq . ( [ eq : def - k ] ) , and a straightforward coordinate transformation of eqs . ( [ eq : uttt])- ( [ eq : utrr ] ) , we obtain @xmath147 this integral is taken over a portion of the future horizon ; @xmath148 is the value of the @xmath149 line which intersects the future horizon at the ( arbitrary ) event where we start our null geodesic . the right - hand - side of eq . ( [ eq : kint ] ) goes to @xmath29 as @xmath150 . however , this would correspond to integrating along the full future horizon of an eternal black hole spacetime . the unruh vacuum state is singular on the past horizon ( i.e. , @xmath151 ) of such a spacetime , so it is not surprising that we get a divergence in this case . physically , this would correspond to a black hole that hawking - radiates for an infinite time . for the unruh vacuum state to be realizable , we must consider only physically realistic collapse spacetimes in which the black hole forms at a finite time in the past . therefore , the smallest allowed value of @xmath152 , which is the value of the @xmath149 line at which the the worldline of the surface of the collapsing star intersects the future event horizon , as shown in fig . ( 2 ) . for such spacetimes , the anec integral along a portion of the future horizon is negative but finite . this result seems to be in agreement with the results of wald and yurtsever @xcite . they find that anec is satisfied along the complete null geodesic comprised of the future horizon plus the null line which would have been the past horizon in an eternal black hole spacetime . in a collapse spacetime , this complete null geodesic originates at past null infinity , propagates through the collapsing star , `` reflects off '' @xmath118 , and then emerges into the vacuum outside the star at @xmath152 ( see fig . ( 2 ) ) . evidently there is a positive contribution to the anec integral from the part of the geodesic prior to its exit from the collapsing star @xcite . this contribution must be greater than or equal to @xmath153 . ( note that the source of the positive contribution we refer to here is the quantized field , not the classical collapsing matter . the latter would presumably make the anec integral even more positive . ) the dimensions of @xmath140 are @xmath154 , whereas @xmath145 , @xmath146 , and @xmath148 are dimensionless . let us now rescale the coordinates to get a more familiar choice of dimensions . let @xmath155 , i.e. , @xmath156 , so @xmath157 . then @xmath158 , and @xmath159 . if we rewrite eq . ( [ eq : kint ] ) in terms of the rescaled coordinates , we obtain @xmath160 compare this result with our result for null rays outside the horizon , i.e. , eq . ( [ eq : anec / nint ] ) , in which @xmath161 is interpreted as the characteristic affine parameter distance along the null geodesic over which the negative energy density falls off . let us now show that a similar interpretation holds _ on _ the horizon . the integrand of eq . ( [ eq : kinthat ] ) is equal to @xmath162 , so @xmath163 note that @xmath164 can be interpreted as the characteristic affine parameter distance over which the integrand falls off , i.e. , as @xmath165 increases from @xmath166 to @xmath167 , @xmath168 falls off to @xmath72 of its initial value . thus the interpretation of the qi for null geodesics on the horizon is the same as that for eq . ( [ eq : anec / nint ] ) . in this section , we examine awec and anec in four - dimensional evaporating black hole spacetimes . in these spacetimes , the components of the renormalized expectation value of the stress - energy tensor are only known numerically . accordingly , we make use of the numerical data of elster @xcite for the conformally - coupled massless scalar field and that of jensen , mclaughlin , and ottewill @xcite for the electromagnetic field , evaluated in the unruh vacuum state . we examine the cases of outgoing radial timelike observers , outgoing radial null geodesics , and orbiting null geodesics . only geodesics outside the horizon are considered , as no numerical data are available for @xmath169 . we find that although awec and anec are violated in all three cases , there appear to be qi - type bounds in each case which constrain the degree of the violations , as in two - dimensional spacetime . these results might at first sight seem to be in disagreement with recent results of visser @xcite . he found that for the renormalized vacuum expectation value of the stress - energy tensor of a conformally - coupled massless scalar field in schwarzschild spacetime , anec is satisfied for all null geodesics that reach spatial infinity , and for orbiting geodesics at @xmath170 . however , his results are applicable for the hartle - hawking vacuum state , which represents a black hole in thermal equilibrium with a surrounding external thermal bath of radiation . it is therefore not too surprising to find that anec holds in that case . by contrast , our analysis is performed in the unruh vacuum state , which corresponds to the situation of a black hole evaporating into empty space . the four - velocity of an outgoing radial timelike observer in 4d schwarzschild spacetime is given by @xmath171 where again @xmath88 . let us adopt the notation of elster @xcite so that : @xmath172 , where the components of the stress - tensor here represent renormalized expectation values in the unruh vacuum state . therefore , we may write @xmath173 if a qi - type bound of the kind found in 2d , i.e. , eq . ( [ eq : awec / qi ] ) , also holds in 4d , then we might expect it @xcite to have the form @xmath174 where again @xmath61 is the characteristic proper time which the observer spends in the negative energy . ( note that since the fields we will be considering are massless , the mass @xmath175 of the black hole is only natural length scale which appears in the problem . ) we want to numerically compare the left and right - hand - sides of eq . ( [ eq : awec / qi-4d ] ) , in order to see if the inequality is satisfied . the integrand in eq . ( [ eq : awec / qi-4d ] ) drops off rapidly with increasing @xmath62 near the horizon ( unlike in the 2d case ) , hence the dominant contribution to the awec integral comes from the initial part of the trajectory . we will therefore approximate the awec integral as @xmath176 where @xmath177 is understood to be evaluated at the starting point @xmath129 ( minimum @xmath62 ) of the trajectory . since the magnitude of @xmath178 rapidly decreases near the horizon , our estimate of using @xmath179 gives a result which is somewhat more negative than the true result . therefore , correction of this estimate only goes in the direction of strengthening the inequality . our results for the electromagnetic field are displayed in table 1 . the values of @xmath180 and @xmath181 for the electromagnetic field are from the numerical data given in ref . @xcite , and the value of @xmath182 used in this case is the photon luminosity given by page @xcite . in the fourth column of table 1 , @xmath183 is chosen such that the magnitude of @xmath184 decreases by at least a factor of two from @xmath129 to @xmath185 . the characteristic proper time , @xmath61 , is computed from @xmath186 the next to last column in the table is our approximation of the awec integral given in eq . ( [ eq:4dawec / prox ] ) , while the last column represents the right - hand - side of our qi given by eq . ( [ eq : awec / qi-4d ] ) . by examining the last two columns of table 1 , we see that the qi , eq . ( [ eq : awec / qi-4d ] ) , is easily satisfied . a similar analysis for the conformally - coupled massless scalar field shows that the qi is easily satisfied in that case as well . numerical values of @xmath187 , and @xmath188 for the scalar field were obtained from the numerical data given in ref . @xcite . .outgoing timelike geodesics moving through the quantized electromagnetic field . [ cols=">,>,>,>,>,>,>,>,>",options="header " , ] in our calculations , we have ignored the backreaction of the hawking radiation on the background spacetime . it might be better to repeat our calculations on an evolving background , such as a vaidya spacetime . however , we can estimate the effects of backreaction on an outgoing null ray just outside the horizon in the following way . our approximation should hold as long as the time of escape of a null ray from near the horizon is small compared to the evaporation time of the black hole . here these times are measured by an observer at infinity . for such a null ray we have @xmath189 solving for @xmath190 yields @xmath191 where @xmath190 is the time for the ray to go from @xmath54 to @xmath192 . for @xmath193 and finite , in the limit of small @xmath55 : @xmath194 by contrast , the evaporation time of the hole is @xmath195 , where @xmath196 , and @xmath197 is the number of species of particles , each of which is assumed to contribute approximately @xmath198 to the black hole s luminosity . our approximation should break down only when @xmath199 $ ] . we now examine orbiting null geodesics in schwarzschild spacetime . the line element is @xmath200 where again @xmath88 . for an orbiting null ray in schwarzschild with @xmath201 , @xmath202 here we choose @xmath203 , for our future - directed ray . from the geodesic equations , we have @xmath204 where @xmath85 and @xmath205 are constants . it follows that @xmath206 at @xmath170 , we have @xmath207 from the numerical values in figs . ( 1 ) and ( 3 ) of ref . @xcite for the quantized electromagnetic field , and eq . ( [ eq : bo / def ] ) , we find that @xmath208 \left|_{r=3 m } \right . \nonumber \\ & \approx & -3.1 \times 10^{-4 } \,{e^2 \over m^4 } \ , , \label{eq : t3m}\end{aligned}\ ] ] where @xmath209 is the tangent to the null geodesic . for one orbit , @xmath210 , so @xmath211 we may therefore rewrite eq . ( [ eq : t3 m ] ) as @xmath212 integration of @xmath213 over one orbit gives @xmath214 thus we see that for orbiting null geodesics @xmath215 the expressions given by eqs . ( [ eq : anec-1norbit ] ) and ( [ eq : anecinq-1norbit ] ) are of the same form as those for radial null geodesics discussed earlier . the difference here is that we integrate only over one orbit , i.e. , we do not consider multiple traversals through the same negative energy region . the inclusion of the contributions of an arbitrary number of orbits would violate the inequality given in eq . ( [ eq : anecinq-1norbit ] ) . ( since the @xmath170 orbit is unstable , realistically we would expect the null ray to at best only make a few orbits . ) however , it is of interest to note that if we consider the null geodesics which generate the boundary of the future of an event @xmath216 with @xmath170 , then the orbiting null geodesics leave this boundary after half an orbit . this is due to the fact that orbiting null geodesics originating at @xmath216 , which initially set off in opposite spatial directions , encounter a conjugate point ( i.e. , the geodesics cross ) on the far side of the black hole after half an orbit @xcite , @xcite . at this point , the geodesics leave the boundary of the future of @xmath216 ( i.e. , an event @xmath217 which lies beyond the conjugate point along either null geodesic can be connected to @xmath216 by a timelike curve ) , and therefore are not achronal . this observation may be relevant in the context of proving singularity theorems in the presence of negative energy . for example , in the proof of a singularity theorem such as the penrose theorem , one uses either the local or the averaged null energy condition to insure the focusing of the null geodesics which generate the boundary of the future of a closed trapped surface . geodesics of the orbiting variety considered here would not relevant , at least in the spherically symmetric case , since they leave the boundary of the future of a point in any case , irrespective of energy condition - focusing arguments . we also note in passing that , at least in this simple case , while these null geodesics are in the boundary they obey the inequality , eq . ( [ eq : anecinq-1norbit ] ) . let us compare the previous results with the following situation . consider a 4d minkowski spacetime with one spatial dimension compactified , and a quantized , minimally - coupled , massless scalar field in the casimir vacuum state on this background . let @xmath218 be the tangent to a null geodesic in the @xmath219-direction . the vacuum expectation value of the stress - tensor in this state is @xmath220 where we have taken @xmath219 as the compactified dimension with circumference @xmath205 . it follows that @xmath221 the affine parameter lapse for one orbit is @xmath222 . if we integrate over one orbit , we find that @xmath223 we see that this result has the same form as eq . ( [ eq : anec-1norbit ] ) for the black hole case . both of these expressions are invariant under rescaling of the affine parameter ( as is eq . ( [ eq:4danec / qi ] ) for radial null geodesics ) . here , as in the black hole case , we have ignored the possibility of multiple orbits , which would violate the inequality . however , it is possible that in the 4d compactified case , as was found previously in 2d @xcite , an anec - type bound might hold on the _ difference _ between the expectation values of @xmath96 in an arbitrary quantum state and in the casimir vacuum state . in this paper , we have examined awec and anec in both two and four- dimensional evaporating schwarzschild black hole spacetimes , for quantized massless scalar and electromagnetic fields . our 2d results are the following : it was discovered that awec is satisfied for outgoing radial timelike geodesics which start just outside the horizon and reach infinity , for the unruh vacuum state . this is due to the positive contribution to the awec integral of the hawking radiation at large distances , which outweighs the contribution of the negative energy near the horizon . by contrast , awec is violated for such geodesics in the boulware vacuum state . similarly , anec is violated for half - complete outgoing radial null geodesics , which start just outside the horizon , in both the unruh and boulware vacuum states . this is also true on the horizon in the unruh vacuum state . however , anec is satisfied along ingoing null geodesics and along outgoing null geodesics inside the horizon in the unruh vacuum state , due to a large positive contribution to the integral near the singularity . in 4d , we examined awec and anec in the unruh vacuum state , using previously calculated numerical values of the stress - tensor components . there it was found that awec is violated for outgoing timelike geodesics which start just outside the horizon and reach infinity . ( note that in 4d , the hawking radiation decays like @xmath224 , so the large positive contribution to the awec integral found in 2d does not arise here . ) similarly , we found that anec is violated for analogous outgoing null geodesics . this condition is violated for orbiting null geodesics as well . however , in all cases where the conditions are violated , there appear to be qi - type bounds which limit the degree of the violations . let @xmath61 be the characteristic proper time for timelike geodesics , and @xmath225 be the affine parameter length for null geodesics , over which the negative energy density changes significantly . in 2d the qi bounds have the form that the integrated negative energy density along the timelike or null geodesic ( i.e. , the awec or anec integral ) is greater than or equal to minus the inverse of @xmath61 or @xmath225 , respectively . in 4d , the corresponding inequalities have additional factors of @xmath226 on the right - hand - sides . can the semi - classical effects of negative energy , in processes such as black hole evaporation , invalidate the singularity theorems before quantum gravity effects become significant ? although we do not have a definite proof in the general case , our results suggest that the answer may be no . it remains possible that even though anec ( or even a weaker energy condition @xcite ) might fail in some regions of a given spacetime , it may hold in enough other regions for the conclusions of the singularity theorems to be maintained . our ( admittedly ) 2d results for anec would seem to support this view . in ref . @xcite , it was shown that if anec is satisfied along the null geodesics which generate the boundary of the future of a trapped surface , then penrose s singularity theorem will still hold . there the local null energy condition was replaced by anec along half - complete null geodesics . however , two points should be emphasized about that result . first , it is not necessary to assume that the anec integral is non - negative to prove the result @xcite . in fact , from lemma 3 of ref . @xcite or eq . ( 5 ) of ref . @xcite and the einstein equations , it is sufficient to have @xmath227 where @xmath228 is the initial expansion of the null geodesic congruence at the trapped surface . the anec integral may be negative as long as it is not more negative than @xmath229 . physically this implies that a singularity will still form provided that the defocusing effects due to the presence of any negative energy are more than compensated for by the initial convergence of the null geodesics produced by the trapped surface . second , the existence of only _ one _ such trapped surface whose orthogonal null geodesics obey either anec or the weaker inequality , eq . ( [ eq : w - anec ] ) , is required to prove the occurrence of a singularity . there may exist other trapped surfaces in the spacetime whose orthogonal null geodesics do not obey either of the inequalities . however , as long as there exists _ at least one _ trapped surface with the desired properties , a singularity is inevitable . consider an evaporating black hole spacetime , with backreaction included . the penrose diagram for the standard scenario is shown in fig . let us assume that this standard picture is correct . the line @xmath230 is the event horizon , the dashed curve @xmath231 is the apparent horizon , and the lines @xmath232 , @xmath233 , and @xmath234 are outgoing null geodesics orthogonal to the trapped surfaces labelled @xmath235 , @xmath236 , and @xmath237 , respectively . the outgoing null rays from @xmath235 do not focus but instead reach future null infinity , so one expects that _ any _ averaged energy condition which would insure focusing , such as anec or eq . ( [ eq : w - anec ] ) , would fail along them . the closest analogs to these rays in our analysis are the outgoing null rays just outside the horizon shown in fig . it was found that in both two and four - dimensional schwarzschild spacetime , anec was violated along these rays , although in each case a qi limits the degree of the violation . one would expect that if an analogous qi holds in more general cases , such as fig . ( 3 ) , it will not be strong enough to guarantee focusing of null rays in the region between @xmath230 and @xmath231 in fig . we could not examine the weaker inequality eq . ( [ eq : w - anec ] ) , since in a static evaporating black hole background there are no trapped surfaces outside the horizon . for the outgoing null rays from @xmath236 in fig . ( 3 ) ( a trapped surface inside the collapsing matter ) , one would suspect that anec probably holds . in this case , the focusing effects of the matter of the star most likely overwhelm any defocusing by negative energy . for the outgoing null rays from @xmath237 ( a trapped surface `` behind '' the event horizon , in the vacuum outside the collapsing matter ) , one might expect either anec or a similar but weaker inequality such as eq . ( [ eq : w - anec ] ) to hold , since the null rays eventually do run into a singularity . this is assuming , of course , that the `` standard '' scenario depicted in fig . ( 3 ) is in fact the correct one - a view which could well be wrong . in our 2d analysis , we found that anec was satisfied along these rays , and also along ingoing null rays . we showed that our approximation of neglecting backreaction should fail only when the time of escape of a null ray from near the horizon is comparable to the evaporation time of the black hole . this implied that such a ray would have to start out extremely close to the horizon . nevertheless , it would be useful to attempt a similar analysis to ours for evaporating black hole spacetimes with backreaction taken into account . the situation is complicated by our ignorance of the detailed form of the stress - tensor components in such cases . recent work of kuo and ford @xcite has shown that negative energy densities in flat spacetime are subject to large fluctuations . in these circumstances , one does not expect the semi - classical einstein equations to be a good approximation . if the same is true of negative energy densities in curved spacetime , and we see no reason to believe otherwise , then the question arises as to whether such fluctuations could cause a gross failure of the semi - classical approximation at scales well above the planck scale . large stress - tensor fluctuations would presumably produce large fluctuations in spacetime geometry . therefore , it is very important to determine whether such fluctuations might render suspect the semi - classical picture of the hawking process , or the conclusions of the singularity theorems in the presence of negative energy . on the other hand it could turn out , for example in the hawking process , that if the timescale of the stress - tensor fluctuations is very short compared to other relevant time scales , such as the evaporation time of the hole , then the standard analysis may still be valid . in the proofs of the singularity theorems , one assumes a fixed classical spacetime background which determines the causal structure . if the background geometry fluctuates , it is not clear that the usual global techniques employed in classical general relativity will still be useful @xcite . these questions are currently under active investigation . 0.2 in * acknowledgements * the authors would like to thank arvind borde , eanna flanagan , matt visser , bob wald , and ulvi yurtsever for helpful discussions . tar would like to thank the members of the tufts institute of cosmology for their kind hospitality and encouragement while this work was being done . this research was supported in part by nsf grant no . phy-9208805 ( tufts ) , by the u.s . department of energy ( d.o.e . ) under cooperative agreement # df - fc02 - 94er40818 ( mit ) , and by a ccsu / aaup faculty research grant . s.w . hawking and g.f.r . ellis , _ the large scale structure of spacetime _ ( cambridge university press , london , 1973 ) , pp . 88 - 96 . h. epstein , v. glaser , and a. jaffe , nuovo cim . * 36 * , 1016 ( 1965 ) . casimir , proc . akad . wet . * b51 * , 793 ( 1948 ) . tipler , phys . d * 17 * , 2521 ( 1978 ) . c. chicone and p. ehrlich , manuscr . math . * 31 * , 297 ( 1980 ) . galloway , manuscripta math . * 35 * , 209 ( 1981 ) . a. borde , class . quantum grav . * 4 * , 343 ( 1987 ) . roman , phys . d * 33 * , 3526 ( 1986 ) . roman , phys . d * 37 * , 546 ( 1988 ) . j. friedman , k. schleich , and d. witt , phys . lett . * 71 * , 1486 ( 1993 ) . note that in a general spacetime , unlike the relation between the local null and weak energy conditions , anec does not simply follow by continuity from awec . an example of this is illustrated in sec . ( [ sec:2devap ] ) for the unruh vacuum state . g. klinkhammer , phys . d * 43 * , 2542 ( 1991 ) . a. folacci , phys . d * 46 * , 2726 ( 1992 ) . r. wald and u. yurtsever , phys . rev d * 44 * , 403 ( 1991 ) . m. visser , phys . b349 * , 443 ( 1995 ) . u. yurtsever , phys . d * 51 * , 5797 ( 1995 ) . u. yurtsever , phys . d * 52 * , r564 ( 1995 ) . m. morris and k. thorne , am . * 56 * , 395 ( 1988 ) . m. morris , k. thorne , and u. yurtsever , phys . lett . * 61 * , 1446 ( 1988 ) . l. ford and t. roman , phys . d , * 51 * , 4277 ( 1995 ) . ford , proc . a * 364 * , 227 ( 1978 ) . ford , phys . d * 43 * , 3972 ( 1991 ) . this particular choice was initially made in ref . @xcite purely for mathematical convenience . presumably , one could prove analogous inequalities with other choices of sampling functions . ford and t.a . roman , phys . d * 41 * , 3662 ( 1990 ) . ford and t.a . roman , phys . d * 46 * , 1328 ( 1992 ) . ford and t.a . roman , _ quantum field theory constrains traversable wormhole geometries _ , manuscript in preparation . u. yurtsever , class . quantum grav . * 7 * , l251 ( 1990 ) . r. penrose , phys . * 14 * , 57 ( 1965 ) . s. w. hawking , commun . phys . * 43 * , 199 ( 1975 ) . a. borde , topology change in classical general relativity " , tufts institute of cosmology preprint tutp 94 - 13 , gr - qc 9406053 . ford and t.a . roman , phys . d * 48 * , 776 ( 1993 ) . unruh , phys . d * 15 * , 365 ( 1977 ) . note that this expression differs somewhat in form from eq . ( [ eq : tlqi2duncp ] ) . however , the latter is an inequality valid for all quantum states in 2d minkowski spacetime , whereas the integral in eq . ( [ eq : intt ] ) is evaluated for a restricted class of geodesics in a specific quantum state in schwarzschild spacetime . note also that the right - hand - side of eq . ( [ eq : intt ] ) is more negative than the right - hand - side of eq . ( [ eq : tlqi2duncp ] ) . this is presumably due to the integration over a half - complete geodesic in eq . ( [ eq : intt ] ) ; even in minkowski spacetime , there exist states for which eq . ( [ eq : tlqi2duncp ] ) would not be true if the integration were taken only over a half - geodesic . davies and w.g . unruh , proc . a * 356 * , 259 ( 1977 ) . r. m. wald , _ general relativity _ ( univ . of chicago press , chicago , 1984 ) . r. m. wald , _ quantum field theory in curved spacetime and black hole thermodynamics _ ( univ . of chicago press , chicago , 1994 ) . u. yurtsever , private communication . t. elster , phys . 94a * , 205 ( 1983 ) . jensen , j.g . mclaughlin , and a.c . ottewill , phys . d * 43 * , 4142 ( 1991 ) . m. visser , _ lorentzian wormholes - from einstein to hawking _ ( american institute of physics press , new york , 1995 ) , sec . 12.3.6 , pp . 128 - 9 . the form of eq . ( [ eq : awec / qi-4d ] ) is motivated by the following considerations : if we imagine a sequence of observers shot outward with increasing values of @xmath38 , then @xmath238 and @xmath239 , so @xmath240 . the factor of @xmath226 is required in order for the dimensions to be correct . this form is also suggested by the form of the inequality found for orbiting null geodesics , eq . ( [ eq : anecinq-1norbit ] ) . d. n. page , phys . * d13 * , 198 ( 1976 ) . a. borde , private communication . , footnote 2 of r. penrose , _ techniques of differential topology in relativity _ ( society for industrial and applied mathematics , philadelphia , 1972 ) , for an interesting example with timelike geodesics . c .- i kuo and l.h . ford , phys . d * 47 * , 4510 ( 1993 ) . c .- i kuo , _ quantum fluctuations and semiclassical gravity theory _ thesis , tufts university ( 1994 ) , unpublished . the authors are grateful to a. borde for this remark . * [ 1 ] radial null geodesics in the spacetime of a black hole formed by gravitational collapse : an ingoing geodesic ( a ) , an outgoing geodesic inside the horizon ( b ) , and an outgoing geodesic outside of the horizon ( c ) . * [ 2 ] the spacetime of a black hole formed by gravitational collapse . the null ray which lies in the future horizon enters the collapsing body ( shaded region ) , and then re - emerges into the vacuum spacetime at @xmath241 . * [ 3 ] the spacetime of an evaporating black hole , including the effects of backreaction .
quantum dots and single - molecule junctions have been considered as possible building blocks for nano - electronics and quantum - information processing.@xcite recent technological progress has made it possible to manufacture and study electron transport trough ultra - small quantum - dot devices , nanotubes or single molecules.@xcite these devices are designed as a small central region comprising of a quantum dot or a single molecule which is coupled to at least two leads where the finite bias voltage is applied to . the investigation of such devices is of fundamental importance for our understanding of open quantum systems out of equilibrium . due to the quantization of the charge , the physical properties of such junctions are dominated by many - body effects at temperatures below the charging energy @xmath2 , where @xmath3 is the capacitance of the device . the experimental devices are often fully controllable by external gate electrodes or elongation of the scanning tunneling microscope tip.@xcite this gives the opportunity to directly study true many - body correlation effects , such as the kondo effect ( see for example ref . and ) , under the influence of an external bias voltage . however , the theoretical understanding of the interplay between coherent transport favored by many - body correlations and current - driven dephasing at finite bias is still at its infancy and further investigations are needed . the present work has two main objectives . on the one hand , we establish the reliability of the recently introduced scattering - states numerical renormalization group ( snrg ) approach@xcite to quantum transport by comparing results for small values of @xmath0 to the diagrammatic kadanoff - baym - keldysh approach , which becomes exact in the limit @xmath1 . on the other hand , we will discuss discrepancies and reveal shortcomings of those diagrammatic approaches at intermediate values of the charging energy . we investigate quantum transport through a quantum - dot device using a minimal model@xcite where the complex interacting region is replaced by a single spinful orbital which is coupled to two noninteracting leads . a single coulomb matrix element @xmath0 accounts for the charging energy of the device . we calculate nonequilibrium spectral functions@xcite and current - voltage ( iv ) characteristics using the snrg as well as different approximations@xcite within the diagrammatic kadanoff - baym - keldysh expansion in the local coulomb interaction @xmath0 . over the past 40 years , the keldysh technique @xcite has proven to be the most successful approach to nonequilibrium dynamics . in the context of quantum transport through nano - junctions direct expansions in the interaction@xcite as well as self - consistent re - summation schemes have been employed.@xcite however , such diagrammatic expansions rely on a small expansion parameter , and are , therefore , confined to weak coupling . but quantum - impurity models@xcite commonly used in the theory of quantum transport on the molecular level often exhibit infra - red divergences in perturbation theory@xcite which also restrict the diagrammatic keldysh approaches to certain parameter regimes usually to high temperature or to large bias . in contrast to equilibrium conditions , where complete and accurate solutions can be obtained using a variety of nonperturbative techniques such as the bethe _ ansatz_,@xcite conformal field theory,@xcite or wilson s numerical renormalization group ( nrg ) approach,@xcite techniques for calculating quantum - transport out of equilibrium remain largely at the development stage . recent advancements on the analytical side @xcite include suitable adaptations of the wegner s flow - equation@xcite and the real - time renormalization - group method.@xcite these methods can successfully access large voltages , but are generally confined to the weak - coupling regime . based on the scattering - states approach to quantum transport@xcite the bethe ansatz was extended to quantum - impurity models out of equilibrium,@xcite but remains limited to a certain class of models . on the numerical side , progress has been made in several directions . currents have been extracted from time - dependent density matrix renormalization group@xcite calculations using finite 1d wires , and the results agree well with bethe ansatz results for certain models.@xcite quantum monte carlo approaches based on scattering states@xcite can access the intermediate coupling regime@xcite at finite bias . recent real - time formulations of continuous - time quantum monte carlo@xcite and an iterative real - time path integral approach@xcite to quantum transport offer the appealing advantage of working directly in the continuum limit , but are confined to relatively short time scales . access to low temperatures and long times is hampered in the former case by a severe sign problem , and by the extrapolation to long memory times in the latter case . hence neither approach can presently be applied to nonequilibrium dynamics of correlated systems with a small underlying energy scale , as is the case with ultra - small quantum dots when tuned to the kondo regime . the usage of lippmann - schwinger scattering states has been well established in quantum - field theory@xcite for over 50 years and also successfully adapted to the description of quantum transport through strongly interacting nano - devices coupled to ballistic leads.@xcite these states fulfill the correct boundary condition of the open quantum system : ( i ) they break time - reversal symmetry and , therefore , are ( ii ) complex and current - carrying and ( iii ) describe ballistic transport in the leads combined with scattering events in the small interacting quantum - dot region . this time - reversal symmetry breaking is required for current carrying systems and reveals itself naturally in all diagrammatic approaches by the occurrence of retarded and advanced green s functions . it is a consequence of any regularization when performing the limit to an infinitely large system . in particular , the work of hershfield @xcite and doyon and andrei @xcite has rigorously shown that these boundary conditions remain unaltered when a local interaction is switched on . the noninteracting current - carrying system evolves into the new steady - state of the _ interacting _ system , and the steady - state density operator retains a boltzmannian form.@xcite the explicit construction of those scattering states allows to exactly solve the dc and ac kondo model at the toulouse point@xcite as well as the interacting resonant level model.@xcite recently , an extension@xcite to wilson s numerical renormalization group has been developed for steady - state quantum transport through nano - devices which is able to deal with the crossover from weak to strong coupling for arbitrary bias voltages . it is based on oguri s idea@xcite of discretizing the single - particle scattering states which are the solutions of the lippman - schwinger equation@xcite for the noninteracting problem and , therefore , fulfill the correct boundary condition of an open quantum system . this scattering - states numerical renormalization group approach@xcite ( snrg ) evolves the analytically known density matrix of a noninteracting system to the density matrix of the fully interacting problem by employing the time - dependent nrg ( td - nrg).@xcite the nrg is ideally suited to the problem , being known to provide accurate solutions of quantum - impurity models on all relevant interaction strengths at zero bias.@xcite since the td - nrg can access exponentially long time scales,@xcite dwell times on the order of the inverse kondo - temperature are easily accessible . this paper is organized as follows . after the model used is defined , we provide the details of the different theoretical approaches in sec . [ sec : theory ] . we summarize the basic ideas of the snrg method introduced in ref . in sec . [ sec : snrg ] and state all necessary equations of the diagrammatic nonequilibrium techniques in sec . [ sec : kbk - approach ] . the main body of the paper is in sec . [ sec : results ] , where we present and discuss the results obtained for the various methods . in order to set the stage for a detailed comparison between the snrg and diagrammatic approaches at finite bias , we begin with a discussion of the magnetic and charge fluctuation scales at zero bias in sec . [ sec : equiscale ] . since the nrg provides an accurate solution in this regime for arbitrary coupling strengths and temperatures , this reveals the validity range of the diagrammatic approaches . we show that in contrast to the nrg some of the diagrammatic expansions fail to produce a single low - energy scale for intermediate and large values of @xmath0 . however , in the weak correlation regime all these approaches agree excellently for arbitrary voltages at small @xmath0 and yield the same nonequilibrium green functions as well as iv characteristics which are presented in sec . [ sec : weak - coupling ] . discrepancies between the different approaches at intermediate values of the coulomb interaction are discussed in sec . [ sec : inter - coupling ] , where the spectral functions and iv curves of a symmetric and an asymmetric junction are considered . we conclude with summary and a short outlook in sec . [ sec : summary ] . quantum impurity models are used to describe quantum transport on the molecular level . their hamiltonian @xmath4 @xmath5 consists of three parts : an impurity part @xmath6 modeling the interacting device with a finite number of degrees of freedom , one or several bosonic or fermionic baths represented by @xmath7 , and the coupling of these subsystems by @xmath8 . throughout this paper , we restrict ourselves to junctions modeled by the single impurity anderson model with one spinful orbital coupled to a left ( l ) and a right ( r ) lead and an on - site coulomb repulsion @xmath0 @xmath9 here , @xmath10 is the single - particle energy of the quantum dot , @xmath11 measures its orbital occupancy and @xmath12 represent the elementary hybridization - matrix elements coupling the dot to the two leads . the different chemical potentials @xmath13 in both leads appear as a shift of the band centers and are functions of the external voltage @xmath14 . for simplicity , we assume that both leads have the same density of states , @xmath15 , characterized by the same band width @xmath16 but different band centers . this hamiltonian is commonly used to model a single coulomb - blockade resonance in ultra - small quantum dots.@xcite in the absence of the local coulomb repulsion @xmath17 , the single - particle problem is diagonalized exactly in the continuum limit@xcite by the following scattering - states creation operators @xmath18 \punkt\end{aligned}\ ] ] @xmath19 labels left ( right ) moving scattering states created by @xmath20 . the local retarded resonant level green s function @xmath21^{-1}\end{aligned}\ ] ] enters as an expansion coefficient . defining @xmath22 , we will use @xmath23 and @xmath24 - i\gamma(\w)\end{aligned}\ ] ] in the following . in the limit of infinitely large leads volume @xmath25 the single - particle spectrum remains unaltered , and these scattering states diagonalize the hamiltonian ( [ eqn : siam ] ) for @xmath26 : @xmath27 the scattering states are solutions of the lippmann - schwinger equation@xcite and therefore break time - reversal symmetry , which constitutes a necessary boundary condition to describe a current carrying open quantum system . this is encoded in the small imaginary part @xmath28 entering eq . ( [ eq : scattering - states - operators ] ) ( [ eq : deltadef ] ) required for convergence when performing the continuum limit @xmath25 in the leads . the complex expansion coefficients in ( [ eq : scattering - states - operators ] ) are given by _ retarded _ functions , e.g. @xmath29 , which causes the scattering states to be complex and current carrying . for zero bias voltage , time - reversal symmetry manifests itself in the identical spectrum for left and right movers which are time - reversal pairs in that limit . to avoid any contribution from bound states , we will implicitly assume a wide band limit : @xmath30 , where @xmath31 and @xmath32 . hershfield has shown that the density operator for such a noninteracting current carrying quantum system retains its boltzmannian form@xcite @xmath33 even for finite bias . the @xmath34 operator accounts for the different occupation of the left- and right - moving scattering states , and @xmath13 for the different chemical potentials of the leads . therefore , all steady - state expectation values of operators can be calculated using @xmath35 which includes the finite bias . in the absence of a coulomb repulsion @xmath0 , this is a trivial and well - understood problem . it was shown@xcite that the current expectation value using this density - operator @xmath36 reproduced the standard result@xcite for noninteracting devices . the knowledge of the analytical form of @xmath35 , however , makes this steady - state model accessible to a nrg approach.@xcite the expansion coefficients of @xmath37 in eq . ( [ eq : scattering - states - operators ] ) contain the complex single - particle green function @xmath38 which we separate in modulus and phase @xmath39 this phase is absorbed into the new scattering states @xmath40 by a local gauge transformation . the impurity operator @xmath41 is expanded into left- and right - mover contributions @xmath42 using the inversion of eq . ( [ eq : scattering - states - operators ] ) . these two new operators @xmath43 are then defined as @xmath44 and obey the anti - commutator relation @xmath45 . the scattering - states numerical renormalization group approach@xcite ( snrg ) starts from a logarithmic discretization of the scattering - states continuum @xmath46 in intervals @xmath47 $ ] and @xmath48 $ ] @xmath49 , controlled by the parameters@xcite @xmath50 and @xmath51 $ ] . the intervals for @xmath52 are defined as @xmath53 $ ] and @xmath54 $ ] . an average over various @xmath55-values@xcite is used to mimic the conduction band continuum . then , the discretized version of the noninteracting hamiltonian ( [ eq : h0-u0 ] ) is mapped onto a semi - infinite wilson chain @xmath56 whose tight - binding matrix elements @xmath57 decay exponentially @xmath58 for large @xmath59 . in contrast to the standard nrg,@xcite the impurity degree of freedom has been included into @xmath60 since not the leads but the full scattering states have been discretized . any complex phase in the tight - binding parameters @xmath57 can be absorbed into the creation ( anihilation ) operators @xmath61 @xmath62 of an electron on the chain link @xmath59 with spin @xmath63 and mover @xmath64 by a local gauge transformation . we use @xmath65 defined in eq . ( [ eqn : d - alpha - cont ] ) as starting vector @xmath66 for the householder transformation@xcite and obtain the tight - binding coefficients of the wilson chain ( [ eqn : h0-lambda ] ) by the usual procedure.@xcite it is straight forward to shown that the energy of the first chain link corresponds to the energy of the original quantum - dot orbital : @xmath67 . in order to include the local coulomb interaction , the density operator @xmath68 must be expanded in the new orbitals @xmath65 . it consist of two contributions : a density term and a backscattering term @xmath69 , where @xmath70 and the backscattering @xmath71 term is defined as @xmath72 the local coulomb interaction @xmath73 @xmath74 leads to a mixing of left and right movers since @xmath71 does not commute with @xmath75 . however , the term @xmath76 , @xmath77 commutes with @xmath34 and can be absorbed into the steady - state density operator @xmath78/z$ ] with @xmath79 using the arguments given in ref . . starting from an equilibrated system for times @xmath80 , the initial hamiltonian @xmath81 is changed to @xmath82 by a sudden quench at @xmath83 . then , the density operator @xmath84 evolves from its initial value @xmath35 at @xmath83 as @xmath85 if @xmath86 describes a quantum impurity problem and @xmath87 is an impurity operator , it was recently shown that the real - time dynamics of the expectation value of @xmath88 can be calculated@xcite by evaluating @xmath89 where @xmath90 denotes the matrix elements of the operator @xmath87 and @xmath91 the nrg eigenenergy of the eigenstate @xmath92 to @xmath93 at nrg iteration @xmath94 . the sum restriction @xmath95 indicates that at least one of the states @xmath96 must be a discard state at iteration @xmath94 . excitations between two retained states will be refined in the following iterations @xmath97 and , therefore , will contribute at a later iteration . @xmath98 labels the environment degrees of freedom of the wilson chain links to be incorporated in subsequent iterations @xmath97 and the @xmath99 are just tensor - product states of the eigenstates of the @xmath94th iteration and the yet uncoupled rest chain . the reduced density matrix @xmath100 traces out all environment degrees of freedom @xmath98 . the initial conditions are encoded into the density operator @xmath35 calculated with the initial hamiltonian @xmath81 . the calculation of the overlap matrix between the nrg eigenstates of @xmath81 and @xmath82 allows for the basis set transformation of @xmath101 into the basis of the final hamiltonian provided that @xmath35 remains restricted to the last wilson shell.@xcite this transformed @xmath101 enters eq . ( [ eqn : o - vs - t - td - nrg ] ) . the discarded states form a complete basis set@xcite for the fock - space of the _ entire _ wilson chain of length @xmath102 , i.e. @xmath103 where @xmath104 labels all discarded states at iteration @xmath94 . the iterative diagonalization thus procures the set of ( approximate ) eigenstates for the _ whole _ energy range from high energies on the order of the bandwidth down to very low energies such as the kondo scale . this is indispensable because nonequilibrium processes usually involve all energy scales and can not be confined to a finite low energy window set by the last wilson shell as in the usual equilibrium nrg . in sec . [ sec : scattering - state ] we have argued that the analytic form the steady - state nonequilibrium density operator is known for the noninteracting case . this allows for applying the nrg approach to construct a faithful representation of @xmath105 . we assume that when switching on the coulomb interaction @xmath73 for infinitely large leads ( i ) a steady state is reached after some characteristic but finite time and ( ii ) it is unique and independent of the initial condition . as described earlier , the boundary condition of time - reversal symmetry breaking is imposed on the scattering states and the nonequilibrium density operator at @xmath83 for @xmath26 . the interaction quench at @xmath83 , i.e. switching on a local scattering potential , and the subsequent unitary time evolution do not affect this boundary condition , and the time - evolved operators characterize the interacting current carrying open quantum system . the time average of the density operator @xmath106 projects out the steady - state contributions to the time - evolved density operator @xmath107 even in a finite - size system : only the energy diagonal terms contribute in accordance with the steady - state condition @xmath108 = 0 $ ] . even though @xmath109 remains unknown analytically , we can construct it systematically using the time - dependent nrg@xcite described above . the steady - state retarded green s function is defined as @xmath110_s } \theta(t ) , \label{eqn : steady - state - gf } \end{aligned}\ ] ] where @xmath111 , @xmath112_s$ ] denotes the commutator ( @xmath113 ) for bosonic , and the anti - commutator ( @xmath114 ) for fermionic correlation functions . this green s function can be calculated using the time - dependent nrg@xcite and extending ideas developed for equilibrium green s functions.@xcite the completeness relation for the basis of discarded states as introduced above is given by @xmath115 where @xmath116 denotes the first iteration at which the nrg truncation is employed , @xmath102 is the total number of iterations ( i.e. the length of the wilson chain ) , and @xmath104 only runs over the states which are discarded at iteration @xmath94 . for each iteration @xmath94 , we can partition the completeness relation ( [ eqn : completness - nrg ] ) into two parts , @xmath117 , where the first part incorporates the iterations @xmath116 to @xmath94 and the second the iterations @xmath118 to @xmath102 . since @xmath119 spans the part of the fock - space which contains all kept states @xmath120 after iteration @xmath94 , the identity @xmath121 must hold . the different contributions to the green s function are calculated for each energy scale @xmath122 by expanding the ( anti-)commutator in eq . ( [ eq : neqgf ] ) and inserting the completeness relations eqs . ( [ eqn : completness - nrg ] ) and ( [ eqn : kept - states ] ) repeatedly . by making use of the fact that local operators @xmath123 and @xmath124 are diagonal in the environment degree of freedom @xmath98 , reduced density matrices @xmath101 occur naturally when tracing out the environment @xmath98 here as well . although the excitation energies remain confined to the same energy scale , terms connecting different energy scales @xmath125 and @xmath126 are implicitly included through the reduced density matrices such as defined in ( [ eqn : red - dm ] ) . similar to the real - time dynamics , the summation over all @xmath94 then ensures that all energy scales @xmath125 contribute to the green s functions.@xcite a detailed derivation is given in ref . . it was shown that the algorithm is identical to the equilibrium algorithm@xcite if @xmath127 . laplace transforming @xmath128 yields the steady - state spectral function for the retarded green s function which is used to calculate the current ( see eq . ( [ eq : ss - current ] ) below ) . we employ the nonequilibrium perturbation theory as formulated by kadanoff and baym@xcite and keldysh @xcite on the usual keldysh time contour , for example , see refs . . since we are only interested in the steady - state properties , the information and correlations of the initial conditions are assumed to be lost . this is archived by sending the initial time @xmath129 and dropping all correlation functions which involve the initial state . it is again assumed , that the system reaches a steady state which is translational invariant in time . therefore , the single - particle green s function does only depend on the difference between the two formerly independent times of particle creation and annihilation . the laplace transform of the time difference then leads to the formulation in frequency space for all green s functions of the steady state . in the nonequilibrium steady - state formulation two independent components of the contour ordered green s function survive which are chosen to be the retarded and lesser green s functions , @xmath130 and @xmath131 respectively . the advanced and greater functions are related via @xmath132 the two relevant green s functions can be expressed as @xcite @xmath133 \\ \feff(\w ) & = & f_l(\w)\gamma_l(\w)+f_r(\w)\gamma_r(\w ) \label{eqn : f - eff } \quad,\end{aligned}\ ] ] where , again , @xmath134 are the hybridization functions of the leads , @xmath135 their imaginary parts ( see eq . ( [ eq : deltadef ] ) ) and @xmath136 are the fermi functions of the corresponding leads . the retarded and lesser self - energies , @xmath137 and @xmath138 respectively , include all correlation effects induced by the coulomb interaction @xmath0 . @xmath139 accounts for the frequency independent hartree energy shift . diagrammatic expansions in the coulomb interaction of the self - energy @xcite have been investigated for systems in equilibrium @xcite as well as in nonequilibrium.@xcite the self - energies can be evaluated either non self - consistently , where bare propagators are used as inner lines , or in terms of skeleton diagrams , where fully - dressed propagators are taken into account . in this study we focus on three different approximations for the self - energy : ( a ) the bare expansion up to second order in @xmath0 , where hartree - fock ( hf ) propagators are used as internal lines . the latter are just the noninteracting propagators , but with a shifted level position @xmath140 . this approximation is labeled @xmath141 and its diagrammatic representation is schematically shown in fig . [ fig : perturb](a ) and ( b ) . ( b ) the self - consistent evaluation of the second - order skeleton diagram of fig . [ fig : perturb](a ) and ( b ) . this approximation is called second born approximation ( 2ba ) but in contrast to the usual 2ba , no exchange contribution exists for the single impurity anderson model ( [ eqn : siam ] ) with only one spinful orbital . ( c ) in the gw approximation@xcite ( gwa ) the bare coulomb interaction @xmath0 is screened by an infinite series of particle - hole excitations , which can be summed as indicated in fig . [ fig : perturb](c ) . no contributions with odd orders in the interaction occur in this series due to the definition of the matrix elements of the coulomb interaction in our model ( [ eqn : siam ] ) , where we set matrix elements between electrons with the same spin explicitly to zero . the 2ba and the gwa are both evaluated self - consistently , and thus the self - energies can be derived from a luttinger - ward functional.@xcite therefore , both constitute conserving approximations in the sense of kadanoff and baym.@xcite it can be shown that elementary sum rules such as charge and current conservation are obeyed.@xcite in contrast , the non self - consistent @xmath142 approximation is not conserving which can lead to the violation of current conservation , as it will be demonstrated later . the hartree shift is produced by the average occupation of the quantum dot @xmath143 and analytic expressions for the self - energies read @xmath144 where the effective interactions are given by @xmath145 and the particle - hole bubbles are @xmath146 in the above expressions the advanced and greater green s functions can be determined via eq . ( [ eq : advgf ] ) and ( [ eq : grtgf ] ) and @xmath147 denotes the opposite spin of @xmath148 . equations ( [ eq : grdef])-([eq : pless ] ) form a closed set , which is solved self - consistently for the 2ba and gwa . for the @xmath142 approximation all particle - hole propagators ( [ eq : pr])-([eq : pless ] ) are evaluated only once with bare green s functions @xmath149 and eq . ( [ eq : wrba ] ) is used as the effective interaction . the hartree shift is included in order to determine the desired filling . the effective fermi function @xmath150 was defined in eq . ( [ eqn : f - eff ] ) . the gwa @xcite has been successfully applied to overcome some shortcomings of local - density calculations and estimate the screening of the coulomb interactions in solid state physics . recently , it has been employed to calculate quantum transport through nanoscale devices.@xcite in the context of the single impurity anderson model it was shown to accurately describe the equilibrium properties in the weakly - interacting regime and in asymmetric situations with a nearly empty or nearly full impurity orbital.@xcite in the strongly interacting kondo regime , i.e. @xmath151 , the gwa produces a narrow peak in the spectral function at the fermi level , which could be interpreted as remnants of the expected many - body resonance.@xcite however , the line shape of this low - energy resonance as well as the high - energy hubbard peaks at @xmath152 and @xmath153 are not correctly reproduced by this approximation.@xcite additionally , for very large interactions strength @xmath154 , all three perturbative approaches favor an unphysical magnetic ground state , which is actually forbidden by the mermin - wagner theorem.@xcite in the nonequilibrium situation , the proximity to bifurcation points of these sets of equations leads to unphysical hysteretic response.@xcite the current flowing from lead @xmath64 onto the impurity region can be expressed as @xcite @xmath155\end{aligned}\ ] ] where @xmath156/\pi$ ] is the frequency and voltage dependent spectral function of the retarded impurity green s function . since the steady - state current onto the interacting region from the left must be equal to the current leaving to the right lead , i.e. @xmath157 , we can symmetrize the left and the right currents with a linear combination@xcite and write it as @xmath158 in the wide band limit , @xmath159 and @xmath160 holds such that the term proportional to @xmath161 drops out of ( [ eq : currentsym - def ] ) and we obtain @xmath162 \pi\gamma \rho^r_\s(\w , v ) \ : . \end{aligned}\ ] ] where we have defined @xmath163 @xmath164 @xmath163 reaches the universal conductance quantum @xmath165 for a symmetric point - contact junction , @xmath166 , and is strongly suppressed in the tunneling regime @xmath167 . for the voltage drop across the two contacts of the impurity to the leads we employ a serial resistor model where the chemical potentials in the leads are given by @xmath168 and @xmath169 . at zero temperature , the zero bias conductance @xmath170 is proportional to the spectral function at the fermi level . in the zero temperature fermi liquid and for a symmetric junction @xmath171 . the conductance is given by its universal value @xmath172 which shows in the slope at zero bias of the iv characteristics , i.e. @xmath173 . we also define a leakage current @xmath174 i g^<_\s(\w ) \\ \nonumber & & + \frac{4e}{h}\sum_\s \int \!d\w\ : \feff(\w ) \pi \rho^r_\s(\w)\end{aligned}\ ] ] which must vanish due to current conservation , @xmath175 , in a physical junction . therefore , deviations from @xmath176 measures shortcomings of an approximation . in this section we compare and discuss the results obtained from the different diagrammatic keldysh approaches with the snrg . for simplicity , we used symmetric structureless leads characterized by a constant density of states with a half - bandwidth @xmath177 , i.e. @xmath178 . the total @xmath32 is used as the energy scale : all energies , voltages and temperatures are measured in units of @xmath179 throughout the paper . for the snrg a rather large @xmath180 was chosen , and @xmath181 states were retained in each nrg - iteration step . @xmath55-averaging@xcite with either @xmath182 or @xmath183 different @xmath55-values was performed , and the broadening parameter for the spectral function@xcite was chosen @xmath184 . for large @xmath0 some unphysical wiggles may emerge in the spectral function , as it is explained below . in principle , these wiggles can be minimized by choosing a smaller @xmath185 , incorporating more states or performing the @xmath55-averaging with a larger number of @xmath55-values . we did not include an external magnetic field , and no magnetic solutions are encountered for the parameter values used in this paper . therefore , we will drop the spin - index from now . the two spin components of the spectral functions and self - energies are identical , e.g. @xmath186 and @xmath187 respectively . before we apply finite bias voltages , we will compare the different equilibrium low - energy scales obtained with the diagrammatic approaches to nrg results . while the diagrammatic approach becomes exact only in the weak - coupling limit @xmath1 , the nrg produces the correct scales for all interaction strengths . we will identify the validity range of the diagrammatic expansion . in that regime the diagrammatic approach produces correct results even in nonequilibrium , and we will therefore use it to benchmark the snrg for finite voltages . the single impurity anderson model in equilibrium for @xmath188 always forms a local fermi liquid.@xcite the spectral function for a symmetric junction approaches the zero temperature limiting value @xmath189 in accordance with the friedel sum rule.@xcite the fermi - liquid formation is associated with a characteristic low - energy scale , which is identified with the kondo temperature @xmath190 at large coulomb repulsions and near half - filling . the snrg coincides with the usual nrg@xcite in equilibrium , which accurately describes the crossover from high to low temperatures and provides the correct low - energy scale @xmath190 depending exponentially on @xmath0.@xcite the @xmath142 approximation , however , predicts a low - energy scale which is perturbative in @xmath0 and too large.@xcite the gwa does produce a narrow many - body resonance in the spectral function at the fermi level . extracting a low - energy scale from the full width at half maximum ( fwhm ) for an asymmetric junction ( @xmath191 ) , as shown in fig . 5 of ref . , suggests an exponential variation with the ionic level position @xmath10 . however , the exponent has the wrong prefactor as compared to the exact analytic form.@xcite in order to extract the low - energy scale from our model calculations we employ two different methods : we calculate the temperature dependent zero bias conductance @xmath192 and fit it to a phenomenological form.@xcite since @xmath193 is directly determined by the spectral function , it is sensitive to the amount of spectral weight in the temperature window @xmath194 . the scale @xmath195 extracted in this way constitutes the energy scale relevant for the zero - bias charge transport in the system . this procedure yields the same result as the aforementioned extraction from the fwhm of the resonance at the fermi level . the second way utilizes the screening of the effective local magnetic moment , @xmath196 , where @xmath197 is the magnetic susceptibility , @xmath198 the magnetization and @xmath199 an external magnetic field . we calculate @xmath198 for a finite but small external magnetic field @xmath200 , and extract the susceptibility via the difference quotient . in the fermi - liquid regime the effective magnetic moment follows an universal curve as function of temperature from which the low - energy scale is determined by defining @xmath201.@xcite the resulting @xmath202 sets the scale for magnetic excitations in the system and is directly linked to the kondo - screening of the local magnetic moment . for large values of the coulomb interaction , the scales @xmath202 and @xmath203 should coincide ( apart from a constant of order one ) and vary as @xmath204 for a symmetric quantum dot . for very small values of the coulomb repulsion @xmath205 both should approach @xmath206 . the charge scale @xmath195 is expected to be roughly constant and on the order of @xmath207 for @xmath208 since for such small interactions charge fluctuations to and from the leads dominate the physics , and the spectral function stays very close to its hf form . on the other hand , the magnetic scale is known to decrease exponentially for all @xmath0 . @xcite extracted from the various approximations as described in the text . a fit to the magnetic scale of the nrg to show the exponential decay @xmath209 is also included in the plot . ] figure [ fig : gfu4-ii ] shows the two scales extracted from nrg and gwa calculations for a symmetric junction in equilibrium . the nrg results show the expected @xmath0-dependencies : the charge scale @xmath195 is on the order of @xmath206 for small @xmath210 and decreases exponentially for large @xmath211 . the magnetic scale @xmath202 decreases exponentially for all @xmath0 as it is evident from the comparison with a fit function @xmath212 also included in the plot . furthermore , there exists only one universality scale for large @xmath0 which manifests itself by @xmath213 ( not shown ) . on the other hand , the scales obtained from the gwa agree with the nrg only for small @xmath0 . the charge scale @xmath195 perfectly agrees with the nrg curve for @xmath214 . significant deviations are observed for larger @xmath0 , where the gwa-@xmath195 decreases faster than the nrg . for @xmath0 significantly larger than the ones shown in the plot , no scales could be extracted due to the artificial symmetry breaking already reported in the literature.@xcite we added a second gwa charge scale @xmath215 to the graph which is obtained from the width of the low - energy feature at @xmath216 of @xmath217 ( and not at the fwhm as for @xmath195 ) . the correspondingly extracted scale should coincide with @xmath195 , apart from a prefactor . but it is found that both scales follow the same trend only for small @xmath0 and already for @xmath218 a much stronger decrease than the expected @xmath204 is observed in @xmath215 . ) zero - temperature spectral functions for @xmath219 and @xmath220 and a quantum - point contact @xmath221 . the inset shows the temperature evolution of the spectral function right at the fermi level , @xmath222 . the nrg - parameters are @xmath223 , @xmath224 , @xmath183 , @xmath225 and @xmath226 nrg iterations were performed . ] therfore , the extraction of the charge scale within the gwa at intermediate @xmath0 is somewhat ambiguous . a comparison of the zero - temperature equilibrium spectral function of the gwa and nrg for @xmath227 is depicted in fig . [ fig : gfu4-iii ] . the low - energy feature of the gwa spectral function is too narrow and exhibits a rather spiky line - shape which suggest at too low charge scale . this is supported by the evolution of @xmath217 as a function of temperature which is shown in the inset of fig . [ fig : gfu4-iii ] . the logarithmic increase of @xmath217 which occurs at temperatures on the order of the relevant charge scale also reveals that the charge scale is predicted as too low in gwa compared to the nrg . however , a considerable broadening occurs away from the fermi level which leads to the same fwhm for the gwa as in the nrg and consequently the larger @xmath195 emerges in thermodynamic quantities like @xmath228 . the magnetic scale @xmath202 extracted from the gwa exhibits some peculiar @xmath0-dependence . for small @xmath229 the scale agrees with the nrg . however , it develops a minimum at @xmath230 and then increases again for increasing @xmath0 ! this clearly indicates a failure of the gwa to describe magnetic properties for intermediate and large interactions . since the gwa effective moments @xmath231 show universality as functions of the dimensionless temperature @xmath232 for low temperatures ( not shown ) , the increase in @xmath202 implies a too strong screening of magnetic moments . the effective coulomb interaction is over - screened by @xmath233 ( see eq . ( [ eq : wrgw ] ) ) . the electrons remain itinerant even at rather large @xmath0 , and the gwa fails to capture the atomic limit . therefore , the magnetic screening scale @xmath202 remains large in the gwa and actually increases with @xmath0 . the scales extracted from the other two diagrammatic approximations all coincide with the gwa for small @xmath234 . for larger @xmath0 the @xmath142 approximation produces the same difference as the gwa between the charge and the magnetic scale , whereas within the 2ba both scales decrease with increasing @xmath0 , but in a polynomial rather than exponential fashion . we have established that the diagrammatic approach produces reliable results for interactions up to the order of the hybridization strength @xmath234 which we will use in the following section to benchmark the snrg in that regime . we study the nonequilibrium properties of a symmetric and an asymmetric junction in the weakly correlated regime @xmath236 . we use a very low temperature @xmath237 , which is sufficiently small compared to all other scales in the problem so it can be considered as @xmath238 with impunity . for such small interactions the diagrammatic approaches and the snrg yield identical results for all voltages . figure [ fig : weakcoupall](a ) shows the nonequilibrium spectral function of a symmetric junction in the quantum - point contact regime , i.e. @xmath239 and @xmath240 . for small voltages @xmath241 the spectra are even indistinguishable from the hartree - fock ( hf ) result . only at larger @xmath242 small deviations around the fermi level as can be observed . and @xmath243 for different voltages . results for the spectral functions are shown for hf , gwa and snrg , while the self - energy is shown for the gwa only . ( b ) current as a function of voltage for a symmetric ( @xmath244 and @xmath243 ) and an asymmetric junction ( @xmath245 , @xmath246 and @xmath247 ) . @xmath248 is normalized to @xmath249 ( symmetric ) and @xmath250 ( asymmetric ) and measured in units of @xmath179 . a small temperature of @xmath251 was used for all calculations . parameters for the snrg calculations are @xmath180 , @xmath252 , @xmath182 , @xmath253 and 12 nrg - iterations were performed . , title="fig : " ] and @xmath243 for different voltages . results for the spectral functions are shown for hf , gwa and snrg , while the self - energy is shown for the gwa only . ( b ) current as a function of voltage for a symmetric ( @xmath244 and @xmath243 ) and an asymmetric junction ( @xmath245 , @xmath246 and @xmath247 ) . @xmath248 is normalized to @xmath249 ( symmetric ) and @xmath250 ( asymmetric ) and measured in units of @xmath179 . a small temperature of @xmath251 was used for all calculations . parameters for the snrg calculations are @xmath180 , @xmath252 , @xmath182 , @xmath253 and 12 nrg - iterations were performed . , title="fig : " ] the imaginary part of the retarded self - energy @xmath254 $ ] for that junction obtained with the gwa is shown in the inset of fig . [ fig : weakcoupall](a ) for various voltages . the overall scale of @xmath254 $ ] is much smaller than @xmath206 and , therefore , the total self - energy @xmath255 is dominated by the charge - fluctuation scale @xmath206 . but the general influence of a finite bias voltage can already be observed here : the quasiparticle scattering amplitude is increased by the interplay between the voltage - induced fluctuations and the interaction . the characteristic fermi - liquid quadratic minimum in @xmath254 $ ] at the fermi level is destroyed with increasing voltage and the local fermi liquid prevailing in equilibrium ( @xmath256 ) is suppressed at large enough bias . the evolution of the minimum in @xmath254 $ ] with voltage bears some resemblance with a temperature evolution . the quasi - particle coherence is destroyed by a finite voltage in a similar fashion as with increasing temperature . the resulting iv characteristics of a symmetric and an asymmetric junction are shown in fig . [ fig : weakcoupall](b ) for @xmath257 . the current is normalized to @xmath258 and measured in units of @xmath206 . the rescaled current always saturates at @xmath259 for large voltages independent of @xmath0 ( not shown ) , as required by eq . ( [ eq : currentsym ] ) . the initial slope at zero voltage of the iv curve remains unaltered for all values of @xmath0 , in accordance with the fermi liquid nature of the model at small bias and zero temperature . while a symmetric junction with symmetric coupling to the leads always results in symmetric spectral functions , @xmath260 , and antisymmetric iv characteristics , @xmath261 ( see eq . ( [ eq : currentsym ] ) ) , an asymmetric junction in combination with asymmetric coupling yields a non - antisymmetric iv characteristics with @xmath262 . this is clearly visible in fig . [ fig : weakcoupall](b ) . the bias window ranges from @xmath263 to @xmath264 and is not symmetric around the fermi level . in combination with the shift of spectral weight to higher energies in @xmath265 the center of the spectral function is at @xmath266 this leads to a smaller contribution to the current for negative voltages . in contrast to the spectral functions , the iv characteristics of the snrg and gwa agree perfectly with the hf results for all voltages . the current is rather insensitive to the detailed distribution of spectral weight and measures only the total amount in the bias window @xmath267 $ ] . the snrg produces the correct results for small values of the interaction @xmath229 , and has thus no principal limitations . therefore , the expectation that it is reliable at arbitrary interaction strengths as well is warranted . as demonstrated in the previous section the interaction plays a minor role for small @xmath269 . on the other hand , with an odd number of electrons on the quantum dot and at very large @xmath269 and @xmath270 , the system develops a kondo effect as @xmath188 ( see for example ref . and ) . the snrg was shown @xcite to correctly describe the strongly correlated kondo regime out of equilibrium . the enhancement of the conductance in the coulomb - blockade region was reproduced for small bias , and the destruction of the many - body resonance at the fermi level with increasing voltage has been studied . in this section we will focus on the intermediate interaction regime , where correlations become increasingly important . since the diagrammatic keldysh approximations already show deficiencies in equilibrium see , for example , in sec . [ sec : equiscale ] or ref . discrepancies will extend to finite voltages . and @xmath271 at @xmath272 . the bias voltages are @xmath273 ( a ) , @xmath274 ( b ) and @xmath275 ( c ) . snrg - parameters are @xmath180 , @xmath181 , @xmath183 , @xmath225 and 8 nrg - iterations were performed . , title="fig : " ] + and @xmath271 at @xmath272 . the bias voltages are @xmath273 ( a ) , @xmath274 ( b ) and @xmath275 ( c ) . snrg - parameters are @xmath180 , @xmath181 , @xmath183 , @xmath225 and 8 nrg - iterations were performed . , title="fig : " ] + and @xmath271 at @xmath272 . the bias voltages are @xmath273 ( a ) , @xmath274 ( b ) and @xmath275 ( c ) . snrg - parameters are @xmath180 , @xmath181 , @xmath183 , @xmath225 and 8 nrg - iterations were performed . , title="fig : " ] the nonequilibrium spectral function for various voltages and intermediate interaction @xmath276 is shown in fig . [ fig : u4-i](a)-(c ) for a symmetric junction ( @xmath277 ) with symmetric coupling to the leads ( @xmath240 ) at @xmath272 . now all diagrammatic approximations yield different results . at low voltages , the snrg and @xmath142 approximation reproduce the slight humps at energies @xmath278 which are the first indicators of upper and lower hubbard satellites forming at large @xmath0 . the gwa only produces the broad high - energy tails , without the indication of forming separate peaks and the 2ba completely fails to produce the enlarged spectral weight at high energies . as the voltage is raised , the coulomb interaction causes additional dephasing , leading to increasingly broadened spectra . however , the @xmath142 approximation produces systematically too broad high - energy tails and an unphysical plateau around @xmath279 , which even develops a slight dip as seen in fig . [ fig : u4-i](c ) . we attribute this to a tendency to overestimate the coulomb repulsion . this might already be guessed from the equilibrium spectral functions , where the @xmath142 approximation unexpectedly produces the high - energy hubbard satellites for arbitrary large coulomb repulsion . these are connected to the ionic many - body states of the isolated atom which are not expected to be described by a second - order perturbation theory . however , the analytic structure of the retarded self - energy , eq . ( [ eq : sigr ] ) , ( [ eq : wrba ] ) , ( [ eq : pr ] ) and ( [ eq : gfu0 ] ) , has two direct consequences : ( i ) for small coupling to the leads or large @xmath0 it favors a @xmath280 behavior . this results in a two - peak structure with the peak - positions and widths roughly given by @xmath281 and @xmath206 , respectively . ( incorporating a screened and dynamic coulomb interaction , as it is done in the 2ba and gwa , leads to a prefactor smaller than @xmath282 and additional imaginary parts enter in the frequency dependence of @xmath283 . the hubbard satellites are then moved to lower energies and broadened . ) ( ii ) the fermi functions entering eq . ( [ eq : pr ] ) through @xmath284 lead to a narrowing of the integration interval for decreasing temperature . at zero temperature this always produces a vanishing imaginary part of the self - energy at the fermi level , @xmath285=0$],@xcite given that the noninteracting propagators are non - singular at @xmath279 . ( this reasoning also holds for the 2ba and gwa ) . the combination of ( i ) and ( ii ) gives rise to the two - peak structure in the spectral function for large @xmath0 and the emergence of an additional peak at the fermi level for low temperatures which is usually interpreted as the kondo resonance . but in principle there is no justification why the @xmath142 approximation should be reliable for large values of @xmath0 under arbitrary conditions . already for the asymmetric model in equilibrium the phase - space argument ( ii ) does not guarantee the correct description of the low - temperature fermi liquid anymore , and it is well - known that the @xmath142 approximation produces unphysical results.@xcite therefore , the large differences to all other methods at finite voltages , as it is observed here for @xmath219 , is not surprising . the snrg tends to produce additional features in the spectral function for large bias voltages at the positions of the chemical potentials of the leads , @xmath286 . these are the humps visible for larger voltages in the curves of fig . [ fig : u4-i ] . they are artifacts of the nrg discretization and dependent on the broadening procedure of the nrg spectral functions.@xcite in equilibrium , the nrg only resolves spectral information above a cutoff frequency @xmath287 where @xmath288 is on the order of the temperature @xmath289 . the nrg broadening parameters @xcite are usually adjusted such that artefacts are minimized . additionally the spectral function is interpolated between @xmath290 . this translates itself to the present implementation of the snrg which does not provide spectral information in the intervals @xmath291 $ ] centered around the two chemical potentials . here , @xmath292 and @xmath293 . furthermore , the time - dependent nrg introduces additional discretization errors@xcite which increase with increasing value of @xmath0 . @xmath55-averaging over different discretizations@xcite improves the spectral functions and these artifacts could be removed by adjusting the broadening parameter depending on the voltage . in this paper , however , we keep the broadening@xcite parameter fixed at @xmath184 independent of the bias and performed @xmath55-averaging with @xmath182 and @xmath294 . let us focus on the different behavior of the spectral functions around @xmath279 depicted in fig . [ fig : u4-i](a)-(c ) . the height @xmath217 is reduced for increasing @xmath242 , and the spectral functions of the snrg , 2ba and gwa approach each other and eventually coincide . for @xmath275 , the 2ba curve is still a little higher compared to snrg and gwa , but at even larger voltages ( not shown ) it also falls on top of the snrg and gwa . in contrast , the @xmath142 spectral function does not approach this large voltage limit , and the zero frequency value of the spectral function is considerably reduced compared to the other approaches . this is in accord with an enhanced scattering amplitude at the fermi level , visible in the imaginary part of the self - energy depicted in the insets . upon increasing the voltage , the @xmath142 approximation does not follow a systematic trend since @xmath217 is larger than the snrg at low voltages and smaller at high @xmath242 . the other approximations show systematic deviations as the 2ba is always larger than the snrg , while the gwa is always smaller . in the present calculations , the temperature @xmath272 is only about on fifth of the equilibrium kondo temperature for these parameter values , i.e. @xmath295 . as discussed in sec . [ sec : equiscale ] , the gwa produces a too small charge scale , which results in an even higher effective temperature . this leads to a reduction of the spectral function at the fermi level in addition to the effect of the small bias . the imaginary part of the self - energy is correspondingly too large compared to the snrg , as can be seen in the inset of fig . [ fig : u4-i](a ) . the 2ba , on the other hand , overestimates the low - energy scale , which explains the trends in @xmath217 and @xmath296 $ ] . increasing the current through the junction by applying a larger bias enhances the charge fluctuations on the local orbital . as already mentioned in section [ sec : weak - coupling ] , theses additional fluctuations introduce dephasing@xcite and destroy the coherent quasiparticles which constitute the low - temperature fermi liquid . the accompanying destruction of the characteristic quadratic minimum in @xmath297 $ ] around @xmath298 is observed in the insets . the system is driven away from the equilibrium fermi - liquid fixed point , and the spectral functions at the fermi level decreases . at very large voltage @xmath299 the coherent quasiparticles are completely suppressed , as it can be seen from the large imaginary part of the self - energy around @xmath300 . the spiky features in the snrg self - energy for @xmath275 are due to the aforementioned discretization errors and have no physical meaning . obtained from the spectral functions presented in fig . [ fig : u4-i ] for a quantum - point contact with @xmath243 . the current is normalized to @xmath249 and measured in units of @xmath179 . ] a coulomb interaction @xmath219 leads to a reduction of the current compared to its hf value as depicted in fig . [ fig : u4-iv ] . this is characteristic for the onset of the coulomb blockade . all approaches predict this reduction but slight differences can be noticed . the @xmath142 approximation overestimates the coulomb blockade resulting in a current which is systematically smaller than the snrg result . even though the @xmath142 spectral function differs strongly from all other approaches for large @xmath242 , this failure to describe the correct single - particle dynamics is concealed in the current as all approaches yield identical results . it again shows the insensitivity of the current to the detailed distribution of spectral weight in @xmath265 . the 2ba slightly overestimates the current for intermediate voltages , which is again explained by the too large low - energy scale @xmath195 and the accompanying underestimation of correlation effects . the gwa current merges with the snrg result for @xmath301 , which together with the satisfactory spectral function for these voltages indicates a good description of the nonequilibrium properties for intermediate to large @xmath242 . the spectral function for a quantum dot with a level position @xmath302 , coulomb interaction @xmath276 and asymmetric coupling @xmath247 is shown in fig . [ fig : u4asymgf ] . the asymmetry between positive and negative voltages is directly visible in the spectral functions . , @xmath302 , with asymmetric coupling @xmath247 for small ( upper ) and large ( lower ) bias , @xmath303 and @xmath304 , respectively . the vertical dashed lines indicate the location of the left and right chemical potentials . the inset shows the total occupation @xmath305 of the impurity as a function of the bias voltage obtained with the diagrammatic approaches . snrg - parameters are the same as for fig . [ fig : u4-i ] . , title="fig : " ] , @xmath302 , with asymmetric coupling @xmath247 for small ( upper ) and large ( lower ) bias , @xmath303 and @xmath304 , respectively . the vertical dashed lines indicate the location of the left and right chemical potentials . the inset shows the total occupation @xmath305 of the impurity as a function of the bias voltage obtained with the diagrammatic approaches . snrg - parameters are the same as for fig . [ fig : u4-i ] . , title="fig : " ] apart from voltage - induced broadening which was already discussed in previous sections , an additional shift of spectral weight in @xmath265 is observed with increasing @xmath242 . while the snrg moves spectral weight to higher energies for positive bias and towards @xmath279 for negative @xmath242 , the diagrammatic approaches produce the opposite trends . due to the stronger coupling to the left lead , @xmath306 , the left - moving scattering states dominate the excitations and the spectral function of the impurity orbital , as can be seen from eq . ( [ eqn : d - lr - expand ] ) and ( [ eqn : d - alpha - cont ] ) . the effective noninteracting single - particle excitation energy of an @xmath64-mover is given by @xmath307 , which implies almost symmetric parameters for the left - movers at the negative voltage @xmath308 since then @xmath309 . therefore , the asymmetry of the spectral function is expected to be reduced and @xmath265 to be closer to that of a symmetric junction , which is indeed observed in the snrg . the diagrammatic approaches underestimate correlations in the ionic many - body states , and the occupancy of the impurity is overestimated for negative voltages . it increases almost linearly with negative voltage as can be seen from the inset of fig . [ fig : u4asymgf](b ) . therefore , the hartree shift ( [ eq : sighartree ] ) also increases , and spectral weight is moved towards higher energies opposite to what would be expected from the physical argument presented above . as a consequence the spectral functions are strongly attracted to @xmath310 . the effective single - particle excitation energy of a left - moving scattering state for a positive voltage @xmath275 is greater than zero , @xmath311 . this produces an intermediate valence situation for the left - movers , where correlations renormalize the effective excitation energies to even larger frequencies@xcite and a shift of spectral weight to higher energies results . an additional drag of spectral weight towards the chemical potential of the weaker coupled right lead , @xmath312 , is expected . the snrg produces such a shift as can be seen in fig . [ fig : u4asymgf](b ) . the diagrammatic approaches , however , underestimate the level - renormalization in the presence of strong valence fluctuations , a tendency already observable in equilibrium ( not shown ) . additionally , the reduced occupancy ( inset ) diminishes the hartree energy which again leads to a shift towards the stronger coupled chemical potential @xmath313 . , @xmath276 and @xmath247 , calculated with the spectral functions depicted in fig . [ fig : u4asymgf ] . the result for the symmetric junction already shown in fig . [ fig : u4-iv ] is included for comparison . ( b ) the leakage current @xmath314 for the keldysh approaches obtained from eq . ( [ eq : currentleak ] ) . the currents are normalized to @xmath250 and measured in units of @xmath179 . , title="fig : " ] , @xmath276 and @xmath247 , calculated with the spectral functions depicted in fig . [ fig : u4asymgf ] . the result for the symmetric junction already shown in fig . [ fig : u4-iv ] is included for comparison . ( b ) the leakage current @xmath314 for the keldysh approaches obtained from eq . ( [ eq : currentleak ] ) . the currents are normalized to @xmath250 and measured in units of @xmath179 . , title="fig : " ] figure [ fig : u4asymiv ] shows the iv characteristics of this junction . the asymmetry of @xmath315 is clearly visible when compared to the result from the symmetric junction ( also included in the plot ) . for @xmath316 , the rescaled current is very close to its values from the symmetric junction and the discrepancies between the 2ba , gwa and snrg follow the already discussed characteristics : the 2ba underestimates correlations and yields a slightly too large current . the gwa has the correct distribution of spectral weight in the bias window and produces a rather good estimate for the current , despite its deficiencies in the description of the single - particle spectra . the failure to produce the correct shifts of spectral weights in @xmath265 causes the current in the diagrammatic approaches to be smaller than the snrg for negative voltages . the @xmath142 approximation , however , reveals its non - conserving nature in the violation of current conservation for this asymmetric junction . this is illustrated in fig . [ fig : u4asymiv](b ) displaying the leakage current @xmath317 of eq . ( [ eq : currentleak ] ) . in contrast to the conserving 2ba and gwa methods and the snrg , @xmath317 does not vanish for the @xmath142 approximation ! thus , left and right current of eq . ( [ eq : currentsym - def ] ) do not have the same magnitude , i.e. @xmath318 , and the current calculated from eq . ( [ eq : ss - current ] ) does not make sense , since different linear combinations @xmath319 ( @xmath320 ) yield different results . therefore , we did not include the calculated iv curves in fig . [ fig : u4asymiv](a ) . increasing the asymmetry further , i.e. @xmath321 , recovers the equilibrium spectral functions of a quantum dot coupled to a single lead in all approaches ( not shown ) . in the snrg , the backscattering term @xmath71 , eq . ( [ eq : back - scattering ] ) , is suppressed , and the model approaches an equilibrium single - channel problem . in the diagrammatic approaches , the nonequilibrium conditions enter only through the effective fermi function @xmath322 , eq . ( [ eqn : f - eff ] ) , which approaches its equilibrium value for @xmath323 . in this regime , all differences in the spectral functions of the presented approaches are given by the known discrepancies already present in equilibrium . in the recently developed snrg approach to open quantum systems the scattering states of a noninteracting quantum impurity model are used to construct the nonequilibrium green s functions for the steady state at finite bias voltage . we have established the reliability of the snrg by benchmarking it against the diagrammatic kadanoff - baym - keldysh approach , which becomes exact in the limit @xmath1 . it has been shown that the spectra and the current - voltage characteristics agree excellently for small coulomb interactions for symmetric and asymmetric junctions at arbitrary bias voltage . for intermediate values of @xmath0 we have compared the snrg to three different approximations obtained from the keldysh approach , namely the second - order perturbation theory ( @xmath142 ) , the fully self - consistent second - order ( 2ba ) and the gw approximation ( gwa ) . as correlation effects play an increasingly important role discrepancies occur between the different methods . these were explained by the insufficient treatment of the coulomb interaction within the diagrammatic approaches . the fermi liquid at zero bias voltages is characterized by a single low - energy scale which is captured accurately by the snrg , but is not properly reproduced by the diagrammatic approaches . no single low - energy scale can be extracted from the @xmath142 approximation and the gwa at intermediate and large @xmath0 . while the scale associated with charge fluctuations decreases with increasing @xmath0 , the magnetic scale exhibits a qualitatively different @xmath0-dependency , since it develops a minimum for intermediate @xmath0 and increases again towards larger @xmath0 ! the gwa shows a tendency to over - screen magnetic moments for increasing values of @xmath0 and fails to reproduce the atomic limit . these deficiencies translate themselves to finite bias and explain the discrepancies at small to intermediate voltages . at large bias voltages the self - consistent diagrammatic approaches ( 2ba and gwa ) reproduce the snrg spectral functions for a symmetric junction , while the second - order perturbation theory yields an unphysical plateau around the fermi level . all diagrammatic approximations and the snrg capture the onset of the coulomb blockade in the iv characteristics of the symmetric junction . the small discrepancies are explained by the deficiencies in the treatment of the interaction . however , the failure of the @xmath142 approximation to correctly describe the single - particle dynamics at large bias is masked in the current , since there only the total spectral weight in the bias window enters . in contrast to the other methods , the @xmath142 approximation reveals its non - conserving nature by producing a finite leakage current for an asymmetric junction , which is unphysical . this raises the question about the reliability of the results obtained within that method or extensions of it,@xcite even for a symmetric junction . they are only well - justified for cases where @xmath324 for all frequencies . the voltage dependent redistribution of spectral weight for an asymmetric junction is not well - reproduced by the diagrammatic approaches . this has been attributed to too large hartree shifts due to the wrong occupation number of the impurity and the inaccurate renormalization of the single - particle level in intermediate - valence situations . it leads to the underestimation of the current for large negative voltages . the snrg provides access to the description of nonequilibrium steady - state properties of nanoscale junctions for arbitrary coulomb interaction and voltages . it opens promising perspectives for future investigations , such as the influence of charge fluctuations when approaching the strongly - correlated regime or the effects of an applied magnetic field . we are grateful to j. e. han , a. millis , a. schiller , p. schmitteckert , g. schn , h. schoeller , p. werner , m. wegewijs and g. zarand for helpful discussions . we acknowledge financial support from the deutsche forschungsgemeinschaft under an 275/6 - 1 and supercomputer support by the nic , forschungszentrum jlich under project no . hhb000 .
stochastic geometry has recently been used for the analysis and performance evaluation of wireless ( ad hoc as well as cellular ) networks ; in this approach , one models node locations as a spatial point process , e.g. , homogeneous poisson point processes , and one computes various network statistics , e.g. , interference , successful transmission probability , coverage ( or , outage ) probability etc . as spatial averages . this often leads to tractable performance metrics that are amenable to parametric optimization with respect to network parameters ( node density , protocol parameters , etc . ) . more precisely , this approach yields spatial averages of the performance metrics for given network parameters ; then the parameters can be chosen to optimize performance . this approach takes a macroscopic view of the network with the underlying assumption that all nodes in the network have identical statistical characteristics . in practice , due to randomness and heterogeneity in networks , nodes need to adapt to local spatial and temporal conditions ( e.g. , channel conditions and topology ) to reach optimum network wide performance . for example , nodes in wireless lans adjust their window sizes based on acknowledgment feedback ; in cellular networks nodes are scheduled based on channel conditions and adapt their transmit powers based on the measured sinrs , which in turn depend on the transmit powers set by other nodes . in all such scenarios , distributed adaptive algorithms are used to reach a desired network wide operating point e.g. that maximizing some utility . while the behavior of such distributed optimization protocols is often well understood on a given topology , there are usually no analytical characterizations of the statistical properties of the optimal state in large random and heterogeneous networks . the main aim of this work is to use stochastic geometry to study spatial adaptations of medium access control in aloha that aim at optimizing certain utilities . while we identify a utility for which stochastic geometry can be used to compute the spatial distribution of map and the expected utility , we are far from being able to do so for all types of utilities within the @xmath0-fair class and we discuss the difficulties to be faced . let us start with a review of the state of the art on aloha . wireless spectrum is well known to be a precious and scarce shared resource . medium access control ( mac ) algorithms are employed to coordinate access to the shared wireless medium . an efficient mac protocol should ensure high system throughput , and should also distribute the available bandwidth fairly among the competing nodes . the simplest of the mac protocols , aloha and slotted aloha , with a `` random access '' spirit , were introduced by abramson @xcite and roberts @xcite respectively . in these protocols , only one node could successfully transmit at a time . reference @xcite modeled node locations as spatial point processes , and also modeled channel fadings , interferences and sinr based reception . this allowed for spatial reuse and multiple simultaneous successful transmissions depending on sinr levels at the corresponding receivers . all the above protocols prescribe identical attempt probabilities for all the nodes . reference @xcite further proposed opportunistic aloha in which nodes transmission attempts are modulated by their channel conditions . among the initial attempts of map adaptation in aloha , reference @xcite analyzed protocol model and proposed stochastic approximation based strategies that were based on receiver feedback and were aimed at stabilizing the network . references @xcite also optimized nodes attempt probabilities ( or thresholds ) in order to maximize the spatial density of successful transmissions . reference @xcite analyzed both plain and opportunistic aloha in a network where all the nodes communicate to one access point . they assumed statistically identical rayleigh faded channels with no dependence on geometry ( i.e. , no path loss components ) . they demonstrated a paradoxical behavior where plain aloha yields better aggregate throughput than the opportunistic one . reference @xcite also studied optimal random access with sinr based reception . however , they considered constant channel gains . they developed a centralized algorithm that maximizes the network throughput , and also an algorithm that leads to max - min fair operation . reference @xcite modeled network as an undirected graph and studied aloha under the protocol model . they designed distributed algorithms that are either proportional fair or max - min fair . reference @xcite built upon the model of @xcite , and formulated the channel access problem as a non - cooperative game among users . they considered throughput and delay as performance metrics and proposed pricing schemes that induce socially optimum behavior at equilibrium . however , they set time average quantities ( e.g. , throughput , delay ) as utilities ( or costs ) , and concentrated on symmetric nash equilibria . consequently , in their analysis , dependence on local conditions vanishes . in none of the above aloha models , nodes account for both wireless channel randomness and local topology for making their random access decisions , as we do in the present paper . there is a vast literature on the modeling of csma by stochastic geometry which will not be reviewed in detail here . the very nature of this mac protocol is adaptive as each node senses the network and acts in order to ensure that certain exclusion rules are satisfied , namely that neighboring nodes do not access the channel simultaneously . however , csma as such is designed to guarantee a reasonable scheduling , not to optimize any utility of the throughput . the closest reference to our work is probably @xcite where the authors study an adaptation of the exclusion range and of the transmit power of a csma node to the location of the closest interferer . this adaptation aims at maximizing the mean number of nodes transmitting per unit time and space ( while respecting the above exclusion rules ) . this mean number is however only a surrogate of the rate . in addition , the adaptation is only w.r.t . to the location of the nearest interferer . we study spatial adaptation of aloha in ad hoc networks . the network setting is described in section [ sec : network - model ] . we consider quasi - static networks in which mobiles learn the topology , and incorporate this information in their medium access probability ( map ) selection . we consider the cases where nodes are benevolent and cooperate in a distributed way to maximize the global network throughput or to reach either a proportional fair or max - min fair sharing of the network resources . we analyze the case where nodes account only for their closest interferers , for all nodes in a given ball around them , or even all nodes in the network . section [ sec : aggr - interf ] is focused on the distributed algorithms that maximize the aggregate throughput or lead to max - min fairness in such networks . in the proportional fair case , we show that nodes can compute the optimal maps as solutions to certain fixed point equations . in the maximum throughput case , the optimal maps are obtained through a gibbs sampling based algorithm . in the max min case , the optimal maps are obtained as the solution of a convex optimization problem . section [ ss : stocg ] contains the stochastic geometry results . the model features nodes forming a realization of a homogeneous poisson point process in the euclidean plane . we compute the map distribution in such a network in the proportional fair case using shot noise field theory . to the best of our knowledge , this distribution is the first example of successful combination of stochastic geometry and adaptive protocol design aimed at optimizing ceratin utility function within this aloha setting . we also show that the mean value of the logarithm of the throughput obtained by a typical node can be derived from this distribution . finally , we discuss the difficulties to be faced in order to extend the result to other types of utilities . the numerical results are gathered in section [ ss : numerics ] . the aim of this section is two - fold : 1 ) check the analytical results against simulation and 2 ) quantify the gains brought by adaption within this setting . we model the ad - hoc wireless network as a set of transmitters and their corresponding receivers , all located in the euclidean plane . this is often referred to as `` bipole model '' ( * ? ? ? * chapter 16 ) . there are @xmath1 transmitter - receiver pairs communicating over a shared channel . the transmitters follow the slotted version of the aloha medium access control ( mac ) protocol ( see figure [ fig : bipole - model ] ) . a transmitter , in each transmission attempt , sends one packet which occupies one slot . each transmitter uses unit transmission power . we assume that each node has an infinite backlog of packets to transmit to its receiver . the euclidean distance between transmitter @xmath2 and receiver @xmath3 is @xmath4 , and the path - loss exponent is @xmath0 ( @xmath5 ) . we also assume rayleigh faded channels with @xmath6 being the random fading between transmitter @xmath2 and receiver @xmath3 . moreover , we assume that the random variables @xmath7 are independent and identically distributed with mean @xmath8 . thus all @xmath6s have cumulative distribution function ( cdf ) @xmath9 with @xmath10 . all the receivers are also subjected to white gaussian thermal noise with variance @xmath11 , which is also constant across slots . we assume that a receiver successfully receives the packet of the corresponding transmitter if the received sinr exceeds a threshold @xmath12 . let @xmath13 be the indicator variable indicating whether transmitter @xmath3 transmits in a slot , and @xmath14 be @xmath3 s medium access probability . thus @xmath15 when node @xmath3 transmits , the received sinr at the corresponding receiver is @xmath16 then the probability of successful reception @xmath17 can be calculated as follows . @xmath18 thus @xmath19 where @xmath20 further simplifying , @xmath21 then , the rate or throughput of transmitter @xmath3 is given by @xmath22 . the thermal noise appears merely as a constant multiplicative factor in the expression for the successful transmission probability ( see ) . moreover , in interference limited networks , the impact of thermal noise is negligible as compared to interference . we focus on such networks , and thus we ignore the thermal noise factor throughout . in this section , we analyze adaptations of spatial aloha that maximize aggregate throughput or achieve proportional fairness or max - min fairness . the throughput maximizing medium access probabilities solve the following optimization problem . we first argue that the optimum in the above optimization problem is attained at one of the vertices of the hypercube formed by the constraint set . to see this , suppose @xmath24^{\mathcal{n}}$ ] is an optimal solution , and @xmath25 for some @xmath26 . clearly , @xmath27 since the partial derivative is independent of @xmath14 , @xmath14 can be set to either @xmath28 or @xmath29 without reducing the value of the objective function . this proves our claim . in the following we focus only on such extreme solutions . then the above problem is equivalent to finding an @xmath30 such that @xmath31 if and only if @xmath32 is an optimal solution . thus we are interested in @xmath33 [ [ an - iterative - solution ] ] an iterative solution + + + + + + + + + + + + + + + + + + + + + we can pose this problem as a strategic form game with the users as players @xcite . for each player its action @xmath34 lies in @xmath35 , and the utility function @xmath36 is given by @xmath37 this is a potential game with the above objective function as the potential function @xcite . thus the best response dynamics converges to a nash equilibrium . this algorithm can be implemented in a distributed fashion if each node @xmath3 knows @xmath38 for all @xmath2 , and also @xmath39 and @xmath40 for all @xmath41 after each iteration . however , a nash equilibrium can be a suboptimal solution to the above optimization problem . to alleviate this problem , we propose a gibbs sampler based distributed algorithm , wherein each node @xmath3 chooses action @xmath29 with probability @xmath42 the parameter @xmath43 is called the temperature . the gibbs sampler dynamics converges to a steady state which is the gibbs distribution associated with the aggregate throughput and the temperature @xmath43 @xcite . in other words , we are led to the following distribution on the action profiles : @xmath44 where @xmath45 is a normalizing constant . when @xmath43 goes to @xmath28 in an appropriate way ( i.e. , as @xmath46 , where @xmath47 is the time ) , the distribution @xmath48 converges to a dirac mass at the action profile @xmath49 with maximum aggregate utility if it is unique . notice that the aggregate utility @xmath50 equals the aggregate throughput . thus the action profile @xmath49 is a solution to the original throughput optimization problem . the first two terms in the utility function @xmath51 can be seen as `` selfish '' part of user @xmath3 , whereas the last summation term is `` altruistic '' part . the user makes a decision based on whether the `` selfish '' part dominates or viceversa . in a quasi - static network where topology continuously changes , although at a slower time scale , different sets of nodes are likely to be scheduled to transmit under different topologies . thus , in terms of long term performance , maximum throughput medium access is not grossly unfair . the proportional fair medium access problem can be formulated as follows . @xmath52 the objective function can be rewritten as @xmath53 we thus have a convex separable optimization problem . the partial derivative of the objective function with respect to @xmath14 is @xmath54 which is continuous and decreasing in @xmath14 over @xmath55 $ ] . we conclude that at optimality , for each @xmath56 , @xmath57 observe that user @xmath3 s optimal attempt probability is independent of the attempt probabilities of other users . in particular , if @xmath58 , user @xmath3 can perform iterations @xmath59 autonomously . furthermore , @xmath60 clearly , @xmath61 , i.e. , @xmath62 is a contraction . thus the fixed point iterations converge to the optimal @xmath14 starting from any @xmath63 $ ] . the characterization of the optimal attempt probabilities reflects the altruistic behavior of users . more precisely , user @xmath3 s attempt probability is a function of @xmath64 which are measures of @xmath3 s interference to all other users . in particular , if @xmath65 , i.e. , if @xmath3 s transmission does not cause significant interference to the other users , then @xmath3 transmits in all the slots . unlike the throughput maximization problem , there is no `` selfish '' component in the decision making rule . as the target sinr @xmath66 , @xmath67 for all @xmath68 , and the proportional fair attempt probabilities satisfy @xmath69 for all @xmath26 . this yields @xmath70 for all @xmath26 . this is expected , because in the limiting case a transmission can succeed if and only if there is no other concurrent transmission . this is hence aloha without spatial reuse , and it is well known that in this case , the optimal access probability is @xmath71 asymptotically @xcite . our analysis in this section follows @xcite . the max - min fair medium access problem can be formulated as @xmath72 where constraint functions are defined for all @xmath73^{\mathcal{n}}$ ] . the following is an equivalent convex optimization problem ( see @xcite for details ) : @xmath74 the lagrange function of this problem is given by @xcite @xmath75 with @xmath76 being the lagrange multipliers . minimization of the lagrange function ( which is concave in @xmath77 and @xmath78 ) gives @xmath79 wang and kar @xcite suggest that the lagrange multipliers be updated using the gradient projection method . more precisely , for all @xmath26 , @xmath80^+ , \label{eqn : aggr - interf - lambda - update}\end{aligned}\ ] ] where @xmath81 is the step size at the @xmath82th iteration . further more , ( * ? ? ? * theorem 2 ) implies that a solution arbitrary close to an optimal solution can be reached via appropriate choice of step sizes . however , all the users need to exchange variables in order to perform updates . finally , the _ directed link graph _ corresponding to our network is a directed graph in which each vertex stands for a user ( i.e. , a transmitter - receiver pair ) in the network . there is an edge from vertex @xmath3 to vertex @xmath2 in the directed link graph if transmission of user @xmath3 affects the success of transmission of user @xmath2 . two vertices @xmath3 and @xmath2 are said to be connected if either of the following two conditions hold : 1 . there is an edge from @xmath3 to @xmath2 or viceversa , 2 . there are vertices @xmath83 such that @xmath84 and @xmath85 are connected for @xmath86 . clearly , the directed link graph for our network model is a complete graph ; for any pair of vertices @xmath3 and @xmath2 there is an edge from @xmath3 to @xmath2 and also from @xmath2 to @xmath3 . in particular , the directed link graph is a single _ strongly connected component _ @xcite . thus ( * ? ? ? * corollary 1 ) implies that the above optimization also obtains the lexicographic max - min fair medium access probabilities that yield identical rates for all the users . note that a user needs to know the entire topology , and in a few cases , also needs to communicate with all the nodes to implement the adaptation rules developed in sections [ sec : aggr - interf - max - thput]-[sec : aggr - interf - max - min - fair ] . in this section , we carry out analysis under the simplifying assumption that the aggregate interference at a receiver is dominated by the transmission from the closest interferer . this is a reasonable approximation in a moderately dense network , specifically when the path loss attenuations are high . throughout this section , we use the notation @xmath87 for all @xmath56 . in words , @xmath88 is the strongest interferer of node @xmath3 , and @xmath89 is the set of nodes to which node @xmath3 is the strongest interferer . we assume that there is always a unique @xmath88 for each @xmath3 . then , accounting only for the nearest interferer , the approximate probability of successful transmission for node @xmath3 is @xmath90 the analysis of sections [ sec : aggr - interf ] can be adapted to this simplified scenario . the throughput maximization problem can now be posed as follows . @xmath91 as in section [ sec : aggr - interf - max - thput ] , we can argue that some @xmath92 attains the optimal throughput . again , an equivalent optimization problem is @xmath93 or alternatively , @xmath94 we now formulate a strategic form game among users , with action sets @xmath35 and utility functions given by @xmath95 again a gibbs sampler based algorithm yields the optimal set of the transmitting users . also , user @xmath3 only needs to know the distances of user @xmath88 and all the receivers in @xmath89 and their actions to make its decision . notice that user @xmath3 must choose @xmath96 if @xmath97 such users can set their actions to @xmath29 , and need not undergo gibbs sampler updates . [ [ discussion ] ] discussion + + + + + + + + + + consider a scenario where a node s closest interferer does not transmit , i.e. , has zero attempt probability . nonetheless , this node always has an active closest interferer ( unless there are no other nodes in the network ) . a better approximation of the success probabilities , and hence of the throughput , is obtained by always accounting for the closest active interferer . towards this , let us define @xmath98 for all @xmath32 . we are now faced with the following optimization problem . @xmath99 we can now define users utility functions as follows . @xmath100 where @xmath101 . the analogous distributed algorithm ( gibbs sampler based ) can again be shown to lead to the optimal solution . we now aim to solve the following optimization problem . @xmath102 following the discussion in section [ sec : aggr - interf - prop - fair ] , we obtain @xmath103 again , if @xmath104 , iterations @xmath59 converge to the optimal @xmath14 starting from any @xmath63 $ ] . if @xmath105 , i.e. , if @xmath3 s transmission does not cause significant interference to the users for whom @xmath3 is closest interferer , then @xmath3 transmits in all the slots . the same user may not transmit ( in any slot ) under the maximum throughput objective if @xmath106 and @xmath88 and users in @xmath89 transmit . we can have explicit formulae for the attempt probabilities in a few special cases . 1 . suppose @xmath89 is singleton for each @xmath3 . if @xmath107 , then @xmath108 2 . _ linear network topology : _ we now consider a scenario where nodes are placed along a line , say @xmath109 , and are indexed sequentially . we also assume that for any node @xmath3 the potential interferers are the two immediate neighbors @xmath110 and @xmath111 . this also amounts to assuming @xmath112 for all @xmath3 . then , assuming @xmath113 , the proportional fair attempt probability of node @xmath3 satisfies @xmath114 this is quadratic equation in @xmath14 , which on solving gives @xmath115 similar to section [ sec : aggr - interf - max - min - fair ] , the max - min fair medium access problem is @xmath116 again , the constraint functions are defined for all @xmath73^{\mathcal{n}}$ ] . let us recall the definition of the directed link graph associated with the network . in this section , we only account for the interference due the closest interferer . thus there is an edge from vertex @xmath3 to vertex @xmath2 if and only if @xmath117 . we assume that the directed link graph is connected ( i.e. , all the vertices in the graph are connected with each other ) . if it is not connected , the max - min fair medium access problem on the entire graph decomposes into separate max - min fair medium access problems on each of the connected subgraphs , which can be solved independently . we now pursue the following convex optimization problem which is equivalent to the above max - min fair medium access optimization problem . @xmath118 the last set of constraints along with the connected assumption ( of the directed link graph ) forces @xmath119 to be equal for all @xmath26 . this confirms equivalence to the initial optimization problem . now the lagrange function is @xmath120 where @xmath121 , are the lagrange multipliers . an approach similar to section [ sec : aggr - interf - max - min - fair ] prescribes the following update rules . for all @xmath122 , @xmath123^+ , \label{eqn : close - interf - lambda - update } \\ \mu_{ij}(n+1 ) = & [ \mu_{ij}(n ) + \beta(n)(\theta_i - \theta_j)]^+ , \label{eqn : close - interf - mu - update}\end{aligned}\ ] ] where @xmath81 is the step size at the @xmath82th iteration as before . observe that any user @xmath3 can perform updates - via local information exchange . more precisely , it needs to communicate only with user @xmath88 and the users in @xmath89 . the aim of the present paper is primarily of theoretical nature and it is beyond our scope to discuss implementation issues . let us however stress that the discussed adaptations of the map are implementable . we will focus on the proportional fair case in view the main focus of the paper . assume each receiver has a distinctive pilot signal with fixed power . since we assumed a quasi - static network in which the nodes move at a slower time scale , each node can then learn the distance that separates it from a given receiver by listening to its pilot signal and by performing a time average over ( so as to smooth out fading ) . once this data is available for all receivers , a given transmitter can then solve the key fixed point equation that characterizes its optimal map . distinctive pilot signals can be obtained by a collection of orthogonal codes chosen at random by the receivers . in practice , it is enough for a transmitter to detect the `` dominant '' receivers ( i.e. , those within a certain distance to it ) , so that the scheme will work in an infinite network with a finite ( properly tuned ) number of such codes . we now assume that the transmitting nodes are scattered on the euclidian plane according to a homogeneous poisson point process of intensity @xmath124 . for each transmitter , its corresponding receiver is at distance @xmath125 in a random direction . the traffic and channel models are the same as in section [ sec : network - model ] . as before the transmitters use slotted aloha to access the channel , and a receiver successfully receives the packet from its transmitter if the received sinr exceeds a threshold @xmath12 . finally , transmitters adapt their attempt probabilities as described in section [ sec : aggr - interf ] . each transmitter is associated with a multi dimensional mark that carries information about the adaptive transmission probability and the transmission status . let @xmath126 denote a marked poisson point , where * @xmath127 denotes the poisson point process of intensity @xmath124 , representing the location of transmitters in the euclidean plane . * @xmath128 denote the marks of the poisson point process @xmath129 , which consist of three components : * * @xmath130 denote the angles from transmitters to receivers . these angles are i.i.d . and uniform on @xmath131 $ ] and independent of @xmath132 . we will call them the primary marks . * * @xmath133 denote the maps of the nodes ; @xmath134 is a secondary mark ( i.e. a functionals of @xmath132 and its primary marks , see below ) . * * @xmath135 are indicator functions that take value one if a given node decides to transmit in a given time slot , and zero otherwise . clearly , @xmath136 . in particular , given @xmath134 , @xmath137 is independent of everything else including @xmath138 . the locations of the receivers will be denoted by @xmath139 with @xmath140 . it follows from the displacement theorem @xcite that @xmath141 is also a homogeneous poisson point process of intensity @xmath124 . the above assumptions will be referred to as the poisson model . we will also consider below a more general case where the above marked point process is just stationary . let us consider response functions @xmath142 defined for each @xmath143 as follows @xmath144 for all @xmath145 , the shot noise field @xmath146 associated with the above response function and the marked point process @xmath129 is @xmath147 notice that this shot noise is not that representing the interference at @xmath148 . it rather measures the effect of the presence of a transmitter at @xmath148 on the whole set of receivers . consider a typical node at the origin , @xmath149 , with marks @xmath150 . let @xmath151 denote the palm distribution of the stationary marked point process @xmath129 ( * ? ? ? * chapter 1 ) . the fixed point equation determining the map of node @xmath152 reads ( see ) @xmath153 we have a similar equation for each node and the sequence @xmath133 is readily seen to be a sequence of marks of @xmath132 and its primary marks . it follows directly from monotonicity arguments that @xmath154 notice that we have not used the specific assumptions on the point process so far . hence we have the following general connection between the optimal map distribution and the shot noise @xmath155 : for all stationary marked point processes @xmath156 ( not necessarily poisson ) , for all @xmath157 , @xmath158 with @xmath151 the palm distribution of @xmath129 . we now use the fact that @xmath129 is an independently marked poisson point process ( * ? ? ? * definition 2.1 ) . from slivnyak s theorem ( * ? ? ? * theorem 1.13 ) , @xmath159 for all @xmath160 . consequently , @xmath161 it follows from ( * ? ? ? * proposition 2.6 ) and from the fact that @xmath162 is a homogeneous poisson point process that one can write the laplace transform @xmath163 of the shot noise @xmath164 as @xmath165 where @xmath166 . [ thm : map - distribution ] under the above poisson assumptions , the attempt probability of the typical node has the distribution @xmath167 with @xmath168 given by ( [ eq : lapl ] ) . let @xmath169 denote the density of the shot noise field @xmath170 . then @xmath171 where @xmath172 if @xmath173 and 0 otherwise . now using parseval s theorem @xmath174 with @xmath175 the fourier transform of the real valued random variable @xmath176 and @xmath177 the complex conjugate of @xmath178 . the claim follows after substituting @xmath179 and @xmath180 . [ lem : laplace - transform ] for @xmath181 , the laplace transform @xmath182 can be simplified as @xmath183 this subsection is devoted to the analysis of the mean value of the logarithm of the throughput of the typical node : @xmath184 with @xmath185 defined in . since we know the cdf @xmath186 of @xmath187 , the first term poses no problem . the second term can be rewritten as ( see ) @xmath188.\ ] ] under the law @xmath151 , the points @xmath189 of @xmath132 form a homogeneous poisson point process of intensity @xmath124 . however , the marks @xmath190 do not have the law identified in the last section . in fact , the mark @xmath134 of a point @xmath191 satisfies the following modified fixed point equation : @xmath192 with the convention that @xmath193 if there is no solution in @xmath55 $ ] . we can use the same argument as above to conclude that @xmath194 iff @xmath195 conditioned on there being two nodes at @xmath28 and @xmath148 , the other points form a homogeneous poisson point process of intensity @xmath124 . this allows one to prove the following . [ thm : map - typical - receiver ] under the above poisson assumptions , given that there is a node at 0 and a node at @xmath196 , the attempt probability of the node at @xmath148 has the distribution @xmath197 with @xmath198 and @xmath140 . due to the circular symmetry , the first integral in the expression of @xmath199 depends on @xmath148 only through @xmath200 . thus the density of @xmath201 also depends on @xmath200 only , and it will be denoted by @xmath202 when @xmath203 . the density of @xmath187 identified in the last subsection will be denoted by @xmath186 . the main result of this section is : [ thm : mean - utility ] under the above poisson assumptions , in the proportional fair case , the mean utility of a typical node is @xmath204 see appendix [ proof : thm - mean - utility ] . a preliminary concern when trying to use euclidean stochastic geometry of the infinite plane in the maximum throughput and max - min fairness cases is that it is not clear whether the associated infinite dimensional optimization problems make sense in the first place . in the proportional fair case , each node computes its optimizing map in one step as the solution of a fixed point equation that is almost surely well defined ( in terms of a shot noise ) even in the infinite poisson population case . unfortunately , this does not extend to the other two cases . this does not mean that there is no hope at all . consider for instance the maximum throughput case accounting only for the closest interferer , and further simplify it by measuring interference at the transmitter rather than at the receiver . in other words , consider the same optimization problem as in section [ sec : close - interf - max - thput ] but with @xmath88 being the closest transmitter to transmitter @xmath3 ( rather than to receiver @xmath3 ) . then , in the poisson case , the infinite dimensional optimization problem can be shown to reduce to a _ countable collection of finite optimization problems_. this follows from the fact that there are no infinite `` descending chains '' in a homogeneous poisson point process ( * ? ? ? * chapter 2 ) ( a _ descending chain _ is a sequence of nodes @xmath205 such that @xmath206 for all @xmath82 ) . as a result , the poisson point process can be decomposed into a countable collection of finite `` descending trees '' , where each path from the leaves to the root is a descending chain . the associated optimization is hence well defined and the problem can be reduced to evaluating the solution of the optimization problem of section [ sec : close - interf - max - thput ] on the typical descending tree . hence , there is hope to progress on this and on related cases . this will however not be pursued here and is left for future research . in this section , we study the proposed adaptive spatial aloha schemes quantitatively . we compute various metrics formulated through the stochastic geometry based analysis , and we also perform simulation . the simulation not only validates the analytical model , but also illustrates the performance of schemes for which we do not have an analytical characterization . we used maple and matlab to evaluate the integrals of section [ ss : stocg ] . the infinite integral that shows up in the expression of the laplace transform ( [ eq : lapl ] ) is handled without truncation by maple and matlab . the singularity at @xmath207 in the contour integrals ( [ eq : parsev ] ) leveraging parseval s theorem is a false singularity and it is also handled without further work by either maple or matlab . the matlab code is particularly efficient and is used throughout the analytical evaluations described below . we consider a two dimensional square plane with side length @xmath208 , and @xmath1 nodes placed independently over the plane according to the uniform distribution ; this corresponds to @xmath209 in the stochastic geometry model . , the poisson random variable with mean @xmath210 is highly concentrated around its mean . thus we can use @xmath210 nodes for all the realizations in our simulation . ] each node has its receiver randomly located on the unit circle around it , again as per the uniform distribution . thus @xmath211 for all @xmath3 . we set @xmath181 and @xmath212 . to nullify the edge effect , we take into account only the nodes falling in the @xmath213 square around the center while computing various metrics . while all other parameters remain , we vary @xmath208 and @xmath1 for different simulations . for each parameter set we calculate the average of the performance metric of interest over @xmath214 independent network realizations . we validate the analytical expression against the simulation for the case of proportional fair medium access . for illustration , we plot the cumulative distribution function ( c.d.f . ) of the map in figure [ fig : map - cdf - analysis ] . here we set @xmath215 and consider two values of @xmath1 , @xmath216 and @xmath217 , which correspond to @xmath218 and @xmath219 respectively . the plots show that the stochastic geometry based formula ( see theorem [ thm : map - distribution ] ) quite accurately predicts the nodes behavior in simulation . we also study the distributions @xmath220 ( also referred to as @xmath202 for @xmath203 ) defined in theorem [ thm : map - typical - receiver ] . figure [ fig : map - cdf - receiver ] shows their plots for @xmath219 and two values of @xmath200 , @xmath221 and @xmath222 . again the plots based on the analytical expression and those based on simulation closely match . clearly , under @xmath151 , nodes closer to the origin are more likely to be inactive . , the map distribution of a node at distance @xmath223 from origin , under @xmath151.,width=288,height=240 ] in this section we illustrate the performance of various adaptive schemes and their benefit over plain aloha . we compute the performance metrics via simulation and also through analytical expressions whenever the latter are available . in such cases the analytical results and the simulation validate each other . first we set @xmath224 and @xmath225 . we consider the maximum throughput medium access , however , only accounting for the closest interferers . in figure [ fig : gibbs - closei ] , we show steady state behavior of our gibbs sampling based algorithms ; we have set the temperature @xmath226 . as expected , the improved maximum throughput medium access ( see the discussion at the end of section [ sec : close - interf - max - thput ] ) insures a better exclusion behavior . under this scheme , a lesser number of nodes transmit , and neighboring nodes are unlikely to transmit simultaneously . so , this is expected to deliver better aggregate throughput . now we keep @xmath208 fixed at @xmath227 , but vary @xmath1 from @xmath228 to @xmath229 ; this corresponds to varying @xmath124 from @xmath230 to @xmath231 in the analytical expressions . we evaluate the aggregate throughputs of various aloha schemes including plain aloha . the average throughputs are plotted in figure [ fig : throughput ] . _ although in some of the schemes we derive the attempt probabilities only considering the closest interferers , we always take into account the aggregate interference while calculating the throughput_. when the number of nodes is small , both the throughput maximizing medium access and plain aloha have identical performance ; both prescribe attempt probabilities close to one for all the nodes . when the number of nodes increases beyond @xmath232 , the throughput maximizing medium access significantly underestimates the interference , and thus its performance deteriorates . on the other hand , the aggregate interference based proportional fair scheme significantly outperforms plain aloha in terms of aggregate throughput also . this benefit is sustained even as the number of nodes increases . we also notice that the improved version of maximum throughput medium access ( see the discussion at the end of section [ sec : close - interf - max - thput ] ) yields best performance among all the schemes , and its performance does not deteriorate until a much higher number of nodes . in figure [ fig : log - throughput ] , we plot the logarithms of the aggregate throughputs corresponding to the two proportional fair medium access schemes . the figure illustrates that as the number of nodes increases , the performance of the closest interferer based medium access worsens in comparison to the performance of the aggregate interference based scheme - this is not visible merely looking at the corresponding aggregate throughputs ( see figure [ fig : throughput ] ) ) . in figure [ fig : map - cdf ] , we plot the c.d.f . of the map for the proportional fair case . our objective is to compare the case when nodes account for the aggregate interference with when they account for the closest interferer only . we set @xmath215 and plot map distributions corresponding to two values of @xmath1 , @xmath233 and @xmath234 ( corresponding to @xmath219 and @xmath218 respectively ) . as expected , the nodes attempt more aggressively when they account for the closest interferer only . while there is a favored probability interval in the aggregate interference case , there are more than one such intervals in the closest interferer case . also , in the latter case , about @xmath235 of nodes attempt in almost all slots , irrespective of @xmath1 , the total number of nodes . this can be understood by noticing that the probability that a node is not the closest interferer to any other node is not sensitive to @xmath1 . we have shown the feasibility of the performance analysis of distributed adaptive protocols that aim at maximizing some global utility in a large random network using stochastic geometry . more precisely , the most natural distributed adaptation of the medium access probability of aloha that aims at proportional fairness optimization was shown to have a tractable optimal map distribution . this distribution is obtained from the law of a certain shot noise field that describes the interference created by a typical node to all receivers but his . in the poisson case , the distribution of the optimal map is obtained as a non - singular contour integral which is amenable to an efficient evaluation using classical numerical tools . the network performance at optimum can in turn be deduced from the latter using campbell s formula . this approach is shown to provide an analytic way of quantifying the gains brought by this proportionally fair adaptive version of aloha compared to plain aloha . this line of thoughts opens several research directions . the first one is the extension to other types of fairness , still in the framework of aloha . we indicate that this is possible at least under certain simplifications of the interference model . the second and broader question is whether this approach can be extended to mac protocols other than aloha . an example would be an adaptation of the exclusion radius of csma / ca to the full environment of a node aiming at maximizing some utility of the throughput . a third general question concerns the evaluation of the `` price of decentralization '' . when the discussed protocols are suboptimal because of their greedy / distributed nature , is it possible to use stochastic geometry to evaluate the typical discrepancy between the performance of the distributed scheme and the optimal centralized one ? this work was carried out at laboratory of information , networking and communication sciences ( lincs ) paris , and was supported by inria - alcatel lucent bell labs joint research center . 10 n. abramson . : another alternative for computer communications . in _ proceedings of the fall joint computer conference ( afips ) _ , montvale , new jersey , november 1970 . f. baccelli and b. blaszczyszyn . , volume 3 issue 3 - 4 of _ foundations and trends in networking_. now publishers , 2009 . f. baccelli and b. blaszczyszyn . , volume 4 issue 1 - 2 of _ foundations and trends in networking_. now publishers , 2009 . f. baccelli , b. blaszczyszyn , and p. muhlethaler . an aloha protocol for multihop mobile wireless networks . , 52(2):421436 , february 2006 . f. baccelli , b. blaszczyszyn , and p. muhlethaler . stochastic analysis of spatial and opportunistic aloha . , 27(7):11051119 , september 2009 . f. baccelli , j. li , t. richardson , s. shakkottai , s. subramanian , and x. wu . on optimizing csma for wide area ad hoc networks . , 72(1 - 2):3168 , 2012 . d. p. bertsekas and r. g. gallager . . prentice hall , 1992 . s. boyd and l. vandenberghe . . cambridge university press , 2004 . p. bremaud . . number 31 in texts in applied mathematics . springer - verlag , 1999 . m. franceschetti and r. meester . . cambridge university press , 2008 . b. hajek . stochastic approximation methods for decentralized control of multiaccess communications . , 31(2):176184 , march 1985 . m. k. hanawal , e. altman , and f. baccelli . stochastic geometry based medium access games . in _ ieee infocom _ , orlando , florida , usa , march 2012 . hsu and h .- j . channel - aware aloha with sinr capture : when is the knowledge of channel not helpful ? in _ proceedings of ieee pimrc _ , toronto , canada , september 2011 . mohsenian - rad , v. w. wong , and r. schober . optimal sinr - based random access . in _ proceedings of ieee infocom _ , san diego , ca , usa , march 2010 . d. monderer and l. s. shapley . potential games . , 14(1):124143 , may 1996 . m. osborne and a. rubinstein . . the mit press , 1999 . l. g. roberts . packet system with and without slots and capture . , 5(2):2842 , april 1975 . x. wang and k. kar . distributed algorithms for max - min fair rate allocation in aloha networks . in _ 42nd annual allerton conference on communication , control , and computing _ , urbana , il , usa , september - october 2004 . x. wang , k. kar , and j .- s . lexicographic max - min fair rate allocation in random access wireless networks . in _ 45th ieee conference on decision and control _ , san diego , ca , usa , december 2006 . let @xmath236 be a point process with marks and @xmath237 be the point process @xmath238 shifted by @xmath239 . given that @xmath240 , @xmath241 where the mapping @xmath242 associates with @xmath243 and the real number @xmath45 the solution of @xmath244 with the usual convention if there is no solution in @xmath55 $ ] . it follows from slivnyak s theorem that @xmath245=\\ & & \hspace{-.7 cm } \mathbb{e}^0 \left[\sum_{n\ne 0 } \log \left(1-\frac{h(||x_n - y_0||,\phi^r\setminus \{y_0,y_n\}\circ\theta_{x_n})}{\frac{\vert x_n - y_0\vert^{\alpha}}{t r_0^{\alpha } } + 1}\right ) \right]=\\ & & \hspace{-.7cm}\frac 1 { 2\pi } \int\limits_0^{2\pi } \mathbb{e } \sum_{n } \log \left(1-\frac{h(||x_n - ( r_0,\phi)||,\phi^r \setminus \{y_n\}\circ\theta_{x_n } ) } { \frac{\vert x_n-(r_0,\phi)\vert^{\alpha}}{t r_0^{\alpha } } + 1}\right ) { \rm d}\phi.\end{aligned}\ ] ] it now follows from campbell s formula that @xmath246\\ & & \hspace{-.6 cm } = \int_{\mathbb{r}^2 } \int_v \log\left(1-\frac{v\overline r_0}{||x-(r_0,\phi)||^\alpha+\overline r_0}\right ) \lambda { \rm d}x\\ & & \hspace{3 cm } \mathbb{p}^0 ( h(||x-(r_0,\phi)||,\phi^r\setminus\{y_0\})= { \rm d}v)\\ & & \hspace{-.6 cm } = \int_{\mathbb{r}^2 } \int_v \log\left(1-\frac{v\overline r_0}{||x-(r_0,\phi)||^\alpha+\overline r_0}\right ) ) \lambda { \rm d}x \\ & & \hspace{3 cm } \mathbb{p } ( h(||x-(r_0,\phi)||,\phi^r)= { \rm d}v),\end{aligned}\ ] ] where the last relation follows from slivnyak s theorem .
recent years have seen an increased interest in many - particle spin dynamics , also termed _ spintronics _ @xcite . most of this research have been devoted to condensed matter systems , where the electron spin constitutes the spin degree of freedom @xcite . however , related or equivalent phenomena may occur by considering the polarization of light @xcite or internal zeeman levels of cold atoms exposed to optical light fields @xcite or spatially varying magnetic fields @xcite . in many of the spintronics schemes developed in the condensed matter community , a spin - orbit ( so ) coupling plays a crucial role for the dynamics . the particular form of the so coupling can be of either rashba @xcite or dresselhaus @xcite type , both being frequently analyzed in terms of an effective gauge force @xcite . in other words , one can interpret the so couplings as giving rise to an effective spin - dependent lorentz - type force . thus , the so coupling can serve as a spin - filter or a stern - gerlach apparatus . as the manufacturing techniques improves , undesired effects deriving from crystal impurities can be greatly suppressed and several successful experiments demonstrating the so coupling and its outings have been reported recently @xcite . ultracold atoms in optical lattices @xcite possess another possibility to implement effective so couplings , and the study of it within a clean environment . while in condensed matter settings where the coupling is an intrinsic property of the systems , for cold atoms it has to be externally created via spatially varying laser fields . ever since the pioneering experiments on bose - einstein condensation ( bec ) @xcite , the field has witnessed a tremendous progress both experimentally and theoretically @xcite . the dispersive dipole coupling between light and atoms enables the possibilities to study matter waves in periodic potentials and the experimental realization of model systems discussed within theoretical solid state physics @xcite . today , both the weakly interacting atomic gases as well as more strongly interacting ones can be realized experimentally . furthermore , utilizing the internal atomic hyperfine levels enable studies of the interplay between internal and external degrees of freedom @xcite . one interesting feature is the realization of artificial gauge potentials by means of coupling of internal atomic zeeman levels @xcite . the technique of light induced gauge potentials allows for non - abelian behavior to be analyzed and the so couplings are one such example @xcite . in this paper , we investigate the dynamics of a spinor bec subjected both to a two - dimensional weakly tilted optical square lattice as well to a set of spatially varying light fields giving rise to an effective so coupling between the internal atomic states . thereby , its dynamics will be driven by two mechanisms , namely ; ( i ) bloch oscillations ( bos ) arising due to the tilting of the lattice , and ( ii ) an intrinsic spin hall effect ( she ) deriving from the effective so coupling . the two phenomena may be analyzed in terms of an extrinsic force along the longitudinal direction of the lattice tilt and an intrinsic transverse force . being orthogonal to each other , these forces render a complex correlated motion of the bec in the lattice potential . furthermore , the particular form of the so coupling renders dirac cones characterized by linear dispersions reminiscent of relativistic electrons . as an outcome of these , the evolution possesses as well a trembling _ zitterbewegung _ motion @xcite . normally , such effect damps out very fast @xcite , but we show that in a lattice situations as in the present paper it may indeed be long - lived and hence more easily observable . moreover , the optical lattice can serve as trapping potential for the atoms in the bo regime and no confining harmonic trap is needed which would typically prevent buildup of spin currents @xcite . while so couplings arise as an intrinsic property in many crystals , in cold atom systems it has to be implemented externally . this , at the one hand may be a source for experimental errors , but at the other it gives a handle for the so coupling ; its strength and form can be tuned experimentally . the same holds for the optical lattice , the lattice strength and geometry are both control parameters , which has made it possible to observe bos for becs in optical lattices @xcite . due to the large atomic mass compared to the mass of the electron , zitterbewegung is expectedly more clearly manifested in atomic systems . the first experiments detecting zitterbewegung in atomic systems were recently carried out @xcite . a different but related setup can be found in ref . @xcite , where a spinor condensate accelerating in a periodic magnetic field was studied . we note that in ref . @xcite , the spatially dependent magnetic field gives rise to an effective gauge force . however , bloch oscillations were not considered , and moreover , in the present work the periodic potential is external and independent of the so coupling . the outline of the paper is as follows . in sec . [ sec2 ] , we introduce the model system and outline how effective so couplings are rendered in such systems . to start with we analyze the free case in which the problem is analytically solvable and demonstrate the she and the zitterbewegung effect . applying the acceleration theorem we also solve the problem under constant acceleration . the dispersions , once an optical lattice has been included , are solved numerically displaying the dirac cones . section [ sec3 ] presents the main analysis . the full numerical results are presented , where we demonstrate the presence of a she and how it is modified by the longitudinal oscillatory motion . in particular , it is shown that the hall current will also oscillate , however with an increasing oscillation amplitude . on top of the oscillations deriving from the motion in a periodic potential , we find a persisting zitterbewegung effect . finally , we conclude with a summary in sec . rashba @xcite and dresselhaus @xcite so couplings appear in a variety of solids , ranging from semiconductors @xcite , thin magnetic films @xcite , to graphene @xcite . the so couplings bring about a degeneracy in the energy dispersions . around these degeneracies , the dispersions are linear , reminiscent to that of relativistic particles , and therefore they are called _ dirac cones_. the relativistic properties have been confirmed experimentally in graphene either by _ angle resolved photoemission spectroscopy _ or by considering the conductivity and cyclotron motion of electrons @xcite . the so coupling and its corresponding degeneracy induces as well a non - trivial effective gauge structure , giving rise to phenomena such as the intrinsic spin and anomalous hall effects @xcite . in the field of cold atoms , the first light induced effective magnetic fields were demonstrated in ref . @xcite , and signatures of non - abelian structures were recently observed , see @xcite and reference therein . bos with ultracold atoms have already been realized experimentally by numerous groups @xcite . we proceed by presenting a scheme which realizes the desired so coupling in an atomic tripod system . the idea of effective gauge potentials arising from adiabatic evolution dates back to the seminal works by mead on vibrations of molecules @xcite . berry @xcite generalized this finding by demonstrating the existence of an effective gauge structure in the context a quantum adiabatic evolution . wilczek and zee @xcite further generalized berry s result and found an effective non - abelian gauge potential accompanying adiabatic evolution of degenerate energy eigenstates . dum and olshanii @xcite showed that the effective gauge potentials can be generated in systems of ultracold atoms subjected to spatially dependent light fields that couple internal atomic states . following an approach for creating non - abelian gauge potentials in time - dependent systems @xcite , ruseckas _ et al . _ @xcite outlined a scheme how to achieve effective non - abelian gauge potentials by considering an atomic tripod system in spatially dependent light fields . the coupled four - level system possesses two degenerate dark states that , within certain parameter regimes , can be decoupled from the remaining two bright states via standard adiabatic elimination . the flexibility of this tripod system , deriving from different laser configurations , allows for a variety of phenomena to be studied , such as , e.g. , relativistic properties @xcite , she s @xcite , novel phases of the bec @xcite , and topological defects as effective magnetic monopoles @xcite or vortices @xcite . in the present paper we consider a tripod setup similar to the one described in refs . the details of atomic and laser configurations rendering the gauge potentials can be seen in fig . the internal dynamics , is driven by three lasers that couple the three lower atomic states @xmath0 , @xmath1 , and @xmath2 to the excited state @xmath3 . parameterizing the laser fields as @xmath4 , @xmath5 , and @xmath6 , such that @xmath7 , the two degenerate dark states of the corresponding hamiltonian can be written as @xmath8 e^{-is_{13 } } |1\rangle \\ \\ & & + \left [ e^{-i\pi/4 } \cos \ ! \theta \sin \ ! \varphi - e^{i\pi/4 } \cos \varphi \right ] e^{-is_{23 } } |2\rangle \\ \\ & & - e^{-i\pi/4 } \sin\!\theta |3\rangle \big ) , \\ \\ \displaystyle{\frac{1}{\sqrt{2 } } } \big ( \left [ e^{-i\pi/4 } \cos \ ! \theta \cos \ ! \varphi - e^{i\pi/4 } \sin\varphi \right ] e^{-is_{13 } } |1\rangle \\ \\ & & + \left [ e^{-i\pi/4 } \cos \ ! \theta \sin \ ! \varphi + e^{i\pi/4 } \cos\varphi \right ] e^{-is_{23 } } |2\rangle \\ \\ & & - e^{-i\pi/4 } \sin \ ! \theta |3\rangle \big ) , \\ \\ \end{array}\ ] ] where @xmath9 . provided the field amplitude @xmath10 is much larger than the atom - light detuning @xmath11 ( as depicted in fig . [ fig4 ] ) , the two dark states can be safely eliminated from the two bright states , resulting in an effective two - level system @xcite . to derive an effective so coupling between the two internal dark states @xmath12 and @xmath13 , we choose @xmath14 , @xmath15 , @xmath16 , and @xmath17 $ ] . this can be achieved by coupling the two ground states @xmath0 and @xmath1 to the excited state @xmath3 with two lasers intersecting with an angle @xmath18 and propagating in the @xmath19-plane . these lasers correspond to a standing wave in the @xmath20-direction and a plane wave in the @xmath21-direction . a third laser , couples the states @xmath2 and @xmath3 , and is a plane wave in the @xmath22-plane with an angle @xmath23 of propagation relative to the @xmath21-axis . in terms of the angles @xmath18 and @xmath23 , and the wave numbers @xmath24 , @xmath25 , we have @xmath26 with @xmath27 . performing the adiabatic elimination , taking into account that @xmath28 and the fact that we consider ultracold atoms so that excitations due to atomic motion can be neglected , the effective single particle two - level hamiltonian becomes in dimensionless variables @xcite @xmath29 where @xmath30 is an effective zeeman splitting , @xmath31 , @xmath32 . the pauli matrices @xmath33 , @xmath34 , operate in the space spanned by the dark states , i.e. , @xmath35 , @xmath36 , and @xmath37 . the @xmath38-dependence of the last two terms originates from non - adiabatic couplings @xcite . we note that these do not include @xmath39 and as a consequence the @xmath40-dependence can be fully factored out . in eq . ( [ ham4 ] ) , we have used dimensionless parameters . explicitly , when we further on introduce a lattice potential whose wave number is @xmath41 , we let the characteristic length scale be @xmath42 and we let the characteristic energy scale be the recoil energy @xmath43 . consequently , time @xmath44 is measured in units of @xmath45 . the superscript @xmath46 indicates the free hamiltonian in absence of a spatially dependent optical lattice potentials . the zeeman shift @xmath30 splits the degeneracy of the two internal states and together with the so coupling it gives rise to a transverse current , an intrinsic anomalous hall effect . we will , however , focus on the intrinsic she obtained when @xmath47 . in this case , the transverse particle current vanishes but , on the other hand , a transverse spin current persists . the @xmath47 condition can be achieved by internal stark shift of the individual bare atomic levels @xcite . thus , in the following we will assume @xmath47 , and furthermore we will also consider the symmetric so coupling characterized by @xmath48 . it should be pointed out , however , that even by lifting these constrains an interplay between transverse and longitudinal atomic motion will still occur . one important prerequisite for successful application of the current scheme is that the system can be prepared in the dark subspace spanned by @xmath49 . this issue has been discussed in @xcite . the idea is to slowly turn on the external lasers @xmath50 , @xmath51 , such that the state follows adiabatically states in the dark subspace , preparing a superposition of @xmath49 but no bright state components . we note that we may rewrite the hamiltonian in eq . ( [ ham4 ] ) ( @xmath47 ) as @xmath52 with the vector potential @xmath53 and the scalar potential @xmath54 these potentials constitute vector and scalar parts of the effective gauge potential . although being constant in space , the vector potential gives rise to a non - trivial effective gauge field due to its non - abelian structure , @xmath55\neq0 $ ] . explicitly , this effective field is given by the antisymmetric field tensor @xmath56 , \,\ , \ k , l = x , y,\ ] ] resulting in @xmath57 and @xmath58 otherwise . as a result , the spin - orbital motion of the electrons can be understood in terms of an effective lorentz - type force , akin to that found by wong @xcite for a color - charged classical particle moving in a non - abelian gauge field . the corresponding force operator may be derived from heisenberg equations of motion according to @xcite @xmath59}\\ \\ & = & \displaystyle{v_{so}^2 ( \hat{\bf{p } } \times { \bf e}_z ) \hat{\sigma}_z } , \end{array}\ ] ] where it is understood that the operators on the right - hand side are all given in the heisenberg representation and the prefactor @xmath60 in front of the acceleration is the mass in the chosen units . note that the force in the @xmath21-direction is proportional to @xmath61 and the force in the @xmath20-direction is proportional to @xmath62 . this demonstrates the transverse character of @xmath63 and its equivalence to a lorentz or coriolis force . the crucial aspect of eq . ( [ lforce ] ) is , however , that the force is state dependent : atoms in dark state @xmath12 feel a force with opposite sign compared to those in dark state @xmath13 . therefore , the motion of the particles in the @xmath19-plane may separate the two spin states implying that a non - zero spin current is generated . in the following , we shall call the @xmath20-direction longitudinal and the @xmath21-direction transverse . the effect of the above lorentz - type gauge force can be examined by computing how the initial ( @xmath64 ) state @xmath65 . we assume that the internal and momentum states of the atom are initially uncorrelated and that the initial momentum states are gaussian wave packets of the form @xmath66{4\pi\delta_p^2 } } e^{-\frac{(p_x - p_0)^2+p_y^2}{2\delta_p^2}},\end{aligned}\ ] ] where @xmath67 sets the momentum uncertainty and @xmath68 is the average momentum along the @xmath20-direction . this choice corresponds to the pure internal atomic state @xmath69 . given this initial state , the time evolved state is readily obtained as @xmath70 with @xmath71 \nonumber \\ & & \times \psi(p_x , p_y ) e^{-i(p_x^2+p_y^2)t } , \nonumber \\ \psi_2(p_x , p_y , t ) & = & -\frac{1}{\sqrt{2}}\left [ \cos(\lambda t ) + i\sin(\lambda t)e^{i\theta}\right ] \nonumber \\ & & \times \psi(p_x , p_y ) e^{-i(p_x^2+p_y^2)t } .\end{aligned}\ ] ] here , we have used the energy dispersions @xmath72 and the energy eigenstates @xmath73 of the hamiltonian in eq . ( [ ham4 ] ) , which read @xmath74 where @xmath75 and @xmath76 . in fig . [ fig1 ] we display the dimensionless averages @xmath77 with @xmath78 . the factor @xmath79 derives from normalization of the individual wave packet components . since the initial velocity is along the longitudinal @xmath20-direction , the deviation from @xmath80 is a direct consequence of the state - dependent lorentz - type gauge force @xmath63 . note that @xmath81 giving a vanishing transverse particle current . this symmetry is broken by keeping the zeeman term @xmath82 in the hamiltonian of eq . ( [ ham4 ] ) . the particle current induced by such a @xmath83-term goes under the name intrinsic anomalous hall effect . after @xmath84 , the separation of the two internal dark states has reached an asymptotic value . until @xmath84 , the transverse spin current shows an oscillating behavior deriving from the same mechanism as the one underlying relativistic zitterbewegung in the dirac equation @xcite . in the dirac equation the trembling motion describes coherent coupling between positive and negative energy solutions , while in the present situation it describes population swapping between the internal states . in solids , the amplitude of these oscillations become extremely small due to the lightness of the electron , while in atomic or ionic systems the heavier masses imply larger amplitudes and possible detection @xcite . despite heavier masses , the rapid damping of the trembling motion as depicted in fig . 1 is a limitation also in atomic / ionic zitterbwegung experiments @xcite . in the next subsection we will argue how this decay may be prevented by studying the effect in the presence of an optical lattice . ( green curve correspond to the dark state @xmath12 and the red curve correspond to the dark state @xmath13 ) as defined in eq . ( [ analyexp ] ) . the dimensionless parameters in eqs . ( [ ham4 ] ) and ( [ gaussin ] ) are @xmath85 ( @xmath86 ) , @xmath47 , @xmath87 , and @xmath88 . the initial state is thus a gaussian with a non - zero average momentum along the @xmath20-direction.,width=302 ] in order to generate a current in the system one applies a constant longitudinal force @xmath89 , e.g. , induced by gravitation in atomic systems . in the particular case of solids , the force derives from a voltage across the solid , and the electrons reach a steady state motion in which the applied force is balanced by scattering against crystal impurities . on the other hand , a system of cold atoms is , to a good approximation , free from impurities and one would therefore achieve constant acceleration . assuming that the force is weak and that the total time of evolution is relatively short , we can perform an adiabatic approximation , consisting in replacing @xmath90 by @xmath91 in eq . ( [ timesol ] ) . hence , the momentum grows linearly in time in accordance with classical motion . how this comes about can be easily understood by considering the time evolution of the hamiltonian @xmath92 for small forces @xmath93 and times @xmath44 . in this regime , the time - evolution operator can be approximated accordingly @xmath94 so that the first exponential gives the coupled time - evolution while the second simply boosts the @xmath20-momentum by @xmath95 . for longer time periods , the large velocity rendered by the acceleration will typically cause non - adiabatic excitations , which we neglect for the moment . similarly , for large values of @xmath96 , the factorization of the exponentials is no longer justified . the resulting averages @xmath97 calculated within this approximation are depicted in fig . [ fig2 ] . as before , we encounter a build - up of a transverse spin current but in this case the current changes sign after some time and finally there is no net spin separation . by comparison with a full numerical integration of the hamiltonian we find that the imposed adiabatic approximation breaks down after relatively short times . nevertheless , numerically we have verified that the dynamics beyond this approximation goes as well asymptotically to zero for large times . we believe that the origin of this result is a complicated interplay between the transverse and longitudinal forces , together with the coupling between the two internal states . for spinless particles , the two forces render a spirally transverse motion and hence a net transverse current . for particles with spin - dependent lorentz forces @xmath63 one would expect a net transverse spin current , but the coupled nature of the two spin states seems to destroy such spin current in the case of constant acceleration . , but with a constant weak force of magnitude @xmath93 in the @xmath20-direction added to the hamiltonian ( [ ham4 ] ) and assuming the validity of the adiabatic approximation . the dimensionless parameters are @xmath85 , @xmath47 , @xmath98 , @xmath88 , and @xmath99.,width=302 ] in the following section we numerically investigate the dynamics in the presence of an optical lattice . insight into the evolution behavior is gained by considering the spectrum of the hamiltonian including a lattice potential while disregarding the linear force potential . adding to the so hamiltonian in eq . ( [ ham4 ] ) a spatially oscillating periodic potential , yields @xmath100 the hamiltonian is periodic and consequently possesses a band - gap spectrum @xmath101 with associated bloch eigenstates @xmath102 , characterized by a discrete band - index @xmath103 and quasi - momentum numbers @xmath104 restricted to the first brillouin zone , i.e. , @xmath105 and @xmath106 . a detailed study of bloch theory in the presence of so couplings can be found in ref . @xcite . the first two bands @xmath107 and @xmath108 , obtained from numerical diagonalization of the hamiltonian , are shown in fig . a similar band structure has been obtained in ref . @xcite for cold atoms in a square optical lattice . we directly see the effect of the so coupling in terms of dirac cones at @xmath109 , where @xmath110 . it is further seen that the ground state is not the @xmath111 bloch state , as for a regular square lattice lacking so couplings . this fact is at the heart of the _ jahn - teller effect _ @xcite , where the dirac cone ( or _ conical intersection _ as it is usually referred to when it appears in position space rather than in momentum space ) induces a symmetry breaking that lowers the ground state energy . furthermore , we note that the two lowest bands are not gapped , while there is a gap between the second and third band . indeed , due to the so coupled two - level structure of the hamiltonian , the bands come in pairs where each pair possesses dirac cones for quasi - momenta @xmath109 . the effective number of dirac cones is four ( easily seen by shifting the brillouin zone by @xmath60 in both directions ) , compared to two dirac cones encountered in honeycomb lattices , like in graphene . we notice that recently it was shown how more complex lattice potentials may render single dirac cones within the first brillouin zone @xcite . for a square lattice with rashba spin - orbit coupling . in this example , @xmath85 and @xmath112 . , width=302 ] in the adiabatic regime , a non - zero extrinsic force @xmath113 is treated by using the same approximation as above , yielding @xmath114 in the presence of a periodic potential , this adiabatic assumption has become known as the _ acceleration theorem _ it implies that no population is transferred between the individual bands @xmath115 . as a direct consequence , an initial state with well localized quasi - momentum within one single band will show an oscillatory motion rather than a constantly accelerating one , once the extrinsic force @xmath113 has been turned on . these are the well studied bloch oscillations @xcite , which have been experimentally verified in a variety of systems , such as cold atomic gases @xcite , becs @xcite , semiconductor superlattices @xcite , waveguide arrays @xcite , and photonic crystals @xcite . bos are clearly an outcome of the periodic potential and its characteristic energy spectrum possessing gaps of forbidden energies at the center and edges of the brillouin zone . as the amplitude of the potential @xmath116 vanishes , the gaps close and the acceleration theorem breaks down . in particular , for weak lattice amplitudes , the gap between the first two energy bands scales as @xmath117 . by linearizing the dispersions around the band gap and replacing @xmath118 by @xmath119 , one obtains a realization of the celebrated _ landau - zener model _ @xcite . here , @xmath120 is the initial quasi - momentum in the @xmath20-direction . this model is analytically solvable , giving the probability of population transfer between the two states of the corresponding bands . it has been shown that the decay of bos can be correctly estimated using the landau - zener theory @xcite . in the present model , the situation becomes qualitatively different . at the dirac points , the gap vanishes and the acceleration theorem can not be imposed . however , from fig . [ fig3 ] it is seen that the dirac cones appear between the first two bands , and not between the second and third band . hence , assuming a state initially prepared on the lowest band one may expect predominant population transfer between the @xmath121 and @xmath122 bands and not to higher bands . thereby , further acceleration is hindered by the gap between the second and third band and spin - dependent bos may still be encountered . in other words , a modified acceleration theorem involving pair of bands could result . in fact , as already mentioned , the acceleration theorem is a specific example of the more general adiabatic theorem , and in molecular and chemical physics it has long been known that adiabaticity breaks down in the vicinity of a conical intersection @xcite . in this context , it should be kept in mind that adiabaticity is a widespread concept in physics , and the general definition concerns systems whose dynamics is governed by an explicitly time - dependent hamiltonian . adiabaticity , as it appears in the acceleration theorem , is in this sense more reminiscent of the _ born - oppenheimer approximation _ @xcite , applicable to time - independent hamiltonians describing light and heavy quantal degrees of freedom corresponding here to the spin and the spatial motion , respectively , of the particles . our numerical analysis will be restricted to a system consisting of weakly interacting atoms , for which a mean - field description is expected to capture the phenomena we are interested in . in the @xmath123-wave scattering regime , we thereby consider dynamics rendered by a spinor gross - pitaevskii equation @xcite @xmath124 \\ \\ & & + \left[\begin{array}{cc } g_{11}n_1(x , y , t)+g_{12}n_2(x , y , t ) & 0 \\ 0 & g_{12}n_1(x , y , t)+g_{22}n_2(x , y , t ) \end{array}\right ] \bigg\}\psi(x , y , t ) . \end{array}\ ] ] here , @xmath125= \left[\begin{array}{c}\psi_1(x , y , t)\\ \psi_2(x , y , t)\end{array}\right]\ ] ] is a two - component spinor wave function of the dark states , @xmath126 , @xmath93 the external longitudinal force , and @xmath127 , @xmath128 and @xmath129 are the effective @xmath123-wave scattering amplitudes related to the scattering lengths @xmath130 as @xmath131 , where @xmath132 is the width of the single site wave function in the @xmath40-direction . this definition of @xmath133 implies that the wave function has been normalized as @xmath134 . in the basis of dark states , the scattering lengths are effective ones different from the regular hyperfine scattering lengths . nevertheless , the ratio between the two scattering amplitudes should approximately be the same regardless of basis , and , in particular , for a spinor condensate such as that of @xmath135 one has @xmath136 @xcite . we will restrict the analysis to repulsive interactions , @xmath137 , and we will furthermore assume @xmath138 @xcite . in the absence of a lattice , the so coupling gives rise to a non - abelian mead - berry curvature @xcite , which results in a transverse splitting of spin components , the intrinsic she . however , in figs . [ fig1 ] and [ fig2 ] the motion along the longitudinal @xmath20-direction was either constant or monotonously accelerating , while for a weakly tilted lattice , where bos dominate the longitudinal motion , it is neither clear how the transverse spin current nor how the trembling zitterbewegung motion will be manifested . for instance , will there be sufficient time for a transverse spin current to be established within a single bloch period ? in ref . @xcite , the intrinsic anomalous hall effect was studied in a single particle system confined by an harmonic potential and hence the motion was as well oscillatory as in the case of bos . the transverse current appeared as a rotation on top of the harmonic oscillatory motion , i.e. , in polar coordinates @xmath139 the gauge potential causes a non - zero current in the @xmath140-direction . such current would indeed be zero for an so coupling possessing an abelian structure rather than non - abelian , e.g. for couplings on the form @xmath141 with @xmath142 . it is clear that a corresponding rotational current can not be strictly encountered in the present system since exact polar symmetry is broken by the tilting of the lattice and by the lattice in itself . the time evolution of the system is solved by employing the split - operator approach @xcite . the method relies on factorizing the time - evolution operator into a spatial part and a momentum part valid in the limit of infinitely small time - step propagation . one convenient feature of this method is that the wave function is in principle obtained at any instant of time in both position @xmath143 and momentum @xmath144 space . furthermore , for a system bounded from below , the approach is capable of giving an approximate ground state of the system by simply propagating it sufficiently long in imaginary time . , @xmath145 , and @xmath146 for all plots , @xmath147 in the upper plots ( a ) and ( b ) , and @xmath85 in the lower plots ( c ) and ( d ) . thus , the lower plots correspond to the dispersions depicted in fig . , width=302 ] our numerical considerations will focus on propagation of an initial ( approximated ) ground state for the untilted ( @xmath148 ) lattice . that is , we consider the ground state of a spin - orbit coupled bec in an optical lattice , and then we introduce a quantum quench by suddenly switch on a linear force in the @xmath20-direction . hence , we begin by propagating an ansatz function in imaginary time with zero extrinsic force @xmath148 . when convergence of the imaginary time evolution has been reached , we switch to real time propagation , including the extrinsic force by adding the potential term @xmath149 to the hamiltonian . as predicted by the acceleration theorem , for short times the weak forces @xmath93 will induce a linear increase in the quasi - momentum @xmath118 . in contrast to bos in a regular square lattice , here the so coupling implies that also the quasi - momentum @xmath150 is indirectly affected by the force @xmath93 . at the same time , the build up of motion in the @xmath21-direction induces an intrinsic force @xmath151 along the longitudinal @xmath20-direction of bos . in addition , the linear dispersions induce a trembling zitterbewegung motion . it is thereby likely to expect a complex motion arising from the interplay between the two forces related to she and bos and the outcomes from linear dispersions . figure [ fig5 ] displays the absolute values of the initial ground state wave function achieved via imaginary time evolution for @xmath148 . the momentum wave functions depicted in ( a ) and ( c ) demonstrate the jahn - teller effect as the ground state populates non - zero momentum states @xcite . the lower two plots ( c ) and ( d ) correspond to the parameters of the two lowest bloch bands shown in fig . the fact that the ground state obtained from this numerical method is not symmetric , as seen from the plots , is believed to have a two - fold cause . first , it seems to derive from the fact that the imaginary time evolution is taken only up to a point where the state evolution has approximately converged , and , secondly , from the fact that even small fluctuations can result in an asymmetric ground state , due the four - fold degeneracy of the ground state of our hamiltonian . in fact , the complete convergence of the ground state has not been reached since the position space wave functions depicted in ( b ) and ( d ) are localized in contradiction to the true ground state having a bloch structure , and furthermore since the momentum wave functions have a finite width . this , however , should not be seen as a shortcoming since normally one would cool the atomic gas in the presence of a confining potential which would result in a localized gas . in addition , if the propagated state extends over the full lattice one would run into troubles when quantities such as the mean spatial position is calculated numerically . note that the momentum wave functions of ( a ) and ( c ) are different from the quasi - momentum as is evident in the multiple bragg components separated by @xmath152 . in other words , even though the momentum wave functions extend beyond the first brillouin zone , it populates , to a very good approximation , only the lowest band . we also note that the position space ground state wave functions , shown in ( b ) and ( d ) , predominantly populate the individual lattice sites . hence , only judging from figs . [ fig5 ] ( b ) and ( d ) it would be difficult to predict the presence of a so coupling and the dirac cones that derive from it , which instead manifest themselves in momentum space in terms of a jahn - teller effect . ( a ) , average longitudinal position @xmath153 ( b ) , and average intrinsic force @xmath151 in the transverse direction ( c ) . red lines correspond to the dark state @xmath12 while green lines to @xmath13 . for @xmath153 the two lines are identical because of symmetry reasons . the dimensionless parameters are the same as in fig . [ fig5 ] ( a ) and ( b ) , but now with a non - zero extrinsic force , @xmath99 . , width=302 ] in order to analyze the interplay between the transverse and longitudinal forces , @xmath154 and @xmath113 , we calculate @xmath155 with @xmath78 corresponding to averages in terms of the two dark states @xmath12 and @xmath13 , respectively , @xmath156 , and the last quantity is the intrinsic force in the @xmath21-direction . we find that @xmath157 ( up to numerical accuracy ) and therefore plot only one of the two components . our numerical results are presented in figs . [ fig6 ] and [ fig7 ] . the difference between the two figures lies in the scattering amplitude @xmath158 . in fig . [ fig7 ] the effective scattering amplitude @xmath158 is already as large as 10 recoil energies @xmath159 and we can thereby conclude that the spin current is rather insensitive to the value of @xmath158 . it is generally known that strong non - linearity can qualitatively change system properties . in becs it leads to phenomena such as soliton and vortex creation . normally the greatest impact of the non - linearity is seen around curve crossings @xcite . beyond a critical value of the non - linearity ( in our case @xmath160 ) , in one dimension swallowtail loops are formed in the vicinity of the curve crossings characterizing additional stationary non - linear states , e.g. solitons . up to this critical value the dynamics seems rather insensitive at least on a mean - field level . however , formation of these loops beyond the critical value implies breakdown of the acceleration theorem @xcite . to date , it seems that analysis of these effects have been restricted to one dimensional situations and it is therefore not fully understood how non - linearity affects the structure of the dirac cones appearing in two dimensions . nonetheless , we have numerically found that the general structure of figs . [ fig6 ] and [ fig7 ] drastically change for very large values of @xmath158 . all present bo experiments , however , are performed in the weakly interacting regime where it is known that loops are absent , and we thereby restrict our analysis to these experimentally more relevant parameter ranges . we see from both fig . [ fig6 ] and [ fig7 ] that the @xmath20 and @xmath21 motions are indeed correlated ; when the bloch oscillation reaches its turning points , the spin separation in the @xmath21-direction is roughly maximal and as the velocity in the longitudinal @xmath20-direction changes sign , the spin separation decreases due to the reversed sign of the intrinsic force . when the longitudinal velocity is the greatest ( equivalent to maxima of the intrinsic force @xmath151 ) , the spin current is reversed . more precisely , as the atoms begin to accelerate in the tilted lattice the quasi - momentum @xmath118 increases linearly in time and a spin current is built up in the transverse direction . however , after half a bloch period the quasi - momentum reaches the brillouin boundary and the @xmath20-motion is stopped and then reversed . consequently , also the spin current is reversed and one encounters a decreasing spin current in the @xmath21-direction . finally , the spin current has dropped to zero and builds up in the opposite direction instead . hence , the spin current follows the same oscillatory behavior as the bos . one may argue that a large spin separation should be hindered by the alternating sign of the velocity in the @xmath20-direction . nonetheless , from our numerical results depicted in figs . [ fig6 ] and [ fig7 ] , we note that the spin separation in the @xmath21-direction is relatively large . indeed , @xmath161 exceeds several site - spacings and furthermore this separation amplitude increases in time . the fact that neither the oscillatory spin nor the oscillatory longitudinal current show a clean perfect harmonic pattern derives from , at the one hand possible non - adiabatic effects and the breakdown of the acceleration theorem , and at the other that the transverse current induces an intrinsic force @xmath162 in the longitudinal direction which adds to the extrinsic force @xmath113 . the full coupled dynamics is therefore rather complex , but nonetheless , our figures make clear the overall in - phase oscillations between the two currents . on top of the oscillations of @xmath163 , there is a slight drift toward negative values of @xmath20 . this is an outcome of non - adiabaticity and breakdown of the acceleration theorem . however , this drift is rather small and the dynamics is therefore predominantly adiabatic . this drift can be suppressed by increasing the lattice amplitude @xmath116 on the cost that the amplitude of longitudinal oscillations decreases . it would not , however , imply a diminishing she since it is related to the longitudinal momentum and not position . a closer look at the two figures reveals that in the transverse motions there is an additional oscillating motion different from the spin current being in phase with the bos . this is believed to be zitterbewegung clearly seen in the freely evolving wavepackets of fig . contrary to that figure , here these oscillations do not collapse suggesting that it might be preferable to detect the zitterbewegung effect within a tilted lattice , compared to the case of freely propagating atoms . the survival of zitterbewegung should as well derive from the alternating motion due to the optical lattice . figures [ fig6 ] and [ fig7 ] have been calculated for an approximated initial ground state of the untilted lattice . as already pointed out , experimentally the condensate might be prepared in the presence of an external trap , and then the optical lattice as well as the coupling of the internal states are turned on in such a way that the condensate is not in a true ground state of the coupled system . to analyze the effect of such excitations , we have redone the simulations of figs . [ fig6 ] and [ fig7 ] for an initial gaussian momentum state of a form depicted in eq . ( [ gaussin ] ) with @xmath98 and @xmath88 . the results are very similar to the ones of figs . [ fig6 ] and [ fig7 ] , and we therefore do not present them here . we have also investigated the effect of a shallow harmonic potential in the transverse direction during the evolution in the tilted lattice . again , a transverse spin current is obtained , however this time the amplitude of transverse oscillations is decreased due to the confining potential . this reduction of spin separation would indeed be present in any experiment utilizing a confining trapping potential . the idea of employing an optical lattice potential as a trapping potential circumvent such shorting . in order to achieve trapping in this case one must stay within the bo regime , i.e. adiabatic . as figs . [ fig6 ] and [ fig7 ] demonstrates , even though the atoms oscillate a relatively large spin current can build up during time . in relation to the measurability of the spin separation , we may note that even though @xmath164 is considerably larger than the lattice spacing , it does not imply that the two wave functions @xmath165 and @xmath166 do not spatially overlap . in fact , the overlap is relatively large and fluorescent measurement of a single experimental run would therefore not contain sufficient information to reveal the spin current . however , this is hardly never the case in true experimental realizations and instead an ensemble of measurements should be considered , where the ensemble average gives @xmath164 . since this value is presumably larger than the lattice spacing , it should indeed be measurable @xcite . in the cold atom community , time - of - flight measurements are often employed , in which it is the atomic momentum density that is being measured . if the separation of the momentum wave functions is non - zero , the position space wave functions @xmath165 and @xmath166 will set apart further during the time - of - flight . if this separation is faster than the intrinsic wave function broadening , a finite time - of - flight measurement may enhance the detection efficiency . , but with an increased atom - atom interaction ; @xmath167 . , width=302 ] in this paper , we have considered an ultracold atomic system possessing three well - studied properties of condensed matter theory , bloch oscillations , intrinsic spin hall currents , and zitterbewegung.our motivation for this study comes from the fact that bloch oscillations have been verified in the same type of setups @xcite , light induced gauge potentials have been demonstrated @xcite , and zitterbewegung have been detected in ion - trap experiments @xcite . the spin - hall effect ( she ) and the bloch oscillations ( bo ) in our system are driven by two perpendicular forces : the extrinsic force appearing due to tilting of the lattice potential and the intrinsic force originating from the effective gauge potential caused by the spin - orbit ( so ) coupling . we demonstrated that the interplay between these two forces gives rise to a correlated motion in the @xmath19-plane , and despite the fact that the bos are expected to prevent a build up of large transverse spin currents , we argued that the she should indeed be detectable . we utilized system parameters that should be achievable in current experiments , e.g. , the amplitude of the lattice potential is five recoil energies compared to the experimentally utilized ones of ref . @xcite ranging from one to 14 recoil energies , and the so coupling two recoil energies compared to 8 recoil energies in ref . @xcite . for a potential amplitude as large as @xmath168 , the dispersions are almost flat and longitudinal motion is almost frozen . the momentum along the longitudinal direction do , however , oscillate within the brillouin size and , interestingly and somewhat surprisingly , one would therefore see an oscillating transverse spin hall current despite lack of longitudinal currents . we furthermore showed that the effects are insensitive to experimentally relevant atom - atom interaction strengths . entering a strongly interacting gas seemed to affect our results and we argue that this is likely to derive from a non - linearity induced breakdown of adiabaticity around the dirac points . an additional interesting aspect being discussed is the analogue of relativistic zitterbewegung . our numerical results indicate that it may be preferable to consider an optical lattice configuration when studying this effect since it persists over longer time periods . the optical lattice convey as well the possibility to avoid confining trapping potentials preventing large spin currents . we should also point out that the system presented in this paper allows for many extensions and different phenomena to be studied . for example , vortex generation in lattices @xcite , bose - einstein condensation into non - zero momentum states @xcite , topologically ordered states of matter @xcite , or more strongly correlated systems , by means of second quantization of the mean - field hamiltonian , may find realization in atoms possessing an internal tripod structure and trapped in optical lattices . the flexibility permits as well for various extensions , for example time - dependent lattices or so couplings . a recent paper considered zitterbewegung of ultracold atoms for a driven so coupling @xcite . the same idea can be utilized in order to increase the transverse spin current in our model . m. onoda , s. murakami , and n. nagaosa , phys . lett . * 93 * , 083901 ( 2004 ) ; k. yu . bliokh and yu . p. bliokh , phys . e * 70 * , 026605 ( 2004 ) ; k. yu . bliokh and yu . p. bliokh , phys . a * 333 * , 181 ( 2004 ) ; k. yu . bliokh and yu . p. bliokh , nat . photonics * 2 * , 748 ( 2008 ) . o. hosten and p. g. kwiat , science * 319 * , 787 ( 2008 ) . a. m. dudarev , r. b. diener , i. carusotto , and q. niu , phys . rev . lett . * 92 * , 153005 ( 2004 ) ; x .- j . liu , x. liu , l. c. kwek , and c. h. oh , phys . rev . lett . * 98 * , 026602 ( 2007 ) ; j. y. vaishnav , j. ruseckas , c. w. clark , and g. juzelunas , phys . lett . * 101 * , 265302 ( 2008 ) ; x .- j . liu , m. f. borunda , x. liu , and j. sinova , phys . lett . * 102 * , 046402 ( 2009 ) . m. h. anderson , j. r. enscher , m. r. matthews , c. e. wieman , and e. a. cornell , science * 269 * , 198 ( 1995 ) ; k. b. davies , m. o. mewes , m. r. andrews , n. j. vandruten , d. s. durfee , d. m. kurn , and w. ketterle , phys . * 75 * , 3969 ( 1995 ) . c. j. myatt , e. a. burt , r. w. ghrist , e. a. cornell , and c. e. wieman , phys . lett . * 78 * , 586 ( 1997 ) ; j. stenger , s. inouye , d. m. stamper - kurn , h .- j . miesner , a. p. chikkatur , and w. ketterle , nature * 396 * , 345 ( 1998 ) ; d. m. stamper - kurn , h. j. miesner , a. p. chikkatur , s. inouye , j. stenger , and w. ketterle , phys . rev . lett . * 83 * , 661 ( 1999 ) ; o. mandel , m. greiner , a. widera , t. rom , t. w. hnsch , and i. bloch , phys . . lett . * 91 * , 010407 ( 2003 ) ; l. e. sadler , j. m. higbie , s. r. leslie , m. vengalottore , and m. s. stamper - kurn , nature * 443 * , 312 ( 2006 ) ; j. larson and j .- p . martikainen , phys . rev . a * 80 * , 033605 ( 2009 ) . lin , r. l. compton , a. r. perry , w. d. phillips , j. v. porto , and i. b. spielman , phys . lett . * 102 * , 130401 ( 2009 ) ; y .- j . lin , r. l. compton , k. jimnez - garcia , j. v. porto , and i. b. spielman , nature * 462 * , 628 ( 2009 ) . it is worth pointing out that even if a confining trap is absent in the actual bo experiments , it is present during the condensation process and as it is turned off before releasing the condensate in the tilted lattice it implies , in the local density approximation , that the chemical potential varies within the extent of the atomic cloud which renders a source for decoherence , see @xcite . o. morsch , j. h. mller , m. cristiani , d. ciampini , and e. arimondo , phys . * 87 * , 140402 ( 2001 ) ; m. cristiani , o. morsch , j. h. mller , d. ciampini , e. arimondo , phys . a * 65 * , 063612 ( 2002 ) ; j. h. denschlag , j. e. simsarian , h. haffner , c. mckenzie , a. browaeys , d. cho , k. helmerson , s. l. rolston , and w. d. phillips , j. phys . b : at . mol . * 35 * , 3095 ( 2002 ) ; m. gustavsson , e. haller , m. j. mark , j. g. danzl , g. rojas - kopeinig , and h .- c . ngerl , phys . lett . * 100 * , 080404 ( 2008 ) . t. ando , j. phys . japan * 69 * , 1757 ( 2000 ) ; c. l. kane and e. j. male , phys * 95 * , 226801 ( 2005 ) ; h. min , j. e. hill , n. a. sinitsyn , b. r. sahu , l. kleinman , and a. h. macdonald , phys . b * 74 * , 165310 ( 2006 ) . v. y. demikhovskii and d. v. khmiisky , j. exp * 83 * , 340 ( 2006 ) ; s. smirnov , d. bercioux , and m. grifoni , europhys . lett . * 80 * , 27003 ( 2007 ) ; d. v. khmiisky , phys . rev . b * 79 * , 205401 ( 2009 ) . k. leo , p. h. bolivar , f. bruggemann , r. schwedler , and k. kohler , solid . . commun . * 84 * , 943 ( 1992 ) ; j. feldmann , k. leo , j. shah , d. a. b. miller , j. e. cunningham , t. meier , g. vonplessen , a. schulze , p. thomas , and s. schmittrink , phys . b * 46 * , 7252 ( 1992 ) . r. morandotti , u. peschel , j. s. aitchison , h. s. eisenberg , and k. silberberg , phys . lett . * 83 * , 4756 ( 1999 ) ; t. pertsch , p. dannberg , w. elflein , a. brauer , and f. lederer , phys . lett . * 83 * , 4752 ( 1999 ) . t. gericke , p. wrtz , d. reitz , t. langen , and h. ott , nature physics * 4 * , 949 ( 2008 ) ; n. gemelke , x. zhang , ch .- l . hung , and ch . chin , nature * 460 * , 995 ( 2009 ) ; w. s. bakr , j. i. gillen , a. peng , s. flling , and m. greiner , nature * 462 * , 74 ( 2009 ) ; j. f. sherson , c. weitenberg , m. endres , m. cheneau , i. bloch , and s. kuhr , nature * 18 * , august ( 2010 ) .
the first evidence for winds from cataclysmic variables ( cvs ) came from the discovery of p cygni profiles in the uv resonance lines of ss cygni ( @xcite ) . the dependence on the observability of cv winds with inclination angle and the apparent similarity between cataclysmic variable line profiles and those of ob stars led to the early suggestion by crdova & mason ( 1982 ) that the winds in cvs originate from the accretion disk and that the line radiation pressure is responsible for the wind . more recently , p cygni profiles have been detected from virtually all nonmagnetic cvs with high mass accretion rates ( @xmath5 ) ( e.g. , @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . the absorption component is most apparent in low inclination systems , and is not detected in eclipsing systems ( e.g. , @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . past efforts in the development of models for cv winds focused on one - dimensional models ( @xcite ; @xcite ) , and on kinematic modeling ( @xcite ; @xcite ) , which succeeded in showing consistency between the assumed polar geometry of a disk wind and observed profiles . icke ( 1980 ) developed a two - dimensional disk wind model but did not take into account the radiation pressure due to line absorption . for a typical white dwarf mass of @xmath6 ( @xcite ) the disk luminosity required to produce such a wind without line radiation pressure ( assuming radiation pressure due to continuum scattering only ) would require a luminosity of @xmath7 ( @xcite ) , several magnitudes above the observed luminosity of cvs which vary between @xmath8 and @xmath9 ( @xcite ) . icke did find that for accretion disk winds in general , for sufficiently high disk luminosity , a biconical disk wind would be produced . icke ( 1981 ) suggested that biconical winds were a general property of accretion disk winds independent of the wind driving mechanism . murray et al . ( 1995 ) found that for line - driven disk winds in active galactic nuclei ( agn ) the wind flow tends to be parallel to the accretion disk and developed a one - dimensional disk wind model for these systems . murray & chiang ( 1996 ) found that the single - peaked optical emission lines seen in agns and high - luminosity cvs could be accounted for by emission at the base of an accretion disk wind . using the disk wind model of murray et al . ( 1995 ) , murray & chiang ( 1997 ) calculated synthetic line profiles of 1550 and found qualitative agreement between their models and observed line profiles . in the case of cvs , as we discussed above , the winds are most apparent in low - inclination systems and not observed in eclipsing systems , indicating that the winds in cvs tend to flow perpendicular to the accretion disk . in an earlier paper ( @xcite , hereafter paper i ) we presented two - dimensional isothermal hydrodynamic models of line - driven accretion disk winds ( ldadw ) in cvs . to our knowledge these isothermal models were the first two - dimensional hydrodynamic models developed for ldadws . results from paper i show , in analogy with line - driven winds from early type stars , that terminal velocities are approximately independent of the luminosity of the disk , although increments in luminosity produce increments in mass - loss rate . in paper i we showed that rotational forces are important in the study of winds from accretion disks , and that they cause the velocity streamlines to collide . the collision of streamlines reduce the speed and increase the density of the wind producing an enhanced density region . in paper i we also showed that the highest absorption occurs in the enhanced density region where density is increased relative to a spherically diverging wind with the same mass loss rate and the velocity is roughly half the terminal velocity . this density increase is necessary in order to produce at least marginally optically thick lines . recently proga , stone , & drew ( 1998 ) also developed two - dimensional models for these systems , obtaining similar results to the models presented here . when evaluating the line - radiation pressure from the disk , in order to make the computations manageable , they assumed that the velocity gradient is primarily along the `` z '' direction , as we also did in paper 1 and in the models presented here . proga , stone , & drew worked out their hydrodynamic model in spherical coordinates , while the models presented here were implemented in cylindrical coordinates . they found similar wind velocities and wind mass loss rates as we do here , although they assumed disk luminosities about an order of magnitude greater . a difference we find with the results of the models presented by proga , stone & drew ( 1998 ) is that they obtain unsteady flows characterized by large amplitude fluctuations in velocity and density , while for the models presented in this paper we find a steady flow . a shortcoming of our previous work , presented in paper i , was that we had assumed an isothermal wind and an isothermal disk . in this work we develop a hydrodynamic model of ldadws in cvs which includes the radial structure of an optically thick accretion disk with the corresponding radiation fields and surface temperature distributions . we have also implemented the corresponding energy conservation equation for the models presented here self consistently ( see equation [ 18 ] ) , including the adiabatic heating and cooling effects due to compression and expansion . for the models presented in this paper we have also taken values presented by abbott ( 1982 ) for the line radiation force multiplier parameters , rather than the earlier values used by castor , abbott , & klein ( 1975 ) as we did in our isothermal models presented in paper i. from the results of the hydrodynamic model presented in this paper , we calculate theoretical line profiles for the 1550 line , and find that the line profiles obtained through our model are consistent with observations in their general form and strong dependence with inclination angle . thus we are able to predict the general observed wind properties of cvs . the boundary layer where the disk intercepts the white dwarf and the white dwarf photosphere may also contribute to the radiation field . the white dwarf by itself ( without considering the effects of the accreting mass ) will have a luminosity of the order of @xmath8(@xcite ) , while the accretion disk will have typical luminosities of the order of @xmath10 . in a steady state accretion disk half of the energy of the accreting mass is converting into radiation emitted by the accretion disk . assuming a relatively slowly rotating compact star , it has been predicted that the boundary layer presents a luminosity approximately equal to the accretion disk , thus accounting for the other half of the energy of the accreting mass ( e.g. , @xcite ) . but there is still uncertainty concerning the existence and spectrum of the boundary layer ( @xcite ; @xcite ; @xcite ) . analysis of observations have found , in some cases , a boundary layer luminosity an order of magnitude less than the disk luminosity ( e.g. , @xcite ) , and in other cases a boundary layer luminosity comparable to disk luminosity ( @xcite ) . in the models presented here we have assumed that the radiation field is generated by the accretion disk alone and have therefore neglected any boundary layer or white dwarf radiation . in the future we plan to explore the effects of boundary layer radiation on our models . in 2 we discuss the radial structure of the accretion disk and the radiation field as implemented in our model . in 3 we derive the expression we use for the treatment of the line radiation pressure . in 4 we present and discuss the hydrodynamic calculations . in 5 we present and discuss theoretical 1550 line profiles and compare them with observations . in 6 we present a summary and conclusions of this work . cataclysmic variables are binary systems composed of a white dwarf , a main sequence star , and an accretion disk about the white dwarf . the disk is formed by mass accreting from the main sequence star to the white dwarf . in some objects strong magnetic fields ( @xmath11 ) may prevent the accreting mass in cvs from forming a disk around the white dwarf ( e.g. , @xcite ) . in this work we study the winds originating in nonmagnetic cvs in which an accretion disk is formed . direct observational evidence for the existence of accretion disks in cvs is found in the spectrum of eclipsing cvs . eclipsing cvs present double peaked h and he emission lines in their optical spectrum . in these systems the eclipse of the blue side of the emission line precedes that of the red side ( @xcite ; @xcite ; @xcite ) , demonstrating that the emission lines occupy an extended region and that it rotates about white dwarf in the direction expected for a disk formed through the accretion of mass from the secondary star . furthermore , synthetic emission lines that have been calculated from models of accretion disks rotating around the white dwarf have succeeded in accounting for the double peak features found in observations ( @xcite , 1981 ; @xcite ; @xcite ) . we wish to note here that single peaked , rather than double peaked , h and he emission lines with the same eclipse behavior ( eclipse of the blue wing preceding eclipse of the red wing ) are observed in many cvs ( e.g. , @xcite ; @xcite ; @xcite ) . these single peaked balmer emission lines may be explained by the stark effect in the disk radiated emission ( @xcite ; @xcite ) or by line emission at the base of an accretion disk wind ( @xcite ) . further evidence for the existence accretion disks has been obtained through models for disk structure that have succeeded in accounting for the observed continuum shapes in many cvs ( e.g. , @xcite ; @xcite ; @xcite ; @xcite ) . in most cases the luminosity of cvs is dominated by the radiation emitted by the accretion disk which is typically of order of the solar luminosity @xmath10 ( @xcite ) . cvs with high mass accretion rates ( dwarf novae in outburst , novae , and nova - like variables ) have luminosities which range between @xmath12 and @xmath9 ( @xcite ) . in cataclysmic variables the disk radiation emission and disk temperature are radially dependent . such disks have been studied in detail by shakura and sunyaev ( 1973 ) and we adopt their formulation in what follows . the accretion disk would have an inner radius approximately equal to the radius of the white dwarf ( @xcite ; @xcite ; @xcite , 1995 ) and an outer radius reaching out to two to three times less than the inner lagrangian point ( @xcite ; @xcite ; @xcite ) ( see figures 1 and 2 ) . assuming that the disk is in a steady state , that the shear stress at the inner radius of the disk is negligible , and that the gas in the disk follows keplerian orbits one finds that the energy emitted per area @xmath13 at radius @xmath14 is given by ( @xcite ) : @xmath15 \;.\ ] ] the luminosity of the disk @xmath16 may be obtained by integrating the energy emission per area @xmath13 over the surface of the disk . assuming that the disk is emitting locally as a blackbody , and considering equation ( 1 ) , we have that the radial temperature distribution of the disk will be @xmath17 \right\}^{1/4 } \ ; .\ ] ] where @xmath18 is the stefan - boltzmann constant . equation ( 1 ) is implemented in our model , taking into account foreshortening and assuming local blackbody radiation , in the calculation of radiation flux throughout the wind , and consequently in the calculation of the continuum and line radiation pressure force . equation ( 2 ) is implemented in our model as a boundary condition for the wind near the disk surface . in our model we assume that the temperature at the base of the wind is equal to the disk surface temperature for each radius . a crucial ingredient of the disk wind models is the description of the radiation pressure force per mass due to line scattering . in this section we derive the expression adopted for the accretion disk wind model . we obtain a more complex expression than in the isothermal models we presented in paper i. the expression obtained here is based upon the radiation emitted by an accretion disk and includes not only the radiation pressure along the direction perpendicular to the disk , but also the radiation pressure parallel to the disk . introducing the approximation that the velocity gradient is primarily in the direction perpendicular to the surface of the disk and using the opacity distribution developed by castor , abbott , & klein ( 1975 ) , we obtain the following expression for the force per mass due to lines ( paper i ) : @xmath19 \hat{n } d\omega \ ; , \ ] ] where @xmath20 is the thomson cross section per unit mass , @xmath21 the specific intensity , @xmath22 the speed of light , @xmath23 and @xmath24 are parameters dependent on temperature and the relative abundance of different elements , @xmath25 is the mass density , @xmath26 is the thermal sound speed , and @xmath27 is the component of the wind velocity in the @xmath28 direction . we wish to note here that although equation ( 3 ) takes into account the line radiation pressure in the `` @xmath14 '' direction , we have included only the velocity gradient in the `` @xmath28 '' direction . as we will show in 4 , the accretion disk winds in cvs tend to flow in the direction perpendicular to the disk , particularly in the region near the surface of the disk , thus justifying the assumption of a velocity gradient primarily in the `` @xmath28 '' direction . although on one hand , incluying the velocity gradients en the `` @xmath14 '' direction may possibly produce interesting results , on the other we do not believe that it will significantly affect the overall results . also , as we show below , the assumption of a velocity gradient primarily in the `` @xmath28 '' direction allows us to obtain an expression for the line radiation pressure that requires the calculation of a spatial integration only once , rather than having to evaluate the spatial integrals at each time step , thus making the computations manageable . abbott ( 1982 ) presented elaborate calculations for the @xmath23 and @xmath24 parameters considering temperatures varying from @xmath29 to @xmath30 . he found that the @xmath23 parameter to be generally an order of magnitude greater than the earlier values presented by castor , abbott , & klein ( 1975 ) in which the line radiation parameters calculations were based on atomic line data of the . thus for the models presented in this paper we have used the values @xmath31 and @xmath32 , rather than the earlier values used by castor , abbott , & klein ( 1975 ) of @xmath33 and @xmath32 as we did in our isothermal models presented in paper i. thus we feel confident that our formulation is approximately correct for disk temperatures throughout this range . we wish to note here that abbott ( 1982 ) found values for @xmath24 which varied roughly between @xmath34 and @xmath35 , and that as we have observed through our isothermal models ( paper 1 ) values of @xmath24 near @xmath34 will reduce significantly both the wind mass loss rate and the wind velocity . in our next paper we will implement local ionization equilibrium and calculate the @xmath23 and @xmath24 parameters at each point of the wind rather than assuming fiducial values for these parameters throughout the wind as we have done up to now . note that the parameter @xmath24 used here and throughout this work is not the same as the parameter often used to describe accretion disk viscosity ( e.g. , @xcite ) . from equation ( 3 ) we evaluate the angular dependence exactly , rather than introduce approximations for the angular dependence as we did for our isothermal models presented in paper 1 . therefore : @xmath36 \hat{n } d\omega \ ; .\ ] ] defining @xmath37 : @xmath38 \hat{n } d\omega \ ; , \ ] ] taking the source of radiation to be the accretion disk , we have @xmath39^{\alpha+4 } } \ ; r ' d\phi dr'\ ] ] and @xmath40^{\alpha+4 } } \ ; r ' d\phi dr ' \;\ ] ] where @xmath41 and @xmath42 are the corresponding components of the vector @xmath43 defined in equation ( 5 ) ( the @xmath44 component is zero due to the axial symmetry of the accretion disk ) and @xmath45 is the radiation emission per area of the disk calculated through equation ( 1 ) . thus we have @xmath46\ ] ] where @xmath41 and @xmath42 are given by equations ( 6 ) and ( 7 ) . we also account for the fact that , as can be seen from the appendix of paper 1 , if the velocity gradient increases to sufficiently high values , or the density of the wind decreases to sufficiently low values , the contribution for the radiation pressure due to lines arrives at a maximum value . abbott ( 1982 ) found that the maximum value of the line radiation pressure was obtained when @xmath47 . in this work we have assumed that the maximum possible value of the line radiation pressure is obtained when @xmath48 . we note that we may be overestimating the maximum value of @xmath49 , and thus the maximum value of the force multiplier . we do not expect this to cause a significant difference in the overall dynamics , particularly in the high density regions where most of the spectral lines are forming . this will be discussed again in 4.3 . in the future we shall include local ionization equilibrium into our models self consistently and calculate the line radiation force multiplier at each point in the wind in a manner equivalent to abbott s calculation for ob stars ( @xcite ) . therefore the expression for the total line radiation force per mass for disk winds which we adopt is @xmath50 \ ; \right)^\alpha [ s_z(r , z ) \hat z + s_r(r , z ) \hat r ] \ ; .\ ] ] for the continuum radiation pressure the total radiation flux is calculated throughout the disk wind @xmath51 \hat{n } d\omega \ ; .\ ] ] taking the source of radiation to be the accretion disk , we have @xmath52 ^ 4 } \ ; r ' d\phi dr'\ ] ] and @xmath53 ^ 4 } \ ; r ' d\phi dr ' \;\ ] ] where @xmath54 and @xmath55 are the corresponding components of the radiation flux vector @xmath56 originating from the disk ( the @xmath57 component is zero due to the axial symmetry of the accretion disk ) and @xmath45 is the radiation emission per area of the disk calculated through equation ( 1 ) . the line radiation pressure depends strongly on the radiation flux ( equations [ 5 ] and [ 8 ] ) , and therefore the radiation flux plays an important role in our models . in figure 3 we present the radiation flux perpendicular to the disk as a function of height for different radii . a peculiarity of the disk winds is that gravity increases along the streamlines . in our accretion disk wind models for large radii ( see figure 3 ) we find an increase in radiation flux with height along the streamlines also . this is necessary for wind driving . for small radii ( @xmath58 ) ( see figure 3 ) we observe little or no increase in radiation flux with height , but we do observe a significant increase in the radiation flux with decreasing radius ( @xmath59 a factor of @xmath60 between the innermost [ @xmath61 and outermost [ @xmath62 radii found in the spatial grids of our computational models ) . for the inner disk radii ( @xmath63 ) the radiation flux is locally super - eddington , and the gravity increase is overcome by the high radiation flux . our hydrodynamic model uses the piece - wise parabolic method ( ppm ) numerical scheme ( @xcite ) . we use the 2.5-dimensional hydrodynamic equations of the wind in cylindrical coordinates . that is , the three - dimensional hydrodynamic equations are reduced to two - dimensional equations by assuming that the derivative of any physical variable with respect to @xmath64 is zero . the first spatial dimension is the height above the disk @xmath28 . the second spatial dimension is the distance between the center of the disk and the projection on the disk @xmath14 . the equations are : the equation of state , @xmath65 the mass conservation equation , @xmath66 the momentum conservation equations , @xmath67 \ ; \right)^\alpha \ ; , \nonumber \end{aligned}\ ] ] @xmath68 @xmath69 \ ; \right)^\alpha \ ; , \nonumber\end{aligned}\ ] ] and the energy conservation equation , @xmath70 \ ; \right)^\alpha v_r \nonumber \\ & & \nonumber \\ & & - \rho{gm_{wd } \over ( r^2+z^2)^2}{z \over ( r^2+z^2)^{1/2 } } v_z + \rho { \sigma f_z(r , z ) \over c } v_z \nonumber \\ & & \nonumber \\ & & + \rho { \sigma s_z(r , z ) \over c } \ ; \times k \left ( \max \left [ { 1 \over \rho \sigma v_{th } } \left| { dv_z \over dz } \right|\ ; , \ ; 10 ^ 8 \right ] \ ; \right)^\alpha v_z \ ; , \nonumber\end{aligned}\ ] ] where @xmath71 is the pressure , @xmath72 is the ratio of specific heats , @xmath25 is the density , @xmath73 is the internal energy per mass , @xmath74 , @xmath75 , and @xmath27 are the corresponding velocity components in cylindrical coordinates , @xmath76 is the gravitational constant , @xmath77 is the mass of the white dwarf , @xmath20 is the thomson cross section per unit mass , @xmath55 and @xmath54 are the corresponding radiation flux components ( see equations [ 11 ] and [ 12 ] ) , @xmath22 is the speed of light , @xmath26 is the ion thermal velocity , and @xmath23 and @xmath24 are defined through equation ( 3 ) , @xmath42 and @xmath41 are defined through equations ( 7 ) and ( 6 ) respectively , and @xmath78 is the total energy per mass . equations ( 14)-(18 ) are solved with the ppm scheme for cylindrical coordinates . we wish to note here that our implementation of the ppm scheme in two dimensions in cylindrical coordinates ( `` z '' and `` r '' ) does not include the third term on the left hand side of equation ( 16 ) , and that we have included and implemented this term in our models explicitly . we note also that the centrifugal and coriolis forces appear naturally under cylindrical coordinates as the third term of the left hand side in equation ( 15 ) and the third term of the left hand side in equation ( 16 ) . in this treatment we have implemented the energy equation self consistently . in particular , the adiabatic heating and cooling effects are implemented through the first and second term on the right hand side of equation ( 18 ) . the emission of the boundary layer is one of the major unsolved problems in the study of cataclysmic variables , and it is very closely related to the emission from the inner regions of the accretion disk . owing to the observational uncertainties discussed in the introduction , we have chosen not to include the boundary layer radiation in our current models . in the model presented here we solve the hydrodynamic equations over a range of radii from @xmath79 ( @xmath80 in our models ) to @xmath81 . we wish to note at this point that we have run additional models over a range of radii from @xmath82 ( @xmath83 in our models ) to @xmath84 finding similar results , comparisons are presented below . from equation ( 1 ) it is found that the energy emission per area @xmath13 will increase slightly as radius decreases from @xmath80 down to @xmath85 and will decrease strongly as radius decreases from @xmath85 down to the white dwarf radii ( @xmath86 ) . the function @xmath87 ( equation [ 1 ] ) was originally derived by shakura and sunyaev ( 1973 ) for binary systems with an accretion disk about a black hole . popham and narayan ( 1995 ) developed detailed models of boundary layers and accretion disks in cataclysmic variables and found that their models supported the same function @xmath87 for the accretion disk in cvs . in paper i we showed that the wind density tends to decrease with decreasing radiation flux . thus the the mass loss rate for radii less that @xmath88 will not alter significantly the total mass loss rate of the wind . as we discovered in paper i , care must be taken with the boundary conditions to ensure not only a numerically stable solution but also a physically stationary one . the ppm scheme ( @xcite ) , in order to obtain the physical parameters ( velocity , density , pressure ) of the wind at a given point of the computational grid @xmath89 for a given time step @xmath90 , requires that at time @xmath91 the values of the physical parameter be known at the given spatial grid point ( @xmath89 ) , at the following two grid points ( @xmath92 and @xmath93 ) , and at the previous two grid points ( @xmath94 and @xmath95 ) . therefore , the boundary condition in ppm must be implemented by not only specifying the values of the physical parameters ( or leaving them free depending on the specific boundary condition ) at each grid point in the boundary of the spatial computational grid , but also by specifying the physical parameters at two grid points beyond each boundary grid point . the latter set of two spatial grid points are named `` ghost zones '' since they are points off the actual computational grid . the lower boundary condition of our model is implemented in the following manner . the lower boundary of our computational grid is set at @xmath96 , which corresponds to the photospheric height at @xmath97 , and which we adopt as a fiducial value over the range of radii of our computational grid ( @xcite , paper i ) . the velocity along the @xmath28 direction is fixed at @xmath98 , which is a fiducial value of the sound speed at the photospheric height over the range of radii implemented in our model . we note that in the inner regions of the disk ( @xmath99 ) , where the temperature will be greater that @xmath100 ( equation [ 2 ] ) , our boundary conditions will be subsonic . since the radial forces balance on the disk surface , the initial wind velocity must be normal to the surface of the disk , therefore the velocity along the @xmath14 direction is fixed at a value of @xmath101 . the velocity along the @xmath64 direction is fixed at keplerian speeds . the density is left free . the pressure at the lower boundary is calculated at each time step according the current the value of density and temperature . the temperature is fixed according to equation ( 2 ) . the photospheric height at the innermost radius of our computational grid ( @xmath102 ) is actually two orders of magnitude less than the photospheric height at @xmath97 ( @xcite ) . we have run our models varying the fiducial photospheric height ( lower computational grid boundary ) from a value of @xmath96 down to a value of @xmath103 and found similar results for lower computational boundaries within that range ( see figure 11 ) . the ghost zones for the lower boundary are set in the following manner . the velocity along the @xmath28 direction are fixed according to the following equation : @xmath104 where @xmath105 is the velocity at the lower boundary ( @xmath106 ) . equation 19 was obtained by an approximate integration of the hydrodynamic equations , in the subsonic region , of the one - dimensional models we presented in paper 1 . zero gradients are implemented for the temperature and for the velocities in the @xmath14 and @xmath64 directions . the density in the ghost zones is calculated such that @xmath107 , to ensure mass conservation . the pressure is calculated at each time step according to the current value of density and temperature . for the left , upper , and right boundaries the physical parameters are left free at each grid point in the computational boundary . zero gradients are implemented at the ghost zones for the physical parameters . the models presented in this paper include several important computational improvements and changed physical assumptions when compared with paper 1 . in order to understand the effects of these separately we present results of intermediate models . as we already indicated in 3 , in the accretion disk wind model presented here , for the force multiplier parameters , we have taken values presented by abbott ( 1982 ) @xmath31 and @xmath32 . in our isothermal models presented in paper 1 , for the force multiplier parameters we took the earlier values used by castor , abbott , and klein ( 1975 ) @xmath33 and @xmath32 . in figure 4 we show results from our isothermal models presented in paper 1 , in figure 5 we show results from a similar model which differs only in the implementation of the higher and more recent value for the @xmath23 parameter . we note that in both models the velocity fields are essentially the same . the mass density distribution is also quite similar except for a factor of @xmath108 in the density values of the newer models . from equation 3 we note that the increase in the value of @xmath23 is equivalent to an increase in luminosity ( with respect to line - radiation pressure ) . in turn from the one - dimensional models presented in paper 1 we showed that , in analogy with early type stars , an increase in luminosity essentially leaves unaltered the wind velocities and increases the overall density in proportion to @xmath109 . thus our models in paper 1 indicate that an increase in a factor of @xmath110 in the @xmath23 parameter should produce an increase in overall density by a factor of @xmath111 which is consistent with the results presented in figure 5 . as seen in equation ( 9 ) , in the accretion disk wind model presented here we have also taken into account the fact that if the velocity gradient increases to sufficiently high values , or the density decreases to sufficiently low values , the contribution for the radiation pressure due to lines arrives at a maximum value . in figure 6 we show the results from a model in which we implement this condition . the models represented in figures 5 and 6 differ only in that former model does not implement a maximum value for the line radiation pressure while the latter does . we wish to note here two of the effects we find in implementing a maximum value for the force multiplier . first , in the region high above the inner disk area ( upper left region in figures 5 and 6 ) a decrease in velocity is obtained . we find that the `` z '' velocities components in the fore mentioned region in figure 5 ( @xmath112 ) are higher than the respective velocities in figure 6 ( @xmath113 ) . this difference is due to the fact that the region high above the inner disk area is a low density region since centrifugal forces have driven most of the mass towards larger `` @xmath14 '' ( towards the right in figures 5 and 6 ) at the corresponding heights ( @xmath114 ) . in turn the decrease in density in this region produces an increase in the line radiation pressure ( see equations [ 3 ] and [ 9 ] ) which is allowed to increase indefinitely in the model of figure 5 , but may only increase up to a maximum value in the model of figure 6 . second , we also find that the implementation of a maximum value for the force multiplier also produces an increase in density in the region near the base of the wind by a factor of @xmath115 . in our accretion disk wind model we also introduced an adiabatic wind rather than an isothermal wind as we assumed in paper 1 . in figure 7 we show results from a model which implements an adiabatic wind . the models represented in figures 6 and 7 differ only in that former model assumes an isothermal wind while the latter assumes an adiabatic wind . we do not find significant changes in the density distributions or the wind velocity field when we implement an adiabatic wind rather than an isothermal wind ( see figures 6 and 7 ) . another element that we have introduced in our accretion disk wind model is the radial structure of radiation emission ( equation [ 1 ] ) and temperature of disk surface ( equation [ 2 ] ) , rather than assume an isothermal disk as we did in the models presented in paper 1 . in figure 8 we show results from our accretion disk wind model . the models represented in figure 7 and figure 8 differ in that former assumes an isothermal disk ( and therefore a uniform temperature and radiation emission distribution over the disk surface ) and neglects the radiation pressure along the `` @xmath14 '' direction , while the latter assumes a radial temperature and radiation emission distribution as described in 2 and includes the radiation pressure along the `` @xmath14 '' direction . comparing the results shown in figures 7 and 8 , we note that the enhanced density region above the disk extends from the mid - region of the disk in figure 7 , while in figure 8 the enhanced density region extends from the inner region of the disk . this difference is due to the assumption of an isothermal disk in the figure 7 model which produces a uniform density distribution at the base of the wind , while in figure 8 the higher radiation emission from the inner disk region ( equation [ 1 ] ) produces in turn a higher density at the inner region than in the outer region at the base of the wind . the radial radiation emission structure of the accretion disk wind model causes the density in the inner region of the disk at the base of the wind to be higher than the corresponding density at the same region in the isothermal disk model ( see figures 7 and 8) . this in turn causes an increase in the overall wind mass loss rate by a factor of @xmath115 in the accretion disk wind model ( figure 8) with respect to the isothermal disk wind model ( figure 7 ) . the results shown in figure 8 allow us to estimate the validity of our neglect of radiative cooling . we present here rough estimates of the radiative cooling times and compare them with dynamical time scales . blondin ( 1994 ) obtained an analytical expression for radiative cooling in high - mass x - ray systems . he found for wind temperatures on the order of @xmath116 a radiative cooling coefficient of @xmath117 . rough estimates for the cooling times may be obtained through the following expression @xmath118 , where @xmath23 is boltzmann s constant , @xmath119 is the temperature , and @xmath120 is the particle density . from the results of figure 8 ( discussed below ) we obtain the following rough estimates for the radiative cooling times : lower inner region ( @xmath121 , @xmath122 ) @xmath123 ; lower outer region ( @xmath124 , @xmath122 ) @xmath125 ; higher inner region ( @xmath121 , @xmath126 ) @xmath127 ; higher outer region ( @xmath124 , @xmath126 ) @xmath128 . a rough estimate of the dynamical time maybe obtained by @xmath129 , @xmath130 . viewing figure 8 we note that radiative cooling time in the high density areas tend to be comparable to or larger than the dynamical time . this suggests that in some parts of this region , radiative cooling may be important and the eos could be described as isothermal . in the lower density areas the radiative cooling time scales tend to be much larger than the dynamic time scale , thus indicating that for this region the eos would be best described as adiabatic . these estimates are uncertain by factors of an order of magnitude or more , owing to the fact that they assume optically thin line cooling and are not tailored to the conditions specific to cv winds . however , they provide motivation for our future work , in which we shall include radiative heating and cooling in our models . the spatial computational grid in the @xmath28 direction has @xmath131 points , ranging from the photospheric height to @xmath132 , distributed such that @xmath133 is uniform . the number of grid points for the spatial computational grid in the @xmath14 direction was set at different values which varied from @xmath134 to @xmath135 distributed uniformly . we obtain similar results for each of the grid resolutions implemented within the above range . a vector field graph of wind velocity superimposed with a contour graph of wind density is presented in figure 8 for the low resolution model . a vector field graph of wind velocity superimposed with a contour graph of wind temperature is presented in figure 9 . the physical parameters used are : @xmath136 , @xmath137 , and @xmath138 . we obtain the following values for the mass - loss rate and terminal velocity for our two - dimensional accretion disk wind model : @xmath3 ( obtained by integrating at the photospheric height ) and @xmath139 . we note that the wind mass loss rate we find for our accretion disk wind model is approximately two orders of magnitude larger than the value we found in paper i for our isothermal models . this is due to three reasons : first , the radiation field above the inner region of disk in the accretion disk wind models presented in this work is stronger than in the models presented in paper i in which we assumed an isothermal disk . second , we have used a more recent and larger value for the @xmath23 force multiplier parameter in our accretion disk wind model . for the line radiation pressure , the effect of increasing @xmath23 is equivalent to increasing luminosity ( see equations [ 6 ] , [ 7 ] , and [ 9 ] ) . as we had shown in paper i , through our one - dimensional isothermal models , an increase in disk luminosity produces a corresponding increase in wind mass - loss rate . third , in our accretion disk wind model we implement the fact that , if the velocity gradient increases to sufficiently high values or the density of the wind decreases to sufficiently low values , the contribution for the radiation pressure due to lines arrives at a maximum value ( see equation [ 9 ] ) . this causes a decrease in the line radiation pressure in the regions of the wind where this phenomena occurs . of these three factors , the second , the use of a larger and more recent value for @xmath23 , is the most important , accounting for slightly more than a order magnitude increase in the wind mass loss rate . in figure 10 we show results obtained from the low resolution model ( @xmath134 computational grid points in the @xmath14 direction ) and from the high resolution model ( @xmath135 computational grid points in the @xmath14 direction ) , and as stated above we find similar results ( the density contour values on both figures 10_a _ and 10_b _ are the same ) . we note structure in the density contours in the @xmath135 computational grid which is not apparent in the density contours in the @xmath134 computational grid . figures 10_a _ and 10_b _ are shown at the same physical times , in which steady flows have been obtained . the motivation for the high resolution models was to ensure that the overall results of our low resolution models were not affected by the selection of vertical computational grid . comparing figures 10_a _ and 10_b _ we note structure in the density contours in the @xmath135 computational grid which is not apparent in the density contours in the @xmath134 computational grid . we have not yet fully explored the density spikes in the high resolution model . however , the average properties of the flow such as total mass loss rate and gross structures , are equivalent in the two models . in figure 11 we show results obtained from a model with a height of @xmath103 at the lower computational grid boundary ( figure 11_b _ ) and compare it with the results from a model with a height of @xmath96 at the lower computational grid boundary ( figure 11_a _ ) ( @xmath96 is the lower computational boundary used in the models presented elsewhere in this paper ) . the density contour values on both figures 11_a _ and 11_b _ are the same . as we indicated in 4.2 , these two heights correspond to the photospheric height in the innermost radius of our computational grid ( @xmath140 ) and to the photospheric height in the radius of @xmath141 respectively . in the models with lower photospheric heights we did not increase the number of grid points in the photospheric region . the motivation for this is that in the region between the original photospheric height and the smaller values , the radiation field is not changing considerably . this , added to the consideration that this region is appreciably smaller than the total spatial computational grid , would indicate that a finer vertical grid should produce no significant difference , while considerably increasing the cpu time required for the actual runs . in figure 12 we present a vector field graph of wind velocity superimposed with a contour graph of density for the accretion disk wind model with the same physical parameters , but implementing a reflecting wall boundary condition for the left side of the computational grid . that is , the values for the physical parameters ( velocity , density , and pressure ) are left free at the grid points in the left computational boundary , and are therefore recalculated at each time step . `` mirrored '' values are implemented at the ghost zones for the physical parameters for the reflecting wall boundary conditions . with the reflecting wall boundary condition we obtain quite similar results ( see figures 8 and 12 ) . we wish to note that in figure 12 the `` r '' components of the velocity in the upper region at the inner left boundary do not indicate that there is a mass flow coming from the inner left boundary , but rather that the gas pressure is overcome by centrifugal forces in this region . we also wish to note that in the reflecting wall boundary condition the grid points on the left computational boundary are not at the `` wall '' , but rather near the `` wall '' . the `` wall '' would be located in between the boundary grid points and the `` ghost zone '' grid points ( see 4.2 ) . in figure 13 we show results obtained from the low resolution model ( figure 13_a _ ) and two additional high resolution models ( figures 13_b _ and 13_c _ ) . as stated above the low resolution model contains @xmath134 computational grid points in the @xmath14 direction ranging from @xmath79 ( @xmath142 ) to @xmath81 , and @xmath131 grid points in the @xmath28 direction ranging from @xmath143 to @xmath144 . in this figure we show the results of the low resolution model up to @xmath84 in both directions in order to present them at the same scale as figures 13_b _ and 13_c_. the high resolution models represented in figure 13_b _ and 13_c _ have @xmath135 grid points in the @xmath14 direction ranging up to @xmath84 , and @xmath145 grid points in the @xmath28 direction ranging from @xmath146 to @xmath84 . figures 13_b _ and 13_c _ differ in that in the former the grid points in the @xmath14 direction range down to @xmath79 ( @xmath142 ) , while in the latter the grid points in the @xmath14 direction range down to @xmath147 ( @xmath148 ) . as we stated above we find similar results between the low and the high resolution models . the velocity fields are essentially the same and the density in the corresponding region are of the same magnitude . although we do find more structure in the density contours of the high resolution models , which are not apparent in the low resolution models . in the high resolution models we also find that near the disk surface towards larger radii lower densities are obtained . the motivation for the additional high resolution models was to ensure that the overall results of our models would not be affected by a more detailed modeling of the wind intiation region through the increase of the physical resolution on one hand and by including in the models the region between @xmath149 on the other . we wish to note here that in the figures 13_a _ , 13_b _ , and 13_c _ the velocity vectors presented with the lowest @xmath28 are the lowest @xmath28 in which the velocity vectors are plotted , the actual computational grid starts at a lower @xmath28 as described above in this paragraph . an important feature of the two - dimensional accretion disk wind model results , the role of rotational forces , is apparent in figures 8 and 9 . the centrifugal forces cause the streamlines from the inner region of the disk to bend and to collide with streamlines of material originating at larger radii . this collision generates a large region of enhanced density which runs diagonally in the @xmath28-@xmath14 plane ( this effect was also observed in paper i for our two - dimensional isothermal models ) . an important consequence of the enhanced density region is the corresponding increase in the line optical depth . icke ( 1981 ) suggested that biconical winds were a general property of accretion disk winds independent of the wind driving mechanism , and that this was caused by the geometry of the disk wind and by the rotation of the disk . in this work we have confirmed this suggestion . we find that the principal driving mechanism for accretion disk winds in cvs is the line radiation pressure and that for our two - dimensional models we find that centrifugal forces , which are caused by the rotation of the disk , produce an enhanced density region which is biconical in nature ( figure 8) . in order to further study line - driven accretion disk winds we calculate 1550 line profiles from our accretion disk wind model . we then study the line profiles calculated at different angles and compare them with observations . the line profiles presented here ( figure 14 ) were calculated assuming single scattering and a relative abundance of carbon with respect to hydrogen of @xmath150 throughout the wind . the radiation was assumed to have originated on the surface of the disk , according to equation ( 1 ) , in consistency with our model . the angles for the line of sight with respect to the disk rotation axis were discretized into eighteen values : from @xmath151 to @xmath152 in @xmath153 increments . we wish to note here that although the three - dimensional hydrodynamic equations were reduced to and implemented in two dimensions through azimuthal symmetry ( 2.5-dimensional ) , the line profiles were calculated in three dimensions . for a given point source on the disk surface and a given direction , the geometric line starting from the given point and having the given direction , was divided into several segments as it intersected the spatial computational grid . a three - dimensional computational grid was formed by rotating the spatial grid used in the hydrodynamic calculations along the @xmath64 direction . continuous values of @xmath64 were used to calculate the intersections . over each of these segments the optical depth for each frequency bin , corresponding to a given velocity bin , was obtained by taking into account the ion thermal speed @xmath26 and applying the expression @xmath154 along a sobolev length ( a sobolev length is defined as the distance along which the wind velocity varies by a value equal to the sound speed ) . that is , for a given point source on the disk surface and a given direction , the following equations are implemented : @xmath155 @xmath156 @xmath157 @xmath158 where @xmath159 is the optical depth for frequency @xmath160 , @xmath161 is the component of the wind velocity along the given direction , @xmath26 is the thermal speed , @xmath25 is the wind density , @xmath162 is the line opacity for 1550 , @xmath163 is the resonance frequency for 1550 , @xmath22 is the speed of light , @xmath164 is the intensity for frequency @xmath160 for non - scattered radiation along the given direction , and @xmath165 is the intensity for frequency @xmath160 along the given direction at the disk surface . we note here that the above calculations were done in the vicinity of the 1550 resonance frequency in order to calculate the corresponding line profiles . the total optical depth , for a given point source on the disk surface , for a given direction , for a given velocity bin , was calculated by summing over the contributions of each segment described above . in the line profiles calculations presented here it was assumed that the radiation absorbed at each segment is scattered isotropically , and to therefore contribute uniformly to the radiation in each direction . for each direction of scattered radiation , the velocity of the wind was taken into account through doppler shifts , and then added on to the corresponding velocity bin . the velocity component along the @xmath64 direction was taken into account consistently throughout the line profile calculations . finally , after implementing all of the corresponding integrals , the contribution of the scattered and non - scattered radiation for each velocity bin was summed for each observation angle and the line profiles obtained . the velocity values were discretized in bins of @xmath166 . although this bin size is greater that the ion thermal speed ( @xmath167 ) it is much less than the wind velocities and emission line widths that characterize line profiles in cvs ( @xmath168 ) . for low inclination angles ( @xmath169 ) our model predicts p cygni profiles . the blue - shifted absorption line presents a maximum absorption at roughly half the terminal velocity , which is consistent with observations ( @xcite ; @xcite ) . this feature distinguishes the p cygni line profiles from cvs from the p cygni line profiles seen from early type stars whose blue - shifted absorption line presents a maximum absorption at the terminal velocity ( @xcite ) . furthermore the absorption line corresponds to velocities in the order of @xmath170 , which also agrees with observations . for example , the line profile in figure 16 shows a strong similarity to the line profiles observed from rw sex ( @xcite , @xcite , @xcite ) which has an inclination angle of @xmath171 ( @xcite ) . for low inclination angles close to the orbital axis ( @xmath172 ) our models predict p cygni profiles with a double peaked absorption lines ( see figure 15 ) . such features have been reported in low inclination cvs . figure 15 may be compared with the line profiles observed in hr del ( @xcite ; @xcite ) , which has an inclination angle of @xmath173 ( @xcite ) . for intermediate and high inclination angles ( @xmath174 ) our model predicts p cygni profiles without the absorption component , which is also consistent with observations ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . for intermediate inclination angles ( @xmath175 ) our model predicts p cygni profiles with double peaked emission lines ( see figure 17 ) . such features have been reported in intermediate inclination cvs . figure 17 may be compared with the line profiles observed in ux uma ( @xcite ; @xcite ) , which has an inclination angle of @xmath176 ( @xcite ) . our models suggest that the double peaked emission line observed in cvs may be interpreted as a wind feature . for high inclination angles ( @xmath177 ) our model predicts p cygni profiles with single peaked emission lines with a fwhm in the order of @xmath178 ( see figure 18 ) . singled peaked emission lines are observed in high inclination cvs . figure 18 may be compared with the line profiles observed in rw tri ( @xcite ) , which has an inclination angle of @xmath179 ( @xcite ) . for nearly perpendicular inclination angles ( @xmath180 ) our model predicts p cygni profiles with single peaked emission lines , with a value of the fwhm about @xmath181 less than high inclinations , presenting two satellite peaks due to emission by material near the disk surface ( see figure 19 ) . figure 19 may be compared with the line profiles observed in dq her ( @xcite ) , which has an inclination angle of @xmath182 ( @xcite ) and which shows a comparable fwhm and two satellite peaks . thus the line profiles for ( 1550 ) obtained through our computational accretion disk wind model are consistent with observations in their general form , in particular in the maximum absorption at roughly half the terminal velocity for the blue - shifted component , in the magnitudes of the wind velocities implied by the absorption components , in the fwhm of the emission components , and in the strong dependence in inclination angle . however , we wish to emphasize that the models presented here do not self consistently calculate the wind ionization balance and therefore can only be considered as a preliminary indication that our models are consistent with observations . we plan to present more accurate models in our next paper . we have developed a 2.5-dimensional hydrodynamic line - driven adiabatic accretion disk wind model . our model solves a complete set of adiabatic hydrodynamic partial differential equations , using the ppm numerical scheme and implementing the radial temperature and radiation emission distributions on the surface of an accretion disk . from our models we calculate wind mass - loss rates , terminal velocities , and line profiles for ( 1550 ) for various angles . for typical cataclysmic variable parameter values of disk luminosity @xmath0 , white dwarf mass @xmath1 , and white dwarf radii @xmath2 , we obtain a wind mass - loss rate of @xmath3 , and a terminal velocity of @xmath183 . studies of observed ultraviolet line profiles from cvs have estimated cv wind mass loss rates to be in the order of @xmath184 to @xmath185 and found cv wind terminal speeds in the order of @xmath186 ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) , thus the models developed in this work predict values for wind mass loss rates and terminal speed consistent with observations . furthermore , the line profiles we obtain through our computational model are also consistent with observations in their general form , in particular in the maximum absorption at roughly half the terminal velocity for the blue - shifted component , in the magnitudes of the wind velocities implied by the absorption components , in the fwhm of the emission components , and in the strong dependence in inclination angle . we have shown that rotational forces are important in the study of winds from accretion disks . they cause the velocity streamlines to collide which results in an enhanced density region . in this region the wind speed is reduced and the wind density is increased . the increase in density caused by the collision of streamlines is important because it permits the appearance of blue - shifted absorption lines as observed in p cygni profiles of low - inclination cvs . a shortcoming of our models is the neglect of radiative cooling and heating and local ionization equilibrium throughout the wind . in the next paper of this series we will present models which include these effects . we also plan to extend our models to the study of the wind dynamics of low mass x - ray binaries and active galactic nuclei , where these processes may play a more important role in determining the overall dynamics . abbott , d. 1982 , , 259 , 282 beuermann , k. , thorstensen , j. , schwope , a. , ringwald , f. , sahin , h. 1992 , , 256 , 442 blondin , j. 1994 , , 435 , 756 castor , j. , abbott , d. , & klein , r. 1975 , , 195 , 157 cheng , f. , sion , e. , szkody , p. , & huang , m. 1997 , , 484 , l149 colella , p. , & woodward , p. 1984 , j. comput . phys . , 54 , 174 crdova , f. , & mason , k. 1982 , , 260 , 716 crdova , f. , & mason , k. 1985 , , 290 , 671 dhillon , v. , marsh , t. , & jones , d. 1991 , , 252 , 342 drew , j. 1987 , , 224 , 595 ferguson , d. 1997 , , 486 , 987 frank , j. , king , a. , sherrington , m. , jameson , r. , & axon , d. , 1981 , , 195 , 505 frank , j. , king , a. , & raine , d. 1992 , accretion power in astrophysics , second edition ( cambridge : university press ) friedjung , m. 1992 , , 262 , 487 friedjung , m. , selvelli , p. , & cassatella , a. 1997 , , 318 , 204 gnsicke , b. , beuermann , k. , & thomas , h. 1997 , , 289 , 388 greenstein , j. , & kraft , r. 1959 , , 130 , 99 greenstein , j. , & oke , j. 1982 , , 258 , 209 guinan , e. , & sion , e. 1982 , , 258 , 217 heap , s. , et al . , 1978 , , 275 , 385 holm , a. , panek , r. , & schiffer iii , f. 1982 , , 252 , l35 honeycutt , r. , schlegel , e. , & kaitchuck , r. 1986 , , 302 , 388 horne , k. , & , marsh , t. 1986 , , 218 , 761 horne , k. , marsh , t. , cheng , f. , hubeny , i. , & lanz , t. 1994 , , 426 , 294 huang , s. 1972 , , 171 , 549 hutchings , j. 1980 , , 92 , 458 icke , v. 1980 , , 85 , 329 icke , v. 1981 , , 247 , 152 kaitchuck , r. , honeycutt , r. , & schlegel , e. 1983 , , 267 , 239 kallman , t. 1988 , adv . space . res . , 8 , 259 kiplinger , a. 1979 , , 234 , 997 knigge , c. , long , k. , blair , w. , & wade , r. 1997 , , 476 , 291 knigge , c. , woods , j. , & drew , j. 1995 , , 273 , 225 krautter , j. , et al . , 1981 , , 102 , 337 leibert , j. 1980 , , 18 , 363 lin , d. , williams , r. , & stover , r. 1988 , , 327 , 234 long , k. , mauche , c. , raymond , j. , szkody , p. , & mattei , j. 1996 , , 469 , 841 marsh , t. , & horne , k. 1990 , , 349 , 593 mason , k. , et al . , 1995 , , 274 , 271 mauche , c. , & raymond , j. 1987 , , 323 , 690 mauche , c. , raymond , j. , & mattei , j. 1995 , , 446 , 842 murray , n. , chiang , j. 1996 , , 382 , 789 murray , n. , chiang , j. 1997 , , 474 , 91 murray , n. , chiang , j. , grossman , s. , voit , g. 1995 , , 451 , 498 paczyski , b. 1991 , , 370 , 597 patterson , j. 1984 , , 54 , 443 pereyra , n. , kallman , t. , & blondin , m. 1997 , , 477 , 368 popham , r. , & narayan , r. 1991 , , 370 , 604 popham , r. , & narayan , r. 1995 , , 442 , 337 pringle , j. 1981 , , 19 , 137 prinja , r. , & rosen , s. 1995 , , 273 , 461 proga , d. , stone , j. , & drew , j. , 1998 , in preparation shakura , n. , & sunyaev , r. 1973 , , 24 , 237 shlosman , i. , & vitello , p. 1993 , , 409 , 372 smak , j. 1969 , acta astr . , 19 , 155 smak , j. 1981 , acta astr . , 31 , 395 smak , j. 1994 , acta astr . , 44 , 59 still , m. , dhillon , v. , & jones , d. 1995 , , 273 , 849 thorstensen , j. , ringwald , f. , wade , r. , schmidt , g. , & norsworthy , j. 1991 , , 102 , 272 vitello , p. , & shlosman , i. 1988 , , 327 , 680 vitello , p. , & shlosman , i. 1993 , , 410 , 815 vrtilek , s. , et al . , 1994 , , 425 , 787 wade , r. 1988 , , 335 , 394 warner , b. 1987 , , 227 , 23 young , p. , schneider , d. , & shectman , s. 1981 , , 245 , 1035
supersymmetric standard model is advertised as a great successful theory in particle physics and cosmology . the hierarchy problem is completely solved by the presence of superpartners . a perturbative nature of the model allows us to calculate gauge coupling evolutions in high energy and remarkably the three coupling constants meet at very high energy . this is a strong indication of supersymmetric grand unification @xcite . in cosmology , the neutralino , the superpartner of the gauge or higgs fields , can account for dark matter of the universe . the thermal relic abundance of the neutralino is in the right ballpark for masses of order 100 gev . when we look at actual models of supersymmetry breaking and its mediation , we encounter many unnatural aspects or even inconsistencies . in gravity mediated supersymmetry breaking scenario @xcite , there is a serious cosmological difficulty @xcite as well as problems of flavor changing neutral current ( fcnc ) and cp violating processes . a supersymmetry breaking sector necessarily requires to contain a gauge singlet field whose vacuum expectation value in the @xmath1-component gives masses to gauginos . after inflation , the scalar component of the singlet field starts coherent oscillation . once it dominates over the energy density of the universe , it either spoils the success of big - bang nucleosynthesis ( bbn ) @xcite or overproduces gravitinos @xcite ( see @xcite for earlier works ) . therefore , for the mechanism of thermal relic neutralinos as dark matter to work , we need extra assumptions on the cosmological history so that we can avoid the oscillation to dominate the energy density . models with anomaly mediation @xcite can prevent such oscillation since singlet fields are not necessary to generate the gaugino masses . however , the annihilation cross section of the lightest neutralino , the wino in this scenario , is too large for the thermal relic to naturally explain the dark matter abundance . there seems to be no coherent picture for cosmology of supersymmetric models . models of gauge mediation @xcite also suffer from serious difficulties . the neutralino can not be a stable particle since the gravitino is always the lightest supersymmetric particle ( lsp ) in this scenario . moreover , enhanced production cross sections of gravitinos bring up a danger of overproduction in the thermal bath @xcite . this restricts the reheating temperature of the universe to be low enough , which makes baryogenesis difficult . there is also a naturalness issue in the electroweak sector , i.e. , the @xmath0-problem . we need to give a supersymmetric mass term , @xmath0-term , for the higgs fields which needs to be similar size to supersymmetry breaking terms . unlike in gravity mediation models , we can not relate this @xmath0-term to the gravitino mass via the giudice - masiero mechanism @xcite because the gravitino mass is too small . it is possible to assume a direct coupling between the higgs sector and a supersymmetry breaking sector to obtain the @xmath0-term , but it often induces too large @xmath2-term . @xcite . we can not say that the minimal supersymmetric standard model ( mssm ) is a successful model without having a consistent picture for supersymmetry breaking / mediation . it has been more than twenty years since discussions of supersymmetry phenomenology started , and until recently it has served as the most promising candidate of physics beyond the standard model . however , recent various progress in astrophysical observations and theoretical calculations such as the measurement of the matter density of the universe @xcite and bbn constraints on the gravitino abundance @xcite put a threat to many supersymmetric models . at the same time , these are new strong hints which guide us to the true theory . in this paper , we find that a simple model of gauge mediation actually provides us with a viable and natural scenario for cosmology and phenomenology . the model we discuss consists of a singlet field @xmath3 and messenger fields @xmath4 and @xmath5 . the singlet field @xmath3 breaks supersymmetry spontaneously and that is mediated to the mssm sector by loop diagrams of the messenger fields . the supersymmetry breaking vacuum is stabilized by gravitational effects @xcite . this is the entire model . under reasonable assumptions , the @xmath3 field starts coherent oscillation in the early universe . if we ignore interactions of @xmath3 with the messenger fields , decays of the @xmath3 field either destroy bbn or overproduce gravitinos , which is the same problem in gravity mediation scenarios . a renormalizable coupling to the messengers , however , drastically modifies the decay property of @xmath3 . especially , @xmath3 can decay into the mssm particles well before bbn , and , moreover , the branching fraction to the gravitinos is significantly suppressed . with gaugino masses of @xmath6 gev , gravitinos produced by the @xmath3 decay can be dark matter of the universe when the gravitino mass is @xmath7 mev or @xmath8 mev , which corresponds to the mass of @xmath3 to be @xmath9 gev and @xmath10 gev , respectively . in fact , this is the most interesting parameter range from the consideration of the @xmath0-problem . the @xmath0-problem can be solved by the mechanism of @xmath0-term driven supersymmetry breaking scenario @xcite when @xmath11 . in the next section , we explain the gauge - mediation model we consider . the concept of @xmath0-term driven supersymmetry breaking is also discussed . in section [ sec : evolution ] , we discuss cosmological evolutions of the @xmath3 field . we show that @xmath3 rolls into a preferable ( supersymmetry breaking ) vacuum in generic situations . the decay property of @xmath3 is discussed in section [ sec : dark - matter ] . the bbn constraint on parameters and the abundance of non - thermally generated gravitino dark matter are calculated . the discussion on the initial condition for the @xmath3 oscillation is given in appendix [ app : amplitude ] . the model we study has a very simple structure . we simply introduce messenger fields to any supersymmetry breaking model of the oraifeartaigh type , i.e. , models in which the superpotential contains a linear term of a gauge - singlet chiral superfield . the low - energy effective lagrangian is defined by the following k " ahler potential and superpotential : @xmath12 @xmath13 where @xmath3 is a gauge singlet field in the oraifeartaigh model and @xmath4 and @xmath5 are the messenger fields which have quantum numbers of the standard model gauge group . we introduced an interaction term @xmath14 where @xmath15 is a coupling constant . the constant term @xmath16 contributes to the cosmological constant through @xmath17 term in a scalar potential . the non - renormalizable term @xmath18 in the k " ahler potential represents the effect of the massive fields ( in the oraifeartaigh model ) which have been integrated out . although this model does not break supersymmetry in the limit of the planck scale @xmath19 to infinity , a supersymmetry breaking vacuum appears far away from the origin of @xmath3 after taking into account supergravity effects @xcite . at the supersymmetry breaking vacuum , @xmath20 and @xmath21 the messenger particles obtain masses , @xmath22 , in this vacuum . this vacuum is meta - stable when @xmath23 where @xmath24 is the number of components in @xmath4 , e.g. , @xmath25 for a pair of @xmath4 and @xmath26 in @xmath27 and @xmath28 representation of su(5 ) . the constant term @xmath16 in the superpotential is chosen so that the cosmological constant is cancelled at the supersymmetry breaking vacuum , i.e. , @xmath29 where the gravitino mass is given by @xmath30 the scalar component of @xmath3 obtains a mass by supersymmetry breaking effects : @xmath31 supersymmetry breaking by @xmath32 induces gaugino masses by one - loop diagrams with messengers @xmath4 and @xmath26 . it is calculated to be : @xmath33 where @xmath34 is the number of the messenger particles ( @xmath35 for a pair of * 5 * and @xmath28 representation of su(5 ) ) . the phenomenological requirement that the gaugino mass should be @xmath6 gev determines the ratio @xmath36 . among three parameters @xmath37 , @xmath38 and @xmath15 in this model , low energy physics is not very sensitive to @xmath15 . it only appears through @xmath39 where @xmath40 is the messenger scale of gauge mediation . with fixed gaugino masses , we are left with a single parameter @xmath38 . there is a particularly interesting scale for @xmath38 from the consideration of uv completions of the theory . among various possible uv completions ( see ref . @xcite for various possibilities ) , the simplest one is to identify the field @xmath3 with a meson field of a strongly coupled gauge theory . for example , in an so(5 ) gauge theory with a quark @xmath41 and a superpotential term : @xmath42 the low energy effective theory below the dynamical scale @xmath43 is simply @xmath44 in one of two branches of the theory . supersymmetry is broken by @xmath45 . this vacuum is found to be meta - stable recently by intriligator , seiberg and shih @xcite . the k " ahler potential is unknown but can be parametrized to be @xmath46 if @xmath47 is positive , this is exactly the same effective theory we discussed in the previous subsection . the scale @xmath38 is identified with @xmath48 . the messenger fields couple to @xmath3 through @xmath49 in the superpotential with @xmath50 the coefficient of the operator . the @xmath15 parameter in eq . ( [ eq : superpotential ] ) is then @xmath51 . in this model , the @xmath0-term in eq . ( [ eq : mass ] ) can actually be the @xmath0-term for the higgs fields @xmath52 in models with composite higgs fields @xcite . the higgs fields @xmath53 and @xmath54 are meson fields @xmath55 and @xmath56 where @xmath57 and @xmath58 carry the standard model quantum numbers . the mass term for @xmath41 in eq . ( [ eq : mass ] ) therefore induces a mass term for mesons . ( for detailed discussion , see @xcite . ) since @xmath3 and the higgs fields @xmath53 and @xmath54 are strongly coupled , we expect to have terms in the k " ahler potential : @xmath59 with @xmath60 , these terms induce the soft supersymmetry breaking terms @xmath61 independent of the dynamical scale @xmath62 . this provides a solution to the @xmath0-problem in gauge mediation . the @xmath0-term is related to the supersymmetry breaking parameters because it is the source of the supersymmetry breaking . a combination of the gravitational gauge mediation and @xmath0-term driven supersymmetry breaking provides a completely natural scenario for gauge mediation . the @xmath37 parameter in eq . ( [ eq : superpotential ] ) is @xmath63 in this scenario ( under a reasonable assumption that @xmath64 ) , and it fixes the dynamical scale @xmath65 by the requirement @xmath66 gev from eq . ( [ eq : gaugino ] ) . for having this relation , the factor @xmath67 needs to compensate the one - loop factor @xmath68 , which determines the dynamical scale to be of the order of the gut scale @xmath69 gev . this is indicating a close connection between dynamics of supersymmetry breaking and spontaneous symmetry breaking in grand unified theories . in fact , the explicit model in ref . @xcite is based on a gut model . a meson field made of @xmath57 and @xmath58 introduced before plays a role of the higgs field which breaks gut symmetry dynamically . this mechanism simultaneously provides us with a solution to the doublet - triplet splitting problem with sufficiently suppressed dimension five operators for proton decays . with @xmath70 , the mass of scalar component of @xmath3 , @xmath71 , is @xmath72 gev ( see eq . ( [ eq : ms ] ) ) , and the gravitino mass is @xmath73 gev . we will use @xmath71 and @xmath74 as free parameters instead of @xmath37 and @xmath38 . relations among them are : @xmath75 @xmath76 the bino mass is expressed in terms of these two by @xmath77 there are no @xmath78 ambiguities in these relations . as we have seen in the previous section , the @xmath3 field is stabilized at the supersymmetry breaking vacuum by supergravity effects . however , there is a true and unwanted supersymmetric vacuum at @xmath79 and @xmath80 . it is thus non - trivial whether @xmath3 can settle to the supersymmetry breaking minimum in the thermal history of the universe . this question is addressed in ref . @xcite where it is concluded that thermal effects make the supersymmetric vacuum more attractive once the messenger particles are thermalized . in our scenario , it is possible that the temperature of the universe never exceeds the mass of messenger particles so that thermal effects are not important . we show in this section that @xmath3 runs into the supersymmetry breaking minimum under reasonable assumptions . it is found to be rather difficult to fall into the supersymmetric minimum @xmath81 because it can be thought of a small hole in the vast @xmath3 space . the relevant scalar potential of the @xmath3 field in the @xmath82 region is given by @xmath83 the potential for @xmath84 region depends on a uv completion , but it is not important for our discussion . the logarithmic term is induced at one - loop level due to the interaction with messenger particles . the supersymmetry breaking vacuum given in eq . ( [ eq : vev ] ) is determined by a balance between the mass term and the linear term . the logarithmic term is important near @xmath81 . the @xmath3 field , however , feels much stronger force toward the supersymmetry breaking vacuum at most points on the complex @xmath3 plane since it is supported by a quadratic potential . this is why the supersymmetry breaking vacuum is more attractive at a low temperature . if the value of @xmath3 is not at the origin after inflation , @xmath3 rolls down to the @xmath85 vacuum . the minimum of the scalar potential during inflation is not at @xmath79 or @xmath86 , but is likely to be around @xmath87 ( see appendix [ app : amplitude ] for detailed discussion ) . once inflation starts , @xmath3 is trapped at the minimum @xmath88 and freezes until the end of inflation . after inflation , the value of @xmath3 remains fixed at @xmath89 for a while until the mass term @xmath71 becomes important compared to the hubble mass term of @xmath90 , and eventually , it starts to oscillate about the supersymmetry breaking minimum . in general , the initial value @xmath91 is not aligned to the @xmath92 direction , which makes motion of @xmath3 almost confined in a straight line on the complex plane . therefore , @xmath3 never approaches to the supersymmetric minimum . the amplitude of the oscillation gradually decays due to the cosmic expansion and @xmath3 settles to the supersymmetry breaking minimum . field ( left ) and time dependence of the amplitude @xmath93 ( right ) for @xmath94gev , @xmath95mev and @xmath96 . , title="fig:",height=181 ] field ( left ) and time dependence of the amplitude @xmath93 ( right ) for @xmath94gev , @xmath95mev and @xmath96 . , title="fig:",height=181 ] in fig . [ fig : oscillation ] , we show a typical evolution of the @xmath3 field for @xmath94gev , @xmath95mev and @xmath97 . here , the initial condition we took is @xmath98 . as we expected @xmath3 oscillates along an almost straight line about the supersymmetry breaking minimum ( see also a closeup in the left panel ) . we also show time dependence of the amplitude @xmath93 during oscillation in the right panel . the figure shows that the impact parameter to @xmath79 during the oscillation is of @xmath99 and the amplitude of the oscillation decays gradually . eventually , @xmath3 settles to the supersymmetry breaking minimum . in the above analysis , effects of the logarithmic potential in eq . ( [ eq : scalarpot ] ) are not important as long as @xmath100 . it does attract @xmath3 to the supersymmetric minimum , but the attractive force is only important when @xmath3 approaches very close to the origin . with an impact parameter of @xmath101 , the effect is negligible . the messenger fields have masses , @xmath102 , when @xmath3 is away from the origin . with the large impact parameter of the oscillation of @xmath3 , @xmath4 and @xmath103 are always stabilized at @xmath104 . although they are always massive , time variation of the masses by the @xmath3 oscillation causes a non - perturbative production of the messenger fields @xcite . once they are copiously produced , it gives rise to a force to attract @xmath3 to origin so that the energy density is reduced by minimizing the mass of @xmath4 @xcite . however , since the variation of @xmath105 is quite adiabatic ( i.e. , @xmath106 ) , the effect is exponentially suppressed . we need to check here that the messenger particles do not enter thermal equilibrium for the above discussion to hold . if they are produced in the thermal bath , it induces a thermal mass term to @xmath3 which attracts to @xmath81 @xcite . however , as long as temperature is always ( including the inflaton oscillation era ) lower than @xmath102 , production of messenger particles is also exponentially suppressed . with the large impact parameter @xmath107 and initial amplitude @xmath108 of this scenario , the condition can be satisfied in most inflation models . we now follow the cosmological evolution after the @xmath3 oscillation . a brief summary of the history is the following . because of a long lifetime of the @xmath3 field , its oscillation energy density eventually dominates over the universe unless the reheating temperature after inflation is very low . decays of @xmath3 reheat the universe to a temperature of @xmath109 . the large entropy production from the @xmath3 decay dilutes preexisting radiation and matter , including thermally produced superparticles , gravitinos and also even baryon asymmetry of the universe . the main decay channel of the @xmath3 field is into two higgs bosons ( @xmath110 ) or two gluons ( @xmath111 ) depending on @xmath71 . radiation produced by the @xmath3 decay takes over the energy density of the universe below this decay temperature and therefore standard bbn can happen after temperature drops down to @xmath112 . decays of @xmath3 produce gravitinos through a rare decay mode @xmath113 . the produced amount is remarkably consistent with the observed dark matter density of the universe in the parameter regions of our interest . we start discussion with deriving the interaction lagrangian of @xmath3 , with which we can calculate the decay width of @xmath3 and the branching ratio of the gravitino mode . we will then discuss the gravitino abundance of the universe . the @xmath3 field couples to the mssm particles only through loop diagrams with messenger particles @xmath4 and @xmath26 . the effect is encoded in the @xmath3 dependence of low energy parameters . the low energy values of the gauge coupling constants depend on @xmath3 through @xmath114 since it changes the scale at which @xmath4 and @xmath26 are integrated out . the kinetic term of the gauge bosons are @xmath115 where @xmath116 with @xmath117 being the gauge coupling constant at a scale @xmath118 . the factor @xmath119 is the beta - function coefficient below the messenger scale and @xmath120 is the renormalization scale . after canonically normalizing the kinetic terms of gauge bosons , we obtain the interaction terms from eq . ( [ eq : gauge ] ) : @xmath121 the interaction terms with gauginos can be obtained in a similar way . from the @xmath3 dependence of the gaugino mass terms : @xmath122 with @xmath123 we obtain @xmath124 comparing with eq . ( [ eq : gauge - int ] ) , this gives larger contribution to the decay width of @xmath3 by a one - loop factor when @xmath125 . the same conclusion can be obtained for scalar fields . the soft mass terms are @xmath126 with @xmath127 ^ 2 \cdot 2 c_2 n \left| \frac{m^2}{s } \right|^2\ . \label{eq : scalar - mass}\end{aligned}\ ] ] @xmath128 is the quadratic casimir factor . we can read off interaction terms as follows : @xmath129 again , this gives larger contribution by a one - loop factor compared to eq . ( [ eq : gauge - int ] ) . for the higgs fields , the scalar masses have two origins , i.e. , one from the gauge mediation and the other from eq . ( [ eq : higgs - kahler ] ) . for the calculation of couplings between @xmath3 and the higgs fields , we should use the gauge - mediation contribution , i.e. , eq . ( [ eq : scalar - mass ] ) , for @xmath130 in eq . ( [ eq : scalar ] ) . it is amusing that the @xmath3 field dominantly decays into supersymmetric particles and the higgs fields if it is kinematically allowed . coupling to gravitinos can be obtained by simply looking at the lagrangian . the largest contribution for the @xmath3 decay comes from the coupling to the longitudinal mode of the gravitino ( the fermionic component of @xmath3 , @xmath131 ) . from the @xmath132 term in the k " ahler potential , we obtain @xmath133 to @xmath134 coupling such as @xmath135 therefore , the decay width is suppressed by @xmath136 compared to the gaugino / scalar modes . also , there is a suppression of @xmath137 compared to the gauge boson mode . this effect is important for the gravitino abundance . in the following we consider two cases ( a ) : @xmath138 , where @xmath3 can decay into two higgs bosons and ( b ) : @xmath139 , where @xmath3 dominantly decays into two gluons . we find that in both cases gravitinos from the @xmath3 decay in early time naturally explain the dark matter component of the universe . * case ( a ) * : the @xmath140 decay is open ( @xmath138 ) . + we further assume here that @xmath3 dominantly decays into two higgs bosons . this assumption is valid as long as @xmath141 is smaller than other su(2)@xmath142 or su(3)@xmath143 charged superparticle masses ( except for the higgsino ) . this will be justified later . from the experimental lower bound on the higgs boson mass @xmath144 gev @xcite , we obtain @xmath145 gev . the decay width of @xmath3 in this case is given by @xmath146 where @xmath147 is a two - loop factor : @xmath148 @xmath149 and @xmath150 are the gauge coupling constants of su(2)@xmath142 and u(1)@xmath151 gauge interactions . the lifetime of @xmath3 is then @xmath152 with this relatively long lifetime and the large initial amplitude of @xmath3 , it is reasonable that the coherent oscillation of @xmath3 dominates over the energy density of the universe before @xmath3 starts to decay . although it is possible that @xmath3 domination does not happen by assuming a presence of another long - lived matter density ( such as the inflaton oscillation ) , we do not consider such a case here . the decay of @xmath3 produces radiation and it reheats the universe . the reheating temperature after the @xmath3 decay is calculated to be @xmath153 this temperature needs to be larger than @xmath154 mev in order for standard bbn to happen @xcite . the baryon asymmetry and the dark matter component of the universe also needs to be generated before bbn . since primordial radiation and matter are significantly diluted by the entropy production from the @xmath3 decays , it is non - trivial whether we can consistently obtain those two components . we explain here the dark matter ( gravitino ) production from the @xmath3 decay and we will discuss a possible mechanism for baryogenesis later . under the assumption that the @xmath3-domination happens , the number density of the gravitinos , @xmath155 , from @xmath3 decays is given by @xmath156 where @xmath157 is the branching fraction into two gravitinos and @xmath158 is the entropy density of the universe . note here that the number density can be calculated only with the two parameters in the lagrangian , @xmath71 and @xmath74 . the partial decay width is calculated from eq . ( [ eq : gravitino ] ) : @xmath159 therefore , the branching fraction @xmath160 is given by @xmath161 the energy density is then @xmath162 comparing with the observed dark matter density of the universe , @xmath163 ( 2@xmath164 ) @xcite , we find that the gravitino is a perfect candidate of dark matter in this scenario . if @xmath3 can decay into two binos and if the bino is the next to lightest supersymmetric particle ( nlsp ) , the produced binos later decay into gravitinos . this possibility is excluded by the bbn constraints on the bino decay @xcite . the conclusion is the same for the case where the higgsino is lighter than the bino . this justifies the assumption that @xmath3 mainly decays into two higgs bosons . other su(2)@xmath142 or su(3)@xmath143 charged supersymmetric particles are heavier than the bino in gauge mediation . in fact , if the scalar tau lepton ( stau ) is lighter than the bino , which is possible if there is a large @xmath165-term from the direct coupling between the higgs fields and @xmath3 in the k " ahler potential , a small portion of the @xmath166 region becomes viable . staus can reduce the number density by the pair annihilation process after the production from @xmath3 decays . the energy density of the stau before its decay is approximately @xmath167 where @xmath168 is the freezing out temperature , @xmath169 , and @xmath170 is the energy density of the stau obtained by the standard calculation of the thermal relic abundance . in the region of our interest , where the gravitinos from the @xmath3 decay explain dark matter of the universe , the above amount is on the border of the bbn constraints . since the viable region appears only in a special circumstance and also the region is quite small , we do not consider this possibility in the rest of the paper . * case ( b ) * : the @xmath140 decay is closed ( @xmath139 ) . + in this case , the dominant decay channel is into two gluons , @xmath171 . the decay rate is calculated to be @xmath172 with a one - loop factor @xmath173 from this , the reheating temperature after the @xmath3 decay is given by @xmath174 the branching fraction of the gravitino mode is @xmath175 by using the same formula in eq . ( [ eq : density ] ) , we obtain @xmath176 the dark matter density can be explained by the non - thermally produced gravitinos also in this case . ) which account for the dark matter component of the universe . the correct amount of gravitino dark matter can be obtained within the allowed range of parameter space . below the dot - dashed line in the left figure , the decay mode of @xmath3 into two higgs boson is closed ( case ( b ) ) . the constraint from the lower bound on the decay temperature @xmath177 mev is indicated . the contour of @xmath178 mev is also shown . in the right figure , the decay mode @xmath140 is open above the dot - dashed line ( case ( a ) ) . the decay mode into two binos is open in the right - lower region . , title="fig:",height=283 ] ) which account for the dark matter component of the universe . the correct amount of gravitino dark matter can be obtained within the allowed range of parameter space . below the dot - dashed line in the left figure , the decay mode of @xmath3 into two higgs boson is closed ( case ( b ) ) . the constraint from the lower bound on the decay temperature @xmath177 mev is indicated . the contour of @xmath178 mev is also shown . in the right figure , the decay mode @xmath140 is open above the dot - dashed line ( case ( a ) ) . the decay mode into two binos is open in the right - lower region . , title="fig:",height=283 ] we show in fig . [ fig : region ] the region of the parameter space where the non - thermally produced gravitino explains the dark matter component of the universe . we took @xmath35 and used @xmath74 and @xmath179 as free parameters of this two - parameter model , and showed contours of @xmath71 ( see eq . ( [ eq : bino ] ) ) as dashed lines . the values of @xmath180 expressed in terms of those two are @xmath181 respectively . the right figure corresponds to the case ( a ) where the higgs boson mode is the dominant decay channel ( above the dot - dashed line ) . the constraint from @xmath177 mev is satisfied in the figure once the higgs mode is open . the lower bound on the bino mass ( @xmath182 gev ) is obtained from the requirement that the bino mode should be kinematically forbidden . the gravitinos can explain the dark matter density when @xmath183 gev , which is exactly the preferred values from the @xmath0-term driven supersymmetry breaking scenario and grand unified theories . another consistent region is found in the left figure , where the gluon mode is the dominant decay channel ( case ( b ) ) . the constraint from bbn , @xmath177 mev , excludes the lower mass region of @xmath179 in this case . the correct amount of dark matter can be obtained with @xmath184 gev . this is also a natural size for the @xmath0-term driven supersymmetry breaking scenario . for a larger value of @xmath34 , we obtain a larger lower bound on @xmath185 , and the allowed region of @xmath74 is not very sensitive to @xmath34 for case ( a ) and moves to larger values linearly for case ( b ) . we comment on other possible sources of gravitinos . thermally produced gravitinos @xcite are significantly diluted by the entropy production from @xmath3 decays . it is diluted by a factor of @xmath186 where @xmath187 the temperature at which @xmath3 oscillation starts to dominate ( see discussion in the subsection [ sec : entropy ] ) . if @xmath3 oscillation starts during the reheating process after inflation , which is the case in our scenario in order not to thermalize the messenger fields , the amount of thermal gravitinos does not depend on the reheating temperature after inflation . it is given in terms of the parameters of the model and the initial amplitude @xmath188 by @xmath189 this is sub - dominant when @xmath190 . thermal relic of nlsps also generates gravitinos through their decays @xcite . again entropy production reduces the amount of nlsps by a factor of @xmath191 $ ] , where @xmath192 is the freezing out temperature of pair annihilation processes @xcite . the superwimp component of gravitino dark matter is therefore negligible in the parameter regions of our interest . since gravitinos are non - thermally produced through the decay of a heavy particle @xmath3 , they are relativistic when produced , but they subsequently slow down as the universe expands . the gravitinos need to be sufficiently non - relativistic before the structure formation to start . the free - streaming scale of the dark matter is constrained to be less than @xmath78 mpc by the matter power spectrum inferred from lyman-@xmath193 forest @xcite . we estimate the free - streaming scale @xmath194 of the gravitinos . we find that it is sufficiently small , but interestingly it may be large enough to be distinguished from that of cold dark matter . the free - streaming scale at the time of matter - radiation equality @xmath195 is defined by @xmath196 \ , \end{aligned}\ ] ] where @xmath197 and @xmath198 are scale factors at matter - radiation equality and at the time of @xmath3 decays , respectively . @xmath199 is the hubble parameter at the current epoch . up to logarithmic dependence , we obtain @xmath200 for case ( b ) , @xmath201 is in an interesting range for the small - scale structure problem in @xmath38cdm model . although not conclusive , it has been reported that there are discrepancies between simulations of structure formation in the cold dark matter scenario and observation at a small scale . the observed number of galactic satellites in the inner region ( @xmath202 ) of halo is an order of magnitude smaller than results of simulations @xcite . also , simulation of the structure formation predicts a cuspy profile in the central region of halo @xcite whereas observations such as studies of gravitational lensing @xcite and rotation curves @xcite indicate nearly uniform density core structure . it has been studied that dark matter with free - streaming scales of @xmath203 mpc may resolve these issues because it can smear out the density fluctuation in small scale @xcite . ( see also @xcite for recent proposals of particle physics models which realize large @xmath201 via non - thermally produced dark matter . ) a free - streaming scale of @xmath204 kpc means that there is a cut - off in matter power spectrum below that scale . it might be possible to probe this small - scale structure in future experiments by using strong gravitational lensing @xcite such as in lsst @xcite and snap @xcite , and also by observations of 21 cm fluctuations in high red - shift region @xcite . in the previous section , we have assumed that the oscillation energy density dominates over the universe before it decays . we discuss the amount of entropy production and possible baryogenesis scenarios here . as we have discussed , the @xmath3 field starts to oscillate when the hubble parameter becomes smaller than @xmath205 . there are two cases which we need to consider separately . if the oscillation starts during the reheating process after inflation ( the inflaton oscillation era ) , @xmath3-domination happens when : @xmath206 where @xmath207 is the reheating temperature after inflation . a different result is obtained if @xmath3 starts to oscillate in a radiation - dominated era ( after reheating ) . the temperature at which @xmath3 starts to oscillate is given by @xmath208 in this case , @xmath187 is @xmath209 the dilution factor @xmath210 is then @xmath211 the primordial radiation and matter are diluted by this amount . ( @xmath3-domination does not happen if @xmath210 is larger than unity . ) with the low decay temperature @xmath212 , dilution effects are quite large . for example , if @xmath213 and @xmath214 , the dilution factor is @xmath215 . this dilutes unwanted relics such as thermally produced neutralinos to a negligible level . since baryon asymmetry is also diluted , primordial baryon asymmetry needs to be larger such as @xmath216 . it is a non - trivial task to obtain such a large asymmetry under a condition that the reheating temperature can not be too high in order not to thermally produce the messenger particles ( min@xmath217 \ll \lambda \langle s \rangle$ ] ) . examples of possible scenarios for baryogenesis are the right - handed sneutrino inflation scenario @xcite and scenario with right - handed sneutrino dominated universe @xcite . in those scenarios , baryon asymmetry is generated via leptogenesis @xcite after the decay of right - handed sneutrinos as follows : @xmath218 where @xmath219 is an effective cp violating phase and @xmath220 corresponds to the heaviest neutrino mass , and @xmath207 denotes the decay temperature of the right handed sneutrino . after dilution by the @xmath3 decay , it reduces to @xmath221 independent of @xmath222 for @xmath223 . therefore , the observed baryon asymmetry @xmath224 @xcite can be consistently explained with a slightly suppressed initial amplitude @xmath225 . note here that if the initial amplitude @xmath91 is suppressed to this level , thermally produced gravitinos can give a non - negligible contribution to the matter density ( see eq . ( [ eq : thermal ] ) ) , and also the decay of thermally produced nlsps during bbn may put constraints on species of the nlsp and/or @xmath222 . taking into account gravity effects in supersymmetry breaking / mediation models opens up a possibility of very simple gauge mediation . we have considered the simplest model of gauge mediation , in which we have only two parameters . one combination of the parameters needs to be fixed so that we obtain the gaugino masses to be @xmath6 gev , and the other has preferable range from a simple solution to the @xmath0-problem and grand unification , i.e. , 10 mev @xmath226 gev . this is the simplest model of gauge mediation , but amazingly free from problems . we have shown that @xmath3 rolls into the supersymmetry breaking vacuum with generic initial conditions . the entropy production by the @xmath3 decay dilutes gravitinos and neutralinos which are thermally ( over)produced . supersymmetric particles do not cause the fcnc or cp problems as the gravitino mass is small enough . of the most amazing is the consistency of the dark - matter abundance . we can calculate the abundance of the non - thermally produced gravitinos from @xmath3 decays in terms of the two parameters in the model . within the parameter region of our interest , the calculated amount can account for the dark matter of the universe . we thank ted baltz and masamune oguri for useful discussion on dark matter . r.k . would like to thank paddy fox , hitoshi murayama and yasunori nomura for stimulating discussion and jay wacker for discussion on non - perturbative particle production during the @xmath3 oscillation . this work was supported by the u.s . department of energy under contract number de - ac02 - 76sf00515 . as we mentioned in section [ sec : evolution ] , the minimum of the scalar potential during inflation is likely to be around @xmath227 . the singlet field @xmath3 may couple to the inflaton sector in the superpotential , @xmath228 where @xmath229 is a coefficient of @xmath230 , and @xmath231 denotes the superpotential of a inflaton sector . the scalar potential of @xmath3 during inflation is then given by @xmath232 where the first term originates from the khler potential in eq . ( 1 ) and the second term from the superpotential in eq . ( [ eq : stoinf ] ) . in the above expression , we have used the hubble equation , @xmath233 assuming that the @xmath1-component of the inflaton dominates the energy density during inflation . this potential has a minimum at @xmath234 . in principle , the @xmath229 parameter can be much larger than @xmath78 depending on a uv completion of the theory . for example , in a model of ref . @xcite , the theory above the scale @xmath38 is a strongly coupled conformal field theory , and the @xmath229 parameter is enhanced to be @xmath235 . if this is the case , the second term in eq . ( [ eq : hubblemass ] ) pushes the minimum to the planck scale . however , the effective theory in terms of @xmath3 does not make sense for @xmath236 , and it is more appropriate to consider the dynamics in terms of the constituent field @xmath41 . the constituent field @xmath41 , in contrast to @xmath3 , may have a positive hubble mass squared around the symmetry enhancement point at @xmath237 . this hubble mass term attracts @xmath41 to the origin , and it prevents @xmath238 from going to the planck scale . therefore , we can expect the actual minimum during inflation does not exceed @xmath239 even if the coefficient @xmath229 is large . 0 s. dimopoulos and h. georgi , nucl . b * 193 * , 150 ( 1981 ) ; s. dimopoulos , s. raby and f. wilczek , phys . d * 24 * , 1681 ( 1981 ) ; n. sakai , z. phys . c * 11 * , 153 ( 1981 ) . a. h. chamseddine , r. arnowitt and p. nath , phys . lett . * 49 * , 970 ( 1982 ) ; r. barbieri , s. ferrara and c. a. savoy , phys . b * 119 * , 343 ( 1982 ) ; l. j. hall , j. d. lykken and s. weinberg , phys . d * 27 * , 2359 ( 1983 ) . g. d. coughlan , w. fischler , e. w. kolb , s. raby and g. g. ross , phys . b * 131 * , 59 ( 1983 ) . t. banks , d. b. kaplan and a. e. nelson , phys . d * 49 * , 779 ( 1994 ) [ arxiv : hep - ph/9308292 ] . m. ibe , y. shinbara and t. t. yanagida , phys . b * 639 * , 534 ( 2006 ) [ arxiv : hep - ph/0605252 ] . m. dine , w. fischler and d. nemeschansky , phys . b * 136 * , 169 ( 1984 ) . i. joichi and m. yamaguchi , phys . b * 342 * , 111 ( 1995 ) [ arxiv : hep - ph/9409266 ] . l. randall and r. sundrum , nucl . b * 557 * , 79 ( 1999 ) [ arxiv : hep - th/9810155 ] . g. f. giudice , m. a. luty , h. murayama and r. rattazzi , jhep * 9812 * , 027 ( 1998 ) [ arxiv : hep - ph/9810442 ] . m. dine , w. fischler and m. srednicki , nucl . b * 189 * , 575 ( 1981 ) ; s. dimopoulos and s. raby , nucl . b * 192 * , 353 ( 1981 ) ; m. dine and w. fischler , phys . b * 110 * , 227 ( 1982 ) ; nucl . phys . b * 204 * , 346 ( 1982 ) ; c. r. nappi and b. a. ovrut , phys . b * 113 * , 175 ( 1982 ) ; l. alvarez - gaume , m. claudson and m. b. wise , nucl . b * 207 * , 96 ( 1982 ) ; s. dimopoulos and s. raby , nucl . phys . b * 219 * , 479 ( 1983 ) . m. dine and a. e. nelson , phys . d * 48 * , 1277 ( 1993 ) [ arxiv : hep - ph/9303230 ] ; m. dine , a. e. nelson , y. nir and y. shirman , phys . rev . d * 53 * , 2658 ( 1996 ) [ arxiv : hep - ph/9507378 ] . m. dine , a. e. nelson and y. shirman , phys . d * 51 * , 1362 ( 1995 ) [ arxiv : hep - ph/9408384 ] . t. moroi , h. murayama and m. yamaguchi , phys . b * 303 * , 289 ( 1993 ) . a. de gouvea , t. moroi and h. murayama , phys . d * 56 * , 1281 ( 1997 ) [ arxiv : hep - ph/9701244 ] . m. bolz , a. brandenburg and w. buchmuller , nucl . b * 606 * , 518 ( 2001 ) [ arxiv : hep - ph/0012052 ] . j. pradler and f. d. steffen , arxiv : hep - ph/0608344 . g. f. giudice and a. masiero , phys . b * 206 * , 480 ( 1988 ) . d. n. spergel _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 148 * , 175 ( 2003 ) [ arxiv : astro - ph/0302209 ] ; d. n. spergel _ et al . _ , arxiv : astro - ph/0603449 . m. kawasaki , k. kohri and t. moroi , phys . d * 71 * , 083502 ( 2005 ) [ arxiv : astro - ph/0408426 ] ; k. jedamzik , arxiv : hep - ph/0604251 . r. kitano , phys . b * 641 * , 203 ( 2006 ) [ arxiv : hep - ph/0607090 ] . r. kitano , arxiv : hep - ph/0606129 . k. intriligator , n. seiberg and d. shih , jhep * 0604 * , 021 ( 2006 ) [ arxiv : hep - th/0602239 ] . n. j. craig , p. j. fox and j. g. wacker , hep - th/0611006 . l. kofman , a. d. linde and a. a. starobinsky , phys . d * 56 * , 3258 ( 1997 ) [ arxiv : hep - ph/9704452 ] . l. kofman , a. linde , x. liu , a. maloney , l. mcallister and e. silverstein , jhep * 0405 * , 030 ( 2004 ) [ arxiv : hep - th/0403001 ] . r. barate _ [ lep working group for higgs boson searches ] , phys . b * 565 * , 61 ( 2003 ) [ arxiv : hep - ex/0306033 ] ; [ lep higgs working group ] , arxiv : hep - ex/0107030 . k. ichikawa , m. kawasaki and f. takahashi , phys . d * 72 * , 043522 ( 2005 ) [ arxiv : astro - ph/0505395 ] . j. l. feng , a. rajaraman and f. takayama , phys . lett . * 91 * , 011302 ( 2003 ) [ arxiv : hep - ph/0302215 ] ; j. l. feng , s. f. su and f. takayama , phys . d * 70 * , 063514 ( 2004 ) [ arxiv : hep - ph/0404198 ] . w. buchmuller , k. hamaguchi , m. ibe and t. t. yanagida , arxiv : hep - ph/0605164 . v. k. narayanan , d. n. spergel , r. dave and c. p. ma , arxiv : astro - ph/0005095 ; m. viel , j. lesgourgues , m. g. haehnelt , s. matarrese and a. riotto , phys . d * 71 * , 063534 ( 2005 ) [ arxiv : astro - ph/0501562 ] ; m. viel , j. lesgourgues , m. g. haehnelt , s. matarrese and a. riotto , phys . lett . * 97 * , 071301 ( 2006 ) [ arxiv : astro - ph/0605706 ] . a. a. klypin , a. v. kravtsov , o. valenzuela and f. prada , astrophys . j. * 522 * , 82 ( 1999 ) [ arxiv : astro - ph/9901240 ] ; b. moore , s. ghigna , f. governato , g. lake , t. quinn , j. stadel and p. tozzi , astrophys . j. * 524 * ( 1999 ) l19 . b. moore , f. governato , t. quinn , j. stadel and g. lake , astrophys . j. * 499 * , l5 ( 1998 ) [ arxiv : astro - ph/9709051 ] . j. a. tyson , g. p. kochanski and i. p. dellantonio , astrophys . j. * 498 * , l107 ( 1998 ) [ arxiv : astro - ph/9801193 ] . j. d. simon , a. d. bolatto , a. leroy , l. blitz and e. l. gates , astrophys . j. * 621 * , 757 ( 2005 ) [ arxiv : astro - ph/0412035 ] . c. j. hogan and j. j. dalcanton , phys . d * 62 * , 063511 ( 2000 ) [ arxiv : astro - ph/0002330 ] . m. kaplinghat , phys . rev . d * 72 * , 063510 ( 2005 ) [ arxiv : astro - ph/0507300 ] . r. kitano and i. low , arxiv : hep - ph/0503112 ; j. a. r. cembranos , j. l. feng , a. rajaraman and f. takayama , phys . lett . * 95 * , 181301 ( 2005 ) [ arxiv : hep - ph/0507150 ] . n. dalal and c. s. kochanek , arxiv : astro - ph/0202290 . lsst , ` http://www.lsst.org/lsst_home.shtml ` a. loeb and m. zaldarriaga , phys . lett . * 92 * , 211301 ( 2004 ) [ arxiv : astro - ph/0312134 ] . h. murayama , h. suzuki , t. yanagida and j. yokoyama , phys . lett . * 70 * , 1912 ( 1993 ) . h. murayama and t. yanagida , phys . b * 322 * , 349 ( 1994 ) [ arxiv : hep - ph/9310297 ] . m. fukugita and t. yanagida , phys . b * 174 * , 45 ( 1986 ) .
various mathematical problems can be formulated as finding a common fixed point of finitely many firmly nonexpansive mappings . these include , for instance , the convex feasibility problem @xcite , which consists of finding a point in the intersection of a finite number of closed convex sets , or the problem of finding a common zero of a finite family of maximal monotone operators @xcite . firmly nonexpansive mappings were introduced by bruck @xcite in order to study sunny nonexpansive retractions onto closed subsets of banach spaces . in hilbert spaces , these mappings coincide with firmly contractive ones defined by browder @xcite . a basic example of firmly nonexpansive mappings is the metric projection onto closed convex subsets of hilbert spaces . this mapping or finite compositions thereof appear in different iterative methods used in the study of convex minimization problems . results on the asymptotic behavior of some such methods follow from analyzing compositions of finitely many firmly nonexpansive mappings , as it is the case of the cyclic projection algorithm employed in solving the convex feasibilty problem ( see @xcite for a deep treatment of this topic ) . another important example of firmly nonexpansive mappings is the resolvent of a monotone operator which was introduced by minty @xcite . since the subdifferential of a proper convex lower semi - continuous function is a maximal monotone operator , some splitting methods which involve resolvents and are used to solve convex minimization problems can be abstracted to compositions of firmly nonexpansive mappings . using this approach , bauschke , combettes and reich @xcite ( see also @xcite ) proved , in particular , that having two proper convex lower semicontinuous functions defined on a hilbert space , one can apply alternatively the two resolvents to obtain weak convergence to a solution ( provided it exists ) of a minimization problem associated to the two functions , see ( [ def - fct - min ] ) . it is noteworthy that the convex feasibilty problem for two sets can be reformulated as a problem of type ( [ def - fct - min ] ) for both the consistent and the inconsistent case . the most straightforward way to find a common fixed point or , more generally , a fixed point of the composition of finitely many firmly nonexpansive mappings is to iterate the mappings cyclically . the particular case of just one mapping was considered in hilbert spaces by browder @xcite who proved that , for any starting point , the picard iterates of a firmly nonexpansive mapping converge weakly to a fixed point , whenever the fixed point set is nonempty . although the class of firmly nonexpansive mappings is not closed under composition , in hilbert spaces firmly nonexpansive mappings are averaged and the composition of averaged mappings is averaged . thus , results for the general case of finitely many firmly nonexpansive mappings can be deduced from the asymptotic behavior of averaged mappings ( see also @xcite for a more general order of composing the mappings that still preserves weak convergence ) . in banach spaces not every firmly nonexpansive mappings is averaged and so extensions of such results to this setting are not immediate . bruck and reich @xcite introduced strongly nonexpansive mappings that generalize firmly nonexpansive ones when the space is uniformly convex . this class of mappings is closed under composition and so one can analyze compositions of firmly nonexpansive mappings using results on the asymptotic behavior of strongly nonexpansive mappings . more recently , problems where the study of compositions of firmly nonexpansive mappings plays an important role ( such as minimization problems @xcite or abstract cauchy problems@xcite ) have also been formulated in a nonlinear setting . since our applications focus on minimizing convex functions in geodesic spaces , we mention some of the work in this direction . the proximal point method and modifications thereof for optimization problems in linear spaces have been extended to riemanninan manifolds in @xcite and references therein . spaces of nonpositive curvature in the sense of alexandrov ( also known as cat@xmath5 spaces ) proved to be an appropriate setting for considering such problems too . jost studied the concept of resolvents for convex functions to develop the theory of harmonic mappings @xcite which motivated bak @xcite to extend the proximal point algorithm to this context . in @xcite , a proximal splitting algorithm was also approached in this setting ( assuming in addition local compactness ) and applied to the computation of the geometric median and the frchet mean of a finite set of points . regarding the convex feasibility problem , results on the asymptotic behavior of the alternating projection method have been obtained in @xcite . the minimization problem ( [ def - fct - min ] ) was studied in this setting in @xcite . the goal of this work is to create a suitable theoretical framework so that previous results can be better understood and in some cases improved . to this end , we present a systematic study of the asymptotic behavior of compositions of a finite number of firmly nonexpansive mappings in @xmath0-uniformly convex geodesic spaces . asymptotic regularity is a very important concept used in the study of the asymptotic behavior of sequences and was formally introduced by browder and petryshyn in @xcite . in section [ section - as - reg ] we give asymptotic regularity results for sequences obtained by applying a finite number of firmly nonexpansive mappings in a cyclic way . we assume first that they have a common fixed point , but consider also the case of two mappings which satisfy the weaker condition that their composition is not fixed point free . moreover , we use techniques which originate from proof mining ( see @xcite for more details ) to provide polynomial rates of asymptotic regularity , thus giving quantitative versions of such results and extending a corresponding result proved in @xcite . section [ section - conv ] deals with @xmath6-convergence results for these problems . assuming appropriate compactness conditions we obtain strong convergence . when the mappings have a common fixed point we apply our findings to the cyclic projection method for finitely many sets in cat@xmath1 spaces improving previous results from @xcite . we show that the results concerning two firmly nonexpansive mappings for which the fixed point set of the composition is nonempty can be used to study the minimization problem ( [ def - fct - min ] ) in cat@xmath5 spaces which partially recovers the main result of @xcite and yields a convergence result for the inconsistent convex feasibility problem for two sets . let @xmath7 be a metric space . a _ geodesic path _ that joins two points @xmath8 is a mapping @xmath9\subseteq{\mathbb{r}}\to x$ ] such that @xmath10 , @xmath11 and @xmath12 for all @xmath13 $ ] . the image @xmath14)$ ] of @xmath15 is called a _ geodesic segment _ from @xmath16 to @xmath17 . we denote a point @xmath18 belonging to such a geodesic segment by @xmath19 , where @xmath20 . @xmath7 is a _ ( uniquely ) geodesic space _ iff every two points in @xmath21 are joined by a ( unique ) geodesic path . a subset @xmath22 of @xmath21 is _ convex _ iff it contains all geodesic segments that join any two points in @xmath22 . for more details on geodesic metric spaces , see @xcite . ball , carlen and lieb introduced in @xcite the notion of @xmath0-uniform convexity which plays an essential role in banach space theory . recall that a normed space @xmath23 is said to be _ convex _ for @xmath24 iff there exists a constant @xmath25 such that for any @xmath26 , @xmath27 recently , naor and silberman @xcite extended this notion to the setting of geodesic spaces in the following way . fix @xmath28 . a metric space @xmath7 is called _ convex _ with parameter @xmath29 iff @xmath7 is a geodesic space and for any three points @xmath30 and all @xmath31 $ ] , @xmath32 note that inequality guarantees that the space @xmath21 is uniquely geodesic ( see ( * ? ? ? * lemma 2.2 ) for a proof in the case @xmath33 ) . also , any closed convex subset of a @xmath0-uniformly convex space is again a @xmath0-uniformly convex space with the same parameter . it is well - known that any @xmath34 space over a measurable space is @xmath0-uniformly convex . in the setting of nonlinear spaces , every @xmath35 space is @xmath36-uniformly convex with parameter @xmath37 . actually , in this case ( [ eq : p - uc ] ) characterizes @xmath35 spaces . for @xmath38 , any @xmath39 space @xmath21 with @xmath40 is a @xmath36-uniformly convex space with parameter @xmath41 for any @xmath42 , see @xcite . a detailed treatment of cat@xmath1 spaces can be found , for example , in @xcite . we recall next a notion of weak convergence in metric spaces . let @xmath43 be a bounded sequence in a metric space @xmath7 . then @xmath44 is called the _ asymptotic radius _ of @xmath43 and the _ asymptotic center _ of @xmath43 is the set @xmath45 an element of @xmath46 will also be referred to as an asymptotic center . if @xmath21 is a complete @xmath0-uniformly convex space , then any bounded sequence has a unique asymptotic center since @xmath21 admits a modulus of uniform convexity that does not depend on the radius of balls ( see @xcite ) . we say that a sequence @xmath43 _ @xmath6-converges _ to @xmath47 iff @xmath16 is the unique asymptotic center of every subsequence of @xmath43 . in this case we call @xmath16 the _ @xmath6-limit _ of @xmath43 . this concept was introduced by lim @xcite . it is easy to see that in a complete @xmath0-uniformly convex space , any bounded sequence has a @xmath6-convergent subsequence . in the setting of @xmath35 spaces , @xmath6-convergence is equivalent to another concept of weak convergence that uses projections on geodesic segments ( see @xcite ) . reasoning as in @xcite , one can see that this equivalence also holds for @xmath0-uniformly convex spaces . let @xmath7 be a metric space . given @xmath22 a nonempty subset of @xmath21 , the _ distance of a point _ @xmath48 to @xmath22 is @xmath49 if @xmath50 and @xmath22 are nonempty subsets of @xmath21 , we denote @xmath51 . the _ metric projection _ @xmath52 onto @xmath22 is the mapping defined by @xmath53 , for every @xmath47 . let @xmath21 be a compete cat@xmath1 space ( with @xmath40 for @xmath54 ) and @xmath22 a closed and convex subset of @xmath21 . then @xmath55 is well - defined and singlevalued ( see ( * ? ? ? * proposition 2.4 , page 176 ) and ( * ? ? ? * proposition 3.5 ) ) . if @xmath56 , @xmath52 is also nonexpansive . for @xmath54 we have that for every @xmath57 , @xmath58 . moreover , in this case @xmath52 is a uniformly continuous mapping because @xmath21 is bounded and admits a modulus of uniform convexity that does not depend on the radius of balls ( see the reasoning in the proof of ( * ? ? ? * proposition 4.5 ) ) . in fact , @xmath52 is a lipschitz mapping ( one can easily derive this from the proof of ( * ? ? ? * theorem 4.1 ) ) . we will also use the following inequality which originates from the work of reshetnyak ( see , for instance , ( * ? ? ? * theorem 2.3.1 ) or ( * ? ? ? * lemma 2.1 ) for a simple proof ) . let @xmath21 be a @xmath35 space . then for every @xmath59 @xmath60 we finish this subsection with two simple results that will be used later on . [ lemma - conv - seq ] let @xmath61 and @xmath62 be two sequences of nonnegative numbers . if for every @xmath63 @xmath64 and @xmath65 , then the sequence @xmath61 is convergent . [ lemma : real.numbers.inequality ] let @xmath66 and @xmath67 . then there exists a constant @xmath68 depending only on @xmath69 and @xmath70 such that for all @xmath71 and @xmath72 , @xmath73 when @xmath74 $ ] , clearly @xmath75 . for @xmath76 , the result is an immediate consequence of a kaczmarz - steinhaus inequality @xcite which states that there exists a constant @xmath77 depending only on @xmath69 such that for all @xmath78 , @xmath79}\binom{l}{i } x^i+\alpha{\left\vertx\right\vert}^l.\ ] ] firmly nonexpansive mappings were first introduced by browder @xcite , under the name of firmly contractive , in the setting of hilbert spaces , and later by bruck @xcite in the context of banach spaces . recently bruck s definition was extended to a nonlinear setting in @xcite ( see also @xcite ) . we recall this definition using the framework of @xmath0-uniformly convex spaces . let @xmath22 be a nonempty subset of a @xmath0-uniformly convex space @xmath7 . we say that a mapping @xmath80 is _ firmly nonexpansive _ iff @xmath81 for all @xmath82 and @xmath83 . as in the linear case , any sunny nonexpansive retraction is firmly nonexpansive . in particular , the metric projection onto closed and convex subsets of a complete cat@xmath5 space is firmly nonexpansive . having @xmath21 a complete cat@xmath5 space and a function @xmath84-\infty,+\infty]$ ] that is convex , lower semi - continuous and proper ( i.e. , not constantly equal to @xmath85 ) , the _ resolvent of @xmath86 _ defined by @xmath87,\ ] ] where @xmath88 , is also firmly nonexpansive . in the subsequent section we will use the following inequality ( for its proof see , for instance , ( * ? ? ? * lemma 3.2 ) ) : for every @xmath8 , @xmath89 bruck extended in @xcite the notion of strongly nonexpansive mappings and introduced the so - called strongly quasi - nonexpansive ones with the aim of proving convergence of iterations built by applying finitely many such mappings . we consider in the following a related property which is useful when studying the asymptotic behavior of sequences obtained by applying a finite number of firmly nonexpansive mappings in the setting of @xmath0-uniformly convex spaces ( see also the notion of strongly attracting mappings in @xcite ) . let @xmath22 be a nonempty subset of a metric space @xmath7 . a mapping @xmath80 is said to satisfy _ property _ iff @xmath90 and there exist @xmath91 and @xmath92 such that @xmath93 for all @xmath94 and @xmath95 . if @xmath21 is @xmath0-uniformly convex with parameter @xmath96 , then every firmly nonexpansive mapping @xmath80 with @xmath90 satisfies with @xmath97 and @xmath98 . indeed , given @xmath94 and @xmath95 , for any @xmath99 we have that @xmath100 hence , @xmath101 . taking limit as @xmath102 , we get the conclusion . a second example of mappings satisfying is the metric projection in spaces of curvature bounded above . let @xmath21 be a complete @xmath39 ( with @xmath103 for @xmath38 ) and @xmath104 closed and convex . recall that if @xmath54 , then the metric projection @xmath105 need not be nonexpansive . however , it satisfies with @xmath106 and @xmath98 . to see this , let @xmath47 , @xmath107 and @xmath1080,1[$ ] . then , @xmath109 hence , @xmath110 and we only need to let @xmath102 . + given finitely many mappings satisfying @xmath111 one can relate the intersection of their fixed point sets with the fixed point set of their composition . [ prop - int - fp - set ] let @xmath7 be a metric space and let @xmath112 be a family of mappings with property . if @xmath113 , then @xmath114 clearly @xmath115 . conversely , let @xmath116 . if @xmath117 and @xmath118 , then @xmath119 notice that @xmath120 . then , by , @xmath121 . since @xmath122 has property , we deduce that @xmath123 , that is , @xmath124 . hence , @xmath125 and @xmath126 . continuing in this way we obtain the result . the following also purely metric condition is more particular than and is still satisfied by any firmly nonexpansive mapping in cat@xmath5 spaces . in this setting every mapping for which the property below holds is nonexpansive . note also that in hilbert spaces , this notion coincides with firm nonexpansivity ( see also the remark in @xcite ) . let @xmath22 be a nonempty subset of a metric space @xmath7 . a mapping @xmath80 satisfies _ property _ iff @xmath127 for all @xmath82 . any mapping that satisfies and has fixed points also satisfies with @xmath106 and @xmath128 . indeed , if @xmath80 satisfies and @xmath90 , then for all @xmath94 and @xmath95 we have that @xmath129 from where @xmath130 . however , the converse implication does not hold . to see this , it suffices to take @xmath131 \to \mathbb{r}$ ] , @xmath132 . then @xmath133 satisfies with @xmath106 and @xmath134 , but fails for @xmath135 and @xmath136 . note that the metric projection and the resolvent are well - defined in less regular frameworks than complete spaces of curvature bounded above , such as complete uniform busemann nonpositively curved spaces ( see @xcite ) . however , in this setting they may fail to be nonexpansive or satisfy property . since we derive our results concerning the convex feasibility problem and the minimization problem ( [ def - fct - min ] ) from the asymptotic behavior of mappings satisfying properties or , we state these results in the context of spaces of curvature bounded above . in this section we analyze the asymptotic behavior ( also from a quantitative point of view ) of the sequence generated by applying iteratively a finite number of firmly nonexpansive type mappings in a cyclic way . we study the case when the considered mappings have a common fixed point , but focus also on the situation when they satisfy the weaker assumption that the fixed point set of their composition is nonempty . let @xmath7 be a metric space . a mapping @xmath137 is said to be _ asymptotically regular at @xmath48 _ iff @xmath138 and it is _ asymptotically regular _ iff it is asymptotically regular at every @xmath48 . likewise , a sequence @xmath139 is _ asymptotically regular _ iff @xmath140 . a rate of convergence of @xmath141 towards @xmath142 will be called a _ rate of asymptotic regularity_. as mentioned in the introductory section , asymptotic regularity is an important tool used in the study of the asymptotic behavior of sequences . in the sequel , having @xmath143 , @xmath144 , we use the notation @xmath145 . let @xmath146 , @xmath144 and @xmath147 . we consider the following cyclic method allowing errors in the evaluation of the values of the mappings @xmath148 : given @xmath149 , define the sequence @xmath43 in @xmath21 by @xmath150 where @xmath151 . we prove below that the sequence @xmath43 is asymptotically regular provided the mappings used to define it satisfy property and have a common fixed point . the result applies , in particular , to firmly nonexpansive mappings in the setting of @xmath0-uniformly convex spaces . [ thm - as - reg - p1 ] let @xmath7 be a metric space and @xmath152 satisfy property . if @xmath113 , then given any starting point @xmath153 , the sequence @xmath43 defined by is asymptotically regular . let @xmath154 and @xmath155 such that @xmath156 for @xmath157 . because each @xmath148 satisfies we get that for all @xmath158 , @xmath159 where @xmath68 is a constant depending only on @xmath69 and @xmath160 thus , for all @xmath161 , @xmath162 since @xmath163 , we deduce that @xmath164 , so @xmath165 . this finally yields that @xmath43 is asymptotically regular . it is easy to see that theorem [ thm - as - reg - p1 ] can be slightly generalized by relaxing property imposed on the mappings @xmath166 to the condition that they satisfy the inequality used to define only for one common fixed point instead of all fixed points . however , we will not focus on such extensions because the final goal of this work is to apply the above result when the mappings are metric projections , which satisfy property in spaces of curvature bounded above . let @xmath7 be a metric space and @xmath152 satisfy property . if @xmath167 then the picard iteration of @xmath168 is asymptotically regular . note that , for @xmath153 and @xmath169 , @xmath170 , where @xmath43 is defined by ( [ it : inexact ] ) with @xmath171 for every @xmath172 . hence , for all @xmath173 , @xmath174 which implies , by theorem [ thm - as - reg - p1 ] , that @xmath133 is asymptotically regular . as a consequence , we have that the sequence obtained by the cyclic projection method for a finite number of sets is asymptotically regular in spaces of curvature bounded above . [ cor - proj - as - ref ] let @xmath175 . suppose @xmath21 is a cat@xmath1 space ( with @xmath176 for @xmath54 ) . let @xmath177 be nonempty closed and convex sets such that @xmath178 . given any starting point @xmath153 , let @xmath43 be defined by ( [ it : inexact ] ) with @xmath179 , for @xmath180 then the sequence @xmath43 and the mapping @xmath181 are asymptotically regular . we give next a rate of asymptotic regularity for the sequence @xmath43 generated by ( [ it : inexact ] ) with @xmath171 for every @xmath172 . for simplicity , we prove the theorem for two mappings , but the proof method works equally well in the general case . the result extends ( * ? ? ? * theorem 5.2 ) where a quadratic rate of asymptotic regularity in @xmath182 was given for the sequence generated by the alternating projection method for two sets in the context of cat@xmath5 spaces . since its proof relies on particular properties of the metric projection , the obtained rate of asymptotic regularity is slightly better than the one we give here when restricting to the setting of cat@xmath5 spaces . however , our result holds not only in a broader framework , but also for more general mappings , namely firmly nonexpansive ones . [ thm - as - reg - fn - quant ] let @xmath7 be a @xmath0-uniformly convex space with parameter @xmath96 and suppose that @xmath183 are firmly nonexpansive mappings with @xmath184 consider @xmath149 , @xmath185 and @xmath186 such that @xmath187 . suppose that @xmath43 is defined by with @xmath171 for every @xmath172 . then for all @xmath188 and @xmath189 , we have that @xmath190 , where @xmath191 & \mbox{if } { \varepsilon}<2b,\\ 0 & \mbox{otherwise}. \end{array } \right.\ ] ] moreover , for all @xmath188 , @xmath192 and @xmath193 , we have that @xmath194 , where @xmath195 since for all @xmath196 , @xmath197 we can assume that @xmath198 . denote @xmath199 take @xmath200 $ ] . for each @xmath201 , @xmath202 from where @xmath203 hence , we have that @xmath204 assume that for all @xmath205 , @xmath206 . then , using , we deduce that @xmath207 a contradiction . this means that there exists @xmath208 such that @xmath209 hence , @xmath210 and @xmath211 from where @xmath212 . this implies that for all @xmath213 @xmath214 by and the mean value theorem , we have that for all @xmath213 @xmath215 that is , @xmath216 . hence , for all @xmath217 , @xmath218 and @xmath219 . thus , for all @xmath220 , @xmath221 it is easy to see that for every @xmath222 and @xmath193 , @xmath223 thus , @xmath224 is a rate of convergence for @xmath225 towards @xmath142 . the above result also holds if we consider a finite family of firmly nonexpansive mappings @xmath226 with @xmath227 in this case , a rate of asymptotic regularity for @xmath43 , defined by with @xmath171 for every @xmath172 , is @xmath228 & \mbox{if } { \varepsilon}<2b,\\ 0 & \mbox{otherwise}. \end{array } \right.\ ] ] moreover , for every @xmath229 , @xmath230 + \theta\left(\frac{{\varepsilon}}{2 \lceil r/2 \rceil-1},b , c , p\right)\ ] ] is a rate of convergence for @xmath225 towards @xmath142 . we prove in the following an asymptotic regularity result for two mappings that satisfy and for which the fixed point set of the composition is nonempty . to simplify the writing , we consider the sequences @xmath43 and @xmath231 defined by @xmath232 where @xmath233 and @xmath234 . [ thm - as - reg - p2 ] let @xmath7 be a cat@xmath5 space and let @xmath183 satisfy property . if @xmath235 , then given any @xmath153 , the sequences @xmath43 and @xmath231 defined by are asymptotically regular . denote @xmath236 . then , for every @xmath63 , @xmath237 hence , @xmath238 we also have that for every @xmath239 , @xmath240 this yields , by lemma [ lemma - conv - seq ] , that @xmath241 converges to some @xmath242 . furthermore , we have that @xmath243 . suppose @xmath244 . we prove by induction that @xmath245 for all @xmath246 . for @xmath247 this is clear . suppose the claim holds true for all @xmath248 . we show that it also holds for @xmath249 . since @xmath122 satisfies , we have that @xmath250 likewise , @xmath251 adding the two above inequalities and using ( [ eq : cat(0).four.points ] ) we obtain that @xmath252 dividing by @xmath253 , using the nonexpansivity of @xmath122 , the triangle inequality and the notation @xmath254 , @xmath255 thus , @xmath256 letting @xmath257 above and using the induction hypothesis and ( [ thm - as - reg - eq2 ] ) , we get that @xmath258 which finishes the induction . thus , @xmath259 is unbounded . let @xmath260 . since , by ( [ thm - as - reg - p2-eq1 ] ) , for all @xmath63 , @xmath261 it follows that @xmath262 , which shows that @xmath263 is bounded , a contradiction . therefore , @xmath264 , that is , @xmath265 which yields , using , that @xmath266 because @xmath267 at the same time , @xmath268 from where @xmath269 . [ rmk - as - reg - p2-bound ] note that in the proof of theorem [ thm - as - reg - p2 ] one can replace the condition that @xmath270 by the assumption that one of the sequences @xmath43 or @xmath231 is bounded ( and so the other sequence will be bounded too ) . we show in the next section ( see theorem [ thm - conv - p2 ] ) that these two conditions are in fact equivalent . we apply in the sequel the above result to the study of a minimization problem which was considered in hilbert spaces in @xcite and was recently taken up in @xcite in the setting of cat@xmath5 spaces . let @xmath7 be a cat@xmath5 space and @xmath271-\infty,+\infty]$ ] convex , lower semi - continuous and proper . let @xmath272-\infty,+\infty]$ ] , @xmath273 where @xmath88 . we focus on the problem @xmath274 denote by @xmath275 the set of its solutions . one way to approach this problem is to apply alternatively the two resolvents corresponding to the functions @xmath86 and @xmath276 . in this process one can also allow errors when computing the values of the resolvents . thus , given a starting point @xmath153 , one can construct the sequences @xmath43 and @xmath231 defined by ( [ it : two ] ) , when considering @xmath277 and @xmath278 . note that if @xmath279 , then @xmath280 and @xmath281 , so @xmath282 . at the same time , if @xmath283 , then @xmath284 . to see this , denote @xmath280 . then , by ( [ ineq - res ] ) , for each @xmath48 , @xmath285 and , similarly , for every @xmath286 , @xmath287 adding these two inequalities and applying ( [ eq : cat(0).four.points ] ) we obtain that @xmath279 . since the resolvents @xmath288 and @xmath289 are firmly nonexpansive mappings , we can directly apply theorem [ thm - as - reg - p2 ] to obtain that the sequences @xmath43 and @xmath231 are asymptotically regular when @xmath290 . this property also follows from ( * ? ? ? * theorem 1 ) , where the given proof relies on particular properties of the resolvent and uses , instead of the assumption that @xmath290 , the weaker condition that @xmath291 is bounded below . in fact , analyzing this proof , one can immediately extract a rate of asymptotic regularity for the sequences @xmath43 and @xmath231 and a rate for the computation of @xmath292-solutions of the minimization problem ( [ def - fct - min ] ) . [ minpro1 ] let @xmath7 be cat@xmath5 space and suppose that @xmath291 is bounded below by @xmath293 . consider @xmath149 and @xmath186 such that @xmath294 . suppose that @xmath43 and @xmath231 are defined by with @xmath277 , @xmath278 , @xmath171 and @xmath295 for every @xmath172 . then for all @xmath188 and @xmath296 , we have that @xmath297 where @xmath298 + 1.\ ] ] moreover , if @xmath279 , then for all @xmath299 and @xmath300 , @xmath301 from the proof of ( * ? ? ? * theorem 1 ) we know that for every @xmath8 and @xmath63 , @xmath302 taking above @xmath303 and @xmath304 we obtain that for @xmath222 , @xmath305 from this , as in the proof of theorem [ thm - as - reg - fn - quant ] , we have that @xmath306 is a rate of asymptotic regularity for @xmath43 . suppose @xmath279 . then @xmath307 is decreasing and bounded above by @xmath308 . letting now @xmath309 and @xmath310 in ( [ cor - minpro - eq1 ] ) and using the fact that , by ( * ? ? ? * corollary 1 ) , @xmath311 for @xmath222 , it follows that for every @xmath299 and @xmath312 , @xmath313 as already noted in @xcite ( see also @xcite for the hilbert space context ) , the alternating projection method for two nonempty , closed and convex sets @xmath314 and @xmath50 in a complete cat@xmath5 space @xmath21 is a particular case of this minimization problem when taking @xmath315 and @xmath316 , where , for @xmath104 nonempty , closed and convex , @xmath317 , \quad \delta_c(x):=\left\ { \begin{array}{ll } 0 & \mbox{if } x \in c,\\ \infty & \mbox{otherwise } , \end{array } \right.\ ] ] is the indicator function . clearly , for every @xmath88 , @xmath318 . thus , the sequences @xmath43 and @xmath231 obtained as described before are asymptotically regular with a rate quadratic in @xmath182 that is given by ( [ cor - min - rate - as ] ) with @xmath319 , @xmath320 and @xmath321 . we point out that , if @xmath322 , one aims to find best approximation pairs with respect to @xmath314 and @xmath50 , that is , find @xmath323 such that @xmath324 . in hilbert spaces , this problem was studied , for instance , in @xcite . using the asymptotic regularity results from the previous section we prove in this section that the sequences generated by ( [ it : inexact ] ) and ( [ it : two ] ) @xmath6-converge . moreover , if we impose the condition that the image of one of the considered mappings is boundedly compact , then we obtain strong convergence . the results that we give below find natural applications to projection methods or the minimization problem defined by ( [ def - fct - min ] ) . [ thm - conv - p1 ] let @xmath7 be a complete @xmath0-uniformly convex space and suppose that @xmath325 are firmly nonexpansive mappings . if @xmath326 , then given any starting point @xmath153 , the sequence @xmath43 defined by is @xmath6-convergent to some @xmath327 . if , in addition , there exists @xmath328 such that @xmath329 is boundedly compact , then @xmath43 converges to @xmath330 . let @xmath63 and @xmath331 . because @xmath332 we have , by lemma [ lemma - conv - seq ] , that the sequence @xmath333 converges . let @xmath328 . for @xmath63 , let @xmath334 with @xmath335 be such that @xmath336 . then , @xmath337 by theorem [ thm - as - reg - p1 ] , @xmath43 is asymptotically regular , so @xmath338 . denote by @xmath330 the unique asymptotic center of @xmath43 . let @xmath339 and assume that its unique asymptotic center is @xmath0 . then , for every @xmath340 @xmath341 taking limit superior as @xmath342 we conclude that @xmath343 . then , @xmath344 which yields , by uniqueness of asymptotic centers , that @xmath345 thus , @xmath43 @xmath6-converges to @xmath330 . we may suppose that @xmath346 is boundedly compact . since @xmath347 is a bounded sequence in @xmath346 it has a convergent subsequence whose limit will be @xmath330 . since @xmath348 converges , this implies that @xmath43 converges to @xmath330 . as a straightforward consequence , the sequence generated by the cyclic projection method for a finite number of sets @xmath6-converges in cat@xmath5 spaces . moreover , if one of the sets is locally compact , then we obtain in fact strong convergence . this result extends ( * ? ? ? * theorem 4.1 ) and a corresponding one mentioned in @xcite as an application of a proximal splitting algorithm . [ cor : projections - convergence ] let @xmath21 be a complete cat@xmath5 space and @xmath177 be nonempty closed and convex sets such that @xmath178 . given any starting point @xmath153 , let @xmath43 be defined by ( [ it : inexact ] ) with @xmath179 , for @xmath349 . then @xmath43 @xmath6-converges to some point @xmath350 . moreover , if there exists @xmath229 such that @xmath351 is locally compact , then @xmath43 converges to @xmath330 . a similar result also holds in spaces of curvature bounded above by some @xmath54 if assuming an appropriate bound on the diameter and local compactness of one of the sets . let @xmath21 be a complete cat@xmath1 space with @xmath54 and for which @xmath176 . let @xmath177 be nonempty closed and convex sets such that @xmath178 and there exists @xmath229 such that @xmath351 is locally compact . given any starting point @xmath153 , let @xmath43 be defined by ( [ it : inexact ] ) with @xmath179 , for @xmath349 . then @xmath43 converges to some point in @xmath352 . as in the proof of theorem [ thm - conv - p1 ] , one has that @xmath333 is convergent for every @xmath353 and @xmath354 for @xmath349 ( here one applies corollary [ cor - proj - as - ref ] and the fact that @xmath355 is uniformly continuous as mentioned in subsection [ subsection - geod - sp ] ) . we can suppose that @xmath356 is locally compact . since @xmath347 is a bounded sequence in @xmath356 it has a convergent subsequence @xmath357 . suppose its limit is @xmath358 . taking into account that , for every @xmath328 , @xmath359 , it follows that @xmath360 . thus , because @xmath348 converges , we have that @xmath361 and we are done . we focus next on the convergence of the sequences defined by ( [ it : two ] ) when the considered mappings satisfy @xmath362 . [ thm - conv - p2 ] let @xmath7 be a complete cat@xmath5 space . suppose @xmath183 satisfy property @xmath362 . given any starting point @xmath153 , consider the sequences @xmath43 and @xmath231 defined by ( [ it : two ] ) . the following statements hold true : * if @xmath363 , then @xmath43 and @xmath231 are unbounded ; * if @xmath364 , then there exists @xmath365 such that @xmath43 and @xmath231 @xmath6-converge to @xmath330 and @xmath366 , respectively . if , in addition , the image of @xmath122 or @xmath367 is boundedly compact , then the convergence is strong . \(i ) suppose @xmath43 is bounded . denote @xmath368 . then , for every @xmath63 and @xmath18 , @xmath369 thus , @xmath370 . denote by @xmath330 the unique asymptotic center of @xmath371 . let @xmath372 and assume that its unique asymptotic center is @xmath0 . then , @xmath373 taking limit superior as @xmath342 and using theorem [ thm - as - reg - p2 ] and remark [ rmk - as - reg - p2-bound ] we conclude that @xmath374 , which shows that @xmath375 . \(ii ) if @xmath376 , clearly @xmath43 and @xmath231 are bounded and we continue the reasoning from ( i ) . applying ( [ thm - conv - p2-eq1 ] ) with @xmath377 we have , by lemma [ lemma - conv - seq ] , that @xmath378 converges . this yields that @xmath379 is convergent too . then , @xmath380 which means that @xmath381 . thus , @xmath371 @xmath6-converges to @xmath330 , from where @xmath382 @xmath6-converges to @xmath330 too . in a similar way one obtains that there exists @xmath383 such that @xmath231 @xmath6-converges to @xmath384 . we need to prove that @xmath385 . since @xmath386 we have that @xmath387 one can also show that @xmath388 , which implies that @xmath389 . strong convergence when the image of @xmath122 or @xmath367 is boundedly compact can be obtained as in the proof of theorem [ thm - conv - p1 ] . it follows immediately that we can recover the convergence result for problem ( [ def - fct - min ] ) that was proved in ( * ? ? ? * theorem 1 ) . [ minpro2 ] let @xmath7 be a complete cat@xmath5 space and @xmath271-\infty,\infty]$ ] convex , lower semi - continuous and proper . suppose ( [ def - fct - min ] ) has a solution . given @xmath153 , consider the sequences @xmath43 and @xmath231 defined by ( [ it : two ] ) , where @xmath277 and @xmath278 . then there exists @xmath358 such that @xmath390 is a solution of ( [ def - fct - min ] ) for which @xmath43 and @xmath231 @xmath6-converge to @xmath330 and @xmath391 , respectively . consequently , one obtains the convergence of the alternating projection method in cat@xmath5 spaces . let @xmath7 be a complete cat@xmath5 space and @xmath392 nonempty , closed and convex . suppose that @xmath393 . given @xmath153 , consider the sequences @xmath43 and @xmath231 defined by ( [ it : two ] ) , when @xmath394 and @xmath395 . then there exists @xmath396 such that @xmath43 and @xmath231 @xmath6-converge to @xmath397 and @xmath398 , respectively . if , in addition , @xmath314 or @xmath50 are locally compact , then the convergence is strong . note that in the above result @xmath290 if , for example , one of the sets @xmath314 or @xmath50 is bounded . we finish this paper by illustrating the behavior of the alternating projection method in the poincar upper half - plane model , see @xcite . using _ mathematica@xmath399 _ version 9.0 , we developed an elementary program , http://cfne.url.ph/ , where one can observe the behavior of the picard iteration for the composition of projections onto two geodesics . using this program we obtain figure [ fig:1 ] which clearly depicts the following possible alternatives : the first one , figure [ fig:1 ] ( a ) , corresponds to the situation where the intersection of the two geodesics is nonempty and illustrates strong convergence of the picard iteration to a point in this intersection , see corollay [ cor : projections - convergence ] ; when the intersection is empty , as in figure [ fig:1 ] ( b ) and ( c ) , the picard iteration either converges to a best approximation pair ( bounded case ) or diverges ( unbounded case ) , as is expected from theorem [ thm - conv - p2 ] . in this paper we have primarily focused on providing an appropriate framework and a general method for the study of the convex feasibility problem for a finite number of sets in a nonlinear setting . for this purpose , it was enough to compose finitely many firmly nonexpansive mappings in a cyclic way . a question which arises naturally is whether one could take infinitely many firmly nonexpansive mappings . note that , in this case , theorem [ thm - conv - p1 ] is no longer true without additional assumptions ( it is enough to consider the proximal point algorithm in hilbert spaces where the condition imposed on the stepsize sequence is essential , see @xcite ) . regarding the minimization problem ( [ def - fct - min ] ) , we have given an explicit rate of asymptotic regularity for the sequences @xmath43 and @xmath231 defined by ( [ it : two ] ) with @xmath277 and @xmath278 . it would be interesting to give a quantitative version of theorem [ thm - as - reg - p2 ] for mappings that satisfy property and from where the asymptotic regularity of the sequences @xmath43 and @xmath231 can be deduced . we have obtained new and more general convergence results to approximate a common fixed point of a finite number of firmly nonexpansive mappings in geodesic spaces . our results , which are both qualitative and quantitative , can be applied to the cyclic projection method for finitely many sets in geodesic spaces of curvature bounded above or to an abstract minimization problem which allows the study of the alternating projection method for the inconsistent convex feasibility problem for two sets . david ariza and genaro lpez were supported by dges ( grant mtm2012 - 34847-c02 - 01 ) , junta de andaluca ( grant p08-fqm-03453 ) . adriana nicolae was supported by a grant of the romanian ministry of education , cncs - uefiscdi , project number pn - ii - ru - pd-2012 - 3 - 0152 . part of this work was carried out while adriana nicolae was visiting the university of seville . she would like to thank the department of mathematical analysis and the institute of mathematics of the university of seville ( imus ) for the hospitality . li , c. , yao , j.c . : variational inequalities for set - valued vector fields on riemannian manifolds : convexity of the solution set and the proximal point algorithm . siam j. control optim . * 50 * , 24862514 ( 2012 )
dynamical friction is the reaction force that a perturber feels from the gravitational wake it generates in the ambient medium . this fundamental interaction has therefore had a wide range of applications in the formation of astronomical structures . applications to collisionless systems include mass segregation in star clusters ( e.g. kim and morris 2003 , portegies et al 2004 ) , galaxy distribution in hierarchical cluster formation ( e.g. frenk et al 1996 , bullock and johnston 2005 ) , the decay of satellites in galactic halos and galaxy mergers ( binney and tremaine 2008 and references therein ) . dynamical friction also occurs in collisional particle systems and was applied to the study of the growth of planetesimals ( hornung et al 1985 , stewart and whetherill 1988 ) , the eccentricity excitation of planetary embryo orbits ( ida 1990 , kokubo and ida 1996 , namouni et al 1996 ) , and the confinement of planetary rings ( goldreich and tremaine 1982 ) . for gaseous systems , applications include the orbital decay of compact stars orbiting supermassive black holes ( narayan 2000 ) , and the heating of intercluster gas by decelerating supersonic galaxies ( el - zant et al . 2004 , faltenbacher et al 2005 , kim et al 2005 ) . following the pioneering work of chandrasekhar ( 1943 ) on the gravitational drag experienced by a perturber in collisionless star systems , the derivations of dynamical friction assume that the perturber has a constant velocity in the ambient medium . the assumption of uniform motion is made for collisionless systems ( binney and tremaine 2008 ) , collisional particle systems ( hornung et al 1985 , stewart and whetherill 1988 , ida 1990 , goldreich and tremaine 1982 ) as well as gaseous media ( kim and kim 2007 , kim et al 2008 , ostriker 1999 ) . the resulting dynamical friction force which depends on the perturber s velocity is subsequently used to model the evolution of the perturber s motion whose velocity varies with time . only the numerical simulations of the nonlinear fluid equations are free from this assumption ( snchez - salcedo and brandenburg 1999 , 2001 ) . the neglect of the perturber s acceleration in dynamical friction is justified by the small amplitude of the force that is proportional to the square of the pertuber s mass . neglecting acceleration is independent of the type of motion and is mainly motivated by the lack of an analytical framework where the perturber s accelerated motion could be taken into account . to overcome this obstacle , we derive in appendix a a new expression of dynamical friction that applies to a general type of motion and depends on the perturber s motion and the boundary conditions but not the density perturbation . in particular , dynamical friction is shown to be non - local as it depends on the whole trajectory of the perturber . although non - locality does not appear if the perturber s velocity is constant , its origin is easy to identify : as the perturber s wake propagates in the medium , a velocity change because of acceleration affects the density enhancement near the perturber and offsets the force from its local value as the far parts of the wake were generated by a different velocity . therefore when acceleration is taken into account , the amplitude of dynamical friction may change with respect to the value derived assuming a constant velocity because in modelling the force on an accelerated perturber as that on a constant velocity perturber , one assumes that the gravitational wake produced by accelerated motion is similar to that of uniform motion . we shall see that this difference between the two wakes makes dynamical friction dependent on the initial velocity of the perturber . another difference between the wakes generated by constant velocity motion and accelerated motion is the absence of a perturbed region with front - back symmetry with respect to a subsonic accelerating perturber . in effect a constant velocity perturber raises a gravitational wake in the ambient medium that is symmetric with respect to it and truncated by the sonic shockwave . this truncation defines two regions , one that lacks front - back symmetry located behind the perturber , and one with front - back symmetry around the perturber that becomes the largest region inside the wavefront for velocities smaller than half the sound speed . dynamical friction arises from the former region as shown by ostriker ( 1999 ) . when acceleration is taken into account , front - back symmetry is lost everywhere . this would intuitively imply that dynamical friction on an accelerated perturber should be much larger than the value derived assuming uniform motion for subsonic perturbers . curiously , we shall show that the force on an accelerating subsonic perturber initially at rest is similar to the uniform friction force . lastly , accelerated motion friction would differ from uniform motion friction because the former never reaches steady state . in effect , dynamical friction on a constant subsonic velocity perturber is known to be time - independent ( ostriker 1999 ) . it is the object of this paper to quantify how taking into account acceleration modifies dynamical friction on a perturber travelling in an infinite homogeneous gaseous medium . the perturber s motion is assumed to have a constant acceleration , a choice that makes the analytical developments simpler . physically such constant acceleration approximates the deceleration felt by the perturber because of dynamical friction as the force amplitude is small and the corresponding instantaneous velocity change is equally small . constant acceleration also models the motion of a young star with an asymmetric jet system moving in interstellar gas ( hirth et al 1994 , hartigan and morse 2007 , namouni 2005 , 2007 ) . in what follows , we characterize the density perturbation associated with accelerated motion in section 2 . we derive the force on an accelerated perturber in section 3 using the general expression of dynamical friction ( derived in appendix a ) that does not depend on the density enhancement caused by the perturber but only on the perturber s motion and the medium s boundary conditions . in particular we show that including acceleration makes dynamical friction dependent on the initial velocity of the perturber . section 4 contains concluding remarks . in appendix b , the new expression of dynamical friction derived in appendix a is validated by applying it to uniform rectilinear motion and recovering the known results on dynamical friction . consider a point - like perturber of mass @xmath0 on an accelerated rectilinear trajectory , of initial velocity @xmath1 and acceleration @xmath2 , that travels inside an inviscid isothermal gaseous medium of density @xmath3 and sound speed @xmath4 . assuming the density perturbation associated with the perturber s wake is small compared to the medium s unperturbed density @xmath5 , the corresponding density enhancement @xmath6 defined as @xmath7 satisfies the standard forced sound wave equation ( appendix a , equation [ wav2a ] ) : @xmath8 { \cal h}(t ) , \label{waveeq}\ ] ] where @xmath9 is the perturber s gravitational potential , @xmath10 is the heaviside function , @xmath11 is the gravitational constant , @xmath12 is the perturber s trajectory chosen along the @xmath13direction and @xmath14 is dirac s delta function . the solution of this equation is the linard - wiechert potential which may be written in integral form using the retarded green function as ( [ dens2a ] ) : @xmath15 { \cal h}(u ) } { |{\bm x}-{\bm \xi}(u)|}\ { \rm d}u . \label{density}\end{aligned}\ ] ] the acceleration @xmath2 introduces two natural time and length scales given by : @xmath16 in these units , the perturber s trajectory may be written as @xmath17 where @xmath18 is the initial mach number . the mach number is a function of time and reads : @xmath19 . the scaling time @xmath20 is the duration to increase the mach number by unity and the scaling length @xmath21 is the corresponding travelled distance . applied to the density enhancement these scalings yield : @xmath22^{1/2}\right ) { \cal h}(u ) } { \left[r^2+\left(z-{\cal m}_0u- u^{2}/2\right)^2\right]^{1/2}}\ { \rm d}u , \label{density2}\ ] ] where @xmath23 is the cylindrical radius ; unless it is stated otherwise , all length and time variables and constants are normalized to ( [ scalings ] ) . in order to derive the density profile we use the property of the @xmath24-function @xmath25=\sum_i \alpha(u_i ) { \delta(u - u_i)}/{|\beta^\prime(u_i)|}$ ] where @xmath26 are the roots of @xmath27 , and @xmath28 is the derivative of @xmath29 . the roots of the argument of the @xmath24function in equation ( [ density2 ] ) satisfy the quartic equation : @xmath30 where @xmath31 is the spherical radius . this equation is the counterpart of the quadratic equation derived by ostriker ( 1999 ) for uniform motion . the former may be reduced to the latter once time and length dimensions are restored . equation ( [ roots ] ) , however , can not be solved analytically because its varying coefficients do not permit simple and unique expressions of the roots . instead we determine the roots numerically and substitute them into the density expression obtained by applying the @xmath24function property cited above . the density expression reduces to : @xmath32 fig . ( 1 ) follows the level curves of @xmath6 for perturbers initially at rest ( @xmath33 ) as they evolve from @xmath34 to mach 6 . the front - back symmetry that characterizes the density enhancement of subsonic constant velocity motion in a uniform medium is absent for both subsonic and supersonic motion . the perturber reaches mach 1 at @xmath35 but remains inside the sonic wavefront of radius @xmath36 as the latter has been propagating faster than the perturber s velocity . once the velocity is supersonic , a mach cone is formed inside the sonic wavefront and a shallow density depression develops and trails the perturber ( @xmath37 ) . eventually as mach 2 is reached the perturber catches up with the sonic shockwave and exits the sonic region to form an external mach cone . ( 2 ) shows the level curves of @xmath6 associated with decelerating perturbers with @xmath38 as a function time until the perturber s velocity vanishes . the density enhancement of subsonic perturbers has a structure similar to that of uniform motion ( ostriker 1999 ) but lacks its front - back symmetry with respect to the perturber . we show in the following section that under certain circumstances , dynamical friction may be not significantly affected by the loss of front - back symmetry even as closed level curves around the perturber contribute to the force unlike those of uniform motion ( see introduction ) . the deceleration of supersonic perturbers modifies the structure of the density enhancement and reduces the extent of the mach cone but does not erase it as the velocity vanishes ( bottom row , right panel ) indicating the presence of a finite friction force . the reaction force exerted on a point - like perturber in a homogeneous gaseous medium by its gravitational wake is given by equation ( [ for3a ] ) and is written in dimensional variables as : @xmath39 , r\leq ct}r\sin\theta{\rm d}r\ , { \rm d}\theta\ , { \rm d}\varphi \ \frac{{\bm y}-{\bm \delta}}{|{\bm y}-{\bm \delta}|^3},\label{force}\ ] ] where @xmath40 , @xmath41 and @xmath42 are the spherical coordinates of the vector @xmath43 defined as the relative position within the medium with respect to the perturber s position at the retarded time @xmath44 . it may be written as @xmath45 where @xmath46 is the position of fluid element in the medium . this definition of @xmath43 influences the force integral through the boundary conditions of the medium denoted by @xmath47 . the phase @xmath48 , the upper radial cutoff at the retarded sonic shockwave @xmath49 , and the boundary conditions are all that is necessary to compute the friction force . in particular , the absence of an explicit expression for the density enhancement is irrelevant for the derivation of the force . the derivation of expression ( [ force ] ) assumes that the perturber is assembled at @xmath34 a fact that introduces the size of the retarded sonic shockwave as an upper cutoff . this assumption is not restrictive as it allows us to study the time - dependent evolution of dynamical friction as well approximate steady state behaviour ( see section 3.2 ) . it also describes turning on a numerical simulation of dynamical friction ( snchez - salcedo and brandenburg 1999 , 2001 ) . for rectilinear motion along the @xmath13direction in an infinite medium , the force is equally along @xmath50 and its expression can be reduced further by integrating over the longitude @xmath51 and the variable @xmath52 as : @xmath53 , r\leq t}\ \frac{{\rm d}r } { \delta^2}\ \left[\frac { r-\tau\delta}{\left|1+(\delta / r)^2 - 2(\delta / r)\tau\right|^{1/2}}\right]^{\tau_{\rm max}}_{\tau_{\rm min}},\label{forcex}\ ] ] where the scalings ( [ scalings ] ) have been applied ; the perturber s trajectory @xmath54 , the time , @xmath36 , and the position @xmath46 and its spherical components are dimensionless hereafter . the expressions of @xmath55 and @xmath56 as a function of the perturber s trajectory and the variable @xmath40 are determined from the boundary conditions . for uniform rectilinear motion , @xmath57 is constant and the phase @xmath58 . calling @xmath59 the gravitational potential s small distance cutoff scaled to @xmath21 ( [ scalings ] ) , the integral that appears in the expression ( [ forcex ] ) may be written as ( appendix b , equation [ ostria ] ; ostriker 1999 ) : @xmath60\left[1+{\rm sign}(1-{\cal m})\right]-\log \left|\frac{1+{\cal m}}{1-{\cal m}}\right| -\left[1-{\rm sign } ( 1-{\cal m})\right ] \ \log\left|\frac{(1-{\cal m})t}{r_0}\right|\label{ostriker}\right).\end{aligned}\ ] ] for accelerated motion , the phase @xmath61 . in order to examine how acceleration modifies the dynamical friction force , we consider the three cases of perturbers initially at rest ( @xmath33 ) , accelerating perturbers ( @xmath62 ) and decelerating perturbers ( @xmath63 ) . before deriving dynamical friction for accelerated motion , a choice of the small distance cutoff @xmath59 must be made . this cutoff is usually taken as the perturber s accretion radius or the system s typical size . the former dimensional cutoff reads @xmath64 and is a function of the mach number whereas the latter is a constant that does not depend on the flow s variables . in this work , the perturber is treated as a compact object whose gravitational potential is truncated at small distances by the perturber s size . to satisfy such a condition and be consistent with a point - like potential , the perturber s size needs to be much smaller than the accretion radius @xmath64 . this choice of cutoff is in agreement with the numerical simulation of the fluid equations in the context of bondi - hoyle - lyttleton accretion ( edgar 2005 ) . for larger size or extended objects , equation ( [ force ] ) no longer applies and dynamical friction may be obtained from the general expression derived for an arbitrary perturbing potential ( appendix a , equation [ forc2a ] ) . in an infinite medium , the only relevant physical boundary condition is that of the small distance cutoff @xmath65 . , @xmath66 and @xmath59 remain scaled to @xmath21 ( [ scalings ] ) . ] for a perturber initially at rest , @xmath67 , @xmath68 and the mach number is @xmath69 . the variables @xmath40 and @xmath70 therefore need to satisfy the three conditions : ( i ) @xmath71 , ( ii ) @xmath72/(2t - r)\equiv f(r , t , r_0)$ ] , and ( iii ) @xmath73 . the last condition implies that @xmath74 . examination of the three conditions shows @xmath75 , and that @xmath56 and the remaining bounds on @xmath40 in the integral ( [ forcex ] ) depend on the value of @xmath36 as the velocity increases . there are five time intervals that need to be considered : @xmath76 case i is that of transitory motion where the density perturbation is in the perturber s vicinity . prior to @xmath77 , the force is zero as the three conditions above can not be satisfied . the integration domain of @xmath40 and @xmath56 are given as : @xmath78^{1/2}. \end{aligned}\ ] ] calling @xmath79 the integral in equation ( [ forcex ] ) and using these boundaries for @xmath40 , we find : @xmath80 _ { r_2^{f}}^{t } , \label{i0}\\ i_{\rm i}&=&-\frac{r_0(2+r^f_2)-(2+r_0+r^f_2)t -r^f_2t^2+t^3}{2r_0t^2}-\frac{2t - r_0}{2t^3}\ , \log\left|\frac{2t - r_0}{tr^f_2-r_0}\right|.\label{case1}\end{aligned}\ ] ] examination of the variation of @xmath81 as a function of time shows that it decays smoothly to zero as @xmath77 is approached . the expression of the force in this time interval is the equivalent of that derived for uniform motion in appendix b ( equation [ transient ] ) when @xmath82 and the wake has not left the perturber s vicinity . in that case , the force read @xmath83+{\cal m}-{r_0}/{2ct}+(1-{\cal m}^2)ct/2r_0\}$ ] ; it helped explain the transient phase observed before saturation in the numerical simulation of the nonlinear fluid equations ( snchez - salcedo and brandenburg 1999 ) . case ii covers subsonic motion and the corresponding integration boundaries are : @xmath84^{1/2 } , \label{dom21}\\ r^c_1\leq r\leq t&\mbox{with } & \tau_{\rm max}=1.\label{dom22}\end{aligned}\ ] ] the latter domain ( [ dom22 ] ) does not contribute to the force . the corresponding integrand is proportional to @xmath85 and the sign argument is positive for @xmath86 . the expression of @xmath79 for domain ( [ dom21 ] ) may be obtained from equation ( [ i0 ] ) by substituting @xmath87 for @xmath36 . the result is : @xmath88 in the limit @xmath89 and @xmath90 , @xmath91 . this is exactly the friction force acting on subsonic perturbers with uniform motion given by equation ( [ ostriker ] ) . it is interesting that although the density enhancements of subsonic accelerated motion and subsonic uniform motion differ significantly , the corresponding friction forces are identical when the limit @xmath92 is taken for the accelerated motion friction . ) is independent of @xmath59 and that no assumptions on @xmath59 are made to derive its expression . ] this result is the more surprising as the density enhancement associated with accelerated motion does not involve a region with front - back symmetry as explained in the introduction leading one to guess wrongly that friction would be stronger for accelerated motion than it is for uniform motion . the explanation of this agreement comes from the friction force s dependence on the phase @xmath93 ( dimensional variables ) . as the contribution of the force for subsonic motion comes mostly from the vicinity of the perturber s retarded position ( @xmath94 ) , we have @xmath95 and the phase becomes @xmath96 . the latter is exactly the phase of uniform motion with a mach number @xmath97 regardless of the type of motion @xmath98 as long as , ( 1 ) @xmath89 which implies in dimensional variables that @xmath99 , ( 2 ) the velocity remains subsonic and ( 3 ) the motion does not have a history with supersonic episodes ( see section 3.3 ) . case iii concerns motion between mach 1 and mach 2 where the perturber travels at supersonic speed but has not yet crossed the sonic shock wave . the integration domains are given as : @xmath100^{1/2 } , \label{dom31}\\ r^c_2\leq r\leq r_3^c&\mbox{with } & \tau_{\rm max}=1,\ \ \mbox{where}\ \ \ r_3^{c}=t-1+[(t-1)^2 - 2r_0]^{1/2},\label{dom32}\\ r^c_3\leq r\leq r^c_1&\mbox{with } & \tau_{\rm max}=f(r , t , r_0),\label{dom33}\\ r^c_1\leq r\leq t&\mbox{with } & \tau_{\rm max}=1.\label{dom34}\end{aligned}\ ] ] the integration radii satisfy @xmath101 . this shows that domain ( [ dom34 ] ) does not contribute a finite term to the force in the same way as domain ( [ dom22 ] ) . domains ( [ dom31 ] ) and ( [ dom33 ] ) contribute terms similar to that of equation ( [ i0 ] ) except that for the former @xmath36 is replaced with @xmath102 and for the latter @xmath36 is replaced with @xmath87 , and @xmath103 with @xmath104 . the second domain ( [ dom32 ] ) contributes the term @xmath105 given as : @xmath106_{r_2^{c}}^{r_3^{c } } , \label{i2}\end{aligned}\ ] ] which when combined with the contributions of domains ( [ dom31 ] ) and ( [ dom33 ] ) yields : @xmath107 case iv defines a small time interval near mach 2 where the perturber exits the sonic wavefront . the corresponding integration domains are : @xmath108 the expression of @xmath79 is obtained by summing up the terms associated with domains ( [ dom31 ] ) and ( [ dom32 ] ) of case iii , and substituting @xmath36 for @xmath109 in domain ( [ dom33 ] ) . the result is : @xmath110 case v concerns supersonic motion that exited the sonic wavefront @xmath111 . the integration domains are given as : @xmath112 the integral @xmath79 is obtained as previously by combining the domains of case iv and making the necessary boundary substitutions . the result is : @xmath113 for @xmath89 and @xmath114 , the scaled force reads : @xmath115.\label{case5l}\ ] ] this expression differs from that of equation ( [ ostriker ] ) through the first term in the bracket @xmath116 implying that the force on accelerated motion of a perturber initially at rest , is larger than that on uniform motion . ( 3 ) shows the friction force ( [ case1 ] ) , ( [ case2]),([case3 ] ) , ( [ case4 ] ) , ( [ case5 ] ) as a function of time , or equivalently the mach number , for the three values of the small distance cutoff @xmath117 , @xmath118 , and @xmath119 . the force obtained by using the uniform motion expression , @xmath120 , ( [ ostriker ] ) with @xmath69 is also shown for the same values of @xmath59 . for subsonic motion , the agreement between the two forces is best when @xmath59 is smallest . the force @xmath120 overestimates maximal drag by about 20% ; it underestimates the drag on supersonic motion by the factor @xmath121 . lastly , we remark that the derivation of dynamical friction in the five time intervals shows that there are no force singularities associated with accelerated motion such as that at mach 1 for uniform motion ( [ ostriker ] ) . to an accelerating perturber of trajectory , @xmath122 with @xmath62 , corresponds the phase @xmath123 where @xmath19 is the mach number . it could therefore seem sufficient to take the force expressions of perturbers initially at rest and substitute @xmath57 for @xmath36 . there is however a key difference between the two cases : the condition from the retarded sonic wavefront , @xmath73 , refers to time regardless of the type of motion . it should therefore be replaced with @xmath124 that we refer to as condition ( iv ) . this difference makes the force dependent on the initial velocity in contrast to the expressions derived assuming a constant velocity ( [ ostriker ] ) . the physical reason for such a dependence was discussed in the introduction : as the velocity varies , the far part of the wake that was launched in the medium corresponds to an earlier velocity . combining the force from the far part of the wake to that near the perturber results in a force that has a memory of the earlier velocity . inspection of the conditions ( i , ii ) of the previous section along with ( iv ) shows that the two cases @xmath125 and @xmath126 may be considered separately . in the following we do not derive the force expressions for the counterparts of cases i and iv as their space extensions vanish when @xmath59 tends to zero . for an accelerated perturber with an initial supersonic speed ( @xmath125 ) , the boundary conditions ( i , ii , iv ) away from the transient phase ( case i ) are identical to those of case v ( [ dom - v ] ) except that in @xmath103 and @xmath102 , @xmath36 is replaced with @xmath57 , and the outer integration domain is @xmath127 . provided that @xmath128^{1/2}$ ] that we call case vi , the friction force is given as : @xmath129{\cal m } + [ r^c_2+r^f_2+{\cal m}_0(r^f_2-r^c_2)]{\cal m}^2+(r^f_2 -r^c_2){\cal m}^3}{2 r_0 { \cal m}^2({\cal m}+{\cal m}_0 ) } \nonumber \\ & & -\frac{2{\cal m}+r_0}{2{\cal m}^3}\,\log\left|\frac{2{\cal m}+r_0}{r_2^{c}{\cal m}-r_0}\right| - \frac{2{\cal m}-r_0}{2{\cal m}^3}\,\log\left|\frac{2{\cal m}-r_0}{r_2^{f}{\cal m}-r_0}\right| -\frac{2}{{\cal m}^2}\log\left|\frac{{\cal m}-{\cal m}_0}{{\cal m}+{\cal m}_0}\right|.\label{case6}\end{aligned}\ ] ] new terms depending the initial mach number appear in the expression of the force . as @xmath18 is set to zero the force reduces to ( [ case5 ] ) with @xmath130 . this shows that including acceleration in the perturber s motion makes the medium store the information about the perturber s initial state which is in turn reflected in the reaction force felt by the perturber . taking the limit @xmath89 , the force becomes : @xmath131 three extra terms appear with respect to the uniform velocity expression ( [ ostriker ] ) . as @xmath132 , the force tends to that of case v ( [ case5l ] ) of perturbers initially at rest as the density perturbation of a larger mach number dominates the initial density perturbation . for an accelerated perturber with an initial subsonic speed , the boundary conditions ( i , ii , iv ) show that the force expression is given as : @xmath133^{1/2}\leq { \cal m}\leq1+(2r_0)^{1/2 } : & & i = i_{\rm ii},\label{case61}\\ 1+(2r_0)^{1/2}\leq { \cal m}\leq 1+\left[(1-{\cal m}_0)^2 - 2r_0\right]^{1/2 } : & & i = i_{\rm iii},\label{case62}\\ 1+\left[(1-{\cal m}_0)^2 + 2r_0\right]^{1/2}\leq { \cal m } : & & i = i_{\rm vi},\label{case63}\end{aligned}\ ] ] where in the expressions of @xmath134 and @xmath135 , @xmath36 is replaced with @xmath57 . we recover the general statement we made about the force on subsonic motion in case ii . in particular , the friction force on an initially subsonic perturber depends on the initial velocity as soon as the velocity is supersonic and becomes larger than @xmath136 . in fig . ( 4 ) we show the friction force with @xmath137 for an initial subsonic velocity with @xmath138 using equations ( [ case61 ] ) , ( [ case62 ] ) and ( [ case63 ] ) , and an initial supersonic velocity with @xmath139 using ( [ case6 ] ) . we remark that the dependence of dynamical friction on the initial velocity of the perturber is not an artifact of choosing to assemble the perturber at @xmath34 . this starting condition can be replaced with @xmath140 and @xmath141 in turn can be made to tend to infinity thus approximating a steady state . examination of the general force expression shows that @xmath142 is equivalent to the condition @xmath34 where the limits @xmath143 and @xmath144 are taken for accelerating and decelerating perturbers respectively . for positively accelerating perturbers ( case iv ) , the force is independent of the initial mach number as @xmath145 ( [ case60 ] ) . in the next section , we show that for decelerating perturbers the dependence on @xmath18 remains as the velocity decays to zero . for decelerating perturbers ( @xmath63 ) , we seek the dynamical friction force from @xmath34 until the velocity vanishes at @xmath146 . the choice of a negative mach number instead of a negative acceleration makes the derivations easier but results in a positive friction force ( opposed to the direction of motion ) . the phase @xmath123 is now negative and the boundary conditions that determine the integration domains of ( [ forcex ] ) are ( i ) , ( vi ) and ( v ) @xmath147 . the latter condition shows that @xmath148 . three cases need to be considered : @xmath149 , @xmath150 and @xmath151 . for subsonic motion ( @xmath149 ) and away from the transient regime ( i.e. @xmath152^{1/2}\leq { \cal m}$ ] ) , the integration domains are given as : @xmath153 as for case ii ( equations [ dom21 ] and [ dom22 ] ) , the latter domain ( [ decesub2 ] ) does not contribute to the force . the corresponding integrand is proportional to @xmath154 and the sign argument is positive for @xmath155 . the contribution of domain ( [ decesub1 ] ) is identical to that of case ii and @xmath156 ( [ case2 ] ) with @xmath36 replaced by @xmath57 . dynamical friction on a decelerating subsonic perturber is therefore independent of its initial velocity . this confirms the general statement we made about subsonic motion in an infinite medium in case ii . for supersonic perturbers with an initial mach number @xmath150 , we need to consider three cases depending on the value of the mach number @xmath57 . the first which we call case vii corresponds to @xmath157^{1/2}\leq { \cal m}\leq -1+(2r_0)^{1/2}$ ] whose boundaries are : @xmath158^{1/2 } , \label{decesup1}\\ r^f_1\leq r\leq { \cal m}-{\cal m}_0 , & \mbox{with } & \tau_{\rm min}=-1 . \label{decesup2}\end{aligned}\ ] ] domain ( [ decesup1 ] ) contributes a term similar to ( [ i0 ] ) with @xmath87 and @xmath159 substituted for @xmath36 and @xmath103 respectively . domain ( [ decesup2 ] ) contributes a term similar to ( [ i2 ] ) with @xmath159 and @xmath160 substituted for @xmath104 and @xmath102 respectively . the force is then given as : @xmath161 { \cal m}+[r^c_1 + r^f_1 + { \cal m}_0 ( r^f_1- r^c_1 ) ] { \cal m}^2 + ( r^f_1-r^c_1 ) { \cal m}^3}{2r_0{\cal m}^2({\cal m}+{\cal m}_0)}+\nonumber\\ & & + \frac{2{\cal m}+r_0}{2 { \cal m}^3}\,\log\left|\frac{2{\cal m}+r_0}{r_1^{f}{\cal m}+r_0}\right| + \frac{2{\cal m}-r_0}{2{\cal m}^3}\,\log\left|\frac{2{\cal m}-r_0}{r_1^{c}{\cal m}+r_0}\right| + \frac{2}{{\cal m}^2}\log\left|\frac{{\cal m}-{\cal m}_0}{{\cal m}+{\cal m}_0}\right| \label{case7}.\end{aligned}\ ] ] the second case which we call case viii concerns @xmath162^{1/2}$ ] . the corresponding integration domains are : @xmath163^{1/2},\label{dom83}\\ r^f_1\leq r\leq { \cal m}-{\cal m}_0 , & \mbox{with } & \tau_{\rm min}=-1.\label{dom84}\end{aligned}\ ] ] the integration radii satisfy @xmath164 . in contrast to case iii , this ordering shows that the second domain ( [ dom82 ] ) does not contribute a finite term to the force . domain ( [ dom81 ] ) contributes a term equal to @xmath134 , and domain ( [ dom83 ] ) a term similar to ( [ i2 ] ) where @xmath165 and @xmath159 are substituted for @xmath36 and @xmath103 respectively . the last domain ( [ dom84 ] ) contributes a term similar to that of ( [ decesup2 ] ) in case vii . summing up all terms , we find : @xmath166{\cal m}}{2r_0{\cal m}^2({\cal m}+{\cal m}_0)}+\nonumber\\ & & + \frac{[r^c_1 + r^f_1 + r^f_2 - r^f_3 + { \cal m}_0 ( r^f_1 + r^f_2 - r^f_3 - r^c_1 ) ] { \cal m}^2 + ( r^f_1 + r^f_2 - r^f_3-r^c_1){\cal m}^3}{2r_0{\cal m}^2({\cal m}+{\cal m}_0)}+\nonumber\\ & & + \frac{2{\cal m}+r_0}{2 { \cal m}^3}\,\log\left|\frac{2{\cal m}+r_0}{r_1^{f}{\cal m}+r_0}\right| + \frac{2{\cal m}-r_0}{2{\cal m}^3}\,\log\left|\frac{2{\cal m}-r_0}{r_3^{f}{\cal m}-r_0}\right| + \frac{2}{{\cal m}^2}\log\left|\frac{{\cal m}-{\cal m}_0}{{\cal m}+{\cal m}_0}\right|-\frac{2{\cal m}-r_0}{2{\cal m}^3}\,\log\left|\frac{r_1^c{\cal m}+r_0}{r_2^f{\cal m}-r_0}\right| . \label{case8}\end{aligned}\ ] ] the third case concerns @xmath152^{1/2}\leq { \cal m}\leq 0 $ ] whose boundaries turn out to be those of subsonic motion @xmath167 . the corresponding force is therefore given by ( [ case2 ] ) where @xmath36 is replaced @xmath57 . ( 5 , left panel ) shows the force on a perturber with an initial velocity @xmath168 using equations ( [ case7 ] ) , ( [ case8 ] ) and ( [ case2 ] ) alongside the uniform motion friction force ( [ ostriker ] ) . in the supersonic regime the uniform motion force overestimates the drag by about 10% at its maximum . in the subsonic regime the force on accelerated motion departs significantly from the uniform velocity force before joining it at @xmath169 to assume the expression ( [ case2 ] ) . in particular , when the velocity vanishes so does the force . lastly , for a supersonic perturber with an initial velocity @xmath151 , we find that : @xmath170^{1/2}\leq { \cal m}\leq -1+(2r_0)^{1/2 } : & & i = i_{\rm vii},\label{case8x } \\ -1+(2r_0)^{1/2}\leq { \cal m}\leq 0 : & & i = i_{\rm viii}.\label{case8x2}\end{aligned}\ ] ] fig . ( 5 , right panel ) shows the force on a perturber with an initial velocity @xmath171 using equations ( [ case8x ] ) and ( [ case8x2 ] ) alongside the uniform velocity friction force ( [ ostriker ] ) . in the supersonic regime the uniform motion force overestimates the drag by about 20% at its maximum . in the subsonic regime the force on accelerated motion departs significantly from that on uniform motion and is finite at @xmath172 . taking the limit where @xmath59 is small , we find @xmath173 and : @xmath174 at @xmath172 , @xmath175 in contrast to the uniform motion force which vanishes with the perturber s velocity . the knowledge of the history of the perturber s motion is therefore required to evaluate dynamical friction . simply applying uniform motion dynamical friction force ( [ ostriker ] ) may yield incorrect results regarding the perturber s evolution in the ambient medium as evidenced in fig . ( 6 ) where we plot the relative error between the accelerated motion and uniform motion forces for decelerating supersonic perturbers . a similar conclusion was previously reached by snchez - salcedo and brandenburg ( 1999 ) who numerically tested the accuracy of the uniform motion friction formula ( [ ostriker ] ) using the full nonlinear fluid equations . following the evolution of a decelerating initial perturber with @xmath176 up to @xmath177 , they found that equation ( [ ostriker ] ) underestimated the friction force as @xmath57 became smaller than unity ( see their fig . 5 ) which corresponds to our cases vii and viii . the authors attributed this discrepancy to the ` memory of the wake ' . in this paper , we studied the influence of acceleration on dynamical friction in a gaseous medium . our main conclusions are stated in the abstract . this work also constitutes an illustration of the use of the general expression of dynamical friction derived in appendix a. in particular , applying that formulation to a perturber with accelerated motion allowed us to illustrate how dynamical friction is not a local force and may depend the perturber s initial state . the meaning of the initial state depends on the problem s configuration . for instance , it may correspond to the time the perturber enters the gaseous medium or to the initial time of a numerical simulation of dynamical friction . we note that non - locality is not peculiar to rectilinear motion and should be true of the dynamical friction associated with rotational ( circular or eccentric ) motion . the main technical result is the set of expressions @xmath81 to @xmath178 that describe dynamical friction as a function of the initial and current velocities as the perturber is accelerated ( positively or negatively ) in the ambient medium . our derivations have assumed a constant acceleration because the calculations are simpler to carry out in this case . dynamical friction for constantly accelerating perturbers applies readily to the motion of young stars with asymmetric jet systems . we believe that the general features we found will hold for a general type of acceleration as long as the instant velocity change remains small . including the deceleration generated by dynamical friction instead of a constant negative acceleration is likely to further reduce the friction force because the acceleration s amplitude increases with time . the dependence of dynamical friction on the perturber s size , @xmath59 , is most important for supersonic motion . in the force expressions , the dimensionless radius , @xmath59 , stands in for the ratio @xmath179 of dimensional parameters . the reaction force therefore depends explicitly on acceleration even in the limit @xmath180 as in case vi . dynamical friction on subsonic motion is independent of @xmath59 when @xmath180 . using the force expression requires the knowledge of the perturber s acceleration and the medium s sound speed . in fig . ( 3 ) to ( 5 ) , the value @xmath137 was used so as to make accelerated motion friction closer to uniform motion friction ( see fig . the assumption of constant velocity in the derivation of dynamical friction is also made for collisionless star systems . although they have different magnitudes , the uniform motion friction forces in a gaseous medium and a collisionless star system behave similarly ( as @xmath181 for supersonic motion and @xmath182 for subsonic motion ; see ostriker 1999 , snchez - salcedo and brandenburg 2001 ) . this similarity suggests that the conclusions we found are likely to hold for collisionless systems . the author thanks an anonymous referee and scott tremaine for useful comments . in this appendix , we derive a general expression of dynamical friction on a perturber of gravitational potential @xmath183 with a general trajectory @xmath54 that travels inside an inviscid homogeneous gaseous medium . the perturber is set in motion at time @xmath34 . the medium s response to the perturber s presence is described by the standard fluid equations : @xmath184 where @xmath185 and @xmath186 are the perturbed velocity and density fields , @xmath187 is the pressure , and @xmath10 is the heaviside function . prior to the perturber s motion , the gaseous medium is assumed to be at rest @xmath188 $ ] , homogeneous of density @xmath5 , pressure @xmath189 and isothermal with a sound speed @xmath190 . the linearization of medium s equation in the vicinity of the rest state yields : @xmath191 where we used @xmath7 and @xmath192 . the combined equations in turn yield the forced sound wave equation satisfied by the density perturbation @xmath6 as : @xmath193 the solution of this equation is known in electromagnetism theory as the linard - wiechert potential ( jackson 1998 , chapter 14 ) and represents the electric potential of a moving electric charge distribution . it is written in integral form using the retarded green function as : @xmath194 \ \nabla^2\phi_p[{\bm x}^\prime - { \bm \xi}(u)]\ { \cal h}(u ) } { |{\bm x}-{\bm x}^\prime|}. \label{densa}\end{aligned}\ ] ] in reaction to the perturber s excitation of the wake , the gaseous medium exerts a force on the perturber given as : @xmath195\ \nabla^2\phi_p[{\bm x}^\prime - { \bm \xi}(u)]\ { \bm \nabla } \phi_p[{\bm x } - { \bm \xi}(t)]\ { \cal h}(u ) } { |{\bm x}-{\bm x}^\prime|}\label{forc1a}\ ] ] the symbol @xmath196 denotes the boundaries of the spacial integration domain defined by the medium s physical boundary as well as the lower distance cutoff in the perturber s vicinity such as its size or its corresponding accretion radius . the boundaries @xmath196 are used to truncate the perturber s potential at small and large distances to avoid divergence . applying the variable changes , @xmath197 and @xmath198 and integrating the force over @xmath199 yields the general expression of dynamical friction as : @xmath200 , |{\bm y}|\leq ct}{\rm d}^3x\frac{{\rm d}^3y}{|{\bm y}| } \ \ \rho_p({\bm x}-{\bm y}+{\bm \delta})\ { \bm \nabla } \phi_p({\bm x } ) , \label{forc2a } \end{aligned}\ ] ] where @xmath201 is the perturber s density distribution , @xmath202 , and @xmath203 . for a given perturbing potential , dynamical friction therefore depends only on the phase @xmath204 as well as the boundary conditions specified by @xmath205 and @xmath206 . the dependence of @xmath204 on the whole trajectory @xmath54 shows that dynamical friction is manifestly non - local . deriving dynamical friction through equation ( [ forc2a ] ) is in contrast with the previous two - step methods of estimating dynamical friction for uniform motion by first calculating explicitly ( [ densa ] ) then applying an analytical or numerical calculation to the first integral in equation ( [ forc1a ] ) to derive the force ( ostriker 1999 , kim and kim 2007 , kim et al 2008 ) . as a result such methods obscure the true dependence of the force on the perturber s motion and the boundary conditions . for a point - like perturber , @xmath207 and @xmath208 , where @xmath0 is the perturber s mass . the forced sound wave equation , density enhancement and friction force are reduced to : @xmath209 { \cal h}(t ) , \label{wav2a}\\ \varrho({\bm x},t ) & = & \frac{gm}{c^2}\int_{-\infty}^{+\infty } \frac{\delta \left[u - t+|{\bm x}-{\bm \xi}(u)|/c\right ] { \cal h}(u ) } { |{\bm x}-{\bm \xi}(u)|}\ { \rm d}u , \label{dens2a}\\ { \bm f}&=&\frac{{\cal h}(t)(gm)^2\rho_0}{c^2}\int_{\partial v[{\bm y}+{\bm \xi}(t - r / c ) ] , r\leq ct}r\sin\theta{\rm d}r\ , { \rm d}\theta\ , { \rm d}\varphi \ \frac{{\bm y}-{\bm \delta}}{|{\bm y}-{\bm \delta}|^3},\label{for3a } \end{aligned}\ ] ] where @xmath45 . the condition on the time of perturbation ( @xmath210 ) represented by the heaviside function in equation ( [ wava ] ) and the expression ( [ forc1a ] ) now implies that the spacial domain that enters the expression of the force is equal to the size of the wavefront , @xmath211 . however , the spacial domain refers to the relative position with respect to the perturber not at time @xmath36 but at the retarded time @xmath44 . this explains why the force ( [ for3a ] ) describes subsonic as well as supersonic motion that perturbs the medium outside the sonic wavefront of radius @xmath211 . in the force expression , the boundaries @xmath212 $ ] are treated as truncation boundaries . in the next appendix , the force expression is applied to a point - like perturber in uniform rectilinear motion to recover the corresponding known results . we validate the expressions ( [ forc2a ] ) and ( [ for3a ] ) by deriving dynamical friction for a perturber set in uniform rectilinear motion in an infinite homogeneous medium . the perturber s trajectory is written as @xmath213 where @xmath50 is the unit vector along @xmath13 , the phase @xmath214 where @xmath57 is the mach number . the friction force is equally along @xmath50 and its expression is simplified as follows : @xmath215 , r\leq ct}r\sin\theta{\rm d}r\ , { \rm d}\theta\ , { \rm d}\varphi \ \frac{r\cos \theta-{\delta}}{|r^2+\delta^2 - 2r\delta\cos\theta|^{3/2}},\\ f&=&\frac{2\pi { \cal h}(t)(gm)^2\rho_0}{v^2}\int_{\partial v[{\bm y}+{\bm \xi}(t - r / c ) ] , r\leq ct } \ \frac{{\rm d}r } { r } \left[\frac{1-{\cal m}\tau}{({1+{\cal m}^2 - 2 { \cal m}\tau})^{1/2}}\right]_{\tau_{\rm min}}^{\tau_{\rm max } } , \label{forc3a}\end{aligned}\ ] ] where in the last expression we performed the integrals over @xmath42 and @xmath52 . the extremal values @xmath216 and @xmath56 are determined from the boundary conditions . calling @xmath59 the perturber s size , the force is truncated at small distances by @xmath217 . expressing these conditions in terms of the variables @xmath40 and @xmath70 that appear in equation ( [ forc3a ] ) shows that we need to find the domains that satisfy : ( i ) @xmath218 , ( ii ) @xmath219 , and ( iii ) @xmath49 . the first two conditions yield the following integration domains : @xmath220 the inner domain ( [ dom1a ] ) corresponds to the vicinity of the perturber s retarded position . the radius @xmath221 defines a region around the perturber that does not contribute to dynamical friction as the first two conditions are not satisfied . the third condition depends on the distance @xmath211 travelled by the retarded wavefront . there are two force regimes : the local perturbation regime ( @xmath222 ) where the perturbation remains in the perturber s vicinity and the distant perturbation regime ( @xmath223 ) where the perturbation has expanded beyond the perturber s vicinity . calling @xmath79 the integral that appears in ( [ forc3a ] ) , the integration for the distant perturbation regime extends over the two domains ( [ dom1a ] ) and ( [ dom2a ] ) with @xmath224 and @xmath79 is given as : @xmath225\ { \rm d}r-\left[1-{\rm sign } ( 1-{\cal m})\right]\int_{\frac{r_0}{|1-{\cal m}|}}^{ct}\frac{{\rm d}r}{r},\\i&=&-\log \left|\frac{1+{\cal m}}{1-{\cal m}}\right|+\frac{1}{2}\left[1+{\cal m}-|1-{\cal m}|\right]\left[1+{\rm sign}(1-{\cal m})\right ] -\left[1-{\rm sign } ( 1-{\cal m})\right ] \ \log\left|\frac{(1-{\cal m})ct}{r_0}\right| . \label{ostria}\end{aligned}\ ] ] this expression agrees with those derived by ostriker ( 1999 ) for subsonic and supersonic perturbers moving in an infinite gaseous medium using explicit expressions of the density perturbation . in particular , the force , @xmath226 , is time - independent and behaves as @xmath227 for small mach numbers . we note the interesting result that only the vicinity of a subsonic perturber s retarded position ( [ dom1a ] ) contributes to dynamical friction . we show in section 3 that this property implies that the force on subsonic perturbers with a general type of motion ( not necessarily rectilinear or uniform ) is similar to ( [ ostria ] ) provided that the perturber s motion history does not involve supersonic episodes and that the perturber s acceleration @xmath228 is much smaller than @xmath229 . for the local perturbation regime ( @xmath222 ) , the force is zero before @xmath230 and only the domain ( [ dom1a ] ) is relevant to the integral @xmath79 which is written as : @xmath231\ { \rm d}r= -\log\left[\frac{ct(1+{\cal m})}{r_0}\right]+{\cal m}-\frac{r_0}{2ct}+\frac{(1-{\cal m}^2)ct}{2r_0}. \label{transient}\end{aligned}\ ] ] in fig . ( 7 ) , we show the friction force as a function of @xmath36 for subsonic and supersonic perturbers as @xmath36 increases smoothly from 0 to a value in the distant perturbation regime . the time dependence of the force agrees with that obtained from the numerical simulation of the nonlinear fluid equations ( snchez - salcedo and brandenburg 1999 , figure 2 ) . however , the attempted fit of the force in the local perturbation regime from numerical simulations ( snchez - salcedo and brandenburg 2001 , equation 22 ) does not reproduce the correct dependence of ( [ transient ] ) on @xmath57 , @xmath59 and time . from a perturber initially at rest . the level curves are determined for values of @xmath232 . top left panel : subsonic motion , ( @xmath233 , @xmath234 ) . the density increases nearer the perturber . note how the level curves do not possess front - back symmetry with respect to the perturber . top right panel : motion at mach 1 . middle left panel : supersonic motion inside the sonic wavefront . the solid level curves correspond to ( @xmath235 , @xmath234 ) and ( @xmath236 , @xmath237 ) near the perturber s mach cone . the dashed level curves correspond to a shallow depression around @xmath238 . middle right panel : the perturber reaches the sonic shockwave . the solid level curves correspond to ( @xmath235 , @xmath234 ) . the dashed level curves correspond to shallow depressions around @xmath239 inside the sonic wavefront and @xmath240 inside the mach cone . bottom left panel : supersonic motion at @xmath241 . the solid level curves correspond to ( @xmath242 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath243 . bottom right panel : supersonic motion at @xmath244 . the solid level curves correspond to ( @xmath245 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath246 . for supersonic motion , larger values of @xmath247 are confined to the sides of the mach cone where the discontinuity with the unperturbed medium is largest . , title="fig:",width=245 ] from a perturber initially at rest . the level curves are determined for values of @xmath232 . top left panel : subsonic motion , ( @xmath233 , @xmath234 ) . the density increases nearer the perturber . note how the level curves do not possess front - back symmetry with respect to the perturber . top right panel : motion at mach 1 . middle left panel : supersonic motion inside the sonic wavefront . the solid level curves correspond to ( @xmath235 , @xmath234 ) and ( @xmath236 , @xmath237 ) near the perturber s mach cone . the dashed level curves correspond to a shallow depression around @xmath238 . middle right panel : the perturber reaches the sonic shockwave . the solid level curves correspond to ( @xmath235 , @xmath234 ) . the dashed level curves correspond to shallow depressions around @xmath239 inside the sonic wavefront and @xmath240 inside the mach cone . bottom left panel : supersonic motion at @xmath241 . the solid level curves correspond to ( @xmath242 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath243 . bottom right panel : supersonic motion at @xmath244 . the solid level curves correspond to ( @xmath245 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath246 . for supersonic motion , larger values of @xmath247 are confined to the sides of the mach cone where the discontinuity with the unperturbed medium is largest . , title="fig:",width=245 ] + from a perturber initially at rest . the level curves are determined for values of @xmath232 . top left panel : subsonic motion , ( @xmath233 , @xmath234 ) . the density increases nearer the perturber . note how the level curves do not possess front - back symmetry with respect to the perturber . top right panel : motion at mach 1 . middle left panel : supersonic motion inside the sonic wavefront . the solid level curves correspond to ( @xmath235 , @xmath234 ) and ( @xmath236 , @xmath237 ) near the perturber s mach cone . the dashed level curves correspond to a shallow depression around @xmath238 . middle right panel : the perturber reaches the sonic shockwave . the solid level curves correspond to ( @xmath235 , @xmath234 ) . the dashed level curves correspond to shallow depressions around @xmath239 inside the sonic wavefront and @xmath240 inside the mach cone . bottom left panel : supersonic motion at @xmath241 . the solid level curves correspond to ( @xmath242 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath243 . bottom right panel : supersonic motion at @xmath244 . the solid level curves correspond to ( @xmath245 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath246 . for supersonic motion , larger values of @xmath247 are confined to the sides of the mach cone where the discontinuity with the unperturbed medium is largest . , title="fig:",width=245 ] from a perturber initially at rest . the level curves are determined for values of @xmath232 . top left panel : subsonic motion , ( @xmath233 , @xmath234 ) . the density increases nearer the perturber . note how the level curves do not possess front - back symmetry with respect to the perturber . top right panel : motion at mach 1 . middle left panel : supersonic motion inside the sonic wavefront . the solid level curves correspond to ( @xmath235 , @xmath234 ) and ( @xmath236 , @xmath237 ) near the perturber s mach cone . the dashed level curves correspond to a shallow depression around @xmath238 . middle right panel : the perturber reaches the sonic shockwave . the solid level curves correspond to ( @xmath235 , @xmath234 ) . the dashed level curves correspond to shallow depressions around @xmath239 inside the sonic wavefront and @xmath240 inside the mach cone . bottom left panel : supersonic motion at @xmath241 . the solid level curves correspond to ( @xmath242 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath243 . bottom right panel : supersonic motion at @xmath244 . the solid level curves correspond to ( @xmath245 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath246 . for supersonic motion , larger values of @xmath247 are confined to the sides of the mach cone where the discontinuity with the unperturbed medium is largest . , title="fig:",width=245 ] + from a perturber initially at rest . the level curves are determined for values of @xmath232 . top left panel : subsonic motion , ( @xmath233 , @xmath234 ) . the density increases nearer the perturber . note how the level curves do not possess front - back symmetry with respect to the perturber . top right panel : motion at mach 1 . middle left panel : supersonic motion inside the sonic wavefront . the solid level curves correspond to ( @xmath235 , @xmath234 ) and ( @xmath236 , @xmath237 ) near the perturber s mach cone . the dashed level curves correspond to a shallow depression around @xmath238 . middle right panel : the perturber reaches the sonic shockwave . the solid level curves correspond to ( @xmath235 , @xmath234 ) . the dashed level curves correspond to shallow depressions around @xmath239 inside the sonic wavefront and @xmath240 inside the mach cone . bottom left panel : supersonic motion at @xmath241 . the solid level curves correspond to ( @xmath242 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath243 . bottom right panel : supersonic motion at @xmath244 . the solid level curves correspond to ( @xmath245 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath246 . for supersonic motion , larger values of @xmath247 are confined to the sides of the mach cone where the discontinuity with the unperturbed medium is largest . , title="fig:",width=245 ] from a perturber initially at rest . the level curves are determined for values of @xmath232 . top left panel : subsonic motion , ( @xmath233 , @xmath234 ) . the density increases nearer the perturber . note how the level curves do not possess front - back symmetry with respect to the perturber . top right panel : motion at mach 1 . middle left panel : supersonic motion inside the sonic wavefront . the solid level curves correspond to ( @xmath235 , @xmath234 ) and ( @xmath236 , @xmath237 ) near the perturber s mach cone . the dashed level curves correspond to a shallow depression around @xmath238 . middle right panel : the perturber reaches the sonic shockwave . the solid level curves correspond to ( @xmath235 , @xmath234 ) . the dashed level curves correspond to shallow depressions around @xmath239 inside the sonic wavefront and @xmath240 inside the mach cone . bottom left panel : supersonic motion at @xmath241 . the solid level curves correspond to ( @xmath242 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath243 . bottom right panel : supersonic motion at @xmath244 . the solid level curves correspond to ( @xmath245 , @xmath234 ) . the dashed level curves correspond to a shallow depression around @xmath246 . for supersonic motion , larger values of @xmath247 are confined to the sides of the mach cone where the discontinuity with the unperturbed medium is largest . , title="fig:",width=245 ] from decelerating perturbers . the level curves are shown for a subsonic perturber with @xmath248 ( left column ) and a supersonic perturber with @xmath249 ( right column ) . note how front - back symmetry is lost everywhere for the subsonic perturber . this however does not significantly modify dynamical friction from its value obtained using a constant velocity . , title="fig:",width=245 ] from decelerating perturbers . the level curves are shown for a subsonic perturber with @xmath248 ( left column ) and a supersonic perturber with @xmath249 ( right column ) . note how front - back symmetry is lost everywhere for the subsonic perturber . this however does not significantly modify dynamical friction from its value obtained using a constant velocity . , title="fig:",width=245 ] + from decelerating perturbers . the level curves are shown for a subsonic perturber with @xmath248 ( left column ) and a supersonic perturber with @xmath249 ( right column ) . note how front - back symmetry is lost everywhere for the subsonic perturber . this however does not significantly modify dynamical friction from its value obtained using a constant velocity . , title="fig:",width=245 ] from decelerating perturbers . the level curves are shown for a subsonic perturber with @xmath248 ( left column ) and a supersonic perturber with @xmath249 ( right column ) . note how front - back symmetry is lost everywhere for the subsonic perturber . this however does not significantly modify dynamical friction from its value obtained using a constant velocity . , title="fig:",width=245 ] + from decelerating perturbers . the level curves are shown for a subsonic perturber with @xmath248 ( left column ) and a supersonic perturber with @xmath249 ( right column ) . note how front - back symmetry is lost everywhere for the subsonic perturber . this however does not significantly modify dynamical friction from its value obtained using a constant velocity . , title="fig:",width=245 ] from decelerating perturbers . the level curves are shown for a subsonic perturber with @xmath248 ( left column ) and a supersonic perturber with @xmath249 ( right column ) . note how front - back symmetry is lost everywhere for the subsonic perturber . this however does not significantly modify dynamical friction from its value obtained using a constant velocity . , title="fig:",width=245 ] , @xmath118 , and @xmath119 . the solid curves correspond to the friction force that takes acceleration into account given by equations ( [ case1 ] ) , ( [ case2 ] ) , ( [ case3 ] ) , ( [ case4 ] ) and ( [ case5 ] ) , and the dashed curves to the uniform motion friction force expression ( [ ostriker]).,width=321 ] ) and an initial supersonic velocity ( right panel , @xmath139 ) . the solid curves correspond to the friction force that takes acceleration into account given by equations ( [ case6]),([case61 ] ) , ( [ case62 ] ) and ( [ case63 ] ) , and the dashed curves to the uniform motion friction force ( [ ostriker ] ) . the small cutoff radius is @xmath137 . the dotted curve in the left panel is the friction force on a perturber initially at rest.,title="fig:",width=321 ] ) and an initial supersonic velocity ( right panel , @xmath139 ) . the solid curves correspond to the friction force that takes acceleration into account given by equations ( [ case6]),([case61 ] ) , ( [ case62 ] ) and ( [ case63 ] ) , and the dashed curves to the uniform motion friction force ( [ ostriker ] ) . the small cutoff radius is @xmath137 . the dotted curve in the left panel is the friction force on a perturber initially at rest.,title="fig:",width=321 ] ( left panel ) , @xmath250 ( right panel ) . the solid curves correspond to the friction force that takes acceleration into account given by equations ( [ case2 ] ) , ( [ case7 ] ) and ( [ case8 ] ) for @xmath251 and by equations ( [ case8x ] ) and ( [ case8x2 ] ) for @xmath250 . the dashed curves correspond to the uniform motion friction force ( [ ostriker ] ) . the small distance cutoff is @xmath137.,title="fig:",width=321 ] ( left panel ) , @xmath250 ( right panel ) . the solid curves correspond to the friction force that takes acceleration into account given by equations ( [ case2 ] ) , ( [ case7 ] ) and ( [ case8 ] ) for @xmath251 and by equations ( [ case8x ] ) and ( [ case8x2 ] ) for @xmath250 . the dashed curves correspond to the uniform motion friction force ( [ ostriker ] ) . the small distance cutoff is @xmath137.,title="fig:",width=321 ] from equations ( [ ostriker ] ) and ( [ case7 ] ) , on the friction force when using the uniform motion expression instead of the accelerated motion expressions for decelerating perturbers with initial supersonic speeds . the small distance cutoff is @xmath137 . the curves correspond to different initial velocities.,width=321 ] , is given by ( [ forc3a ] ) , ( [ ostria ] ) , and ( [ transient ] ) and scaled to @xmath252 ; time is scaled to @xmath253 . the perturber s motion is initiated at @xmath34 . the mach number is indicated to the left of each curve . the time dependence is similar to that obtained from the simulation of the non - linear fluid equations in figure 2 of snchez - salcedo and brandenburg ( 1999).,width=283 ]
the third author of this paper still vividly recollects his visit to victor klee at the university of washington ( seattle ) in august 2000 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ we had a memorable drive in vic s car from the hotel to the math department . vic skipped all the small talk and immediately asked : `` do you think that the simplex method is polynomial ? '' i still remember how awkward i felt to be asked this question by the very person who had provided the first and seminal insights into it . my ( then as now quite irrelevant ) opinion shall remain between vic and me forever . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ the seminal work referred to above is of course the 1972 paper of klee and minty entitled `` how good is the simplex algorithm ? '' it deals with the number of iterations that may be required in the worst case to solve a linear program ( lp ) by dantzig s _ simplex method _ @xcite , which was at that time the only available practical method . klee and minty exhibited a family of lps ( now known as the _ klee - minty cubes _ ) for which the number of iterations is _ exponential _ in the number of variables and constraints . dantzig s specific _ pivot rule _ for selecting the next step is just one of many conceivable rules . the question left open by the work of klee and minty is whether there is some _ other _ rule that provably leads to a small number of iterations . this question is as open today as it was in 1972 . nowadays , the simplex method is no longer the only available method to solve lps . in particular , there _ are _ proven polynomial - time algorithms for solving lps khachiyan s _ ellipsoid method _ @xcite and karmarkar s _ interior point method _ @xcite that are based on techniques developed originally for nonlinear optimization . it is still unknown , though , whether there is a _ strongly _ polynomial - time algorithm for solving lps , that is , an algorithm for which the number of arithmetic operations does not depend on the bit lengths of the input numbers . [ [ the - p - matrix - linear - complementarity - problem . ] ] the p - matrix linear complementarity problem . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in this paper we are concerned with pivoting methods for the _ p - matrix linear complementarity problem _ ( p - lcp ) , a prominent problem for which neither polynomial - time algorithms nor hardness results are available . in general , an lcp is given by a matrix @xmath0 and a vector @xmath1 , and the problem is to find vectors @xmath2 such that @xmath3 it is @xmath4-complete to decide whether such vectors exist @xcite . however , in a p - lcp , @xmath5 is a p - matrix ( meaning that all principal minors are positive ) , and then there are unique solution vectors @xmath6 for every right - hand side @xmath7 @xcite . the problem of efficiently finding them is unsolved , though . megiddo has shown that the p - lcp is unlikely to be np - hard . for this , he considers the following ( more general ) variant : given @xmath5 and @xmath7 , either find @xmath6 that satisfy ( [ eq : lcpdef ] ) , or exhibit a nonpositive principal minor of @xmath5 . np - hardness of the latter variant would imply that @xmath8 @xcite . there is a different notion of hardness that might apply to the p - lcp as a member of the complexity class @xmath9 @xcite . this class has complete problems , and no polynomial - time algorithm is known for any of them . it has recently been shown that the famous problem of finding a nash equilibrium in a bimatrix game is @xmath9-complete @xcite . some researchers consider this to be an explanation why ( despite many efforts ) no polynomial - time algorithm has been found so far . incidentally , this is the _ second _ problem for which megiddo showed in his technical report @xcite that it is unlikely to be np - hard . however , it is currently not known whether the p - lcp is also ppad - complete . there are various algorithms for solving p - lcps , among them the classic method by lemke @xcite as well as interior point approaches @xcite . still , there is no known polynomial - time algorithm for p - lcps . for example , the complexity of interior point algorithms depends linearly on a matrix parameter @xmath10 @xcite that is not bounded for p - matrices . in this work , we focus on combinatorial methods for p - lcps and in particular on _ simple principal pivoting methods _ that share their essential idea with the simplex method . [ [ simple - principal - pivoting - methods . ] ] simple principal pivoting methods . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath11 , and let @xmath12 be the @xmath13 matrix whose @xmath14th column is the @xmath14th column of @xmath15 if @xmath16 , and the @xmath14th column of the @xmath13 identity matrix @xmath17 otherwise . if @xmath5 is a p - matrix , then @xmath12 is invertible for every set @xmath18 . we call @xmath18 a _ basis_. if @xmath19 , we have discovered the solution : let @xmath20 if on the other hand @xmath21 , then @xmath22 and @xmath23 defined above satisfy @xmath24 and @xmath25 , but @xmath26 fails . in _ principal pivoting _ , one tries to improve the situation by replacing the basis @xmath18 with the symmetric difference @xmath27 , where @xmath28 is some nonempty subset of the `` bad indices '' @xmath29 . the greedy choice is to let @xmath28 be the set of all bad indices in every step . for some p - lcps , however , this algorithm cycles and never terminates @xcite . , one such example is the p - lcp ( [ eq : morrismq ] ) whose digraph model ( see next paragraph ) appears in figure [ fig : uso ] . cycling occurs for any of the three start bases just below the top vertex in figure [ fig : uso ] . ] in _ simple principal pivoting _ , @xmath28 is of size one , and a _ pivot rule _ is employed to select the bad index . simple principal pivoting methods are sometimes called _ bard - type _ methods and were first studied by zoutendijk @xcite and bard @xcite . for a detailed exposition on simple principal pivoting methods see , for instance , ( * ? ? ? * section 4.2 ) or ( * ? ? ? * chapter 4 ) . [ [ the - digraph - model - and - unique - sink - orientations . ] ] the digraph model and unique sink orientations . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + there is a natural digraph model behind principal pivoting , first studied by stickney and watson @xcite . the graph s vertices are all bases @xmath18 , with a directed edge going from @xmath18 to @xmath30 if @xmath31 and @xmath32 . to be nondegenerate , meaning that @xmath33 for all @xmath18 and @xmath14 this is no serious restriction since a nondegenerate problem can always be obtained from a degenerate one by a symbolic perturbation of @xmath7 . ] the underlying undirected graph is the graph of the @xmath34-dimensional cube . a principal pivot step considers the cube that is spanned by the outgoing edges at the current vertex , and it jumps to the antipodal vertex of some subcube of it . a simple principal pivot step can be interpreted as the traversal of an outgoing edge , and this is the analogy with the simplex method for linear programming . the above digraph has the following specific property : every nonempty subcube ( including the whole cube ) has a unique sink , and the global sink corresponds to the solution of the p - lcp @xcite . in the terminology of szab and welzl @xcite , we are dealing with a _ unique - sink orientation _ ( uso ) . the goal of this paper is to deepen the understanding of the uso description of simple principal pivoting methods for the p - lcp . on a general level , we want to understand whether and how algebraic properties of a p - lcp ( which are well studied ) translate to combinatorial properties of the corresponding uso ( which are much less studied ) , and what the algorithmic implications of such translations are . the principal motivation behind this research is the continuing quest for a polynomial - time p - lcp algorithm . in this paper , the concept of a uso serves as the main tool to obtain two more positive results for simple principal pivoting . we believe that this combinatorial approach has some untapped potential for the theory of ( simple ) principal pivoting methods for lcps . [ [ the - randomized - method - of - murty - and - the - morris - orientations . ] ] the ( randomized ) method of murty and the morris orientations . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the simple principal pivoting method of murty ( also known as the _ least - index _ rule ) @xcite works for all p - lcp instances , but it may take exponentially many iterations in the worst case @xcite . as a possible remedy , researchers have considered _ randomized _ methods . the two most prominent ones are randomizedmurty ( which is just the least - index rule , after randomly reshuffling the indices at the beginning ) , and randomedge ( which performs a purely random walk in the uso ) . however , morris found a family of p - lcp instances ( we call their underlying digraphs the _ morris orientations _ ) on which randomedge spends a long time running in cycles and thus performs much worse than even the exhaustive search algorithm @xcite . for randomizedmurty there are as yet no such negative results on p - lcp instances . in particular , on murty s worst - case example , this algorithm takes an expected _ linear _ number of steps @xcite ; the expected number of steps becomes quadratic if we allow arbitrary start vertices @xcite . on general p - lcp instances , the expected number of iterations taken by randomizedmurty is _ subexponential _ if the underlying digraph is acyclic @xcite , but it is unknown whether this also holds under the presence of directed cycles ( as they appear in the morris orientations , for example ) . we prove that not only the randomized variant , but actually _ any _ variant of the least - index rule obtained by some initial reshuffling of indices leads to a _ number of iterations on the morris orientations . in particular , this `` kills '' another family of potentially bad instances for the randomizedmurty rule . [ [ k - matrix - linear - complementarity - problems . ] ] k - matrix linear complementarity problems . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + a _ k - matrix _ is a p - matrix so that all off - diagonal elements are non - positive . lcps with k - matrices ( k - lcps ) appear for example in free boundary problems ( * ? ? ? * section 5.1 ) and in american put option pricing @xcite . it is known that every k - lcp instance can be solved in polynomial time ( even by a simple principal pivoting method @xcite , see also ( * ? ? ? * section 4.7 ) ) , but we prove something stronger : _ every _ simple principal pivoting method takes only a linear number of iterations . we obtain this result by showing that in k - lcp - induced usos , _ all _ directed paths are short . our approach is to extract combinatorial structure from the algebraic properties of k - matrices . subsequently , we use the distilled combinatorial information to get our structural results . [ [ lcps - with - sufficient - matrices - and - the - criss - cross - method . ] ] lcps with sufficient matrices and the criss - cross method . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let us step back and take a broader view that brings together lp , the p - lcp , pivoting methods , and computational complexity : linear complementarity problems with _ sufficient _ matrices generalize both lp and the p - lcp while still being amenable to pivoting approaches . the _ criss - cross method _ @xcite may be considered an extension of murty s algorithm . in fact , the results of @xcite make lcps with sufficient matrices the largest known class of lcps for which megiddo s techniques of @xcite still apply , so that np - hardness is unlikely . determining the complexity of sufficient matrix lcp remains a tough challenge for the future . now we formally introduce the digraph model behind p - lcps and show some of its basic properties . the model was first described by stickney and watson in 1978 @xcite . we use the following notation . let @xmath35:=\{1,2,\ldots , n\}$ ] . for a bit vector @xmath36 and @xmath37 $ ] , let @xmath38 be the element of @xmath39 defined by @xmath40 this means that @xmath38 is obtained from @xmath41 by flipping all entries with coordinates in @xmath42 . instead of @xmath43 we simply write @xmath44 . under this notation , the ( undirected ) _ @xmath34-cube _ is the graph @xmath45 with @xmath46\}.\ ] ] a _ subcube _ of @xmath47 is a subgraph @xmath48 of @xmath47 where @xmath49 for some vertex @xmath41 and some set @xmath50 $ ] , and @xmath51 . [ def : uso ] let @xmath52 be an orientation of the @xmath34-cube ( a digraph with underlying undirected graph @xmath47 ) . we call @xmath52 a _ unique - sink orientation _ ( uso ) if every nonempty subcube has a unique sink in @xmath52 . figure [ fig : uso ] depicts a uso of the 3-cube . the conditions in definition [ def : uso ] require the orientation to have a unique global sink , but in addition , all proper nonempty subcubes must have unique sinks as well . in figure [ fig : uso ] there are six 2-dimensional subcubes , twelve 1-dimensional subcubes ( edges ) and eight 0-dimensional subcubes ( vertices ) . for edges and vertices , the unique - sink condition is trivial . the figure also shows that usos may contain directed cycles . ( @xmath53 for an outgoing edge ; @xmath54 for an incoming edge ) . the unique global sink and a directed cycle are highlighted . [ fig : uso ] ] if @xmath55 is the vertex set of a subcube ( as defined above ) , then the directed subgraph of @xmath56 induced by @xmath55 is denoted by @xmath57 . it is convenient to view the orientation of the @xmath34-cube @xmath47 as a mapping @xmath58 , where @xmath59 if the edge between @xmath41 and @xmath44 is oriented towards @xmath44 , and @xmath60 if it is oriented towards @xmath41 . to encode a vertex @xmath41 along with the orientations of its incident edges , we can then use a _ configuration _ of @xmath34 bits and @xmath34 signs , where the @xmath14th sign is @xmath61 ( see figure [ fig : uso ] ) . if an orientation @xmath52 of @xmath47 contains the directed edge @xmath62 , we write @xmath63 , or simply @xmath64 if @xmath52 is clear from the context . let @xmath52 be an orientation of the @xmath34-cube and let @xmath65 $ ] . then @xmath66 is the orientation of the @xmath34-cube obtained by reversing all edges in coordinates contained in @xmath67 ; formally @xmath68 if @xmath52 is a uso , then @xmath66 is also a uso for an arbitrary subset @xmath65 $ ] . this proposition can be proved by induction on the size of @xmath67 ; it essentially suffices to show that under @xmath69 , the whole cube still has a unique sink . now we can derive the _ unique - completion property_. its meaning is that we can prescribe the bits for some coordinates and the signs for the remaining coordinates , and in a uso there will always be a unique vertex that satisfies the prescription . in particular , it follows that any uso ( and any of its subcubes ) also has a _ unique source_. in fact , the unique - completion property characterizes unique - sink orientations . [ lem : ucp ] an orientation @xmath52 of the @xmath34-cube @xmath47 is a unique - sink orientation if and only if for every partition @xmath35=a\cup b$ ] and every pair of maps @xmath70 , @xmath71 there exists a unique vertex @xmath41 such that [ eq : ucp ] @xmath72 first suppose that @xmath52 is a uso . let @xmath73 and consider the subcube @xmath74 . let @xmath75 . as we have just noted , @xmath66 is a unique - sink orientation , and thus the subcube @xmath74 has a unique sink @xmath41 with respect to @xmath66 . this vertex @xmath41 is the only vertex that satisfies ( [ eq : ucp]a ) and ( [ eq : ucp]b ) . conversely , let @xmath52 be an orientation of the @xmath34-cube which satisfies the unique - completion property and let @xmath76 . set @xmath77 and @xmath78\setminus c$ ] , and let @xmath79 for @xmath80 and @xmath81 for @xmath16 . then the unique vertex @xmath41 satisfying ( [ eq : ucp]a ) and ( [ eq : ucp]b ) is the unique sink of the subcube @xmath74 . [ [ p - lcp - induced - usos . ] ] p - lcp - induced usos . + + + + + + + + + + + + + + + + + + + we now formally define the uso that is induced by a nondegenerate p - lcp instance @xmath82 @xcite . for @xmath36 , let @xmath83 : v_j=1\}$ ] be the canonical `` set representation '' of @xmath41 . then the unique - sink orientation @xmath52 induced by @xmath82 is @xmath84 recall that @xmath12 is the ( invertible ) @xmath13 matrix whose @xmath14th column is the @xmath14th column of @xmath15 if @xmath16 , and the @xmath14th column of the @xmath13 identity matrix @xmath17 otherwise . in this way , we have reduced the p - lcp to the problem of finding the sink in an implicitly given uso . access to the uso is gained through a _ vertex evaluation oracle _ that for a given vertex @xmath36 returns the orientations @xmath85 of all incident edges . in this setting , the aim is to find the sink of the uso so that the number of queries to the oracle would be bounded by a polynomial in the dimension @xmath34 . in the case of a p - lcp - induced uso , a ( strongly ) polynomial - time implementation of the vertex evaluation oracle is immediate from . from the sink of @xmath52 , we can reconstruct solution vectors @xmath22 and @xmath23 as in . the fact that the @xmath34-cube orientation defined in this way is indeed a unique - sink orientation was proved by stickney and watson @xcite . not every uso is p - lcp - induced ; for @xmath86 , the p - lcp - induced usos are characterized in @xcite , but for @xmath87 , we have only a necessary condition . this condition was originally proved by holt and klee for polytope graphs oriented by linear functions @xcite . every p - lcp - induced uso of the @xmath34-cube satisfies the _ holt - klee condition _ , meaning that there are @xmath34 directed paths from the source to the sink with pairwise disjoint sets of interior vertices.[thm : hk ] in the uso setting resulting from p - lcps , simple principal pivoting can be interpreted as follows : start from any vertex , and then proceed along outgoing edges until the global sink ( and thus the solution to the p - lcp ) is reached . a _ pivot rule _ r determines which outgoing edge to choose if there is more than one option . here is the generic algorithm , parameterized with the rule r. it outputs the unique sink of the given uso @xmath52 . @xmath88 choose @xmath36 choose @xmath89 according to the pivot rule r @xmath90 @xmath41 [ alg : pivot ] note that when @xmath52 contains directed cycles as in figure [ fig : uso ] , this algorithm may enter an infinite loop for certain rules r , but we consider only rules for which this does not happen . in the uso setting we are restricted to _ combinatorial _ rules . these are rules that may access only the orientations of edges @xmath91 ( as given by the vertex evaluation oracle ) but not any numerical values such as @xmath92 that define these orientations . let us now introduce the combinatorial pivot rules that are of interest to us . we write them as functions of the set @xmath93 of candidate indices . the first one is murty s least - index rule @xmath94 that simply chooses the smallest candidate . in the cube , this may be interpreted as a `` greedy '' approach that always traverses the first outgoing edge that it finds ( in the order of the cube dimensions ) . to it is easy to prove by induction on @xmath34 that this rule leads to a finite number of iterations , even if the uso contains directed cycles . the rule has a straightforward generalization , using a fixed permutation @xmath95 : choose the index @xmath96 with the smallest @xmath14 such that @xmath97 . this just reshuffles the cube dimensions by applying @xmath98 , and with @xmath99 , we recover murty s least - index rule . the natural randomized variant of @xmath100 is randomizedmurty : at the beginning , choose the permutation @xmath98 uniformly at random from the set @xmath101 of all permutations , and then use @xmath100 throughout . finally , randomedge stands for the plain random walk : in each iteration , simply choose a random candidate . to unlike the previous rules , randomedge may lead to cycling in @xmath102 , but the algorithm still terminates with probability @xmath103 , and with an expected number of at most @xmath104 iterations . this is a simple consequence of the fact that there is always a short directed path to the sink . [ lem : hamming ] let @xmath105 be the global sink of an @xmath34-cube uso @xmath52 , and let @xmath36 be any vertex . if @xmath106 is the hamming distance between @xmath41 and @xmath107 , then @xmath52 contains a directed path of length @xmath106 from @xmath41 to @xmath107 . morris proved that under @xmath108 , algorithm [ alg : pivot ] may be forced to perform an expected _ exponential _ number of iterations @xcite . more precisely , at least @xmath109 iterations are required on average to find the sink of the @xmath34-cube uso generated ( as described in section [ sec : uso ] ) by the @xmath110 , with @xmath111 here , @xmath34 must be an odd integer in order for @xmath5 to be a p - matrix . to prove this result morris first extracted the relevant structure of the underlying uso , and then showed that on this uso , randomedge runs in cycles for a long time before it finally reaches the global sink . it was left as an open problem to determine the expected performance of randomizedmurty , the other natural randomized rule , on this particular uso . we solve this open problem by showing that for _ any _ permutation @xmath98 , @xmath100 incurs only @xmath112 iterations on the @xmath34-dimensional morris orientation . in particular , this bound then also holds for randomizedmurty . [ thm : general ] for any @xmath113 , algorithm [ alg : pivot ] with @xmath114 finds the sink of the @xmath34-dimensional morris orientation after at most @xmath115 iterations , from any start vertex . computational experiments suggest that the worst running time is attained for the identity permutation if the algorithm starts from @xmath116 ( the global source ) . therefore we analyze this case thoroughly and compute the precise number of iterations . [ thm : id ] algorithm [ alg : pivot ] with @xmath117 finds the sink of the @xmath34-dimensional morris orientation after @xmath118 iterations , starting from @xmath116 . experimental data suggest that this is an upper bound on the number of iterations for any permutation @xmath98 and start vertex @xmath116 . allowing start vertices different from @xmath116 , we get slightly higher iteration numbers , but @xmath99 continues to be the worst permutation . in view of this , we conjecture that @xmath119 iterations actually suffice for all @xmath98 and all start vertices . this would mean that the bound in theorem [ thm : general ] is still off by a factor of @xmath120 . [ [ the - combinatorial - structure . ] ] the combinatorial structure . + + + + + + + + + + + + + + + + + + + + + + + + + + + + the @xmath34-cube uso resulting from ( [ eq : morrismq ] ) can be described in purely combinatorial terms ( * ? ? ? * lemma 1 ) . here we make its structure even more transparent by exhibiting a _ finite state transducer _ ( finite automaton with output ) with just two states that can be used to describe the uso , see figure [ fig : morrisfst ] . for the remainder of this section , we fix @xmath34 to be an odd integer and we are considering the @xmath34-dimensional morris orientation @xmath56 . ] the orientation is then determined for each vertex @xmath121 as follows . if @xmath122 , then @xmath41 is the global sink . otherwise choose some @xmath14 in @xmath35 $ ] so that @xmath123 and consider the bit string @xmath124 as the input string to the transducer . in other words , the transducer is reading the input from right to left , starting immediately to the left of some @xmath116 . the output string @xmath125 then determines the orientation of each edge incident to @xmath41 . the choice of @xmath14 does not matter , because after reading any zero , the transducer is in the state @xmath126 . for instance , let @xmath127 and let @xmath128 . then the transducer takes @xmath129 , outputs @xmath54 and switches to state @xmath130 ; reads @xmath131 , outputs @xmath53 and switches to @xmath126 ; reads @xmath132 , outputs @xmath53 and stays in @xmath126 ; reads @xmath133 , outputs @xmath54 and switches to @xmath130 ; and finally reads @xmath134 , outputs @xmath54 and switches to @xmath126 . the resulting configuration for @xmath41 is @xmath135 . figure [ fig : uso ] depicts the morris orientation for @xmath86 . it can easily be checked that the above procedural definition of the orientation is equivalent to the one given in ( * ? ? ? * lemma 1 ) . in particular , we obtain a uso . it has the remarkable symmetry that the set of configurations is closed under cyclic shifts . [ [ pivoting . ] ] pivoting . + + + + + + + + + from the transducer we can easily derive the sign changes in the configuration @xmath136 that are caused by a pivot step @xmath137 . we always have @xmath138 and @xmath139 . if @xmath140 , we get @xmath141 . in processing @xmath41 and @xmath142 , the transducer then performs exactly the same steps , except that at the @xmath14th coordinate different transitions are used to get from @xmath130 to @xmath126 . but this implies @xmath143 for @xmath144 . here is an example for such a pivot step ( with @xmath145 ; affected signs are indicated ) : @xmath146 if @xmath123 , we get @xmath147 , meaning that at the @xmath14th coordinate the transducer stays in @xmath126 for @xmath41 but switches to @xmath130 for @xmath142 ( assuming that @xmath142 is not the sink ) . this implies that signs change at all coordinates @xmath148 , @xmath149 down to ( and including ) the next coordinate @xmath106 ( wrap around possible ) for which @xmath150 . in our example we have such a step for @xmath151 : @xmath152 in both @xmath41 and @xmath142 , the signs in the block of @xmath103 s at indices @xmath153 alternate . [ [ levels . ] ] levels . + + + + + + + it will be useful to define the following function on the vertices of the cube . let @xmath154 be the number of coordinates where @xmath155 appears . formally , @xmath156 the number @xmath154 is called the _ level _ of @xmath41 . from the two types of pivots , it is easy to see that the value of @xmath157 does not increase along any directed edge ( * ? ? * lemma 3 ) . indeed , if @xmath142 is an out - neighbor of @xmath41 , then either @xmath158 ( this happens in every pivot of type , and in pivots of type with an odd block of @xmath103 s at indices @xmath153 ) , or @xmath157 decreases to the next possible value : @xmath159 , or @xmath160 if @xmath161 . in particular , the values of @xmath157 lie in the set @xmath162 . the sink is the only vertex at level 0 . let us make a small digression and briefly explain why randomedge is slow on @xmath52 . let @xmath163 . so the outdegree of @xmath41 is @xmath164 . now let @xmath41 be a vertex at level @xmath103 . in pivots of type , @xmath165 decreases by one , whereas in pivots of type that do not reach the sink , @xmath165 increases by one . the latter occurs if and only if randomedge pivots at the unique @xmath155 , which happens with probability @xmath166 . at level @xmath103 , we can thus interpret randomedge as a random walk on the integers , where @xmath167 with probability @xmath168 , and @xmath169 with probability @xmath170 . therefore the walk is strongly biased towards @xmath116 , but in order to reach a neighbor @xmath142 of the sink , we need to get to @xmath171 . this takes exponentially long in expectation @xcite . [ [ a - quadratic - bound - for - textscmurty_pi . ] ] a quadratic bound for @xmath100 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + to prove theorem [ thm : general ] we first derive a ( somewhat surprising ) more general result : for _ any _ pivot rule , starting from any vertex @xmath41 with @xmath172 for some @xmath106 , it takes @xmath112 iterations to reach a vertex @xmath142 with @xmath173 . this can be viewed as a statement about the uso induced by the @xmath34-dimensional morris orientation on the cube facet @xmath174 . the statement is that in this induced uso , all directed paths are short ( in particular , the induced uso is acyclic ) . [ lem : poten ] let @xmath36 be a vertex , and let @xmath106 be any index for which @xmath175 . starting from @xmath41 , algorithm [ alg : pivot ] ( with an arbitrary pivot rule r ) reaches a vertex @xmath142 satisfying @xmath176 after at most @xmath177 iterations . we employ a nonnegative potential that decreases with every pivot step . by a cyclic shift of coordinates we may assume that @xmath178 , and thus @xmath179 . let us define @xmath180 : v_j = 0 \text { and } \phi(v)_j={-}\},\\ n_1(v ) & : = \{j\in[n ] : v_j = 1 \text { and } \phi(v)_j={-}\}. \end{aligned}\ ] ] the _ potential _ of @xmath41 is the number @xmath181 now we will observe how the potential changes during a pivot step @xmath182 . we always have @xmath138 , @xmath139 . there are two cases . if @xmath140 , as in , the configurations of @xmath41 and @xmath142 differ only at index @xmath14 . hence @xmath183 and @xmath184 . therefore @xmath185 . in the other case we have @xmath123 , as in . if @xmath186 , we are done , since then @xmath187 . otherwise , @xmath41 is of the form @xmath188 with @xmath189 . if @xmath190 is odd , there is an even number of @xmath103 s between @xmath191 and @xmath192 , implying that @xmath193 and @xmath194 . we then have @xmath195 . if @xmath190 is even , there is an odd number of @xmath103 s between @xmath191 and @xmath192 , implying that @xmath196 and @xmath197 . since @xmath198 in this case , we get @xmath199 . to summarize , the potential decreases by at least 1 in every pivot step @xmath200 , and as long as @xmath41 is not the sink , we have @xmath201 . the highest potential is attained by the source , @xmath202 . therefore after at most @xmath177 steps we have @xmath133 . if @xmath41 happens to be at level 1 already , we obtain a better bound , since @xmath203 in this case . in fact , the vertex with largest potential in level 1 such that @xmath179 is @xmath204 , and its potential is @xmath205 . [ cor : poten ] let @xmath36 be a vertex at level 1 , and let @xmath106 be any index for which @xmath175 . starting from @xmath41 , algorithm [ alg : pivot ] ( with arbitrary pivot rule r ) reaches a vertex @xmath142 satisfying @xmath176 after at most @xmath206 iterations . now @xmath100 comes in : we will apply either lemma [ lem : poten ] or corollary [ cor : poten ] to vertices @xmath41 with @xmath207 and @xmath208 , for some @xmath14 . the next lemma shows that the @xmath103 s at coordinates @xmath209 will stay . [ lem : visonly ] let @xmath210 $ ] . after algorithm [ alg : pivot ] with @xmath211 visits a vertex @xmath41 with @xmath208 , it will visit only vertices @xmath142 satisfying @xmath212 . at every iteration , the algorithm replaces the current vertex @xmath41 with @xmath213 , where @xmath214 is the smallest index such that @xmath215 . it follows that coordinates @xmath209 will be touched only if @xmath216 for all @xmath217 . when this holds for the first time , we have @xmath218 the unique - completion property ( lemma [ lem : ucp ] ) then implies that we have already reached the sink . now we can put it all together . let @xmath41 be the start vertex , and let @xmath14 be the largest index such that @xmath207 ( if no such index exists , @xmath41 is the sink , and we are done ) . lemmas [ lem : poten ] and [ lem : visonly ] show that after at most @xmath219 iterations , we additionally have @xmath220 ( since the next pivot step flips @xmath221 ) . at this point @xmath41 is at level 1 , since there is a unique @xmath155 : at coordinate @xmath96 . after the flip at coordinate @xmath96 , we repeatedly apply corollary [ cor : poten ] together with lemma [ lem : visonly ] to also produce @xmath103 s at coordinates @xmath222 , at which point we have reached the sink . this takes at most @xmath223 more iterations , summing up to a total of @xmath224 a better bound in the case of the identity permutation can be achieved by thoroughly examining the run of the algorithm , as we do next . [ [ the - identity - permutation . ] ] the identity permutation . + + + + + + + + + + + + + + + + + + + + + + + + + let us now go on to prove theorem [ thm : id ] . for this , we first identify certain `` milestone '' vertices that are visited by algorithm [ alg : pivot ] with @xmath117 , and then we count the number of iterations to get from one milestone to the next . let us define @xmath225 [ lem : l4 ] if algorithm [ alg : pivot ] with @xmath117 starts at the vertex @xmath226 , @xmath227 , it gets to the vertex @xmath228 in @xmath229 iterations . the algorithm starts at @xmath230 , and then proceeds by the rules of pivoting and in four phases ( recall that @xmath231 always pivots on the leftmost @xmath53 ) : * it pivots on the coordinates @xmath232 , @xmath233 , , @xmath103 and thus reaches the vertex + @xmath234 after @xmath235 iterations . * it pivots on the coordinates @xmath236 , @xmath120 , @xmath237 , , @xmath238 , @xmath239 and thus reaches the vertex + @xmath240 after @xmath235 iterations ; * it pivots on the coordinates @xmath238 , @xmath241 , , @xmath236 and thus reaches the vertex + @xmath242 after @xmath14 iterations ; * it pivots on the coordinates @xmath103 , @xmath243 , @xmath244 , , @xmath232 and thus reaches the vertex + @xmath245 after @xmath235 iterations . therefore it takes @xmath229 iterations to get from @xmath226 to @xmath228 . the previous lemma allows us to count the number of iterations from @xmath246 to @xmath247 . the following lemma takes care of the remaining iterations . [ lem : l6 ] if algorithm [ alg : pivot ] with @xmath117 starts at the vertex @xmath247 , it gets to the global sink in @xmath248 iterations . the proof is similar to the proof of lemma [ lem : l4 ] , except that here only the first phase takes place : if the algorithm starts at the vertex @xmath249 , it pivots on coordinates @xmath34 , @xmath250 , , @xmath103 and thus reaches the sink after @xmath248 iterations . by lemmas [ lem : l4 ] and [ lem : l6 ] , the overall number of iterations from the source to the sink is @xmath251 in this section we examine _ k - lcps _ , the linear complementarity problems @xmath110 where @xmath5 is a k - matrix . we introduce two simple combinatorial conditions on unique - sink orientations and prove that one of them implies that all directed paths from @xmath116 are short , and the other one implies that all directed paths in the orientation are short . we show that k - lcp orientations satisfy these conditions and conclude that a simple principal pivoting method with an arbitrary pivot rule solves a k - lcp in linearly many iterations , starting from any vertex . a unique - sink orientation @xmath52 is _ 2-up - uniform _ if whenever @xmath253 is such that @xmath254 and @xmath142 is the source of @xmath255 , then the orientation of the subcube @xmath74 is uniform ( see figure [ fig:2uu ] ) . the essential fact we use is that a subcube of a k - uso is also a k - uso , as was observed already by stickney and watson ( * ? ? ? * lemma 1 ) . thus it suffices to prove 2-up - uniformity for k - lcps of dimension 2 . hence suppose that @xmath257 is a @xmath258 k - matrix , so @xmath259 and @xmath260 . then @xmath261 and nonsingular . now 0 is the source of the orientation induced by the @xmath110 if and only if @xmath262 . hence @xmath263 , which proves uniformity . one particular nice property of a 2-up - uniform orientation is that all paths from the vertex 0 to the sink @xmath107 have the shortest possible length , equal to the hamming distance @xmath264 of @xmath107 from @xmath116 . to prove this statement , we use the following lemma . suppose for the sake of contradiction that @xmath140 , @xmath266 and @xmath269 , and that @xmath267 , @xmath270 and @xmath271 . let @xmath272 and consider the subcube @xmath74 ( see figure [ fig : uni - proof1 ] , left ) : we observe that @xmath273 because otherwise the subcube would not contain a sink . but then it follows from 2-up - uniformity of @xmath52 that @xmath274 , a contradiction . it follows from lemma [ lem : crufa - up ] by induction that if @xmath142 is a vertex on any directed path starting from @xmath116 , and if @xmath147 , then @xmath274 . hence all edges on a directed path from @xmath116 to @xmath107 are oriented from 0 to 1 , and therefore the length of a directed path from @xmath116 to @xmath107 is @xmath264 . what is the actual strength of this theorem ? we know that from any vertex of a unique - sink orientation there _ exists _ a path to the sink of length at most @xmath34 ( lemma [ lem : hamming ] ) . it has also long been known how to find such a path from the vertex 0 in the case of a k - uso @xcite . the novelty is that _ any _ directed path starting from 0 reaches the sink after the least possible number of iterations . hence a simple principal pivoting method with an _ arbitrary _ pivot rule finds the solution to a k - lcp in at most @xmath34 iterations . in view of this result , it is natural to ask whether it also holds if the path starts in some vertex different from 0 . imposing 2-up - uniformity is insufficient , as we can observe by looking at the morris orientations . if we swap 0 s and 1 s in any morris orientation , we get a 2-up - uniform orientation . hence all directed paths from 0 to the sink are short ; indeed , 0 is the sink of the orientation after swapping , so all such paths have length 0 . nevertheless , long directed paths ( and even directed cycles ) do exist in this orientation . therefore we introduce the following stronger combinatorial property of usos , which turns out to be satisfied by k - usos as well . [ cor:2n ] algorithm [ alg : pivot ] with an arbitrary pivot rule r , starting at an arbitrary vertex of the corresponding uso , finds the solution to an @xmath34-dimensional k - lcp in at most @xmath280 iterations . as a consequence , we get a result for a larger class of lcps . if @xmath5 is a _ principal pivotal transform _ * section 2.3 ) of a k - matrix @xmath281 , then @xmath5 is a p - matrix ( * ? ? ? * theorem 4.1.3 ) , and the uso @xmath52 induced by the @xmath110 is isomorphic to the uso @xmath282 induced by the @xmath283 for suitable @xmath284 . it follows that corollary [ cor:2n ] also applies to the @xmath110 , even though @xmath5 is not necessarily a k - matrix itself . our proof of theorem [ thm:2lu ] is based on the following crucial fact , which extends lemma [ lem : crufa - up ] . informally , it asserts that once we have a @xmath285 in some coordinate , we will always have a @xmath285 . if @xmath266 , the claim follows from lemma [ lem : crufa - up ] . so let us suppose that we have @xmath140 , @xmath286 and @xmath269 , and that @xmath267 , @xmath270 and @xmath287 . let @xmath288 . we observe that @xmath273 so that the subcube @xmath74 contains a sink . but then it follows from 2-up - uniformity of @xmath278)}$ ] that @xmath64 , a contradiction ( see figure [ fig : uni - proof1 ] , right ) . let @xmath14 be a fixed coordinate . suppose in the considered directed path there is an edge @xmath269 such that @xmath140 . as a consequence of lemma [ lem : crufa ] , all vertices @xmath142 on a directed path starting at @xmath41 satisfy @xmath147 . it follows that in any directed path no more than two edges appear of the form @xmath289 for any fixed @xmath14 : possibly one with @xmath123 and one with @xmath140 . this fact then implies that the length of any directed path is at most @xmath280 . the strength of 2-uniformity can perhaps be explained by its being equivalent to what we call _ local uniformity_. a uso @xmath52 is _ locally up - uniform _ if for every @xmath290 such that @xmath291 and @xmath292 for all @xmath293 , the orientation of the subcube @xmath74 is uniform . a uso @xmath52 is _ locally uniform _ if both @xmath52 and @xmath278)}$ ] are locally up - uniform . local up - uniformity allows for a slight improvement in solving a k - lcp . when following a directed path starting in 0 , instead of traversing one edge at a time we can perform a `` greedy '' principal pivot : jump straight to the sink of the subcube induced by outgoing edges . unlike the general p - matrix case , it can not lead to cycling . due to local up - uniformity , it will never increase the number of iterations and may sometimes reduce it . the answer to this question is negative . it can be shown that the number of @xmath34-dimensional locally uniform usos is asymptotically much larger than the number of k - usos . a lower bound of @xmath294 on the number of locally uniform usos can be derived by an adaptation of develin s construction @xcite of many orientations satisfying the holt - klee condition . an upper bound of @xmath295 on the number of p - usos ( and thus also k - usos ) follows from develin s proof of his upper bound on the number of lp - realizable cube orientations . a. borici and h .- j . pricing american put options by fast solutions of the linear complementarity problem . in e. j. kontoghiorghes , b. rustem , and s. siokos , editors , _ computational methods in decision - making , economics and finance _ , volume 74 of _ applied optimization_. kluwer acad . dordrecht , 2002 . f. holt and v. klee . a proof of the strict monotone @xmath120-step conjecture . in _ advances in discrete and computational geometry _ , volume 223 of _ contemp . _ , pages 201216 .
in recent years , systems where strong correlations between particles are induced by pair - wise scattering , have attracted much attention . in the canonical example , which is the parametric down - conversion , photon pairs are generated during the propagation of a laser beam through a non - linear medium . the outcoming pairs of photons are entangled , and can serve as a probe of fundamental properties of quantum mechanics @xcite , such as the einstein - podolsky - rosen paradox , or violation of the bell inequalities @xcite . on the other hand , entanglement can be exploited in practical applications , such as teleportation @xcite or metrology beyond the shot - noise limit ( snl ) @xcite . in this latter context , recent experiments with entangled states of atoms were a major breakthrough @xcite . in @xcite , two - body interactions were utilized to prepare non - classical squeezed states of atoms trapped in a double - well potential , which implies presence of many - body entanglement @xcite . a similar idea was exploited to generate squeezing in the internal @xcite degrees of freedom . in @xcite , squeezing of a large spin of a collection of two - level atoms was achieved , using an intense laser field interacting with particles trapped in an optical cavity . simultaneously , a substantial experimental effort was put in order to generate entangled pairs of atoms scattered out of a bose - einstein condensate ( bec ) . in @xcite , a collision of two becs lead to weak scattering of correlated atomic pairs onto a three - dimensional sphere of initially unoccupied modes . although moderate number - squeezing between the opposite regions of the halo , and the related violation of the cauchy - schwarz inequality were experimentally demonstrated @xcite , entanglement was never directly observed . alternatively , pair - production schemes were developed , where only few modes are strongly populated in a stimulated process , making the system somewhat easier to handle . stimulated four - wave - mixing processes have been implemented using different spin states of atoms @xcite or bragg scattering @xcite . also , dynamic instabilities in moving optical lattices , populating modes with opposite quasi - momenta , have been used @xcite . in @xcite , a bec was transferred into the first excited state of a trapping potential and subsequent two - body collisions created a `` twin - beam '' system , where stronger - than - classical correlations could directly be observed . analogous schemes have been implemented in internal atomic states , building upon spin - changing collisions @xcite . furthermore , in @xcite it was shown that particles scattered in this process into a pair of @xmath0 zeeman sub - levels are usefully entangled from the metrological point of view . in this work we develop a theoretical model for the generic type of experiments , where particles scatter in pairs into two well - separated regions . if these regions are separated in the momentum space , they could also be set apart by a sufficiently long expansion of the cloud . on the other hand , in cases when the regions are defined as two zeeman sub - levels , they can be separated in space using a stern - gerlach scheme . our model applies to any such possible configuration . therefore , the general conclusions of this study , concerning the correlations , number - squeezing and entanglement , are valid for recent experiments @xcite in the regime , where the depletion of the source bec is low . we derive the bogoliubov equations governing the dynamics of pair formation , and applying the bloch - messiah reduction @xcite , we write the state in terms of pairs of independently squeezed modes . we calculate the density and the number of scattered atoms , and the two body correlation between them . we demonstrate the presence of ideal number - squeezing between the opposite regions , violation of the cauchy - schwarz inequality and , using the fisher information criterion known from quantum metrology , we show that the atoms from the twin - beam system are entangled . we also provide a simple yet useful lower bound for the fisher information in terms of the average number of scattered atoms , and the number of modes they occupy . finally , we apply the above formalism to the twin - beam experiment of @xcite . this paper is organized as follows . in sec . [ universal ] we discuss the general properties of the solutions of the bogoliubov equations . these observation allow to easily calculate the density and the second order coherence of the system in sec . [ density ] and the fluctuations of the population imbalance between the opposite regions in sec . [ squeezing ] . in sec . [ sec_csi ] we take the first step towards the demonstration of particle entanglement present in the system , by showing that the second order correlation function violates the cauchy - schwarz inequality . in sec . [ entanglement ] we demonstrate that the scattered atoms are usefully entangled from the metrological point of view . in sec . [ outline ] we briefly describe the experimental setup of @xcite and in [ theory ] derive the corresponding effective bogoliubov equations . finally , in sec . [ results ] we review the most relevant properties of the twin - beam system by showing the results of the numerical simulation . we conclude in sec . [ conclusions ] . we first present the general properties of the solution of the bogoliubov equation , in cases where particles scatter pair - wise into well - separated regions . our theoretical description of the pair - production process starts with a many - body hamiltonian with contact two - body interactions [ ham ] @xmath1 here @xmath2 is an external trapping potential and @xmath3 is the strength of the two - body interactions , @xmath4 is the scattering length , @xmath5 is the atomic mass and the field operator @xmath6 satisfies the bosonic commutation relations . to derive the bogoliubov equation , we first find the c - number ( mean field ) wave function of the bec using the gross - pitaevskii equation ( gpe ) @xmath7 we then write the field operator as a sum of the c - number part and the bogoliubov correction , @xmath8 and insert this expression into ( [ ham ] ) . by keeping only the terms up to quadratic in @xmath9 we obtain the bogoliubov hamiltonian @xmath10 the resulting bogoliubov equation of motion is linear @xmath11 usually , a numerical solution of this equation is found in a following way . the field operator is expanded in a basis of wave - functions @xmath12 which match the geometry of the scattering problem @xmath13 this expression is inserted into eq . ( [ bog ] ) , the resulting equation is multiplied by @xmath14 and the outcome is integrated by sides over the whole space . in effect , what we obtain is an equation of motion , which , through the matrices @xmath15 and @xmath16 , couples the evolution of the @xmath17-th operator @xmath18 , with ( in general ) all others operators @xmath19 this equation is linear a consequence of the linearity of the bogoliubov equation ( [ bog ] ) so the general solution of ( [ eq_mot ] ) reads @xmath20 where the matrices @xmath21 and @xmath22 satisfy @xmath23 and @xmath24 . later , we will apply this method to solve the bogoliubov dynamics of the twin - beam production . however , we will show in the following , that in cases where the detailed form of the hamiltonian ( [ bog_ham ] ) drives the scattering of atomic pairs into opposite regions ( as indeed happens in twin - beam experiments ) , the basic properties of the system can be deduced analytically if an appropriate set of mode functions @xmath25 is chosen . let us denote the two separate regions into which the particles are scattered by l ( left ) and r ( right ) . particles populate l and r in a process of elastic scattering , so the regions are usually separated in momentum space . from this point of view , it is convenient to switch to the space of wave - vectors @xmath26 and decompose the field operator as follows @xmath27 the operators @xmath28 annihilate a particle in a mode characterized by the time - dependent wave function @xmath29 , which is localized in the right / left region in momentum space . we underline , that this kind of separation is also present in position space after expansion of the cloud , or after application of a stern - gerlach pulse in internal - state experiments , respectively . moreover , the vector @xmath26 might denote the quasi - momentum , if the scattering takes place in an optical lattice . formally , the only difference between the formulation ( [ op_nondiag ] ) and ( [ op_diag ] ) is the splitting of the field operator into the r and l modes . however , for a linear equation of motion such as ( [ bog ] ) , there exists a unique basis of mode functions for which the evolution equations of the mode pairs decouple from each other : [ pair_evolution ] @xmath30 where @xmath31 . this form of the bogoliubov equation has a simple physical interpretation : atoms scatter pair - wise into opposite regions , and the total field operator ( [ op_diag ] ) is a sum of independent mode pairs , which are squeezed in their relative population fluctuations , as will be explained in detail below . although the diagonal form ( [ pair_evolution ] ) is much clearer than ( [ nondiag ] ) , it is not obvious at the moment how this particular basis ( [ op_diag ] ) can be found . this is done in two steps , applying the procedure of the bloch - messiah reduction @xcite . first , using equations ( [ op_diag ] ) and ( [ pair_evolution ] ) , we evaluate the one - body density matrix ( first - order correlation function ) and obtain @xmath32 where @xmath33 . note that @xmath34 is the average occupation of the @xmath35-th eigen - mode , and that a pair of modes @xmath36 and @xmath37 is degenerate ( has the same eigen - value @xmath34 ) due to the assumed symmetry between the left and the right region . since we are using the heisenberg picture , the average value in eq . ( [ den_mat ] ) and all equations that follow are calculated in the initial vacuum state of scattered atoms . in any practical approach , if the basis ( [ op_diag ] ) is not known a priori , this step first requires a numerical evaluation of the density matrix ( [ den_mat ] ) in any convenient basis ( [ op_nondiag ] ) , and subsequent diagonalization . once this is done , then according to eq . ( [ den_mat ] ) , the basis functions @xmath29 are the momentary eigen - functions of the one - body density matrix ( natural orbitals ) . however , a second step is necessary to fully determine the functions @xmath29 , because the density matrix contrary to the field operator ( [ op_diag ] ) is insensitive to global phases of the mode functions . to retrieve this additional information , we calculate the anomalous density @xmath38 multiply it by sides with the eigen - functions of the density matrix and integrate over space . as a result , we retrive the information about the phases and obtain the full form of the mode functions @xmath29 of the diagonal basis . to summarize , we have outlined the structure of the solution of the bogoliubov equation for cases where atoms are scattered into two opposite regions . we will now show that the extra step , which is the transition from the `` numerical approach '' ( [ op_nondiag ] ) to the diagonal basis ( [ op_diag ] ) , allows to easily determine the basic properties of the system of scattered atoms , like its density or higher correlation functions . the simplest observable characterizing the pair - production process is the density @xmath39 which is , consistently with our derivation , localized in the two opposite regions . by integrating the above function over space , we obtain the information about the expected number of scattered atoms as a function of time @xmath40 additional information about the system is carried by the correlations between the scattered particles . the probability of simultaneous detection of two atoms at momenta @xmath41 and @xmath42 can be obtained from the normalized second - order correlation function @xmath43 according to the wick s theorem , this function can be written in terms of the one - body density matrix ( [ den_mat ] ) and the anomalous density ( [ anom ] ) as follows @xmath44 the transition from eq . ( [ g2 ] ) to ( [ g2_w ] ) might seem an unnecessary complication , however we will argue that it allows for a simple and intuitive interpretation of the second - order correlation function . according to eq . ( [ den_mat ] ) , the density matrix is non - vanishing only when @xmath41 and @xmath42 are both either in the right or left region , so @xmath45 governs the hanbury - brown and twiss ( hbt ) type of local correlations . on the other hand , as can be seen from eq . ( [ anom ] ) , the anomalous density is non - zero only when @xmath41 and @xmath42 are in the opposite regions , so it describes the cross - correlations between the two members of the scattered pair . clearly , this simple interpretation of the second order correlation function as a sum of local- and opposite - momentum correlations would have been much more difficult if we had not applied the diagonalization procedure and the wick s theorem . another property characterizing the scattering process are the fluctuations of the population imbalance between the two regions . if these fluctuations are suppressed below the properly defined shot - noise level , the system is _ number squeezed _ , which proves that the atoms scatter in pairs rather then independently to the left and to the right region . a quantitative description of the number squeezing involves the left and right atom number operators defined as the integrals of the density operators over the corresponding volumes , i.e. @xmath46 the population imbalance operator is then simply defined as @xmath47 and using eq . ( [ op_diag ] ) we obtain @xmath48 the number squeezing factor is defined as @xmath49 where @xmath50 is the variance of the population imbalance operator . if the fluctuations between the two regions are suppressed below the shot - noise level defined as @xmath51 , the system is called `` number - squeezed '' . in our case , since @xmath52 does not depend on time and the initial state is a vacuum , we obtain that @xmath53 . therefore , the two - region bogoliubov system is perfectly number - squeezed , as anticipated in the previous section . the ideal number - squeezing is a result of clear separation of the two scattering regions . in such case , it is natural to define the local atom - number operators ( [ at_num ] ) and the population imbalance operator ( [ pop_imb ] ) . it is important to note that not all systems , where particles are scattered in pairs are perfectly number squeezed . for instance , when two bose - einstein condensates collide , they produce a halo of atoms due to two - body elastic scattering into the initially unoccupied modes @xcite . in this system however , there is no simple way to define two separate regions . one can instead measure the number of atoms in two bins lying on the opposite sides of the halo . moderate number - squeezing of the atom number difference between these bins has been observed experimentally @xcite , but it is impossible to reach the limit @xmath54 @xcite . in contrast , the twin matter wave configurations @xcite , are ideal sources of correlated atomic pairs occupying two well - defined areas . apart from the number squeezing , the twin - region system can be characterized by another expression , which is called the cauchy - schwarz inequality . it relates the strength of the local and opposite correlations to witness the pair - scattering process . following @xcite , we define averaged second - order correlations as @xmath55 where @xmath56 . in the symmetric case , the cauchy - schwarz inequality @xmath57 can now be re - written as @xmath58 using expressions ( [ den_mat ] ) and ( [ anom ] ) we obtain @xmath59 thus the cauchy - schwarz inequality reads @xmath60 which is true only for all @xmath61 . as soon as particles start to scatter into the two regions , the cauchy - schwarz inequality is clearly violated . to quantify the degree of violation , a coefficient @xmath62 was introduced in @xcite , which reads @xmath63 when @xmath64 the system is in the `` classical '' regime , while @xmath65 signify correlations which are stronger than allowed by the classical physics . in our case this coefficient reads @xmath66 clearly , always @xmath65 , because it is a sum of unity and a non - negative part . for high mode populations @xmath34 , the second term , which is inversely proportional to the number of scattered particles tends to zero , restoring the classical limit nevertheless , as demonstrated with photons in @xcite , the confidence by which the cauchy - schwarz inequality can be violated in the presence of classical noise still _ increases _ with more strongly populated modes . however , it is the fisher information , which is the quantity highly sensitive to particle entanglement in the high - gain regime , as we show in the following section . this measure quantifies the potential for sub - shot - noise interferometry , and increases with rising mode population , in spite of the decreasing `` granularity '' of the matter wave @xcite that leads to all second - order correlation functions approaching equal values . we now show that atoms occupying the two regions are entangled , and could be used as an input of a quantum interferometer operating below the shot - noise level . we first recall how the precision of the phase estimation is related to the entanglement of input states using as an example the standard two - mode mach - zehnder interferometer ( mzi ) . then , we extend these concepts to the case , where the interferometer operates between two regions , each having a multi - mode structure determined by the bogoliubov equations . when speaking about two - mode interferometers , it is convenient to introduce a set of three operators [ ang ] @xmath67 which obey the same commutation relations as the angular momentum operators . the mzi , which is an interferometric device , where the imprint of the phase @xmath68 onto the input state is preceded and followed by a pair of symmetric beam - splitters , can be represented by a unitary evolution operator @xmath69 . if the phase is estimated in a series of @xmath70 measurements performed on the output state , the precision of the phase estimation is limited by the cramer - rao lower bound ( crlb ) @xcite , @xmath71 here , @xmath72 is the quantum fisher information ( qfi ) , which is related to the unitary transformation @xmath73 . for pure states transformed by the mzi it is equal to @xmath74 , where the variance is calculated in the input state of the interferometer @xcite . the crlb states , that if @xmath68 is determined using any possible type of measurement and estimator , then the precision @xmath75 is bounded as in eq . ( [ crlb ] ) . apart from providing a lower bound for the error of the phase estimation , the @xmath76 is an entanglement measure . namely , when the input state has an average number of @xmath77 particles , then if @xmath78 , the state is particle - entangled @xcite . we now show , that a natural extension of the two - mode picture allows to employ the concept of the qfi as an entanglement measure also in our multi - mode system of interest . to this end , we introduce the following analog of the two - mode angular momentum operators ( [ ang ] ) , [ ang_twin ] @xmath79 where we dropped the explicit time - dependence of the @xmath80 to simplify the notation . also , for simplicity , we choose the well - separated regions r and l to be localized symmetrically on the opposite sites of @xmath81 . the construction of these operators , which satisfy the same commutation relations as ( [ ang ] ) , is based on the analogy between the two - mode systems and the twin - beam configuration . in the former case , the operators connect the right and left modes , while in the latter the left and right sub - spaces . such a definition ( [ ang_twin ] ) is meaningful only in situations , where the system consists of two well - separated sub - systems . using the decomposition of the field operator into the set of independent modes , equations ( [ op_diag ] ) and ( [ pair_evolution ] ) , the above integrals yield , that each angular momentum operator is a sum of operators acting on each mode independently , that is [ ang2 ] @xmath82 these expressions show again that it is natural to describe the two - region system using the diagonal basis ( [ op_diag ] ) . in this language , the angular momentum operators are simply a sum of operators acting on each pair of modes independently , which vastly simplifies the further analysis . to establish a direct relation between the two - mode and two - region case , we now assume that the system is transformed in the multi - mode analog of the mach - zehnder interferometer . as outlined above , to demonstrate the presence of useful entanglement between the atoms in the left and in the right , it is necessary to calculate the qfi . using eq . ( [ ang2 ] ) we obtain that @xmath83 since the construction of the basis ( [ op_diag ] ) explicitly assumes that each mode is independent from all others , the second term in the last equality is @xmath84 , because the symmetry between the @xmath85 and @xmath86 regions implies that @xmath87 for all @xmath35 . therefore we obtain that the qfi is equal to @xmath88 where the last equality comes directly from the substitution of ( [ pair_evolution ] ) into the definition of the @xmath89 operator . also , we used @xmath90 , according to eq . ( [ nscat ] ) . clearly @xmath78 , so the system is entangled . moreover , one can refer the qfi to the ultimate bound for the precision of the parameter estimation , which is the heisenberg limit . for a system with fluctuating number of particles , this upper bound is equal to @xmath91 . using ( [ pair_evolution ] ) again , we obtain , that @xcite @xmath92 for a large number of scattered particles , when @xmath93 , we obtain @xmath94 . the value of the qfi , which is only one - half smaller than the heisenberg limit is a clear indication of very strong entanglement present in the system in the high - gain regime . at intermediate times , @xmath95 due to mode competition , which has a negative impact on the entanglement as witnessed by the qfi @xcite . to picture this , consider a `` frustrated case '' , where all atoms scatter uniformly into @xmath96 pairs of modes , so that all @xmath97 are equal . in this case , the number of scattered atoms is simply @xmath98 , and the qfi is @xmath99 . the qfi normalized to the snl is @xmath100 when , on average , there is less than a particle per a set of modes , i.e. @xmath101 , the qfi surpasses the snl only by a factor of 2 , a natural reminiscence of atoms being scattered in pairs . equation ( [ snl ] ) is a simple yet intuitive estimation of the lower bound of useful entanglement in terms of the number of scattered atoms and occupied modes . we now apply the above formalism to the twin - beam system of @xcite . first , we describe the physical mechanism which leads to the creation of the two correlated beams . as shown below , some basic information about the dynamics of the pair production allow to construct a simple one - dimensional bogoliubov model , which can be easily solved numerically . the experimental sequence applied in @xcite to produce correlated atom pairs was following . first , an almost pure bose - einstein condensate ( bec ) of @xmath102 @xmath103rb atoms with scattering length equal to @xmath104 nm was created at temperature @xmath105nk . the cloud was trapped in an approximately harmonic potential @xmath106 where atomic mass is equal to @xmath107 , and the frequency @xmath108 is much smaller than @xmath109 and @xmath110 , so the bec is strongly elongated along the @xmath111-axis . after the bec was created , the trapping potential was shaken in a controlled way , so the atoms were transferred to the first excited state along the @xmath112-direction . in order to achieve the maximal transfer efficiency , the shaking was optimized using quantum optimal control theory @xcite . afterwards , binary collisions transfered particle pairs to the ground state of the potential , and the excess energy @xmath113 was converted into back - to - back movement of the two atoms along @xmath111 . momentum conservation ensured , that their wave vectors had equal lengths @xmath114 and point in opposite directions . small corrections to the value of @xmath115 may arise from an effective mean - field potential , as will be discussed below . neglecting thermal phase fluctuations along the elongated direction @xmath111 , which is valid at very low temperatures only @xcite , the condensate wave function acting as a source for the pair - production can be found by solving the stationary gpe @xmath116 where @xmath117 is the chemical potential . this function can be evaluated numerically , by referring to the description of the experiment from the previous section , and noting that after the shaking of the trap , the bec is in the first excited state @xmath118 along the @xmath112 axis and in the ground state @xmath119 along @xmath120 . however , this can be approximated by an analytical expression , as argued below . first note , that since the characteristic energies @xmath121 and @xmath122 are large , and the number of atoms in the bec is small , the non - linear term can be safely neglected in evaluation of the eigenstates along @xmath112 and @xmath120 . as a result , assuming that the total wave - function @xmath123 separates in three directions ( which has been confirmed numerically ) , we obtain @xmath124 where the functions @xmath125 and @xmath126 are the eigen - states of the one - dimensional harmonic potential in @xmath112 and @xmath120 correspondingly . the function @xmath127 is found by inserting the above expression into eq . ( [ stat_gpe ] ) and integrating out the orthogonal directions . as a result , we obtain an effective equation @xmath128 where zero - point energy equals @xmath129 and the non - linearity reads @xmath130\times \left[\int\!\ ! \mathrm{d}z\ , \left(\psi^{\rm ( ho)}_{n_z=0}(z)\right)^4\right]\nonumber\\ & = & \frac{3}{8\pi}\frac g{a_{\mathrm{ho},y}\ , a_{\mathrm{ho},z}}.\end{aligned}\ ] ] here @xmath131 are the harmonic oscillator lengths for @xmath132 . since the trap is shallow in the @xmath111 direction , the solution of the stationary gpe ( [ gpex ] ) can be well approximated by the thomas - fermi ( tf ) formula @xcite @xmath133 where the effective chemical potential is @xmath134 leading to a tf radius of @xmath135 . within the approximation of neglecting thermal phase fluctuations , we have fully determined the wave - function of the bec , which we insert into the bogoliubov hamiltonian ( [ bog_ham ] ) . next , we expand the field operator @xmath136 in an orthonormal basis . along the @xmath112 and @xmath120 directions , it is natural to use the eigen - states of the harmonic oscillator as the basis functions , since it matches the geometry of the source bec . along the @xmath111 direction , we use a plane - wave basis , and get @xmath137 since the atom pairs are emitted into the ground state along @xmath112 only ( which is ensured by the anisotropy and anharmonicity of the potential ) , the sum over the eigen - states can be safely truncated at @xmath138 and @xmath119 . this reduces the dynamics of the pair - production to one - dimensional problem along the @xmath111 axis , with the orthogonal directions frozen out , i.e. @xmath139 where @xmath86 is the quantization volume . we insert this field operator into eq . ( [ bog_ham ] ) , evaluate the spatial integrals and upon the change of variables @xmath140 obtain @xmath141 where @xmath142 . we solve the resulting bogoliubov equation numerically , using a fourth order runge - kutta method , and find the matrices @xmath21 and @xmath22 as defined in eq . ( [ nondiag ] ) . using the above hamiltonian , one can also analytically determine @xmath115 , i.e. the position of the central peak . to this end , we employ a two - mode approximation by replacing the function @xmath143 with a dirac delta , and obtain the bogoliubov equation @xmath144 where @xmath115 is shifted with respect to the harmonic excitation energy due to the mean - field repulsion and reads @xmath145 this result is in good agreement with the experimentally measured position of the peak density , i.e. @xmath146 . in this section , we display the most important characteristics of the twin - beam system , starting from the solution of the eigen - problem of the density matrix ( [ den_mat ] ) . in fig . [ fig_vals ] we plot the first four eigen - values of the density matrix , as a function of time . the inset shows the total number of scattered atoms @xmath77 normalized to the occupation of the bec , as a function of time . the bogoliubov approximation is valid for as long as @xmath147 , so we interrupt the simulation at @xmath148ms , when @xmath149 . for longer times , when the depletion of the bec can not be neglected , a atom - number conserving method , such as the one introduced in @xcite must be used . ms , when @xmath150 . ] in fig . [ fig_vecs ] we plot the first four eigen - vectors of @xmath151 localized in the right half - space , i.e. @xmath152 with @xmath153 , calculated at an early time @xmath154ms and at @xmath148ms . due to the time - energy uncertainty relation , the eigen - modes localize around @xmath155 at later times . with @xmath153 . the solid black lines are results of diagonalization of the density - matrix at @xmath154ms while the dashed red lines at @xmath148ms . the figure shows how due to the time - energy uncertainty relation , the eigen - vectors narrow in the course of time around the central wave - vector @xmath115 . ] this can be seen even more clearly , by plotting the density @xmath156 at these two instants , as shown in fig . [ fig_den ] ( dashed lines ) . at @xmath154ms , two broad beams start to form on top of the uniform density . later , at @xmath148ms , strongly localized peaks clearly dominate over the flat background . on top of these curves , we plot the normalized second - order correlation function as defined in eq . ( [ g2 ] ) , with one of the arguments set equal to the resonant wave - vector @xmath115 , i.e. @xmath157 . at @xmath154ms , the cross - correlation , which is governed by the anomalous density , is very large , i.e. @xmath158 . this is a characteristic property of the bogoliubov system in the low - occupation regime @xcite , and indicates strong violation of the cauchy - schwarz inequality ( [ csi_one_body ] ) . also , for this early time , the width of both @xmath159 peaks are much more narrow than the beam size . this is consistent with the results shown in fig . [ fig_vals ] , where at early times many eigen - mode pairs of the density matrix are almost equally occupied . at later times , when a single pair of modes start to become dominant , the width of the peak in @xmath159 and the system size approach each other . while this corresponds to beams that are single - mode with respect to their local one - body properties , the local averaged correlation function as introduced in eq . ( [ eq : summed_g2 ] ) reaches the limit of @xmath160 , exceeding the number fluctuations of a coherent state by a factor of two . for fixed @xmath161 ( solid lines , left @xmath112-axis ) , and density profiles @xmath162 ( dashed lines , right axis ) in momentum space . the results are calculated at @xmath154ms ( a ) and @xmath148ms ( b ) . at early times , many momentum modes are occupied and the width of @xmath159 is much smaller than the beam size . later , two distinct peaks emerge , which are almost single - mode . ] . contributions of @xmath151 and @xmath96 in eq . ( [ g2_w ] ) are shown in blue and red hues , respectively . the dashed line indicates the position of the cut shown in fig . [ fig_den](b ) . arrows indicate the axes of fig . [ fig_avgd ] . ] , as obtained in experiments . solid line : momentum - space peak near @xmath163 , shown along the difference coordinate @xmath164 as indicated by the arrow in fig . [ fig_k1k2 ] . dotted line : peak near @xmath165 , along the sum coordinate @xmath166 . dashed and dash - dotted lines : respective functions , taking into account the finite expansion time in time - of - flight momentum measurements . the grey shaded area is proportional to the normalization @xmath167 . ] the typical tool to capture local second - order correlations in experiments are collinearly integrated functions of the type @xmath168 where the integrals run over an appropriately chosen momentum region @xcite . for symmetric , non - local correlations , back - to - back integration of the type @xmath169 is used . the corresponding normalized functions for our system @xmath170 are shown in fig . [ fig_avgd ] at @xmath171 . for both local ( solid ) and non - local ( dotted ) functions , correlations peaks , which do not span the entire populated range ( grey area ) and follow a gaussian shape , are clearly present . in the next step we take towards future comparison with experiments , we present the results not in momentum space , but rather using real - space data calculated after some finite time @xmath172 of ballistic expansion . only in the limit of @xmath173 ( far field ) , the real - space data is equivalent to the initial momentum space distribution ( if the expanding clouds are sufficiently dilute , so that the mean - field repulsion can be safely neglected ) . in @xcite , the expansion time was @xmath174 , which was sufficient to resolve the twin - beam peaks . nevertheless , the system was not fully in the far - field regime yet , which may have some impact on the correlation functions . as shown in fig . [ fig_avgd ] , the finite expansion time affects the back - to - back peak at @xmath175 much more strongly than the collinear hbt peak , leading to smearing of the measured @xmath176 ( dash - dotted line ) over the entire size of the twin - beam packets . this observation is consistent with some previous results @xcite . intuitively , the broadening effect is related to the random position of scattering events along @xmath111 within the size of the initial cloud , which is non - vanishing with respect to the expanded size of the twin - beam peaks . on the other hand , the local correlation function @xmath177 ( dashed line ) remains largely unaffected . note that although at every instant of the evolution , the field operator @xmath80 can be written as a sum of independently squeezed modes , at very early times the division between the right and left modes is unjustified , because the two peaks are not yet fully separated . however , at @xmath154ms when the density distribution is broad , the number of scattered atoms is @xmath178 . therefore , the system at such early time is hardly accessible experimentally so the quantum state of much less then a single particle is not of interest . as soon as the two peaks are well - formed , at @xmath179ms , with @xmath180 scattered atoms , all the general considerations from sec . [ general ] apply . finally , in fig . [ fig_qfi ] we plot the qfi from eq . ( [ qfi ] ) as a function of time and normalized to the heisenberg limit , i.e. @xmath181 . instead of interrupting the simulation at @xmath182ms , where the scattered fraction of atoms becomes non - negligible and particle number conservation is strongly violated , we extend the calculation up to @xmath183ms , when the number of scattered atoms significantly exceeds 15% of @xmath184 . this is done solely to illustrate that , once the population of one of the modes dominates , @xmath185 , as argued in sec . [ entanglement ] . note that the dominance of a single mode pair at long times is also predicted by the number - conserving theory @xcite , justifying this proceeding . indeed , in the inset , we show the number of pairs of right / left modes which have an occupation bigger or equal to 10% of the largest mode . this approximately tells , how many modes are significantly occupied in the system . at early times , there are over 100 pairs of modes . at @xmath182ms there are still 5 significantly occupied pairs , and only around 4.2ms a single pair of modes starts to dominate . at the same time the qfi approaches its upper bound . . the horizontal grey dashed line denotes the best possible value for the bogoliubov system , which is achieved in a regime , where only a single pair of left / right modes is relevant . the vertical grey dashed line denotes the time @xmath148ms , when the bogoliubov simulation should be interrupted . in the inset , we show the number of pairs of left / right modes that have at least 10% of occupation of the largest modes . we see that only around 4.2ms , the two - mode approximation is valid , as denoted by the horizontal dashed line . ] we have developed a simple bogoliubov model describing twin - atom beam experiments similar to refs . @xcite . due to the elongated geometry of the trapping potential , the dynamics is one - dimensional . as a consequence , the final step of our method can be easily solved numerically without the need for any stochastic method . furthermore , basic information about the scattered particles can directly be drawn from the general properties of the solution of the bogoliubov equations . in this way , we can quantitatively characterize the mode structure and correlation functions of the scattered atoms . also , quite generally , we can show that the population imbalance between the two beams is ideally squeezed and that the system is strongly entangled . these general observations can be applied to most recent experiments , where the atomic pairs scatter into two well - separated regions . finally , using the notion of the quantum fisher information , we have derived a simple lower bound for the useful entanglement of the system . this expression employs only the average number of scattered particles and the number of occupied modes . having understood the fundamental properties of few - mode twin beams , further steps can be made to take into account more specific issues of experimental implementations . a general feature of strongly elongated bose gases at realistic temperatures , such as the source cloud in @xcite , is the quasi - condensation @xcite , where the coherence along the @xmath111-axis is limited due to thermal phase fluctuations . although these fluctuations do not alter the general considerations of sec . [ general ] , they affect the emission dynamics @xcite , and also might have influence on the spatial properties of both the density and the correlation functions @xcite . in future , our method could be applied to the analysis of some more complicated schemes , basing on the twin - beam setup , such as the rarity - tapster - type experiments @xcite . also , according to our results , the two - region state could be used as an input to the mach - zehnder - like interferometer , similarly to @xcite . finally , note that the bogoliubov approximation neglects the secondary collisions between the scattered particles and the atoms from the source cloud . when a scattered atom propagates through the bec , the number of secondary collisions is bounded from above by @xmath186 , where @xmath187 is the peak density of the bec . by plugging in the experimental numbers , we obtain @xmath188 , which well justifies the use of the bogoliubov approximation j. ch . acknowledges the foundation for polish science international team programme co - financed by the eu european regional development fund . t.w . and p. sz . acknowledge the foundation for polish science international ph.d . projects programme co - financed by the eu european regional development fund . r.b . acknowledges support from the austrian science fund ( fwf ) projects cap ( 1607-n16 ) , and atom chip ( z118-n16 ) , and the fwf doctoral programme coqus ( w1210 ) . this research was supported by the national science center grants no . dec-2011/03/d / st2/00200 and n202 167840 . paul g. kwiat , klaus mattle , harald weinfurter , anton zeilinger , alexander v. sergienko , and yanhua shih , phys . lett . * 75 * , 4337 ( 1995 ) r. horodecki , p. horodecki , m. horodecki , and k. horodecki , rev . phys . * 81 * , 865 ( 2009 ) charles h. bennett , gilles brassard , claude crpeau , richard jozsa , asher peres , and william k. wootters , phys . 70 * , 1895 ( 1993 ) charles h. bennett , gilles brassard , sandu popescu , benjamin schumacher , john a. smolin , and william k. wootters , phys . lett . * 76 * , 722 ( 1996 ) a. perrin , h. chang , v. krachmalnicoff , m. schellekens , d. boiron , a. aspect , c.i . westbrook , phys . lett . * 99 * , 150405 ( 2007 ) . v. krachmalnicoff , j .- c . jaskula , m. bonneau , v. leung , g. b. partridge , d. boiron , c. i. westbrook , p. deuar , p. zi , m. trippenbach , k. v. kheruntsyan , phys . lett . * 104 * , 150402 ( 2010 ) . jaskula , m. bonneau , g. b. partridge , v. krachmalnicoff , p. deuar , k. v. kheruntsyan , a. aspect , d. boiron , c. i. westbrook , phys . 105 * , 190402 ( 2010 ) k.v . kheruntsyan , j .- c . jaskula , p. deuar , m. bonneau , g.b . partridge , j. ruaudel , r. lopes , d. boiron , c.i . westbrook , phys . lett . * 108 * , 260401 ( 2012 ) . r. g. dall , l. j. byron , a. g. truscott , g. r. dennis , m. t. johnsson , and j. j. hope , phys . a * 79 * , 011601 ( 2009 ) d. pertot , b. gadway , and d. schneble , phys . * 104 * , 200402 ( 2010 ) wu rugway , s. s. hodgman , r. g. dall , m. t. johnsson , and a. g. truscott , phys . lett . * 107 * , 075301 ( 2011 ) r. bcker , j. grond , s. manz , t. berrada , t. betz , c. koller , u. hohenester , t. schumm , a. perrin and j. schmiedmayer , nat . phys . * 7 * , 608 ( 2011 ) b. lcke , m. scherer , j. kruse , l. pezz , f. deuretzbacher , p. hyllus , o. topic , j. peise , w. ertmer , j. arlt , l. santos , a. smerzi and c. klempt , science * 11 * , 773 ( 2011 ) j. m. vogels , k. xu , and w. ketterle , phys . lett . * 89 * , 020401 ( 2002 ) a. perrin , h. chang , v. krachmalnicoff , m. schellekens , d. boiron , a. aspect , and c. i. westbrook , phys . lett . * 99 * , 150405 ( 2007 ) r. bcker , t. berrada , s. van frank , j .- f . schaff , t. schumm , j. schmiedmayer , g. jger , j. grond and u. hohenester , j. phys . b * 46 * , 104012 ( 2013 ) d.s . petrov , g.v . shlyapnikov and j.t.m . walraven , phys . lett . * 87 * , 050404 ( 2001 ) f. dalfovo , s. giorgini , l. p. pitaevskii and s. stringari , rev . 71 * , 463 ( 1999 )
there is overwhelming evidence for an exotic , non - baryonic , component to the cosmic matter density . however , the detailed nature of this dark matter is so far unknown . a possibility is for the dm density to be set in a similar way as the baryonic density @xcite . an asymmetry between particle and antiparticle number density is first created in a dynamical process of baryogenesis @xcite . the symmetric component of the particle and antiparticle population is then efficiently annihilated into radiation degrees - of - freedom whose energy is redshifted away by the hubble expansion @xcite . the asymmetric component is left over , which forms the matter density observed today . various methods to produce an asymmetry in both the dark and visible sectors concurrently have been proposed . the focus of this paper will be a model proposed by petraki , trodden and volkas @xcite , in which the asymmetries in the dark and visible sectors are produced during a strong first order phase transition , in analogy with the electroweak baryogenesis mechanism @xcite . while we focus on a particular realisation , the possibility of observational gravitational waves in such models is generic . this is because they all rely on beyond - the - standard - model ( bsm ) chiral fermions undergoing cp violating interactions with the broken phase bubbles of an exotic non - abelian gauge group in order to generate the asymmetry @xcite . the presence of a non - minimal dark sector is also generic in such models . since @xcite appeared , much progress has been made on both the experimental and theoretical fronts , in physics relevant to the phenomenology of such models . on the experimental side , a sm - like higgs boson has been discovered and studied at the lhc @xcite . the same machine has led to improved limits being placed on @xmath0 bosons @xcite . planck has set improved limits on exotic radiation present during the cmb ( @xmath1 ) @xcite . ligo has detected gravitational waves @xcite and the lisa pathfinder has flown which returned promising results on the prospects for space - based gravitational wave interferometers @xcite . on the theoretical side , simulations of the sound waves produced in the thermal plasma during a first order phase transition have found these can be a strong source of gravitational waves , which improves the possibility of such a detection @xcite . production of gravitational waves by exotic phase transitions has been discussed in @xcite . a phase transition in a _ generative _ sector , producing an asymmetry which is then communicated to the dark and visible matter sectors , is one such possibility . unlike the electroweak sector , the exotic generative sector higgs mass , vev , gauge and fermionic couplings have not been observed . hence there is a greater range of possibilities as to the mass scale of such a sector and therefore also more possibilities for the phase transition temperature and peak gravitational wave frequency . the cp violation required in these scenarios occurs in the generative sector yukawa interactions and so edm constraints are easily satisfied @xcite . it is therefore timely to revisit the model proposed in @xcite . in sec . 2 we briefly review the model . in sec . 3 we present the main investigation of this paper , we calculate the strength of the phase transition , which allows us to find the stochastic gravitational wave background in terms of the underlying parameters of the model . in sec . 4 we investigate other relevant signals , i.e. @xmath2 at the lhc , constraints from halo ellipticity due to dm self interactions , @xmath1 and direct detection prospects . we then briefly discuss and conclude . we now give an overview of the model presented in @xcite . the symmetry structure of the model includes three new gauge groups and an anomalous global symmetry @xmath3 . the field content is summarised in table [ tab : fieldcontent ] . the model may be organised into three sectors . .the field content and charges of the model . the three right - handed neutrinos are introduced to cancel the cubic @xmath4 anomaly . the @xmath5 and @xmath6 symmetries are broken spontaneously at a high , @xmath7tev ) , scale . the @xmath8 symmetry can either remain exact or be broken spontaneously at a suitably low scale , to allow the @xmath9(gev ) scale dm to annihilate into dark photons . [ cols="^,<,^,^,^,^ " , ] a. the _ generative _ sector , which consists of chiral fermions , @xmath10 , @xmath11 , @xmath12 , and a generative sector higgs , @xmath13 . this higgs breaks the generative sector @xmath14 gauge group and generates mass for the sector s chiral fermions through yukawa terms of the form ( for one generation ) @xmath15 we require an even number of generations of @xmath16 fermions to be free of the witten anomaly @xcite . in our calculations below we always take the minimal choice of two generations of @xmath16 fermions corresponding to 16 degrees of freedom . the yukawa couplings of eq . ( [ eq : genyuk ] ) can be complex and after rotating to the mass basis , we find the choice of two generations gives us seven physical cp violating phases , so that the second sakharov condition is fulfilled . components a separate mass term with @xmath17 and @xmath18 in eq . ( [ eq : genyuk ] ) . majorana mass terms are prohibited by imposing the global @xmath3 symmetry . ] the fermions carry charge under an anomalous @xmath19 , which is violated by rapid @xmath14 sphalerons in the symmetric phase . cp violating interactions of the fermions with the bubble walls of broken @xmath14 phase during a first order phase transition can lead to an @xmath3 asymmetry forming outside the bubble walls . this asymmetry is swept into the expanding bubble in which , if the phase transition is strong enough , the @xmath14 sphalerons are suppressed and the asymmetry is frozen in . the generated asymmetry is transfered to the visible and dark sectors through a chain of yukawa interactions beginning with @xmath20 where the fermion @xmath21 ( scalar @xmath22 ) will communicate with the visible ( dark ) sector , as explained below . b. the _ visible _ sector consists of the sm together with vector - like fermions , @xmath23 and @xmath21 , neutral under the sm gauge group . these transfer the asymmetry , created in the @xmath16 s during the generative phase transition , into the sm via the interactions @xmath24 where @xmath25 ( @xmath26 ) is the sm lepton ( higgs ) doublet . c. the _ dark _ sector consists of the scalar @xmath22 together with vector - like fermions @xmath27 and @xmath28 which make up the dm . the asymmetry is transfered via the interaction @xmath29 the scalar @xmath22 retains a zero vev . the fermions @xmath27 and @xmath28 are charged under a local @xmath30 and may form atomic bound states if the symmetry remains unbroken or is mildly broken . the symmetric components of the @xmath27 and @xmath28 number densities are efficiently annihilated away into dark @xmath30 photons . the @xmath30 gauge symmetry may either remain unbroken , or be broken spontaneously at a suitably low scale , in order to still allow for efficient annihilation of the dark fermions which , assuming the asymmetry is shared efficiently have masses @xmath31 gev . the above review of the model suffices for the casual reader interested in the discussion below . more discussion is provided in @xcite for those interested in understanding the in - depth details of the model . in order to study the strength of the phase transition we must make some assumptions about the tree level potential . here we assume terms up to dimension six term in the tree level potential @xcite @xmath32 such a potential can correspond to the low energy effective theory if there is a heavier scalar to which @xmath13 couples sufficiently strongly @xcite . is relegated to appendix a , as its finite temperature perturbative expansion suffers pathological issues due to the large gauge couplings required in order to achieve a strong first order phase transition . ] taking @xmath33 and @xmath34 , the vev is given by @xmath35 and the field dependent higgs mass by @xmath36 the field dependent masses of the @xmath16 fermions due to the yukawa interactions of eq . ( [ eq : genyuk ] ) and the @xmath14 gauge bosons are given by @xmath37 where @xmath38 is the @xmath14 gauge coupling and we assume a universal , representative , yukawa coupling in our calculations for simplicity . in order to analyse the strength of the phase transition we use the finite - temperature effective potential , @xmath39 , at one loop order with the daisy correction @xcite . the details can be found in appendix b. the integrals which enter the effective potential are evaluated numerically . ( thick grey lines ) for the @xmath40 tree level potential of eq . ( [ eq : v6pot ] ) . the washout condition of eq . ( [ eq : washoutcondition ] ) is satisfied for @xmath41 . the red area does not lead to symmetry breaking as @xmath42 . the orange region is either meta - stable and does not lead to a thermal transition or it leads to a runaway transition . the green area returns a gravitational wave spectrum with a bbo signal to noise ratio ( snr ) @xmath43 . the snr for gravitational waves has been calculated with @xmath44 gev , while all other contours are independent of @xmath45 , see secs . [ sec : gw ] and [ sec : scaling ] for details . note we have used an optimistic value for the wall velocity , @xmath46 , in calculating the snr . the blue points are where we also find a lisa snr @xmath43 . these always occur close to runaway transitions in this model . hence lisa is only sensitive to a very small portion of the parameter space of this model . ] a first order phase transition is characterised by a barrier in the effective potential , between the symmetric and broken phase minima , when the two have the same energy , i.e. at the critical temperature @xmath47 . we denote the broken phase vev at the critical temperature as @xmath48 . sometime after @xmath47 is reached , bubbles of broken phase can begin to nucleate . in order to avoid washout of the asymmetry , the @xmath14 sphalerons have to be sufficiently suppressed in the broken phase . this requires a sphaleron energy @xmath49 where @xmath50 is the bubble nucleation temperature , @xmath51 is the field value inside the critical bubble , @xmath52 is an order @xmath53 number which depends on the couplings of the theory and the range given on the right hand side of the inequality is due to the uncertainty in the sphaleron transition rate prefactor @xcite . determining the @xmath14 sphaleron energy , i.e. calculating @xmath52 , is beyond the scope of this work . in the sm , @xmath54 @xcite . the addition to the sm of a higher dimensional higgs term , @xmath55 , decreases the sphaleron energy by only a few percent for @xmath56 gev @xcite . compared to the sm , the main difference to @xmath57 is therefore expected to come from @xmath38 . the washout avoidance condition becomes @xmath58 as we increase the gauge coupling @xmath38 we therefore require a slightly larger @xmath59 ratio than is usually the case . as is well known , thermal one loop corrections from the gauge bosons and fermions lead to a @xmath60 term in the effective potential , which makes the symmetric minimum the global one at high temperatures . as the temperature drops , with the potential of eq . ( [ eq : v6pot ] ) , the phase transition can be strongly first order and satisfy the washout avoidance condition if the tree level barrier is large enough . an additional effect comes from the cancellation between the @xmath60 term and the tree level @xmath61 term , which can lead to a large barrier forming between the symmetric and broken phase minima . a scan over the parameters of the model showing contours of @xmath59 is shown in fig . [ fig : phi6_scan ] . as can be seen from the figure , for small @xmath62 , the tree level barrier is important , while for large @xmath62 , the cancellation between the @xmath60 and @xmath61 terms becomes the dominant effect . also shown in the figure are regions of possible observable gravitational waves . details of the calculation are given in secs . [ sec : critbub][sec : scaling ] . the first step in our calculation is to determine the nucleation temperature . the probability of creating a bubble within the hubble volume , for @xmath63 tev , reaches unity at @xmath64 where @xmath65 is the three - dimensional euclidean action for an o(3)-symmetric bubble @xcite . the temperature when this occurs is defined as the nucleation temperature , @xmath50 , and @xmath51 is the field value at the centre of the bubble at @xmath50 . the action is given by @xmath66 evaluated for a @xmath67 path with @xmath68 . the first term in the action corresponds to the cost of the surface tension , the second term to the favourability in terms of energy in residing in the deeper , broken phase , minimum . the resulting equation of motion is given by @xmath69 with the boundary conditions @xmath70 we solve the equation of motion numerically . in practical terms , we start with some choice for the release point , @xmath71 , and then use an overshoot / undershoot method to determine the value @xmath71 which returns @xmath72 . one then substitutes the bubble profile , @xmath67 , one has found back into eq . ( [ eq : s3 ] ) , in order to determine @xmath73 . this allows us to calculate the contours of @xmath59 shown in fig . [ fig : phi6_scan ] . the dominant source of gravitational waves in this model comes from sound waves generated in the plasma by the bubbles . the frequency spectrum of gravitational waves , @xmath74 where @xmath75 is the energy density of gravitational waves and @xmath76 is the critical density , has been determined by fitting numerical simulations @xcite . ( an analytic approach has been pursued in @xcite . ) we give the key expressions below . an important quantity is related to how quickly the phase transition occurs @xmath77 phase transitions which take longer result in larger , fewer , and more powerful bubbles , these produce sound waves at lower frequencies and with stronger amplitudes , than the smaller and more common bubbles found in faster transitions . the amplitude of gravitational waves also depends on the vacuum energy density released by the phase transition , compared with the radiation density of the thermal plasma , @xmath78 where @xmath79 counts the effective number of relativistic degrees of freedom . given the field content of this model , @xmath80 when all species are relativistic and we have used this value throughout our calculations . using the method described to calculate the critical bubble , we not only find @xmath50 , but also @xmath81 and @xmath82 . we also need the fraction of vacuum energy converted into bulk motion of the fluid , which for wall velocities @xmath83 @xcite , is given by @xmath84 where @xmath85 the numerical results are approximated by @xcite @xmath86 where the spectral shape is @xmath87 with peak frequency @xmath88 the only missing ingredient required for the gravitational wave spectrum is the wall velocity , @xmath89 . this depends on the strength of the phase transition and the friction of the plasma on the expanding bubble wall . this quantity is notoriously difficult to determine @xcite . we will use an optimistic value by setting @xmath90 in our scan . this results in the largest gravitational wave amplitude possible while remaining consistent with electroweak - style baryogenesis , which requires subsonic wall velocities @xcite . the true @xmath91 will therefore be below our estimate ( or the phase transition will be too strong for baryogenesis ) . hence , our results should be seen as an optimistic estimate for the possibility of such a signal , until a detailed determination of @xmath89 can be performed for this scenario . in fig . [ fig : gw ] we illustrate the sensitivity of the spectra to @xmath89 . though we have not calculated the wall velocity , we can use a simple criterion to determine whether the phase transition is a runaway , in which @xmath92 . the phase transition is necessarily a runaway if the mean field potential , @xmath93+\frac{1}{2}\sum_{\rm fermions}\left[m_{f}^{2}(\phi)-m_{f}^{2}(0 ) \right ] \right),\ ] ] drops below zero for @xmath94 at @xmath50 @xcite . if this is the case , the friction on the wall in the limit @xmath92 , is insufficient to balance the pressure on the wall from the free energy difference between the symmetric and broken phases . clearly , for consistency , we are interested in non - runaway phase transitions . to produce gravitational waves observable at lisa and bbo requires relatively strong transitions . the detectability of the gravitational waves can be estimated by the signal to noise ratio @xcite @xmath95^{2 } df},\ ] ] where @xmath96 is the time of observation in years and @xmath97 is the sensitivity of the gravitational wave detector to stochastic backgrounds for one year of observation . sensitivity curves for lisa and bbo for @xmath98 year are shown in fig . [ fig : gw ] . using @xmath99 years we show parameter space with snr @xmath43 for bbo in fig . [ fig : phi6_scan ] . note due to the fermionic and gauge couplings even some phase transitions strong enough to enter the lisa sensitivity region can remain outside the runaway regime . however , only a small portion of the parameter space can be probed by lisa . another crucial issue for such searches will be the foreground of gravitational waves due to unresolved points sources @xcite . note that the only mass scales which enter the generative sector potential are @xmath45 and @xmath100 . particularly useful scaling relations therefore hold when considering the parameter space of the model . a rescaling @xmath101 leaves many important quantities invariant . in particular @xmath102 which leaves the ratio unchanged . furthermore , when performing the rescaling in eq . ( [ eq : rescale ] ) , the absolute value of the potential , e.g. at the barrier height , increases as @xmath103 , but the bubble size decreases as @xmath104 , which results in @xmath105 again leaving the ratio unchanged . ) , which comes from the temperature dependence of the hubble rate . ] we also find @xmath106 this can be understood from eq . ( [ eq : rescaling2 ] ) . the increase in the magnitude of @xmath107 is cancelled by the increase in nucleation temperature in eq . ( [ eq : alpha ] ) , leaving @xmath81 unchanged . in the symmetric phase plasma , which we take to be constant for simplicity . ] similarly , the increase in @xmath50 is exactly cancelled by the smaller @xmath73 gradient in eq . ( [ eq : beta ] ) , leaving @xmath82 unchanged . however , note the peak gravitational wave frequency , eq . ( [ eq : peakfreq ] ) , does change , due to its linear dependence on @xmath50 . these relations allow one to more efficiently scan the parameter space of the model , because the key quantities , @xmath81 , @xmath82 and @xmath50 can be determined in terms of a fixed values of @xmath45 and @xmath100 and then rescaled in a computationally efficient way . dilepton signal and limit from the atlas @xcite ( cms gives similar limits @xcite ) . the intrinsic width of the resonance is @xmath108 of @xmath109 for gauge couplings @xmath110 respectively . ] the model introduces a @xmath6 gauge group , which explains the presence of a global @xmath4 symmetry at low energy , crucial for dm - baryon symmetric models @xcite . the associated @xmath0 can be searched for at the lhc @xcite . the limit on narrow @xmath0 resonances decaying to @xmath112 , with @xmath113 , from a search using 13 tev atlas data is shown in fig . [ fig : atlas13 ] . relevant formulas for determining the cross section for @xmath0 production can be found e.g. in @xcite . we use the mstw 2008 parton distribution functions @xcite and an estimate for the higher order qcd corrections @xmath114 , as suggested by atlas for sequential standard model @xmath0 masses of 2 and 3 tev . from fig . [ fig : atlas13 ] we see that the @xmath4 breaking scale should be @xmath115 tev , where @xmath116 is the charge of the @xmath6 breaking higgs . we now turn to the dm sector . the phenomenology of dm sectors with gauged @xmath30 symmetries displays rich variations depending on the area of parameter space one is in . here we will consider relatively large dark fine structure constants , @xmath117 , and dm constituent masses min@xmath118\gtrsim 0.1 $ ] gev , for which the dm is sufficiently tightly bound to approach the collisionless limit . the binding energy is given by @xmath119 the most stringent constraints then come from halo ellipticity observations . for @xmath120 gev there is significant tension with observations even with relatively large binding energies , @xmath121 mev @xcite . in this regime the ionized fraction of dm is completely negligible and the hyperfine transition energy is too large for dissipative cooling to be effective @xcite . if future numerical studies confirm the stringent halo ellipticity constraints such a scenario will be ruled out . the complicated phenomenology of dissipative regions of parameter space is beyond the scope of this work @xcite . the constraint can be evaded if the @xmath30 symmetry is broken and we discuss this possibility further in sec . [ sec : dd ] . we now turn to independent constraints on the model , coming from bbn and the cmb , which were derived in @xcite . since then , planck has improved the limit on radiation at the cmb epoch , giving the effective number of neutrino species @xmath122 @xcite . from this and the sm value @xmath123 , we take @xmath124 as a conservative ( essentially @xmath125 ) limit on a bsm contribution . if @xmath30 is unbroken , the dark photons will contribute to @xmath1 at the cmb epoch . the contribution depends on the degrees - of - freedom of the dark photon ( two ) and the dark radiation temperature . this allows one to translate the limit on extra radiation into a bound on the number of effective degrees of freedom of the dark sector , at the temperature it decouples from the visible sector radiation bath , @xmath126 where @xmath127 are the effective visible sector degrees - of - freedom ( 110.25 when including the @xmath128 states ) . the dark sector has @xmath129 @xmath130 including ( excluding ) the contribution of @xmath22 . from which we learn that @xmath22 should be heavy enough to not contribute to @xmath131 or additional degrees of freedom that contribute to @xmath127 must exist . if the couplings between the sectors are not too small , the former possibility will typically be the case , as the decoupling will be driven by the temperature falling sufficiently below the @xmath22 mass making the communication between the sectors ( which proceeds through interactions with @xmath22 ) inefficient . furthermore , future improvements on @xmath1 , estimated to reach uncertainties of @xmath132 for a stage - iv cmb experiment @xcite , should be able to rule out the unbroken @xmath30 scenario independently of the halo ellipticity observations . in order for the above conclusions to hold , we require the kinetic mixing parameter between the sm @xmath133 and dm @xmath30 field strength tensors , @xmath134 , to be sufficiently small . otherwise interactions between the lightest dark matter particle min@xmath135 $ ] and the electrons can keep the sectors in equilibrium . here , for notational clarity , we will assume @xmath136 as the subsequent phenomenology does not depend on the mass ordering and we will also make the mild assumption @xmath137 . to keep the sectors unequilibriated , we then require @xcite @xmath138 this estimate is sufficiently accurate for our purposes and broadly consistent with bounds coming from detailed studies of the energy transfer between the visible and dark sectors in such models @xcite . we now update the direct detection constraints discuss in @xcite . the spin - independent dm - nucleon scattering cross section via @xmath0 exchange is given by @xmath139 where @xmath140 is the dm - nucleon reduced mass . the current limit from cresst - ii , for @xmath141 gev , requires @xmath142 @xcite . however , taking into account the limit on @xmath109 from the lhc , this cross section is constrained to be below the floor for coherent neutrino scattering on cawo@xmath143 , which for this @xmath144 , is @xmath145 @xcite . as the dm forms tightly bound atomic states , the em millicharge of the dm , induced by the kinetic mixing with the photon , is screened and the interaction becomes effectively contact type . the dm - proton scattering cross section becomes suppressed by the bohr radius of the dm bound state to the fourth power , @xmath146 , and can be approximated by @xcite @xmath147 this can be above the neutrino floor . however , note that large signals , which depend on small values of @xmath148 , are in greater tension with the halo ellipticity constraints . we now discuss the possibility of a broken @xmath30 symmetry . for simplicity , here we consider parameter space in which the dm is no longer in a bound state but made up of unbound @xmath149 and @xmath150 particles . is broken , but with a sufficiently light mediator for bound states to form was considered in @xcite . ] the dm with @xmath151 gev scale mass can still annihilate efficiently into dark photons with mass @xmath152 mev , which decay before bbn provided that @xmath153 . beam dump and supernova constraints can be evaded provided that @xmath154 mev ( though some areas of parameter space below this are still allowed @xcite ) . limits from the cmb @xcite do not apply if the dm antiparticle population is sufficiently depleted . the spin - independent dm - nucleon scattering cross section is then @xcite mev , where @xmath155 kev is the nuclear recoil and @xmath156 gev is the relevant target mass . for smaller @xmath157 the interaction becomes long range . ] @xmath158 we have checked that this approximately reproduces the limits presented in @xcite , in which a more careful treatment of the energy spectrum of nuclear recoils and the sensitivity of cresst - ii to low energy nuclear recoils is taken into account . the detection of gravitational waves has opened up a new avenue for understanding the universe . future space based gravitational wave observatories will have sensitivity to first order phase transitions occurring at @xmath159(tev ) . one interesting possibility is a signal from a strong electroweak phase transition associated with electroweak baryogenesis @xcite . here we have considered a more exotic possibility , in which both dark matter and baryon asymmetries are generated during a strong phase transition associated with the spontaneous breaking of a yet undiscovered gauge symmetry . such a scenario necessitates a non - minimal structure , following the requirements for generating an asymmetry and transferring it to the visible and dark sectors . it is interesting to note that the large couplings which are required for cp violation and to strengthen the phase transition help to provide sufficient friction to keep the bubble wall in the non - runaway regime . this can allow for relatively strong phase transitions with subsonic walls which are both detectable by future gravitational wave observatories and self - consistent in the framework of asymmetry generation via an electroweak baryogenesis type mechanism . the possibility of gravitational wave signatures is expected to be a generic feature of such models , as they all require a strong first order phase transition associated with the breaking of an exotic non - abelian gauge symmetry , together with chiral bsm fermions which give the cp violation and help to provide friction @xcite . the dark sector of this model also exhibits an interesting structure , which can contribute non - negligibly to @xmath1 or lead to dm - nucleon scattering . the case with an unbroken @xmath30 gauge symmetry is in tension with halo ellipticity constraints on dm self interactions . the scenario evades edm constraints as the cp violating interactions occur in the generative sector . it is encouraging that upcoming experiments are sensitive to such exotic , non - minimal , scenarios . while the theory space is too large to pursue every possibility , it may be that non - null experimental results will in the near future help us uncover unexpected new physics . of particular interest for the type of models considered here are space based gravitational wave observatories with sensitivity to frequencies in the range @xmath160 hz . the lower range will be probed by lisa and the higher range can be probed by follow - up - missions such as bbo and decigo @xcite . the greater sensitivity expected of second generation space based interferometers will also greatly improve the prospects of detecting such phase transitions . i would like to thank thomas konstandin , kalliopi petraki and graldine servant for useful discussion . i would also like to thank the organisers of the 2016 gravitational waves and cosmology workshop for an inspiring event . the simplest choice of tree level potential corresponds to a single field potential up to dimension four @xmath161 such a potential can arise if the cross couplings of the generative scalar to the other scalars in the theory are negligible . the minimum of the potential is given by the usual expression @xmath162 . the relevant field dependent masses are those of the generative sector higgs mass , @xmath163 the @xmath16 fermions due to the yukawa interactions of eq . ( [ eq : genyuk ] ) and the @xmath14 gauge bosons . here the barrier in the effective potential can be created by the one loop thermal term for the gauge bosons . ( if the gauge coupling is large enough a barrier can also be present at zero temperature . ) the phase transition is strongly first order if the gauge coupling is large enough and the scalar quartic @xmath164 is small enough @xcite . the fermions help delay and hence strengthen the phase transition @xcite . however , the large gauge couplings , @xmath165 for @xmath166 , required in order to obtain a strong first order phase transition with the tree level potential in eq . ( [ eq : v4pot ] ) , lead to pathological issues with the one - loop effective potential . the nucleation temperature becomes sensitive to the method used in implementing the daisy resummation . quantitatively , we find large @xmath167 differences in the nucleation temperature when we include the daisy resummation as a separate term as opposed to when we include the thermal masses in the thermal functions directly @xcite ( see appendix b ) . hence the quantitative results for this potential are somewhat unreliable and we therefore study the @xmath40 potential for which these issues are not present because strong first order phase transitions can then be achieved for smaller values of @xmath38 and larger values of @xmath168 . for completeness we include here the form of the effective potential we have used in our calculations . we have used the one - loop effective potential in the on - shell renormalization scheme . the one - loop @xmath169 contribution is given by @xmath170 - \frac{3}{2}\right ) + 2m_{i}^{2}(\phi)m_{i}^{2}(v)\bigg\ } , \label{eq : oneloop } \end{aligned}\ ] ] where @xmath171 for the @xmath13 , @xmath14 gauge bosons and two generations of the @xmath16 fermions , @xmath172 for bosons ( fermions ) and we have ignored the contribution of the goldstone bosons . the one - loop finite @xmath173 contribution is given by @xmath174\right ) dy . \label{eq : finitet}\ ] ] the thermal masses for @xmath13 and the @xmath14 gauge bosons are given by @xmath175 in order to take into account the resummation of the matsubara zero modes one can include the daisy term @xmath176^{3/2}\big\ } \label{eq : daisy}\ ] ] where the sum runs only over scalars and the longitudinal degrees of freedom of the vector bosons , i.e @xmath177 for @xmath13 and the @xmath14 gauge bosons . the effective potential is then given by the sum of the tree level potential and eqs . ( [ eq : oneloop ] ) , ( [ eq : finitet ] ) and ( [ eq : daisy ] ) . an alternative prescription is to substitute @xmath178 in eq . ( [ eq : finitet ] ) for scalars and longitudinal gauge bosons @xcite . the latter form has the advantage in that it correctly captures the decoupling of heavy degrees of freedom from the effective potential . however , it has the disadvantage in that it effectively resums all matsubara modes , not just the zero modes . hence for the @xmath179 potential , where the two alternative approaches return significantly different results ( see appendix a ) , we can no longer necessarily trust the quantitative predictions coming from the effective potential . h. davoudiasl and r. n. mohapatra , _ on relating the genesis of cosmic baryons and dark matter _ , http://dx.doi.org/10.1088/1367-2630/14/9/095011[_new j. phys . _ * 14 * ( 2012 ) 095011 ] , [ http://arxiv.org/abs/1203.1247 [ 1203.1247 ] ] . k. m. zurek , _ asymmetric dark matter : theories , signatures , and constraints _ , http://dx.doi.org/10.1016/j.physrep.2013.12.001[_phys . * 537 * ( 2014 ) 91121 ] , [ http://arxiv.org/abs/1308.0338 [ 1308.0338 ] ] . k. petraki , m. trodden and r. r. volkas , _ visible and dark matter from a first - order phase transition in a baryon - symmetric universe _ , http://dx.doi.org/10.1088/1475-7516/2012/02/044[_jcap_ * 1202 * ( 2012 ) 044 ] , [ http://arxiv.org/abs/1111.4786[1111.4786 ] ] . v. a. kuzmin , v. a. rubakov and m. e. shaposhnikov , _ on the anomalous electroweak baryon number nonconservation in the early universe _ , http://dx.doi.org/10.1016/0370-2693(85)91028-7[_phys . lett . _ * b155 * ( 1985 ) 36 ] . a. g. cohen , d. b. kaplan and a. e. nelson , _ progress in electroweak baryogenesis _ , http://dx.doi.org/10.1146/annurev.ns.43.120193.000331[_ann nucl . part . _ * 43 * ( 1993 ) 2770 ] , [ http://arxiv.org/abs/hep-ph/9302210[hep-ph/9302210 ] ] . j. shelton and k. m. zurek , _ darkogenesis : a baryon asymmetry from the dark matter sector _ , http://dx.doi.org/10.1103/physrevd.82.123512[_phys . _ * d82 * ( 2010 ) 123512 ] , [ http://arxiv.org/abs/1008.1997 [ 1008.1997 ] ] . b. dutta and j. kumar , _ asymmetric dark matter from hidden sector baryogenesis _ , http://dx.doi.org/10.1016/j.physletb.2011.04.036[_phys . lett . _ * b699 * ( 2011 ) 364367 ] , [ http://arxiv.org/abs/1012.1341 [ 1012.1341 ] ] . collaboration , s. chatrchyan et al . , _ observation of a new boson at a mass of 125 gev with the cms experiment at the lhc _ , http://dx.doi.org/10.1016/j.physletb.2012.08.021[_phys . lett . _ * b716 * ( 2012 ) 3061 ] , [ http://arxiv.org/abs/1207.7235 [ 1207.7235 ] ] . collaboration , g. aad et al . , _ observation of a new particle in the search for the standard model higgs boson with the atlas detector at the lhc _ , http://dx.doi.org/10.1016/j.physletb.2012.08.020[_phys . * b716 * ( 2012 ) 129 ] , [ http://arxiv.org/abs/1207.7214 [ 1207.7214 ] ] . collaboration , _ search for new high - mass resonances in the dilepton final state using proton - proton collisions at @xmath181 = 13 tev with the atlas detector _ , tech . atlas - conf-2016 - 045 , cern , geneva , aug , 2016 . collaboration , v. khachatryan et al . , _ search for narrow resonances in dilepton mass spectra in proton - proton collisions at @xmath181 = 13 tev and combination with 8 tev data _ , http://arxiv.org/abs/1609.05391[1609.05391 ] . collaboration , p. a. r. ade et al . , _ planck 2015 results . cosmological parameters _ , http://dx.doi.org/10.1051/0004-6361/201525830[_astron . astrophys . _ * 594 * ( 2016 ) a13 ] , [ http://arxiv.org/abs/1502.01589[1502.01589 ] ] . collaboration , b. p. abbott et al . , _ observation of gravitational waves from a binary black hole merger _ , http://dx.doi.org/10.1103/physrevlett.116.061102[_phys . _ * 116 * ( 2016 ) 061102 ] , [ http://arxiv.org/abs/1602.03837 [ 1602.03837 ] ] . collaboration , b. p. abbott et al . , _ gw151226 : observation of gravitational waves from a 22-solar - mass binary black hole coalescence _ , http://dx.doi.org/10.1103/physrevlett.116.241103[_phys . * 116 * ( 2016 ) 241103 ] , [ http://arxiv.org/abs/1606.04855 [ 1606.04855 ] ] . m. armano et al . , _ sub - femto - g free fall for space - based gravitational wave observatories : lisa pathfinder results _ , http://dx.doi.org/10.1103/physrevlett.116.231101[_phys . lett . _ * 116 * ( 2016 ) 231101 ] . m. hindmarsh , s. j. huber , k. rummukainen and d. j. weir , _ gravitational waves from the sound of a first order phase transition _ , http://dx.doi.org/10.1103/physrevlett.112.041301[_phys . * 112 * ( 2014 ) 041301 ] , [ http://arxiv.org/abs/1304.2433 [ 1304.2433 ] ] . m. hindmarsh , s. j. huber , k. rummukainen and d. j. weir , _ numerical simulations of acoustically generated gravitational waves at a first order phase transition _ , http://dx.doi.org/10.1103/physrevd.92.123009[_phys . rev . _ * d92 * ( 2015 ) 123009 ] , [ http://arxiv.org/abs/1504.03291 [ 1504.03291 ] ] . c. grojean and g. servant , _ gravitational waves from phase transitions at the electroweak scale and beyond _ , http://dx.doi.org/10.1103/physrevd.75.043507[_phys . * d75 * ( 2007 ) 043507 ] , [ http://arxiv.org/abs/hep-ph/0607107 [ hep - ph/0607107 ] ] . w. buchmller , v. domcke , k. kamada and k. schmitz , _ the gravitational wave spectrum from cosmological @xmath4 breaking _ , http://dx.doi.org/10.1088/1475-7516/2013/10/003[_jcap_ * 1310 * ( 2013 ) 003 ] , [ http://arxiv.org/abs/1305.3392[1305.3392 ] ] . p. schwaller , _ gravitational waves from a dark phase transition _ , http://dx.doi.org/10.1103/physrevlett.115.181101[_phys . * 115 * ( 2015 ) 181101 ] , [ http://arxiv.org/abs/1504.07263 [ 1504.07263 ] ] . j. jaeckel , v. v. khoze and m. spannowsky , _ hearing the signals of dark sectors with gravitational wave detectors _ , http://dx.doi.org/10.1103/physrevd.94.103519[_phys . rev . _ * d94 * ( 2016 ) 103519 ] , [ http://arxiv.org/abs/1602.03901 [ 1602.03901 ] ] . p. s. b. dev and a. mazumdar , _ probing the scale of new physics by advanced ligo / virgo _ , http://dx.doi.org/10.1103/physrevd.93.104001[_phys . rev . _ * d93 * ( 2016 ) 104001 ] , [ http://arxiv.org/abs/1602.04203 [ 1602.04203 ] ] . h. davoudiasl , p. p. giardino and c. zhang , _ higgs - like boson at 750 gev and genesis of baryons _ , http://dx.doi.org/10.1103/physrevd.94.015006[_phys . rev . _ * d94 * ( 2016 ) 015006 ] , [ http://arxiv.org/abs/1605.00037 [ 1605.00037 ] ] . a. katz and a. riotto , _ baryogenesis and gravitational waves from runaway bubble collisions _ , http://dx.doi.org/10.1088/1475-7516/2016/11/011[_jcap_ * 1611 * ( 2016 ) 011 ] , [ http://arxiv.org/abs/1608.00583[1608.00583 ] ] . c. balazs , a. fowlie , a. mazumdar and g. white , _ gravitational waves at aligo and vacuum stability with a scalar singlet extension of the standard model _ , http://dx.doi.org/10.1103/physrevd.95.043505[_phys . rev . _ * d95 * ( 2017 ) 043505 ] , [ http://arxiv.org/abs/1611.01617 [ 1611.01617 ] ] . c. a. baker et al . , _ an improved experimental limit on the electric dipole moment of the neutron _ , http://dx.doi.org/10.1103/physrevlett.97.131801[_phys _ * 97 * ( 2006 ) 131801 ] , [ http://arxiv.org/abs/hep-ex/0602020[hep-ex/0602020 ] ] . j. j. hudson , d. m. kara , i. j. smallman , b. e. sauer , m. r. tarbutt and e. a. hinds , _ improved measurement of the shape of the electron _ , http://dx.doi.org/10.1038/nature10104[_nature_ * 473 * ( 2011 ) 493496 ] . collaboration , j. baron et al . , _ order of magnitude smaller limit on the electric dipole moment of the electron _ , http://dx.doi.org/10.1126/science.1248213[_science_ * 343 * ( 2014 ) 269272 ] , [ http://arxiv.org/abs/1310.7534[1310.7534 ] ] . w. chao and m. j. ramsey - musolf , _ electroweak baryogenesis , electric dipole moments , and higgs diphoton decays _ , http://dx.doi.org/10.1007/jhep10(2014)180[_jhep_ * 10 * ( 2014 ) 180 ] , [ http://arxiv.org/abs/1406.0517[1406.0517 ] ] . c. grojean , g. servant and j. d. wells , _ first - order electroweak phase transition in the standard model with a low cutoff _ , http://dx.doi.org/10.1103/physrevd.71.036001[_phys . rev . _ * d71 * ( 2005 ) 036001 ] , [ http://arxiv.org/abs/hep-ph/0407019 [ hep - ph/0407019 ] ] . d. bodeker , l. fromme , s. j. huber and m. seniuch , _ the baryon asymmetry in the standard model with a low cut - off _ , http://dx.doi.org/10.1088/1126-6708/2005/02/026[_jhep_ * 02 * ( 2005 ) 026 ] , [ http://arxiv.org/abs/hep-ph/0412366 [ hep - ph/0412366 ] ] . c. delaunay , c. grojean and j. d. wells , _ dynamics of non - renormalizable electroweak symmetry breaking _ , http://dx.doi.org/10.1088/1126-6708/2008/04/029[_jhep_ * 04 * ( 2008 ) 029 ] , [ http://arxiv.org/abs/0711.2511[0711.2511 ] ] . p. h. damgaard , a. haarr , d. oconnell and a. tranberg , _ effective field theory and electroweak baryogenesis in the singlet - extended standard model _ , http://dx.doi.org/10.1007/jhep02(2016)107[_jhep_ * 02 * ( 2016 ) 107 ] , [ http://arxiv.org/abs/1512.01963[1512.01963 ] ] . m. quiros , _ finite temperature field theory and phase transitions _ , in _ proceedings , summer school in high - energy physics and cosmology : trieste , italy , june 29-july 17 , 1998 _ , pp . 187259 , 1999 . http://arxiv.org/abs/hep-ph/9901312[hep-ph/9901312 ] . l. carson , x. li , l. d. mclerran and r .- wang , _ exact computation of the small fluctuation determinant around a sphaleron _ , http://dx.doi.org/10.1103/physrevd.42.2127[_phys . rev . _ * d42 * ( 1990 ) 21272143 ] . j. r. espinosa , t. konstandin , j. m. no and g. servant , _ energy budget of cosmological first - order phase transitions _ , http://dx.doi.org/10.1088/1475-7516/2010/06/028[_jcap_ * 1006 * ( 2010 ) 028 ] , [ http://arxiv.org/abs/1004.4187[1004.4187 ] ] . c. caprini et al . , _ science with the space - based interferometer elisa . ii : gravitational waves from cosmological phase transitions _ , http://dx.doi.org/10.1088/1475-7516/2016/04/001[_jcap_ * 1604 * ( 2016 ) 001 ] , [ http://arxiv.org/abs/1512.06239[1512.06239 ] ] . e. thrane and j. d. romano , _ sensitivity curves for searches for gravitational - wave backgrounds _ , http://dx.doi.org/10.1103/physrevd.88.124032[_phys . rev . _ * d88 * ( 2013 ) 124032 ] , [ http://arxiv.org/abs/1310.5300 [ 1310.5300 ] ] . g. d. moore and t. prokopec , _ how fast can the wall move ? a study of the electroweak phase transition dynamics _ , http://dx.doi.org/10.1103/physrevd.52.7182[_phys . rev . _ * d52 * ( 1995 ) 71827204 ] , [ http://arxiv.org/abs/hep-ph/9506475 [ hep - ph/9506475 ] ] . a. megevand and a. d. sanchez , _ velocity of electroweak bubble walls _ , http://dx.doi.org/10.1016/j.nuclphysb.2009.09.019[_nucl . _ * b825 * ( 2010 ) 151176 ] , [ http://arxiv.org/abs/0908.3663 [ 0908.3663 ] ] . s. j. huber and m. sopena , _ the bubble wall velocity in the minimal supersymmetric light stop scenario _ , http://dx.doi.org/10.1103/physrevd.85.103507[_phys . rev . _ * d85 * ( 2012 ) 103507 ] , [ http://arxiv.org/abs/1112.1888 [ 1112.1888 ] ] . j. kozaczuk , s. profumo , l. s. haskins and c. l. wainwright , _ cosmological phase transitions and their properties in the nmssm _ , http://dx.doi.org/10.1007/jhep01(2015)144[_jhep_ * 01 * ( 2015 ) 144 ] , [ http://arxiv.org/abs/1407.4134[1407.4134 ] ] . j. kozaczuk , _ bubble expansion and the viability of singlet - driven electroweak baryogenesis _ , http://dx.doi.org/10.1007/jhep10(2015)135[_jhep_ * 10 * ( 2015 ) 135 ] , [ http://arxiv.org/abs/1506.04741[1506.04741 ] ] . j. m. no , _ large gravitational wave background signals in electroweak baryogenesis scenarios _ , http://dx.doi.org/10.1103/physrevd.84.124025[_phys . _ * d84 * ( 2011 ) 124025 ] , [ http://arxiv.org/abs/1103.2159 [ 1103.2159 ] ] . t. konstandin , g. nardini and m. quiros , _ gravitational backreaction effects on the holographic phase transition _ , http://dx.doi.org/10.1103/physrevd.82.083513[_phys . rev . _ * d82 * ( 2010 ) 083513 ] , [ http://arxiv.org/abs/1007.1468 [ 1007.1468 ] ] . t. konstandin and g. servant , _ cosmological consequences of nearly conformal dynamics at the tev scale _ , http://dx.doi.org/10.1088/1475-7516/2011/12/009[_jcap_ * 1112 * ( 2011 ) 009 ] , [ http://arxiv.org/abs/1104.4791[1104.4791 ] ] . t. konstandin and g. servant , _ natural cold baryogenesis from strongly interacting electroweak symmetry breaking _ , http://dx.doi.org/10.1088/1475-7516/2011/07/024[_jcap_ * 1107 * ( 2011 ) 024 ] , [ http://arxiv.org/abs/1104.4793[1104.4793 ] ] . a. j. farmer and e. s. phinney , _ the gravitational wave background from cosmological compact binaries _ , http://dx.doi.org/10.1111/j.1365-2966.2003.07176.x[_mon . not . * 346 * ( 2003 ) 1197 ] , [ http://arxiv.org/abs/astro-ph/0304393[astro-ph/0304393 ] ] . m. r. adams and n. j. cornish , _ detecting a stochastic gravitational wave background in the presence of a galactic foreground and instrument noise _ , http://dx.doi.org/10.1103/physrevd.89.022001[_phys . * d89 * ( 2014 ) 022001 ] , [ http://arxiv.org/abs/1307.4116 [ 1307.4116 ] ] . j. d. clarke , r. foot and r. r. volkas , _ quark - lepton symmetric model at the lhc _ , http://dx.doi.org/10.1103/physrevd.85.074012[_phys . rev . _ * d85 * ( 2012 ) 074012 ] , [ http://arxiv.org/abs/1112.3405 [ 1112.3405 ] ] . a. d. martin , w. j. stirling , r. s. thorne and g. watt , _ parton distributions for the lhc _ , http://dx.doi.org/10.1140/epjc/s10052-009-1072-5[_eur . phys . j. _ * c63 * ( 2009 ) 189285 ] , [ http://arxiv.org/abs/0901.0002 [ 0901.0002 ] ] . k. petraki , l. pearce and a. kusenko , _ self - interacting asymmetric dark matter coupled to a light massive dark photon _ , http://dx.doi.org/10.1088/1475-7516/2014/07/039[_jcap_ * 1407 * ( 2014 ) 039 ] , [ http://arxiv.org/abs/1403.1077[1403.1077 ] ] . k. k. boddy , m. kaplinghat , a. kwa and a. h. g. peter , _ hidden sector hydrogen as dark matter : small - scale structure formation predictions and the importance of hyperfine interactions _ , http://dx.doi.org/10.1103/physrevd.94.123017[_phys . rev . _ * d94 * ( 2016 ) 123017 ] , [ http://arxiv.org/abs/1609.03592 [ 1609.03592 ] ] . k. n. abazajian et al . , _ neutrino physics from the cosmic microwave background and large scale structure _ , http://dx.doi.org/10.1016/j.astropartphys.2014.05.014[_astropart . * 63 * ( 2015 ) 6680 ] , [ http://arxiv.org/abs/1309.5383 [ 1309.5383 ] ] . h. vogel and j. redondo , _ dark radiation constraints on minicharged particles in models with a hidden photon _ , http://dx.doi.org/10.1088/1475-7516/2014/02/029[_jcap_ * 1402 * ( 2014 ) 029 ] , [ http://arxiv.org/abs/1311.2600[1311.2600 ] ] . collaboration , g. angloher et al . , _ results on light dark matter particles with a low - threshold cresst - ii detector _ , http://dx.doi.org/10.1140/epjc/s10052-016-3877-3[_eur . phys . j. _ * c76 * ( 2016 ) 25 ] , [ http://arxiv.org/abs/1509.01515 [ 1509.01515 ] ] . a. gtlein et al . , _ impact of coherent neutrino nucleus scattering on direct dark matter searches based on cawo@xmath182 crystals _ , http://dx.doi.org/10.1016/j.astropartphys.2015.03.010[_astropart . * 69 * ( 2015 ) 4449 ] , [ http://arxiv.org/abs/1408.2357 [ 1408.2357 ] ] . c. delaunay , c. grojean and g. servant , _ the higgs in the sky : production of gravitational waves during a first - order phase transition _ , http://dx.doi.org/10.1063/1.2735127[_aip conf . * 903 * ( 2007 ) 2431 ] . f. p. huang , y. wan , d .- wang , y .- f . cai and x. zhang , _ hearing the echoes of electroweak baryogenesis with gravitational wave detectors _ , http://dx.doi.org/10.1103/physrevd.94.041702[_phys . rev . _ * d94 * ( 2016 ) 041702 ] , [ http://arxiv.org/abs/1601.01640 [ 1601.01640 ] ] . j. crowder and n. j. cornish , _ beyond lisa : exploring future gravitational wave missions _ , http://dx.doi.org/10.1103/physrevd.72.083005[_phys . * d72 * ( 2005 ) 083005 ] , [ http://arxiv.org/abs/gr-qc/0506015 [ gr - qc/0506015 ] ] . h. kudoh , a. taruya , t. hiramatsu and y. himemoto , _ detecting a gravitational - wave background with next - generation space interferometers _ , http://dx.doi.org/10.1103/physrevd.73.064006[_phys . rev . _ * d73 * ( 2006 ) 064006 ] , [ http://arxiv.org/abs/gr-qc/0511145 [ gr - qc/0511145 ] ] . m. carena , a. megevand , m. quiros and c. e. m. wagner , _ electroweak baryogenesis and new tev fermions _ , http://dx.doi.org/10.1016/j.nuclphysb.2005.03.025[_nucl . * b716 * ( 2005 ) 319351 ] , [ http://arxiv.org/abs/hep-ph/0410352 [ hep - ph/0410352 ] ] .
in the course of a nova eruption , a significant amount of material ( @xmath2 m@xmath3 ) is ejected , which can form a visible shell around the system . observation of this shell can lead to accurate distances to the novae via the expansion parallax method . for this reason , searches for nova shells have been undertaken by various groups ( e.g. @xcite , @xcite ) . with the distance in hand , one can then determine the luminosity of the system at maximum light , which can be used to calibrate the maximum magnitude - rate of decline relation . use of this relation can allow for determinations of the distance to objects such as the virgo cluster @xcite . when we embarked on a survey of nova remnants with small telescopes to measure nova shell line fluxes , we did not expect to make a contribution to the study of nova shell morphology and the nova luminosity calibration . nevertheless , the results ( including the discovery of 4 previously undetected shells ) , combined with data collected at larger earthbound telescopes and with _ hubble space telescope _ , encouraged us to attempt a derivation of nova distances and luminosities at maximum and minimum . we constructed maximum magnitude - rate of decline relations , based on the standard light curve parameters @xmath4 , @xmath5 and @xmath6 , and also studied the distribution of minimum magnitudes . the use of novae as distance indicators poses more problems than that of other methods based on variable stars . while novae are brighter than pulsating variables , their light curve characteristics require a monitoring with high time resolution ( a day or better ) , and the determination of the decline rate requires close monitoring until a given object declines by two or three magnitudes . the nova luminosity calibration might be done on a purely photometric basis , using ( unreddened ) novae in extragalactic systems ( m31 , magellanic clouds ) . however , we would need to tie the novae into the distance ladder by pulsating variables or binaries , thus introducing possible systematic errors of the distance scale . the other approach is the expansion parallax method , which combines the angular expansion rate ( observed several years or decades after outburst ) with radial velocity observations during outburst or at later stages . this approach is also open to systematic errors . well - observed novae give evidence that mass ejection is not spherical , and inclination effects play an important role . furthermore , the radial velocity of the shell of a given nova is often variable with time , even if we restrict ourselves to the observation of the `` principal system '' . high and low velocity material is present , the slower material sometimes shows signs of acceleration . similar problems arise in the determination of angular expansion rates , since the bright nebular structures are often immersed in a more extended fuzz . the more rapidly expanding shell becomes increasingly fainter ( a phenomenon which is also seen in the ` sharpening ' of the nebular emission line profiles in nova spectra in the course of the outburst ) , making the measurement of the shell size more difficult . if we restrict ourselves to the spectral emission line maxima and brightest features in the resolved shells , we still encounter a major problem , since these condensations are often arranged in an inclined `` equatorial ring '' . a combination of both methods is the derivation of the form of the maximum magnitude - rate of decline ( mmrd ) calibration from an extragalactic sample , while determining its zero point calibration using galactic novae . additional problems are the determination of the proper light maximum ( sometimes it is uncertain whether the observed maximum is the true maximum , especially when spectral observations in early stages are lacking ) and the amount of interstellar absorption ( most galactic novae are concentrated towards the galactic plane ) . galactic novae with a well - covered maximum ( to obtain the rate of decline ) , uv observations at the nebular or minimum stage ( to derive the reddening from line ratios or the 2200 band ) , and a well - resolved shell ( to derive the kinematic characteristics ) , are extremely rare . only a handful of novae fulfill these criteria , and a luminosity calibration based on small number statistics may suffer from systematic effects , such as the effect of white dwarf masses or chemical composition on the luminosity . thus we tried to use all novae for which expansion parallax data are available , by making at some points `` educated guesses '' or using `` brute averaging '' . this certainly puts a lot of arbitrariness in the calibration , but we hope that this fairly extensive data base , which will not be augmented considerably for a long time , gives a reasonable insight in the present state of the art of novae as distance calibrators , if one deals with available , and not ideal , samples . the observations were obtained on 1998 march 21 - 23 with dutch 0.9 m telescope at the eso la silla observatory , using a tek tk512cb chip ( @xmath7 pixels ) with a scale of @xmath8/pixel . further observations were obtained at the kitt peak national observatory on 1998 may 28 - june 1 with the 2.1 m telescope using the tek `` t1ka '' chip ( @xmath9 pixels with a scale of @xmath10/pixel ) and on 1998 june 30 with the 0.9 m telescope using the tek `` t2ka '' chip ( @xmath11 pixels with a scale of @xmath12/pixel ) . the data were obtained with narrow band filters centered at h@xmath0 ( 80 fwhm at eso , 36 at kpno ) and [ o iii ] @xmath15007 ( 55 at eso , 31 at kpno ) , as well as off - band and _ ubvr _ filters ; see table [ tbl-1 ] for details . the data were reduced in the standard manner . the search for shells was performed using two different techniques . first , a psf was generated for each image ( with the routines in daophot ) using , whenever possible , stars of comparable brightness to the nova . the psf was then subtracted from each star in the image , and the resultant image examined for evidence of extended emission ( i.e. residual flux ) around both the nova and nearby stars ( as a control sample ) . in almost every case , this technique was definitive . the second method was to perform a radial profile fit of the nova and several stars of comparable magnitude . the fits were compared to determine if the nova had a broader profile . when a shell was detected , a contour plot was generated from which the size of the shell was measured by determining the edge of the shell ( i.e. where the shell merged into the sky ) . for shell sizes obtained from the literature , we note in table [ tbl-2 ] the measurement technique ( direct image or radial profile ) used . radial profiles were quoted for only four objects . in two cases ( rw umi and v351 pup ) , the two techniques yielded consistent answers , while in the remaining two cases ( pw vul and qu vul ) , the radial profile values were considerably smaller than the direct image values . to maintain as consistent a dataset as possible , these discrepant radial profile values were not used in our analysis . we have detected shells around 10 objects , and in table [ tbl-2 ] we present the measured diameters from our data , as well as summarize the previous measurements from the literature . we have obtained data on another 5 objects in which we fail to detect a shell , but for which a shell has been reported in the literature or can be seen in wfpc2 data . since different measurement techniques can produce different results , we include in the table a code indicating the measurement technique used ( radial profile fit or direct image ) ; @xmath13 of the measurements are from direct images where the shell is clearly resolved . for the radial profile fits , a correction for the psf has been made , while for the direct images , not such correction was made . we have not corrected our measurements of direct images for geometric effects caused by the ellipticity of the shell ( see below ) . in the following sections 4 and 5 , we present the results of our new observations ( shell sizes , expansion rates ) , and derive distances and luminosities ; we include the supporting data necessary for these derivations ( i.e. interstellar extinction , light curve parameters ) . nebular expansion parallaxes are a fundamental method for determining the distances to novae if the three basic parameters age , expansion velocity , and size can be properly determined . while age can ( usually ) be determined accurately , size ( geometry ) and expansion velocity are less certain , and will dominate the uncertainty in the derived distance . to derive a luminosity furthermore requires an accurate light curve and a trustworthy reddening estimate . projected nova shells sometimes appear spherically symmetric , but in many cases are oval . furthermore , the densest gas clouds are often concentrated in a ring , which is inclined to the line of sight of the observer . thus , projection effects should be taken into account ( see @xcite ) . there is general agreement that if nova shells are non - spherical , they are prolate ellipsoids . from the shells of the slow novae t aur , dq her , bt mon , rr pic , and fh ser , whose inclinations can be estimated with some confidence , an average ellipticity @xmath14 is deduced ; if major diameters , or some average diameters are used ( as in our diameter and radial profile data ) , errors should on the average be within 10% of the true distance . other novae like gk per or v603 aql tend to have more spherical shells , or at least ` condensations ' arranged in a circular pattern . radial velocities show a very large dispersion , not only in the different `` spectral systems '' ( see , e.g. @xcite ) , but also for a given system . if a p cygni line profile is observed soon after outburst , the radial velocity of the absorption line is strongly influenced by the superimposed emission , but such radial velocities are the only information available for some objects . in the nebular stage , `` castellated '' emission line profiles are observed . they have broad wings that generally disappear at later phases . direct images of nova shells often show the associated condensations ( rings or blobs ) , surrounded by a faint fuzz , which originates in faster moving material ( as seen , e.g , deep images of nova shells in @xcite ) . for several objects in our survey , bright condensations are still close to the star and are easily resolved on hst images , while more extended faint fuzz shows up on images taken with earthbound telescopes . ideally , one should be able to associate maxima of emission lines with given condensations in the shell , i.e. one should have a clear idea about the geometry of the shell . however , the clumps most easily seen on direct images are those moving at right angles to the line of sight , and do not contribute to the width or `` castellation '' of the emission line . in most cases , spatial information of shell spectra , which is essential for reconstructing the shell geometry , is missing . if there was no deceleration of the expanding shell , then measurements at any age would suffice for an accurate distance measurement . however , depending on the neighboring environment , nova shells could suffer from deceleration . @xcite presented a method to approximate the temporal variation of the shell radius , which can be applied if observations at many epochs are available . as only hr del and v603 aql have sufficient data for a deceleration determination , we have derived distances to all objects assuming no deceleration . our estimates are given in table [ tbl-3 ] . the information needed to determine the absolute magnitude at maximum @xmath15 or 15 days after maximum @xmath16 , which is essential for the use of a nova as a `` standard candle '' , is : \1 . a trustworthy apparent magnitude at maximum ( either a photoelectric @xmath17 , @xmath18 , visual ( @xmath19 ) or photographic ( pg ) magnitude ) , as well as a well - defined light curve . in all cases , original sources have been re - examined , and light curve parameters have been re - determined , rather than taken from data collections like @xcite . we prefer photoelectric @xmath18 magnitudes over visual @xmath19 magnitudes , and only quote photographic ( pg ) magnitudes if no other ones are available . the value of the interstellar reddening can be derived with sufficient accuracy from multi - wavelength observations of objects in decline ( e.g. from emission line ratios or the strength of the 2200 band ) , or from uv observations of old novae ( strength of the 2200 band ) . in cases where no such information was available , @xmath20 was determined by the help of the galactic extinction program of @xcite , which interpolates and averages data based on various studies of galactic extinction . in addition , the determination of the absolute magnitude of postnovae requires a reliable magnitude in the minimum stage . we took @xmath18-magnitudes from our survey , as well as from @xcite and a forthcoming statistical study of postnova luminosities @xcite . in the following , absolute magnitudes of novae at maximum light are usually given only with an accuracy of @xmath21 , and at minimum with an accuracy of @xmath22 . calculations were often done with higher accuracy , and when corrections ( photographic to visual magnitudes , effects of accretion disc inclination ) were applied , the result was rounded to the same accuracy . cp pup was a fast nova that reached a maximum of @xmath23 in november 1942 . visual and photographic light curves were published by @xcite and @xcite , from which we derive @xmath4 = 6 , @xmath24 days ( averaged over the visual and photographic curves ) , and @xmath25 , where @xmath4 is the time for a nova to decline 2 magnitudes from maximum , @xmath5 is the time to decline by 3 magnitudes , and @xmath6 is the magnitude 15 days after maximum . the first reported detection of a shell was by zwicky ( see @xcite ) , who noted a diameter of @xmath26 in 1955 in red light . @xcite observed the shell in 1980 ( h@xmath0/[n ii ] ) and found a diameter of 14 , while @xcite found the same shell size ( h@xmath0/[n ii ] ) in 1981 . more recently , @xcite found a diameter of 20 ( h@xmath0/[n ii ] ) in 1995 . a _ hubble space telescope _ image , obtained with the wfpc2 and f656n ( h@xmath0 ) filter in 1995 ( program 6060 ) , shows the stellar remnant surrounded by a circle of bright knots of diameter @xmath27 ; the average surface brightness in the shell is @xmath28 . our h@xmath0 image ( figure [ fig1 ] ) reveals about 11 blobs arranged in a circular shell @xmath29 in diameter ( average surface brightness is @xmath30 ) . the same arrangement is seen in our [ o iii ] data ( average surface brightness of @xmath31 ) . the clumpiness seen in the 1980 and 1995 frames is still present . the expansion velocity ( @xmath32 ) for cp pup has been determined by @xcite and @xcite , who derived a value of 1500 km / s shortly after outburst . spectra in the nebular stage @xcite show a profile consisting of 11 emission components , with a velocity spread between @xmath33 and + 570 km / s . similar results were obtained by @xcite ( @xmath34 and + 590 km / s ) . spectroscopy of the resolved blobs in the shell seen in the ground - based images yields a similar range in velocities @xcite . we have assumed a @xmath35 km / s , based on gratton s result . previous distances estimates ( @xcite , @xcite , and @xcite ) place cp pup at distances of 1.5 , 1.6 and @xmath36 kpc , respectively , while @xcite give 0.85 kpc , by neglecting the shell geometry and assuming a much smaller expansion velocity . since this is a well - observed object , we will illustrate the uncertainties in the expansion rate that directly enters into the distance determination . we compare the ground - based deep h@xmath0 frame , obtained with the eso / mpia 2.2 m telescope in 1987 @xcite , the snapshot hst image of 1995 , and our h@xmath0 frame , obtained 1998 with the eso / dutch 0.92 m telescope . the expansion rate derived from the centers of the blobs that are arranged in the circular shell is 0.151 , 0.153 and 0.151 arcsec / year , respectively . in spite of the fact that the limiting magnitude and the seeing effects are very different for the three frames , the results are in good agreement . if the size outlined by the outer edges of the blobs is used , the expansion rates are 0.193 , 0.187 , and 0.178 ; furthermore , the 1987 image reaches to fainter surface magnitudes , and the extent of the `` faint fuzz '' leads to an expansion rate of @xmath37/yr . the hst exposure reveals that instead of eleven blobs , the shell consists of about 30 blobs with typical diameters of @xmath38 ( some are more extended and show special shapes ) the image also shows a symmetry axis that goes through several blobs which may be interpreted as `` polar blobs '' , while the remainder form an ring that is fragmented into @xmath39 20 single clouds . if the ring is circular , its ellipticity indicates an inclination of @xmath40 , and the blob - ring structure would form a prolate ellipsoid with @xmath41 . correspondingly , the radial velocities observed in high - resolution spectra a few years after outburst , assumed to form principally in the fragments of the ring , must be corrected for inclination . thus , while the measured radial velocity is @xmath42 km / s , the true @xmath43 km / s , and the distance , derived form the expansion rate of the blob centers ( along the major axis ) , is 1140 pc . if this distance and the expansion rate of the faint fuzz are used to calculate the radial velocity of the `` fast material '' , a value of 1415 km / s is derived , which is similar to that observed in the absorption lines in the spectrum of the nova observed shortly after maximum . such a dichotomy of fast and slow moving material , forming shells of different diameters , also occurs in pw vul , qu vul , v842 cen and v1974 cyg ( see below ) . @xcite derived an @xmath44 value of @xmath45 . thus , the absolute magnitude at maximum is @xmath46 , @xmath47 , and the absolute magnitude at minimum is @xmath48 . summing up , if the true expansion velocity is known , the use of `` outer edges '' would decrease the distance by 20% , the neglect of the inclination angle would decrease the distance by 32% , and a `` wrong '' expansion velocity of 1400 - 1500 km / s would increase the distance by @xmath49 . if these errors are taken to be typical ones , luminosities of objects for which details on shell structure and extent , inclination and radial velocity are not known , may suffer errors of up to @xmath50 . ct ser was discovered well past maximum light in early 1948 , and was monitored intensely during decline . the nova was observed again by @xcite , who report a typical cv spectrum containing balmer and he ii @xmath14686 emission . there have been no reports of a detection of a shell for this object . our h@xmath0 image , shown in figure [ fig2 ] , clearly reveals a nearly circular shell in the psf - subtracted image with a diameter of @xmath51 in both directions ( with an average surface brightness of @xmath52 ) . the shell does not appear in the [ o iii ] data ( with a @xmath53 limiting surface brightness of @xmath54 ) . although they do not report on the existence of a shell , @xcite do provide @xmath55 km / s for ct ser . there are no previous distances estimates for ct ser . based on the above expansion velocity , we derive a distance of 1.4 kpc . according to the galactic extinction model of @xcite , the extinction is low ( @xmath56 ) , and is neglected here . the observed apparent maximum magnitude , @xmath57 , is likely @xmath58 below the true maximum . the observed absolute magnitude at maximum , @xmath59 , supports the assumption that ct ser was a slow nova discovered long after maximum . the absolute magnitude at minimum is @xmath60 . rw umi was a slow nova at high galactic latitude with a large outburst amplitude . the photographic light curve by @xcite yields @xmath61 at maximum , @xmath62 days , @xmath63 days , and @xmath64 , although it is not certain that the true maximum was observed , since there is a gap of 16 days before the first positive observation , and there is no spectroscopic data taken during outburst . a marginally resolved shell was originally detected by @xcite with a diameter of @xmath65 . more recently , @xcite observed a diffuse shell with a diameter of @xmath66 , as well as two regions of nebulous emission unconnected to the shell . our h@xmath0 image , shown in figure [ fig3 ] , fails to detect either the diffuse shell nor the two unconnected regions seen by @xcite ; the @xmath53 limiting surface brightness is @xmath67 . the shell is also not visible in our [ o iii ] data to a limiting surface brightness of @xmath68 . a wfpc2 h@xmath0 image obtained about five months prior to our observations ( for details see @xcite program 7386 ) also failed to detect the shell to a @xmath53 limiting surface brightness of @xmath69 . in the five years between the observations , it appears the shell has faded below the detection limit . @xcite estimated @xmath70 km / s . previous distance estimates place rw umi at a distance of @xmath71 kpc ( @xcite estimated 5.6 kpc , @xcite @xmath72 kpc and @xcite @xmath73 kpc ) . we adopt the latter result . a value of @xmath74 is derived from the program of @xcite . with an apparent maximum magnitude @xmath75 ( which may not correspond to true outburst maximum ) , the absolute magnitude at maximum is @xmath76 ( error from distance ) , @xmath77 , and the absolute magnitude at postoutburst minimum is @xmath78 . hr del was an unusually slow nova with a well - observed outburst . the @xmath18 light curve compiled by @xcite yields @xmath79 , @xmath80 days , @xmath81 days , and @xmath82 . @xcite first obtained spatially resolved spectroscopy of the nebular shell , followed by a more detailed study by @xcite . @xcite obtained the first photograph of the resolved shell in [ o iii ] , finding a oval remnant @xmath83 in size in 1981 . @xcite obtained ccd imaging in both [ o iii ] and h@xmath0/[n ii ] in 1992 , which revealed a ring - like structure of diameter @xmath84 in h@xmath0/[n ii ] , and a bipolar structure of similar diameter in [ o iii ] . the h@xmath0/[n ii ] also revealed enhancements that lie at the same positions as the [ o iii ] components , although whether this implies emission from the same material , or a simple line - of - sight alignment , is not clear . in 1993 , @xcite re - observed hr del , revealing more extended emission ( @xmath85 ) than previously seen . a _ hubble space telescope _ image , obtained with the wfpc2 and f656n ( h@xmath0 ) filter ( program 6770 ) , does not show the enhancements seen by @xcite , and has a smaller extent ( @xmath86 ) ; the average surface brightness is @xmath87 . our images , shown in figure [ fig4 ] , are similar to that in @xcite , with h@xmath0 having a complete shell ( @xmath88 ) , and a surface brightness ranging from @xmath89 in the outer region to @xmath90 in the inner region . the [ o iii ] image shows a bi - polar structure , with an average surface brightness of @xmath91 . the enhancements seen by @xcite in the h@xmath0/[n ii ] image are not present in our image , and the emission in the 2 lines appear to be growing at slightly different rates ( i.e. the major axis in [ o iii ] is @xmath92 ) . the expansion velocity for hr del was measured by @xcite ( 470 km / s ) , @xcite ( 540 km / s ) , @xcite ( 520 km / s ) , and @xcite ( 560 km / s ) . previous expansion parallax measurements ( @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite ) yield a distance for hr del of @xmath93 pc . distances based on other techniques ( e.g. interstellar lines , reddening , light curve decline ) yield comparable values of @xmath94 pc @xcite . based on an expansion velocity of 525 km / s ( similar to that used by the papers cited above ) , we derive a distance of 750 pc . note that we attempted to fit the hr del data to derive the deceleration , but the best fit resulted in an acceleration of the shell . an acceleration is possible if there is a fast wind from the stellar remnant , and p cygni profiles ( indicative of outflow ) have been detected in the spectrum of hr del @xcite . the expansion rates for hr del show significant variations , with the measurement of @xcite being the most discrepant . that observation was obtained with a similar filter and exposure time to our data , but with a telescope twice as large , and therefore goes much fainter . this can be seen by overlaying our data on the @xcite h@xmath0 image , which shows that our shell is slightly larger than the main emission in their image , and that we fail to detect their lowest ( @xmath95 ) contour . we adopt a distance of @xmath96 kpc . @xcite derived an @xmath44 value of @xmath97 , and with an apparent maximum magnitude @xmath98 , the absolute magnitude at maximum is @xmath99 , @xmath100 , and the absolute magnitude at minimum is @xmath101 . v3888 sgr was discovered in late 1974 during its decline from observed maximum , which was rapid ( @xmath102 days according to @xcite ) . while some photometric @xcite and spectroscopic @xcite observations were obtained during decline , there have been no published observations of the object in quiescence . our h@xmath0 image , shown in figure [ fig5 ] , shows a nearly complete shell of size @xmath103 ( with a surface brightness of @xmath104 ) ; the object is too faint to yield a reliable profile fit . radial velocity measurements made shortly after discovery gave an average @xmath32 of @xmath105 km / s @xcite . previous estimates place v3888 sgr at a distance of 3.5 kpc @xcite to @xmath106 kpc @xcite , based on photometric and spectroscopic evidence . our derived distance , based on a @xmath32 of 1300 km / s , is 2.5 kpc . the galactic extinction program of @xcite yields a value @xmath107 , however , @xcite argue for a value @xmath108 . the observed apparent maximum magnitude , @xmath109 , is likely @xmath110 below the true maximum , as indicated from the spectroscopic appearance of the nova . if @xmath111 is used , the observed absolute magnitude , @xmath112 would point at a slow nova with @xmath113 . in this case , the absolute magnitude at minimum , @xmath114 , appears to be unusually faint . if the second value , @xmath108 is used , the observed absolute magnitude , @xmath115 , yields an extrapolated maximum luminosity which assigns v3888 sgr to the group of fast novae ( an assumption that also entered the derivation of the reddening value of @xcite ) . the absolute magnitude at minimum , @xmath116 , would be in better agreement with other novae ( this result was also used in table [ tbl-6 ] ) . thus , either the galactic extinction program underestimates the reddening value , or the exnova is indeed unusually faint ( e.g. the exnova lies in the period gap ) . nq vul was a moderately fast nova that has been well - studied . the light curve is somewhat peculiar ( see the compilation by @xcite ) ; the delayed maximum of @xmath117 ( with a scatter of 0.25 ) is followed by a short , deep minimum , so that a derived @xmath4-time of 2.5 days is of no significance : if one disregards this fluctuation , @xmath118 days , @xmath119 days , and @xmath120 . observations by @xcite , obtained in 1993 , revealed a circular shell with an average diameter of @xmath121 , a knot at the northern end , and an inner shell with a diameter of @xmath122 . there is also evidence of a bar - like feature extending across the stellar remnant at a position angle of @xmath123 . a wfpc2 h@xmath0 image obtained in 1997 ( program 7386 ) did not detect the shell to a @xmath53 limiting surface brightness of @xmath124 . however , due to its extremely high resolution , wfpc2 is not efficient at detecting extended , diffuse objects , so the lack of detection is not unexpected . our h@xmath0 image , shown in figure [ fig6 ] , reveals an almost circular shell of size @xmath125 , with an average surface brightness of @xmath126 . the northern extension seen by @xcite is detected , and while a bar - like structure is seen , comparison with field stars shows that it is an artifact of the psf subtraction . the shell does not appear in the [ o iii ] @xmath15007 data to a @xmath53 limiting surface brightness of @xmath127 . the absorption lines of nq vul were measured soon after outburst ( @xcite , @xcite , and @xcite ) , and a wide range of expansion velocities are given by these authors . @xcite noted the conspicuous high - velocity systems ; they derived for the principal spectrum an average value @xmath128 , for the diffuse enhanced @xmath129 , and for the orion spectrum @xmath130 km / s . the total half - width at zero intensity of the strong [ o iii ] @xmath15007 and balmer lines is 1150 and 900 km / s , respectively . one year after outburst , @xcite observed a four - peaked structure with components at @xmath131 and @xmath132 km / s , and @xcite derived a @xmath32 of 705 km / s 5 - 6 years after outburst . as usual , the absorption lines have higher velocities than the emission lines , which may indicate a noticeable inclination of the principal part of the shell . the expansion velocity of @xcite may be an acceptable compromise between the absorption and emission velocities measured by @xcite . previous distances estimates place nq vul at a distance of 1.7 kpc @xcite , 1.2 kpc @xcite and @xmath133 kpc @xcite . if we associate the inner shell of @xcite with the slowly expanding material observed by @xcite ( average @xmath32 of the fast components = @xmath134 km / s ) , and the larger shell , as well as our observation , with the material that is evident in the total width of the nebular and balmer lines ( @xmath135 km / s ) , we derive an average distance @xmath136 kpc . the nova polarization @xcite indicates 1.8 kpc . in the following , we adopt the distance of 1.16 kpc . a value of @xmath137 was derived from @xcite , while @xcite find a value of 2.79 from interstellar polarization . we have used @xmath138 in the following . the absolute magnitude at maximum is @xmath139 , @xmath140 , and the absolute magnitude at minimum is @xmath141 . pw vul was a moderately fast nova @xcite with a well - studied outburst . the light curve is very structured , so that the derivation of reliable decline times is difficult , as noted by @xcite ; we adopt a maximum magnitude @xmath142 , @xmath143 days , @xmath144 days , and @xmath145 . a spectrum of the object , obtained in 1991 by @xcite , shows strong h@xmath0 and [ o iii ] @xmath15007 emission . the shell of pw vul was first detected by @xcite in 1993 , who found a diameter of @xmath146 by examining the radial profile of the object . our h@xmath0 image , shown in figure [ fig7 ] , reveals a clearly resolved , circular shell of size @xmath147 , with an average surface brightness of @xmath148 . the shell does not appear in the [ o iii ] data to a @xmath53 limiting surface brightness of @xmath149 . however , a wfpc2 exposure ( program 7386 ) obtained shortly after our observations reveals a faint , diffuse shell of size @xmath150 , considerably smaller than our ground - based observations , with an average surface brightness of @xmath151 . the wfpc2 data is detecting emission `` hidden '' in the ground - based psf , while the ground - based data shows diffuse emission which wfpc2 can not detect . in order to investigate the large difference in shell sizes between the @xcite data and our data , the former authors kindly provided us with a copy of their data . we were able to confirm their radial profile measurement , however a contour plot reveals a shell of size @xmath152 . while this measurement has not been corrected for the intrinsic psf , it does illustrate the difference in determining the shell size from radial profiles ( fwhm ) and from contour plots ( fwzi ) . the expansion velocity was measured by @xcite shortly after outburst to be 1200 km / s . @xcite measured the velocity 9 years after outburst from the structure of the h@xmath0/[n ii ] @xmath16584 profile , and found a much lower value of @xmath153 km / s . we associate the shell found by hst with the material expanding with 470 km / s , and the more extended shell with the `` fuzz '' that expands with 1200 km / s , and obtain a consistent distance of @xmath154 kpc . previous distances estimates ( @xcite , @xcite , @xcite and @xcite ) place pw vul at a distance of 1.2 , 1.5 - 3.0 , 1.3 , and 1.6 kpc . the first three determinations use techniques other than expansion parallax ( e.g. interstellar reddening measurements ) , while @xcite adopted a small expansion velocity , in combination with an underestimated shell size , to derive a result which is , nevertheless , in agreement with the others . we adopt a distance of @xmath154 kpc , and an interstellar extinction @xmath155 @xcite . the absolute magnitude at maximum is @xmath156 , @xmath157 , and the absolute magnitude at minimum is @xmath158 . qu vul was a moderately fast nova of the onemg group , which was well studied from the x - ray to the radio . from the visual light curve from the afoev archive , we derive a maximum magnitude @xmath159 , decline times @xmath160 and @xmath161 days , and @xmath162 . radio observations by @xcite first detected the shell about one year after outburst , and yielded a diameter of @xmath163 at an age of 497 days . the first reports of an optical shell were presented by @xcite , who detected a larger radial profile for the nova than for field stars in h@xmath0 , while a continuum image showed no differences . those authors derived a diameter for the shell of @xmath164 in 1994 . @xcite obtained infrared images of the object in 1996 , and found a shell diameter of @xmath165 at @xmath166 , again by comparing the radial profile of the nova with field stars . data was obtained in 1998 with the _ hubble space telescope _ using both the wfpc2 ( program 7386 ) and nicmos ( for details see @xcite program 7381 ) . our measurements of that data reveal shell sizes of @xmath167 at @xmath166 and @xmath168 ( a bright , filled - in ring with an average surface brightness of @xmath169 ) at h@xmath0 . our h@xmath0 image , shown in figure [ fig8 ] , reveals a clearly resolved , circular shell of size @xmath26 and an average surface brightness of @xmath170 . the shell may be weakly seen in the [ o iii ] data , with an average surface brightness of @xmath171 ; the radial profile plot clearly shows the presence of the shell . the gross discrepancies in the expansion rates between our data ( based on a resolved shell ) and previous data ( based on radial profiles ) may indicate an over - correction for the intrinsic psf ( hence an underestimation of the shell size ) when determining shell sizes based on profiles . as noted by @xcite , the measurement technique of @xcite is adequate for demonstrating the existence of a shell , but it is not appropriate for determining the shell size , as the technique incorrectly assumes a gaussian profile for the shell , and fails to account for light from the central star . the large discrepancy between the ground - based and hst data may be caused by the fact that the first method traces extended , rapidly moving , low - surface brightness material , while the second one traces dense , clumpy , slowly moving material near the central object . the expansion velocity was measured shortly after outburst to be 790 km / s @xcite and 1200 km / s @xcite . a study of the profiles of the [ o iii ] @xmath15007 and h@xmath0 emission lines in early july , 1985 , by @xcite yielded fwhm values of 1700 and 1360 km / s , respectively ( his fwzi is @xmath172 km / s , the width of the flat - topped section of the [ o iii ] @xmath15007 line 1140 km / s ) . @xcite derived a fwzi value of @xmath173 km / s in 1994 . previous distances estimates , based on expansion parallaxes , place qu vul at distances of 3.6 and 2.6 kpc , respectively , @xcite , while photometric and reddening techniques yield distances of 3.0 kpc @xcite , 1.2 kpc @xcite , and 1.6 kpc @xcite . based on an expansion velocity of 1500 km / s ( the fastest - moving material ) , we derive a distance , based on our observations only , of 1.5 kpc . if the shell - structure of the hst observations is associated with the material that formed the flat - topped structure in the lines observed by @xcite , a distance of 2.0 kpc is derived . averaging these results leads to @xmath174 kpc , in reasonable agreement with estimates based on other methods ( @xmath175 kpc ) . with a distance of 1.75 kpc , and interstellar extinction @xmath176 @xcite , the absolute magnitude at maximum is @xmath177 , @xmath178 , and the absolute magnitude at minimum is @xmath179 . v1819 cyg was a slow nova that has been little studied ; see @xcite for observations of the nova in the first two years after outburst . the light curve has been constructed by @xcite , who give @xmath180 ; their @xmath181 value appears to be not reliable . they derived times @xmath182 days and @xmath183 days , respectively , although other authors give different values . we have evaluated the afoev visual light curve , and derive @xmath184 at maximum , @xmath185 and @xmath186 days , and @xmath187 . while no ground - based imaging observations have been reported , a wfpc2 observation ( program 7386 ) in the f658n ( [ n ii]@xmath188 ) filter reveals a marginally resolved shell with a diameter of @xmath189 . our ground - based observations are unable to resolve a shell of that size . @xcite has published a finding chart for the object , but our images ( see figure [ fig9 ] ) have resolved the szkody object into two stars , with the nova being the southeastern object . from low - resolution spectral scans , @xcite derived `` velocity widths '' of @xmath190 km / s . @xcite made spectroscopic observations at moderate dispersion one month after maximum and observed wide emission lines with a `` half - width '' of 1400 km / s ( which we interpret as `` full width at half maximum '' ) , with the principal and diffuse - enhanced absorption components at @xmath191 and @xmath192 km / s . @xcite find a reddening @xmath193 , and an absolute magnitude at maximum @xmath194 from various relationships between maximum magnitude and speed of decline , and derive a distance of @xmath195 kpc . if we assume , in agreement with @xcite , a @xmath196 km / s for the bulk of material , we derive a distance of 7.4 kpc , in good agreement with the above result . this yields an absolute magnitude @xmath197 , @xmath198 , and the absolute magnitude at minimum is @xmath199 , with the error based on the uncertainty of the extinction only . the much larger value @xmath200 from the galaxy extinction model of @xcite would increase the brightness by @xmath201 . v842 cen was a moderately fast nova ( @xmath202 days , @xmath203 days , @xcite ) , which showed a dust - forming event . observations by @xcite revealed a shell of diameter @xmath204 ( h@xmath0/[n ii ] ) in 1995 . our h@xmath0 image , shown in figure [ fig10 ] , reveals an incomplete , circular shell of size @xmath205 , with an average surface brightness of @xmath206 . the shell is also seen in the [ o iii ] data with an average surface brightness of @xmath207 . the expansion velocity was measured by @xcite after recovery from the dust - forming event in may 1988 . the h lines showed a saddle - shaped structure , with the two peaks at @xmath208 and @xmath209 km / s . the line profiles showed broad wings which extended to @xmath210 km / s ( hwzi ) . a study of the emission lines of [ o iii ] @xmath15007 , [ n ii ] @xmath16584 and h@xmath0 by @xcite over the years 19881991 yielded a ` castellated ' profile , whose initially broad wings faded , while two maxima at about @xmath211 and @xmath212 km / s sharpened over that time interval . a high - resolution spectrum of [ o iii ] @xmath15007 , obtained on 1988 july 17 , is shown in @xcite . its maxima are at @xmath213 and @xmath214 . the expansion velocity of the dense regions of the shell is thus @xmath215 km / s , while low - density material with @xmath216 km / s is also present . based on an expansion velocity of 525 km / s , @xcite derived a distance of @xmath217 kpc . the size of the shell in our data is inconsistent with the size in @xcite , and our distance of 0.42 kpc falls way below their value . the reddening - based distances of 0.9 kpc @xcite and 1.0 kpc @xcite tend to favor the larger distance . if we assume that the value of @xcite is correct , the expansion velocity of the shell observed by us is 1600 km / s , in reasonable agreement with the velocity of the material that caused the broad wings observed in the first years following the outburst . @xcite gave a light curve with a maximum magnitude @xmath218 , and @xmath219 ; @xmath220 is @xmath221 . a somewhat revised expansion rate and velocity of the shell observed by @xcite leads to a distance of 1.15 kpc , and the absolute magnitude at maximum is @xmath222 ( using the error estimate of the above authors ) , and @xmath223 , with the same error bars . the nova has not yet returned to minimum light . qv vul was a moderately fast nova , which was studied spectroscopically by @xcite . the afoev visual light curve yields @xmath224 at maximum , @xmath225 days , @xmath226 days , and @xmath227 . while no ground - based imaging observations have been reported , data was obtained in 1998 with the _ hubble space telescope _ using both the wfpc2 ( program 7386 ) and nicmos ( program 7381 ) . our measurements of that data reveal shell sizes of @xmath228 at @xmath166 , and @xmath229 at [ n ii]@xmath188 ( with an average surface brightness of @xmath230 ) . we failed to detect a shell in our ground - based images ( to @xmath53 limiting surface brightnesses of @xmath231 at h@xmath0 and @xmath232 at [ o iii ] ) . there is no published finding chart of the quiescent object , so we show , in figure [ fig9 ] , our h@xmath0 on - band and off - band images , which clearly identify the nova . an analysis of the shapes of [ o iii ] @xmath15007 and [ n ii ] @xmath16584 , observed in 1989 by @xcite , yields expansion ( hwhm ) velocities of @xmath233 and @xmath234 , respectively . based on an averaged hwhm velocity of the [ o iii ] @xmath15007 and [ n ii ] @xmath16584 lines , @xmath235 km / s , we derive a distance to qv vul of 2.7 kpc . @xcite gave a value of @xmath236 . for a distance of 2.7 kpc , the absolute magnitude at maximum is @xmath237 , @xmath238 , and the absolute magnitude at minimum is @xmath179 . v351 pup was a moderately fast nova which reached a maximum of at least @xmath239 , with @xmath240 days , @xmath241 days , and @xmath242 , as based on the aavso visual light curve @xcite . its outburst was studied in detail as part of the tololo nova survey @xcite . a marginal ground - based detection of a shell ( diameter @xmath243 ) was reported by @xcite . data was obtained in 1998 with the _ hubble space telescope _ using the wfpc2 ( program 7386 ) , and our measurement of that data reveals a h@xmath0 shell @xmath244 in diameter , and an average surface brightness of @xmath245 . our ground - based observations , shown in figure [ fig11 ] , do not show any indication of a shell in the psf - subtracted images . however , in the radial profile , the h@xmath0 data does show that the nova ( fhwm = 4.37 pixels ) is broader than field stars ( @xmath246 pixels ) , while the off - band image does not show any broadening of the nova profile . the expansion velocity quoted by @xcite , 2000 km / s , is based on iue spectra . a detailed analysis of the mg ii 2800 line , based on high - resolution iue spectra , resulted in an expansion velocity of 1200 km / s @xcite . the expansion rate is about @xmath247/yr , based on the ground - based and hst observations . if the expansion velocity of the mg ii 2800 line is assumed to be representative for the shell , a distance of @xmath248 kpc is derived ; the higher velocities quoted by @xcite would lead to a distance around 4 kpc . maximum light is poorly covered by observations . @xcite derived a reddening @xmath249 from recombination lines . the distance of 2.7 kpc leads to @xmath250 , while the distance of 4 kpc @xcite results in @xmath251 ; the values of @xmath252 are @xmath253 and @xmath254 , respectively . we will accept the average values as uncertain , but reasonable , estimates . the nova has not yet returned to minimum light . v1974 cyg was a moderately fast nova , which has been extensively studied . the afoev visual light curve yields @xmath255 at maximum , @xmath256 days , @xmath257 days , and @xmath258 . the first report of a visible shell was by @xcite , who used the faint object camera ( foc ) on _ hubble space telescope _ 467 days after outburst to detect a circular , inhomogeneous ring of diameter @xmath37 in the f278 m ( mg ii ) filter ; this observation was further analyzed by @xcite and they found the shell to be @xmath259 in size . further foc observations @xcite at [ o iii ] @xmath15007 ( f501n ) were obtained at 690 and 818 days after outburst , and revealed shells sizes of @xmath260 and @xmath261 , respectively . @xcite detected extended h@xmath0 emission around the object ( 3 - 4 ) in 1994 - 1995 , which they attribute to excitation of the interstellar medium by uv photons from the outburst . our h@xmath0 image , shown in figure [ fig12 ] , reveals a complete , circular shell of size @xmath262 , with an average surface brightness of @xmath263 . the shell is also seen in the [ o iii ] data , with an average surface brightness of @xmath264 . further _ hubble space telescope _ observations were obtained in 1998 using both the wfpc2 ( program 7386 ) and nicmos ( program 7381 ) . our measurements of that data reveal shell sizes of @xmath265 at @xmath166 , and @xmath266 ( a bright ring with an average surface brightness of @xmath267 ) at h@xmath0 . the expansion velocity for v1974 cyg was determined by several authors to be 1100 km / s @xcite , 1500 km / s @xcite , 830 - 895 km / s @xcite , and 700 - 950 km / s @xcite . previous distances estimates , based on expansion parallaxes from early radio @xcite and _ hubble space telescope _ observations ( @xcite , @xcite ) , place v1974 cyg at a distance of @xmath268 kpc . these observations were re - analyzed by @xcite , who revised the distance estimates downward to @xmath269 kpc , using expansion parallaxes based on hst and radio data , and making them more compatible with a distance of 1.8 kpc derived from other techniques ( @xcite , @xcite ) . based on an expansion velocity of 1070 km / s ( which is the average value of the expansion velocities given in the previous paragraph ) , we derive a distance for the _ hubble space telescope _ data of 2.0 kpc . however , if we use our ground - based observations , where we see a clearly resolved shell , we derive a distance of only 310 pc . this discrepancy can be removed in the same way as in the case of v842 cen and other objects . the ground - based observations record faint , fast - moving material , while the hst frames record the bulk of dense clouds that also show up as maxima in the emission lines . if we assume a distance of 1.8 kpc , the extended shell corresponds to an expansion velocity of 5700 km / s . the maximum expansion velocity seen in early p cygni lines of mg ii was 4500 km / s @xcite . the experimentum crucis would be to carry out spatially resolved spectroscopy of the extended optical shell to see the velocity splitting . we will adopt a distance of @xmath270 kpc , based on chochol s detailed study , and an @xmath271 @xcite , so that @xmath272 , and @xmath273 . the nova has not yet returned to minimum light . hy lup was discovered as a star of magnitude @xmath274 by @xcite , and has been little studied since . a spectrum taken by the discoverer indicated that the nova was already some days after maximum . photometry from the aavso archives @xcite reveals @xmath275 days . while no ground - based observations have been reported since shortly after outburst , a wfpc2 observation ( program 7386 ) in the f656n ( h@xmath0 ) filter reveals a marginally resolved shell with a diameter of @xmath189 . our ground - based observations are unable to resolve a shell of that size . there is no published finding chart of the quiescent object , so we show , in figure [ fig9 ] , our h@xmath0 on - band and off - band images , which clearly identify the nova . the expansion velocity for hy lup was measured immediately after discovery by @xcite to be 2700 km / s . a spectrum taken three years after outburst shows the object in the nebular stage ; the emission lines of h@xmath276 , [ o i ] and [ o iii ] appear double - peaked , with a separation of @xmath277 km / s and a fwhm of 340 km / s of the single components @xcite . there are no previous distances estimates for hy lup . based on an expansion velocity of 400 km / s , we derive a distance of 1.8 kpc . this estimate is based solely on _ hubble space telescope _ data , under the assumption that it shows slow - moving material . if the _ hubble space telescope _ data also images fainter , faster material , a larger velocity ( up to 2700 km / s ) would result in a much larger distance . the galactic extinction model yields a value @xmath278 . the resulting absolute magnitude is @xmath279 , but the expansion velocity and the apparent magnitude may both be lower limits , so that the absolute magnitude is likely some magnitudes brighter . the nova has not yet returned to minimum light . it can not be used for maximum or minimum luminosity determinations , unless additional observations become available . cp cru was a fast nova ( @xmath280 days , based on data from vsnet ) , which showed strong he ii emission @xcite , making it a member of the he / n class @xcite . its maximum observed magnitude was @xmath281 , but true maximum may easily have been missed . while no ground - based imaging observations have been reported , data was obtained in 1998 with the _ hubble space telescope _ using the wfpc2 ( program 7386 ) , and our measurement of that data reveals a h@xmath0 shell with a size @xmath282 . our ground - based observations are unable to resolve a shell of that size . there is no published finding chart of the quiescent object , so we show , in figure [ fig9 ] , our h@xmath0 on - band and off - band images , which clearly identify the nova . the expansion velocity for cp cru has measured shortly after outburst by @xcite to be @xmath283 km / s there are no previous distances estimate for cp cru . based on the expansion velocity of 2000 km / s , a distance of 3.2 kpc is derived . as with hy lup ( also with only _ hubble space telescope _ data ) , this distance may be an overestimate to the true value . the galactic extinction model yields a value @xmath284 . the resulting absolute magnitude is @xmath285 , which is also an indication that the true maximum was missed . the nova has not yet returned to minimum light . it can not be used for maximum or minimum luminosity determinations , unless additional observations become available . we failed to detect a shell around 15 objects , which are listed in table [ tbl-4 ] . three of the objects ( gq mus , v827 her , and v838 her ) were also observed with wfpc2 ( program 7386 ) , and no shell was detected . a shell around gq mus was also not detected in ground - based data by @xcite . four objects v868 cen , by cir , v4361 sgr , and n sco 97 do not have published finding charts , and we show , in figure [ fig13 ] , our h@xmath0 on - band and off - band images , which clearly identify the quiescent object . @xcite have identified a new candidate for v2104 oph based on positional coincidence ( the field in the atlas of @xcite is blank ) , and our image ( figure [ fig13 ] ) shows a faint ( @xmath286 ) object , slightly h@xmath0 bright , at the nominal position ( which is identical to the @xcite candidate ) ; the coordinates of our candidate are @xmath287 and @xmath288 as measured from our image . v888 cen does not have a published chart , while @xcite identifies candidates for v3645 sgr and n car 1972 . our observations of v888 cen reveal that the marked object ( the southern object of a close pair ) is h@xmath0 bright , while for the latter two objects , there are no h@xmath0 -bright candidates ( see figure [ fig13 ] ) . our observations of n car 1971 ( not shown ) show a blank field at the nominal position of the object . however , examination of archival material @xcite shows that the object was not detected on a direct plate obtained 2 days after the discovery spectrum . thus , this object is likely an emulsion defect , and also calls into question the reality of n car 1972 , which was found in a similar manner to n car 1971 . @xcite noted a correlation ( at the @xmath289 confidence level ) between speed class ( @xmath5 ) and the axial ratio of nova shells . such a correlation is not unexpected , as the effects of common - envelope evolution thought to occur during outbursts can remove varying amounts of angular momentum from the system depending on the length of time spent in this phase . we have added our dataset to that of @xcite ( for the three objects in common , we have averaged our ratio with theirs ) , and show , in figure [ fig14 ] , the merged plot . we have fit a second - order polynomial to the data , and find that the data is correlated at the @xmath290 level . however , the fit is driven by 3 objects ( dq her , t aur , and rr pic ) , so the reality of the correlation is still quite speculative . in the previous section , distances to several well - investigated and new novae , and their absolute magnitudes , were derived . we can use these absolute magnitudes , in conjunction with the @xmath4 and @xmath5 times , to investigate a maximum magnitude - rate of decline ( mmrd ) relationship . there is no generally accepted light curve decline parameter ; some authors prefer @xmath4 , to extend the mmrd to faint , extragalactic novae , but the fluctuations of the light curve at maximum can easily reach @xmath291 . the value of @xmath5 is much more stable in this respect . nevertheless , we will derive mmrd relations for both parameters . another difference is the use of the @xmath18 ( or visual ) or the @xmath17 ( or photographic ) light curve to derive the value of @xmath292 . since novae are reddest at maximum , @xmath293 should be larger than @xmath294 . this is generally the case , but the general uncertainty of pinning down the time and magnitude of maximum is so large that errors in using the `` wrong '' color band are almost negligible . since visual or @xmath18 magnitudes are nowadays more common than blue magnitudes ( which were in the past often derived from photographic sky surveys ) , we will restrict our analysis on @xmath18 magnitudes ( see also section 5.3 ) . we have enhanced our sample of objects by adding expansion parallax results from other authors . for these objects , a critical evaluation on assumed light curve data , expansion velocities and extinction values was made . the most complete light curve of t aur is based on harvard photographs , which show @xmath295 at maximum , @xmath296 days , @xmath297 days , and @xmath298 . @xcite derived a distance of @xmath299 kpc , and @xcite derived an @xmath20 value of @xmath300 . the absolute magnitude at maximum is @xmath301 , @xmath302 , and the absolute magnitude at minimum is @xmath158 . the visual light curve was compiled by @xcite : @xmath303 at maximum , @xmath304 days , @xmath305 days , and @xmath306 . @xcite derived a distance of @xmath307 kpc , and @xcite derived an @xmath44 value of @xmath308 . the absolute magnitude is @xmath309 @xmath310 , and the absolute magnitude at minimum is @xmath311 . the brightest nova of the 20th century reached @xmath312 at maximum . a detailed visual light curve @xcite gives single data , and different averaged data points . these yield different decline rates , and we adopt here @xmath313 at maximum , @xmath314 days , @xmath315 days , and @xmath316 . a value of @xmath44 of @xmath317 was derived by @xcite . the expansion rate was determined by @xcite to be @xmath318/yr , with an indication that the rate decreases with time . with an expansion velocity between @xmath319 km / s , a distance of @xmath320 kpc is derived . the absolute magnitude is @xmath321 , @xmath322 , and the absolute magnitude at minimum is @xmath48 . the nova reached at maximum @xmath323 . a light curve by @xcite yields @xmath324 days , @xmath325 days , and @xmath326 . @xcite derived a distance of @xmath327 kpc , and a value of @xmath328 was calculated from the galactic extinction model of @xcite . the absolute magnitude at maximum is @xmath329 , @xmath330 , and the absolute magnitude at minimum is @xmath331 . the visual light curve , compiled by @xcite , reached @xmath332 , and has @xmath333 days , @xmath334 days , and @xmath335 . @xcite derived a distance of @xmath336 kpc from the ring geometry and projection effects of the velocities . they assume an inclination angle of @xmath337 , and from the projected velocity of @xmath338 km , they arrive at an intrinsic velocity of 850 km / s . we derived a projected velocity of 405 km / s , an expansion rate of @xmath339/yr , and at a distance of 0.58 kpc , in good agreement with @xcite . @xcite derived an @xmath44 value of @xmath340 . the absolute magnitude at maximum is @xmath341 , @xmath342 , and the absolute magnitude at minimum is @xmath343 . the photographic light curve , based on harvard photographs , reached @xmath344 , and has @xmath345 days , @xmath346 days , and @xmath347 . @xcite derived an expansion rate of @xmath348/yr , and estimated an expansion velocity of 650 km / s , which yield a distance of 3.5 kpc . a value of @xmath349 was derived from the galactic extinction model of @xcite . the absolute magnitude at maximum is @xmath350 , and @xmath351 , and the absolute magnitude at minimum is @xmath352 . a visual light curve , compiled by @xcite , reached @xmath353 , and has @xmath354 days , @xmath355 days , and @xmath356 . @xcite derived a distance of @xmath357 kpc . a careful study of @xcite , based on nebular slit spectra taken in the 1970s , yields a distance of @xmath358 kpc , while other recent determinations , as quoted by @xcite , yield values near @xmath359 kpc . we adopt the result of @xcite . @xcite derived an @xmath44 value of @xmath360 , which is based on the strength of the 2200 band and is a good average of other determinations . the absolute magnitude at maximum is @xmath361 , @xmath362 , and the absolute magnitude at minimum is @xmath141 . a visual light curve , compiled by @xcite , reached @xmath363 , and has @xmath364 days , @xmath365 days , and @xmath366 . @xcite derived a distance of 1.35 kpc , while @xcite derived an @xmath20 value of @xmath300 . the galactic extinction model of @xcite yields a value which is three times higher . the absolute magnitude at maximum is @xmath367 , @xmath368 , and the absolute magnitude at minimum is @xmath158 . all these values might be brighter by more than @xmath369 if the extinction value based on uv spectroscopy is in error . maximum light was missed for this object . @xcite derived a distance of @xmath370 kpc , in perfect agreement with an earlier result by @xcite . @xcite derived a value of @xmath371 . the observed maximum magnitude is @xmath372 , confirming the fact that the nova was discovered long after maximum light . the absolute magnitude at minimum is @xmath373 . a photographic light curve is based on observations of ahnert , which are scattered over various issues of the beobachtungs - zirkular of the astronomische nachrichten . it reached @xmath374 , and has @xmath375 days , @xmath376 days , and @xmath377 . @xcite derived a distance of @xmath378 kpc , and a value of @xmath379 is based on the galactic extinction model of @xcite . the absolute magnitude at maximum is @xmath380 , @xmath381 . the magnitude at minimum is uncertain ( @xmath382 ) , and the absolute magnitude is about @xmath383 . a photographic light curve , based on sonneberg and harvard plates , peaks at @xmath384 , and has @xmath385 , @xmath386 days , and @xmath387 . the nebular expansion parallax of @xcite yields a distance of 5.9 kpc , and a value @xmath388 is based on the galactic extinction model of @xcite . the absolute magnitude at maximum is @xmath389 , @xmath390 . @xcite derived @xmath391 as minimum magnitude , the absolute magnitude at minimum is @xmath158 . this nova fits nicely into the mmrd relations ( see below ) . @xcite had assumed an @xmath20 of @xmath392 , which is unlikely because of the fairly blue color of the postnova . her result , @xmath393 , was taken by @xcite as evidence for including it in a hypothetical class of super - bright novae . a photographic light curve is based on observations by @xcite . maximum is poorly covered : an observation of @xmath394 belongs to the rising branch of the light curve . two days later , a poor observation gave @xmath395 : , and in the following days , the nova declined smoothly by @xmath396/day . fitting the light curve of a fast nova through the points leads to a maximum magnitude @xmath397 . the light curve has @xmath398 days , @xmath399 days , and @xmath400 . @xcite derived a distance of @xmath401 kpc , in good agreement with @xcite , and a value of @xmath402 was derived from the galactic extinction model of @xcite . the absolute magnitude at maximum is @xmath403 , @xmath404 , and the absolute magnitude at minimum is @xmath405 . the nova reached a maximum of about @xmath406 . a photographic light curve , compiled by @xcite , reached @xmath407 on 1960 march 4 . outburst maximum is poorly covered by observations ; according to @xcite , spectroscopic evidence indicates that the real maximum of @xmath408 was reached on march 1 . this assumption yields @xmath304 days , @xmath409 days , and @xmath410 . @xcite derived a distance of 1.35 kpc , and a value of @xmath411 was derived by @xcite . the absolute magnitude at maximum is @xmath412 , @xmath413 , and the absolute magnitude at minimum is @xmath331 ( quiescent light is superimposed by dwarf - nova like outbursts ) . a photographic observation near maximum has @xmath414 . follow - on observations were compiled by @xcite , and we derive @xmath415 days , @xmath416 days , and @xmath417 . @xcite derived a distance of @xmath418 kpc , based on a @xmath419 km / s as given by @xcite . @xcite , using @xmath420 km / s , as determined from outburst spectra , derived a distance of 1.32 kpc . since nebular expansion velocities may suffer inclination effects ( see our discussion for cp pup ) , the latter value appears more likely . indeed recent observations by @xcite found a present - day @xmath421 km / s and a new distance estimate of @xmath422 kpc . @xcite derived an @xmath44 value of @xmath423 , and @xcite a value @xmath424 . the new results lead to an absolute magnitude at maximum @xmath425 , @xmath47 , and the absolute magnitude at minimum is @xmath352 . a photoelectric light curve was compiled by @xcite . the nova reached @xmath426 , and has @xmath427 days , @xmath346 days , and @xmath428 . @xcite derived a distance of @xmath429 kpc , and a value of @xmath430 was derived from the galactic extinction model of @xcite . the absolute magnitude at maximum is @xmath431 , @xmath432 , and the absolute magnitude at minimum is @xmath433 . a light curve was published by @xcite . maximum was approximately @xmath434 , @xmath435 days , @xmath436 : days , and @xmath437 for the photographic v light curve . @xcite analyzed the stellar profile and the structure of the h@xmath0 + [ n ii ] @xmath16584 -blend , and derived an annual expansion rate of @xmath438/yr , an expansion velocity of 750 km / s , an average @xmath44 value of @xmath439 , and a distance of @xmath440 kpc . the absolute magnitude at maximum is @xmath441 , @xmath442 . since the postnova is a close optical triple star , no accurate absolute magnitude at minimum can be given . a photoelectric light curve was compiled by @xcite . the nova reached @xmath426 , and has @xmath443 days , @xmath444 days , and @xmath445 . @xcite derived a distance of @xmath446 kpc , in good agreement with determinations of @xmath447 kpc ( @xcite ) and @xmath448 kpc ( @xcite ) . new hst and ground - based observations @xcite yield a distance of @xmath449 kpc . adopting this result and the @xmath450 of @xcite , we arrive at an absolute magnitude at maximum of @xmath451 , @xmath452 , and the absolute magnitude at minimum is @xmath453 . a photoelectric light curve was compiled by @xcite . the nova reached @xmath454 , and has @xmath455 days , @xmath456 days , and @xmath457 . @xcite derived a distance of @xmath458 kpc , which is in good agreement with a previous estimate by @xcite . the @xmath459 was taken from @xcite . the absolute magnitude at maximum is @xmath460 , @xmath461 , and the absolute magnitude at postoutburst minimum is @xmath462 . the results of the preceding sections are summarized in table [ tbl-5 ] . some novae are only well - observed photographically , and the @xmath463 and @xmath464 need to be transformed into the @xmath18 system . @xcite find that novae at maximum have an intrinsic @xmath465 . a sample of nine objects with trustworthy reddening indicates a similar value , @xmath466 . the very fast nova v1500 cyg is much bluer at maximum ( @xmath467 ) , but the objects considered here are all fairly slow , and were corrected using the value @xmath468 . since novae become bluer in later stages , we have assumed @xmath469 . table [ tbl-5 ] also give the spectroscopic types of novae according to the classification of @xcite ( he / n , fe ii , and `` hybrid '' objects marked as fe / he / n ) , the light curve types ( a , ao , b , c , d ) , as defined by @xcite , and mentions , for some recent objects , whether an object is a onemg nova . the mmrd relations are shown in figures [ fig15 ] , [ fig16 ] , and [ fig17 ] . most of the fast , bright objects with smooth `` a - type '' light curves are he / n novae or hybrid objects , i.e. fe ii novae that evolve into he / n novae . the three fastest pure fe ii novae are v476 cyg , dk lac and v351 pup ; they also show type a light curves , but their @xmath5 times are between 15 and 30 days , larger than those of he / n or hybrid novae ( @xmath470 days ) . there appears to be no luminosity difference between the he / n novae and the fe ii novae , the only possible feature seems to be the relative faintness of the ` hybrid ' objects relative to the ` pure ' ones . all novae with structured light curves ( types b , c , d ) have @xmath471 days , and they are , likely without exception , fe ii novae . the onemg novae are thought to originate on massive white dwarfs , although their luminosities appear to be consistent than the other novae of a comparable rate of decline . the three objects in our diagram show light curve types a or b , and their @xmath5 times lie between 25 and 50 days . if one sticks to the `` classical '' mmrd relations which hold for the ensemble of points , the following statements can be made : using the @xmath5 time as the lightcurve parameter , the relation appears almost linear , @xmath472 a similar relation results when using the @xmath4 time as the lightcurve parameter : @xmath473 in both cases , the residual for a single object is about @xmath474 , and a linear equation appears to be a good fit to the data . the most recent fit of this type is by @xcite : @xmath475 who , after selecting a `` high quality sample '' and correcting the other novae accordingly , was able to reduce the dispersion to @xmath476 . it should be noted that a restriction on a near , less reddened sample , or a correction of absolute magnitudes at maximum by assuming that @xmath477 for all objects , as done by cohen , does not lead to a decisive improvement of the scatter in our data set , and thus we refrain from any correction which may introduce systematic effects . at an average @xmath478 days , cohen s calibration is @xmath479 fainter than ours . another classical relation uses a specific stretched s - shaped curve , which is apparent in samples of extragalactic novae . it was first suggested by @xcite , and was recently revised by @xcite . we have adopted their turn - over point , but have re - determined the zero - point and the amplitude : @xmath480 our zero - point is almost identical to that of @xcite , but the `` contrast '' between bright and faint novae has increased from @xmath481 to @xmath482 . a possible explanation is that in the extragalactic sample of @xcite , very bright and fast novae of the v1500 cyg and cp pup - type appear fainter because of insufficient sampling time , and the faintest among the slow novae are missed ( malmquist bias ) . finally , it was shown by @xcite that the absolute ( pg ) magnitude 15 days after outburst is similar for novae of all speed classes . @xcite derived for 11 objects @xmath483 , while we obtain for 28 objects : @xmath484 indicating a similar zero - point difference ( @xmath485 ) between cohen s and our sample . a regression shows that there is absolutely no trend in @xmath252 versus , e.g. , @xmath5-time , however , the standard deviation ( @xmath486 ) is quite large . there is one more way to look at the mmrd relation , if one takes the light curve type as a second parameter . the dichotomy between the a / ao `` super - eddington '' and bcd `` eddington '' groups , introduced in the investigation by @xcite , appears well expressed . the ( @xmath487 ) diagram can be divided in two fields : the a / ao group has @xmath488 and @xmath489 , the bcd group has @xmath490 and @xmath491 ( two poorly investigated objects , v500 aql and xx tau , may have poorly determined luminosities and/or decline times , and are not considered in the following discussion ) . in the ( @xmath492 ) diagram , the two groups are separated as well ; the dividing line in time is now at @xmath493 . if one examines the `` super - eddington '' and `` eddington '' groups separately , mmrd relations are quite weakly expressed ; in the @xmath492 diagram , the a / ao group follows the relation @xmath494 with a correlation coefficient of 0.61 . the bcd group shows only a weak correlation coefficient of 0.21 between magnitude and decline time ; @xmath495 . a linear relation for these objects is : @xmath496 in the @xmath487 diagram , the a / ao group follows the relation @xmath497 with a correlation coefficient of 0.45 , and the bcd group yields @xmath498 with a correlation coefficient of 0.46 ( of course , the @xmath499 is the same ) . such a type of fit in two separate regions mimics the stretched s - shaped curve used for a fit of the complete sample , but in the present case we have a criterion ( the light curve form ) to separate two regions , and to use different linear relations ( or simple averages ) in them . the overall scatter in the mmrd relation which is of the order of @xmath500 in all these relations may indicate the presence of hidden second order parameters ( spectroscopic type , light curve type , mass of underlying white dwarf ) which should influence the luminosity . however , the individual errors are too large , so the sample is not able to indicate which parameter is most decisive . thus , the mmrd relation can be used as a reliable distance indicator only in a statistical way ; for a single object , the absolute magnitude can be off by @xmath501 . if samples are taken from different stellar populations , systematic luminosity differences may be present . this does not devaluate the use of novae as distance indicators , it simply means that caution must be used when such relations are applied ( similar problems occur in other distance indicators like supernovae , cepheids , and rr lyrae stars ) . a better understanding of the mmrd relation is also the basis of a better understanding of the physics of nova eruptions . another result of our study is the derivation of absolute magnitudes of classical novae after eruption . table [ tbl-6 ] lists the absolute visual magnitudes of novae , which have settled to a more or less constant minimum brightness . some of the objects have known orbital periods and inclination angles ; relevant data were taken from the catalog of @xcite , and these magnitudes @xmath502 were corrected for inclination effects , using the formula given by @xcite ; the corrected magnitudes @xmath503 are also listed . all novae for which no inclination information was available , were corrected by assuming an average inclination @xmath504 . recent results on orbital periods are available for v500 aql @xcite and v446 her and v533 her @xcite . in addition to the nova data , absolute magnitudes of dwarf novae at maximum , based on hst fgs parallaxes , and of novalike stars , mainly based on hipparcos parallaxes , were derived ( the absolute magnitudes supersede those given by @xcite ) . results are summarized in fig . [ fig18 ] . if one discards novae in or below the period gap ( cp pup , rw umi , qu vul ) and the long - period system gk per , as well as the peculiar novalike ae aqr , the remaining systems ( with @xmath505 < -0.4 $ ] ) indicate a dependence of absolute magnitude on period : @xmath506\ ] ] ( correlation coefficient 0.47 ) . if the period ( in hours ) is used , the relation is @xmath507\ ] ] which , at @xmath508 hours , yields absolute magnitudes which are @xmath509 brighter than those predicted @xcite s formula for dwarf novae at maximum . summing up , no obvious separation of absolute magnitudes of objects belonging to different groups ( novae at minimum , dwarf novae at outburst , and novalike systems at normal light ) can be seen . this again shows the intimate connection of these groups , which is also obvious from space density considerations , and the hibernation scenario , which indicates that only by including novalike stars and dwarf novae to the space densities of novae , realistic recurrence times of novae can be obtained @xcite . we have obtained optical imaging of 30 recent novae , and have detected shells , in combination with _ hubble space telescope _ data , for 14 objects . we fail to detect the previously observed shell around rw umi , while pw vul , v1974 cyg , and v842 cen appear to show two distinct shells . expansion parallaxes for the novae have been derived , and combined with revised distances for other novae , have been used to derive absolute magnitudes . a study of various maximum magnitude - rate of decline relations is in general agreement with previous `` linear '' determinations . however , a relation making use of the light curve type may yield a somewhat better fit to the data . it clearly shows the existence of two types of novae : ( 1 ) the fast ( @xmath510 days ) super - eddington novae , which may be he / n , `` hybrid '' fe ii novae , and , if their evolution is somewhat slower , fe ii novae . they all show smooth light curves with well - defined maxima ( a ) , and some of them may show quasi - periodic light oscillations in their later evolution ( ao ) . ( 2 ) the slow @xmath511 days ) eddington - novae with structured light curves ( ba ) , double maxima ( bb ) , standstills at maximum ( ca ) , dust formations at later stages ( ca , cb ) , and long premaximum stages ( d ) , seem to belong exclusively to the fe ii spectroscopic type . the merging of both groups leads to a well - expressed mmrd ; if both groups are considered separately , their mmrds show smaller slopes , and a jump of @xmath512 occurs at @xmath513 days and @xmath514 days . at minimum , most novae have similar magnitudes as those of dwarf novae at maximum light and novalike stars . we wish to thank the aavso for providing data from their international database , which is based on observations submitted to the aavso by variable star observers worldwide , and the afoev for providing data from their internet database . we also thank tim naylor and fred ringwald for providing a copy of their pw vul data , jeff robertson for providing a copy of his v2014 oph data , jack macconnell for investigating the reality of nova car 1971 and 1972 , and pierluigi selvelli for information on interstellar extinction data . hwd also gratefully acknowledges the hospitality and support of space telescope science institute , where he could carry out part of the work on which this investigation is based . we wish to thank an anonymous referee for many useful comments on this manuscript . almr , i. , ills - almr , e. 1966 , mitt . budapest 60 andre , j. 1991 , phd thesis , erlangen - nrnberg university andre , j. , drechsel , h. , snijders , m.a.j . , and cassatella , a. 1991 , , 244 , 111 andre , j. , drechsel , h. , and starrfield , s. 1994 , , 291 , 869 andrillat , y. and houziaux , l. 1989 , , 238 , 29p belle , k.e . , woodward , c. e. , evans , a. , eyres , s. , gehrz , r. d. , schuster , m. , greenhouse , m. a. , krautter , j. , starrfield , s. g. , truran , j. 1999 , , 31 , 977 bergner , y.k . , bondarenko , s.i . , miroshnichenko , a.s . , yudin , r.v . , yutanov , n.y . , kuratov , k.s . , and mukanov , d.b . 1985 , astron . , 1374 bertaud , c. 1945 , ann . paris obs . 9 , fasc . 1 bertaud , c. 1962 , j. obs . , 45 , 321 bowen , i.s . 1956 , , 61 , 336 buscombe , w. , de vaucouleurs , g. 1955 , observatory , 75 , 170 campbell , l. 1903 , harvard ann . , 48 , 39 campbell , l. 1919 , harvard ann . , 81 , 113 ciatti , f. and rosino , l. 1974 , , 16 , 305 chochol , d. , grygar , j. , pribulla , t. , komk , r. , hric , l. , and elkin , v. 1997 , , 318 , 908 chochol , d. , hric , l. , urban , z. , komk , r. , grygar , j. , and papousek , j. 1993 , , 277 , 103 chochol , d. , komk , r. , hric , l. , and grygar , j. 1995 , in cataclysmic variables , ed . a. bianchini et al . , ( dordrecht , kluwer academic publishers ) , p. 153 cohen , j.g . 1985 , , 292 , 90 cohen , j.g . and rosenthal , a.j . 1983 , , 268 , 689 cottrell , m.j . and smith , s.e . 1978 , , 90 , 441 della valle , m. 1991 , , 252 , l9 della valle , m. 1993 , , 5870 della valle , m. and benetti , s. 1996 , , 6532 della valle , m. and duerbeck , h.w . 1993 , , 275 , 239 della valle , m. and livio , m. 1995 , , 452 , 704 della valle , m. , gilmozzi , r. , bianchini , a. , and esenoglu , h. 1997 , , 325 , 1151 drechsel , h. , rahe , j. , duerbeck , h.w . , kohoutek , l. , and seitter , w. 1977 , , 30 , 323 duerbeck , h.w . 1981 , , 93 , 165 duerbeck , h.w . 1987a , , 131 , 461 duerbeck , h.w . 1987b , , 45 , 1 duerbeck , h.w . 1987c , the messenger ( eso ) , 50 , 8 duerbeck , h.w . 1992 , acta astron . , 42 , 85 duerbeck , h.w . 1996 , in light curves of variable stars a pictorial atlas , ed . c. sterken and c. jaschek , ( cambridge , cambridge university press ) , p. 134 duerbeck , h.w . 1999 , ibvs , 4731 duerbeck , h.w . and covarrubias , r. 1995 , in iau coll . 151 , flares and flashes , eds . j. greiner et al . , ( berlin , springer ) , p. 264 duerbeck , h.w . and zwitter , t. 2000 , in preparation duerbeck , h.w . , geffert , m. , nelles , b. , dmmler , r. , and nolte , m. 1984 , ibvs , 2641 duerbeck , h.w . , shara , m.m . , leibowitz , e.m . , and seitter , w.c . 2000 , in preparation gaposchkin , s. 1946 , harvard bull . , 918 gehrz , r.d . , grasdalen , g.l . , and hackwell , j.a . 1985 , , 298 , l47 gill , c.d . and obrien , t.j . 1998 , , 300 , 221 gill , c.d . and obrien , t.j . 2000 , , 314 , 175 gilmozzi , r. , selvelli , p. , cassatella , a. 1994 , , 65 , 199 gratton , l. 1953 , , 118 , 568 haefner , r. 1999 , ibvs , 4706 hakkila , j. , myers , j.m . , stidham , b.j . , hartmann , d.h . 1997 , , 114 , 2043 herbig , g.h . and smak , j.i . 1992 , acta astron . , 42 , 17 hjellming . r.m . 1994 , talk presented at the 184th aas meeting kaluzny , j. and chlebowski , t. 1989 , acta astron . , 39 , 35 kenyon , s.j . and wade , r.a . 1986 , , 98 , 935 klare , g. and wolf , b. 1978 , , 33 , 327 kohoutek , l. 1981 , , 196 , 87p krautter , j. , klare , g. , wolf , b. , duerbeck , h.w . , rahe , j. , vogt , n. , and wargau , w. 1981 , , 102 , 337 kukarkin , b.w . and gitz , h.k . 1937 , astr . su 14 , 220 lance , c.m . , mccall , m.l . and uomoto , a.k . 1988 , , 66 , 151 larsson - leander , g. 1954 , an . stockholm obs . 18 , no . 3 leibowitz , e.m . , wyckoff , s. , and vidal , n.v . 1976 , , 88 , 750 liller , w. 1993 , , 5867 macconnell , d.j . 1999 , private communication malakpur , i. 1975 , , 48 , 403 marsh , t.r . , oke , j.b . , and wade , r.a . 1983 , , 205 , 33p mattei , j. a. , 1999 , observations from the aavso international database , private communication mclaughlin , d.b . 1943 , pub . aas 10 , 310 mclaughlin , d.b . 1960 , in stellar atmospheres , ed . greenstein , ( chicago , university of chicago press ) , p. 585 orio , m. , balman , s. , della valle , m. , gallagher , j. and gelman , h. 1996 , , 466 , 410 pachoulakis , i. and saizar , p. 1995 , in cataclysmic variables , ed . a. bianchini et al . , ( dordrecht , kluwer academic publishers ) , p. 153 paresce , f. 1994 , , 282 , l13 paresce , f. , livio , m. , hack , w. , and korista , k. 1995 , , 299 , 823 pettit , e. 1949 , , 61,41 ringwald , f.a . and naylor , t. 1996 , , 278 , 808 ringwald , f.a . naylor , t. and mukai . k. 1996 , , 281 , 192 ringwald , f.a . , wade , r.a . , orosz , j.a . , and ciardullo , r.b . 1998 , , 30 , 893 ritter , h. and kolb , u. 1998 , , 129 , 83 robb , r.m . and scarfe , c.d . 1995 , , 273 , 347 robertson , j.w . , honeycutt , r.k . , and henden , a.h . 1999 , , 31 , 1247 rosino , l. and iijima , t. 1987 , , 130 , 157 rosino , l. , iijima , t. , rafanelli , p. , radovich , m. , esenoglu , h. and della valle 1996 , , 315 , 463 saizar , p. , starrfield , s. , ferland , g. j. , wagner , r. m. , truran , j. w. , kenyon , s. j. , sparks , w. m. , williams , r. e. , stryker , l. l. 1991 , , 367 , 310 saizar , p. , pachoulakis , i. , shore , s.n . , starrfield , s. , williams , r.e . , rothschild , e. and sonneborn , g. , 1996 , , 279 , 280 sanford , r.f . 1945 , , 102 , 357 satyvaldiev , v. 1963 , byull . an tadj . ssr ( dushanbe ) 36 , 37 seitter , w.c . 1990 , in iau coll . 122 , physics of classical novae , ed . a. cassatella and r. viotti , ( berlin , springer - verlag ) , p. 79 sekiguchi , k. feast , m.w . , fairall , a.p . , winkler , h. 1989 , , 241 , 311 shin , j .- y . , gehrz , r.d . , jones , t.j . , krautter , j. , heidt , j. , and hjellming , r.m . 1998 , , 116 , 1966 shore , s.n . , sonneborn , g. , starrfield , s. , gonzalez - riestra , r. and ake , t.b . 1993 , , 106 , 2408 slavin , a.j . , obrien , t.j . and dunlop , j.s . 1994 , , 266 , l55 slavin , a.j . , obrien , t.j . and dunlop , j.s . 1995 , 276 , 353 soderblom , d. 1976 , , 88 , 517 solf , j. 1983 , , 273 , 647 spencer jones , h. 1931 , ann . 10 szkody , p. 1994 , , 108 , 639 szkody , p. and howell , s.b . 1992 , , 78 , 537 taylor , a.r . , seaquist , e.r . , hollis , j.m . , and pottasch , s.r . 1987 , , 183 , 38 thorstensen , j.r . , and taylor , c.j . 2000 , , 312 , 629 van den bergh , s. and pritchet , c.j . 1986 , , 98 , 110 van den bergh , s. and younger p.f . 1987 , , 70 , 125 verbunt , f. 1987 , , 71 , 339 vogt , n. and maitzen , h.m . 1977 , , 61 , 601 wade , r.a . , ciardullo , r. , jacoby , g.h . , and sharp , n.a . 1991 , , 102 , 1738 wade , r.a . , harlow , j.j.b . , and ciardullo , r. 2000 , , 112 , 614 warner , b. 1987 , , 227,83 warner , b. 1995 , cataclysmic variable stars , ( cambridge , cambridge university press ) , p. 146 williams , r.e . 1982 , , 261 , 170 williams , r.e . 1992 , , 104 , 725 williams , r.e . , phillips , m.m . , and hamuy , m. 1994 , , 90 , 297 whitney , b.a . and clayton , g.c . 1989 , , 98 , 297 wu , c .- c . , panek , r.j . , holm , a.v . , raymond , j.c . , hartmann , l.w . , and swank , j.h . 1989 , , 339 , 443 younger , j.w . 1980 , , 85 , 1233 zakharov , g.p . 1953 , perem . zvezdy , 9 , 175 gq mus & eso dutch-0.9 m & 21 march 1998 & 900s & 1200s & 18.37 + v351 pup & & & 600s & 1200s & 19.55 + hy lup & & & 600s & 1200s & 18.86 + cp cru & & & 600s & 1200s & 19.48 + n car 72 & & & 1200s & & unident . + v365 car & & 22 march 1998 & 600s & & 18.47 + v868 cen & & & 600s & 900s & 19.88 + v888 cen & & & 600s & 900s & 16.38 + cp pup & & & 600s & & 14.99 + v842 cen & & 23 march 1998 & 450s & 600s & 15.82 + v812 cen & & & 600s & 900s & 20.44 + by cir & & & 450s & 600s & 15.86 + v4361 sgr & & & 180s & 180s & 16.85 + n sco 97 & & & 180s & 180s & 17.58 + cp pup & & & & 1200s & + n car 71 & & & 900s & & unident . + + rw umi & kpno-2.1 m & 28 may 1998 & 1050s & 1050s & 18.95 + v827 her & & & 720s & 600s & 18.08 + hr del & & & 60s & 60s & 12.16 + v1819 cyg & & & & 600s & 20.33 + + ct ser & & 29 may 1998 & 1050s & 1050s & - + v794 oph & & & 1050s & 1050s & - + v838 her & & & 660s & 900s & 19.07 + qv vul & & & 660s & 900s & 18.61 + v1819 cyg & & & 660s & & + v1974 cyg & & 31 may 1998 & 660s & 660s & 16.06 + pw vul & & & & 360s & 17.51 + v2104 oph & & & 3600s & 2700s & 20.50 + v3888 sgr & & 01 june 1998 & 3600s & & 20.96 + v3645 sgr & & & 150s & & unident . + nq vul & & & 660s & 600s & 17.70 + pw vul & & & 360s & & + + qu vul & kpno-0.9 m & 30 june 1998 & 1550s & 1800s & 18.31 + cp pup ( 1942.86 ) & d & 16.9 & 1998.23 & 0 & 815 & 12 + & d & 16.0 & 1995.75 & 1 & + & d & 13.4 & 1987.20 & 24 & + ct ser ( 1948.27 ) & d & @xmath515 & 1998.23 & 0 & 535 & 8 + rw umi ( 1956.73 ) & & not seen & 1998.41 & 0 & 950 & 11 + & & not seen & 1997.98 & 1 & + & d & 3.0 & 1993.69 & 27 & + & r & 2.0 & 1983.5 & 11 & + hr del ( 1967.52 ) & d & @xmath516 & 1998.41 & 0 & 470,540 & 32,34 + & d & @xmath517 & 1997.40 & 1 & 520,560 & 8,35 + & d & @xmath518 & 1993.69 & 27 & + & d & 5.5 & 1992.58 & 7 & + & d & @xmath519 & 1981.33 & 13 & + v3888 sgr ( 1974.76 ) & d & @xmath520 & 1998.41 & 0 & 1300 & 22 + nq vul ( 1976.81 ) & d & @xmath521 & 1998.41 & 0 & 1300,1700 & 15,16 + & & not seen & 1997.92 & 1 & 950,705 & 14,8 + & d & 8 & 1993.69 & 27 & + pw vul ( 1984.57 ) & d & @xmath522 & 1998.79 & 1 & 1200,470 & 17,3 + & d & @xmath523 & 1998.41 & 0 & + & r & 1.1 & 1993.64 & 3 & + qu vul ( 1984.99 ) & d & @xmath524 & 1998.71 & 1,6 & 790,1200 & 19,26 + & d & @xmath525 & 1998.50 & 0 & 570,1190 & 37,4 + & d & @xmath526 & 1998.18 & 2 + & r & 0.8 & 1996.35 & 5 + & r & 1.9 & 1994.52 & 4 & + & d & @xmath527 & 1986.34 & 26 & + v1819 cyg ( 1986.57 ) & d & @xmath528 & 1999.18 & 1 & 600 & 38 + & & not seen & 1998.41 & 0 & 700 & 39 + v842 cen ( 1986.89 ) & d & @xmath529 & 1998.23 & 0 & 535 & 23 + & d & 1.6 & 1995.15 & 10 & 525 & 33 + qv vul ( 1987.87 ) & d & @xmath530 & 1998.95 & 1 & 515 & 37 + & & visible & 1998.95 & 1 & + & d & @xmath531 & 1998.62 & 2 + & & not seen & 1998.41 & 0 & + v351 pup ( 1991.99 ) & d & 1.3 & 1998.23 & 1,6 & 2000 & 36 + & & visible & 1998.12 & 0 & 1200 & 40 + & r & 0.58 & 1995.17 & 36 & + v1974 cyg ( 1992.14 ) & d & @xmath532 & 1998.41 & 0 & 825,1100 & 9,20 + & d & @xmath533 & 1998.12 & 1 & 1500,860 & 21,28 + & d & @xmath534 & 1998.11 & 2 + & d & @xmath535 & 1994.38 & 30 + & d & @xmath536 & 1994.03 & 30 + & d & @xmath537 & 1993.41 & 31 + hy lup ( 1993.71 ) & d & @xmath528 & 1999.12 & 1 & 2700,400 & 25,18 + & & not seen & 1998.23 & 0 + cp cru ( 1996.65 ) & d & @xmath538 & 1998.91 & 1,6 & 2000 & 29 + & & not seen & 1998.23 & 0 + cp pup & 55.37 & 815 & 0.153 & 1120 + & 52.89 & & 0.151 & 1140 + & 44.34 & & 0.151 & 1140 + ct ser & 49.96 & 535 & 0.079 & 1430 + rw umi & 36.96 & 950 & 0.037 & 4900 + & 26.8 & & 0.041 & 5400 + hr del & 30.89 & 525 & 0.149 & 750 + & 29.88 & & 0.146 & 760 + & 26.17 & & 0.220 & 505 + & 25.06 & & 0.110 & 1010 + & 13.81 & & 0.134 & 825 + v3888 sgr & 23.65 & 1300 & 0.110 & 2495 + nq vul & 21.60 & 1025 & 0.169 & 1280 + & 16.88 & 1025 & 0.237 & 910 + & 16.88 & 545 & 0.090 & 1280 + pw vul & 14.22 & 470 & 0.053 & 1880 + & 13.84 & 1200 & 0.145 & 1750 + & 9.07 & 470 & 0.061 & 1635 + qu vul & 13.72 & 570 & 0.060 & 2005 + & 13.51 & 1500 & 0.207 & 1530 + v1819 cyg & 12.61 & 700 & 0.020 & 7385 + v842 cen & 11.34 & ( 1600 ) & 0.255 & ( 1325 ) + & 8.26 & 525 & 0.097 & 1140 + qv vul & 11.08 & 515 & 0.041 & 2675 + v351 pup & 6.24 & 1200 & 0.100 & 2530 + & 3.18 & & 0.091 & 2775 + v1974 cyg & 6.27 & ( 4500 ) & 0.718 & ( 1320 ) + & 2.24 & 1060 & 0.112 & 2000 + & 1.89 & & 0.106 & 2115 + & 1.30 & & 0.115 & 1940 + hy lup & 5.41 & 400 & 0.046 & 1800 + cp cru & 2.26 & 2000 & 0.133 & 3180 + v794 oph & 1939.45 & 58.78 & 4.95 + v365 car & 1948.74 & 49.49 & 52.2 + v812 cen & 1973.25 & 24.98 & 69.0 + v2104 oph & 1976.76 & 21.68 & 0.44 + gq mus & 1983.05 & 15.18 & 26.4 2.57 + v827 her & 1987.05 & 12.41 & 0.86 1.32 + v838 her & 1991.23 & 7.18 & 2.05 1.17 + v868 cen & 1991.25 & 6.98 & 32.1 + by cir & 1995.07 & 3.16 & + v888 cen & 1995.15 & 3.08 & + v4361 sgr & 1996.53 & 1.70 & + n sco 1997 & 1997.43 & 0.80 & + v500 aql & he ? & ao & 17 & 42 & 0.118 & @xmath539 & @xmath540 & pg + v603 aql & hy & ao & 4 & 9 & 0.500 & @xmath541 & @xmath542 & + v1229 aql & fe & b ? & 20 & 38 & 0.100 & @xmath543 & @xmath542 & + t aur & fe & ca & 45 & 50 & 0.044 & @xmath544 & @xmath545 & pg + v842 cen & fe & cb & 35 & 48 & 0.057 & @xmath546 & @xmath547 & + v450 cyg & fe & ca & 88 & 108 & 0.023 & @xmath548 & @xmath549 & + v476 cyg & fe & a & 6 & 15 & 0.333 & @xmath550 & @xmath551 & + v1500 cyg & hy & a & 2.4 & 3.7 & 1.111 & @xmath552 & @xmath553 & + v1819 cyg & fe & bb & 37 & 89 & 0.054 & @xmath544 & @xmath554 & + v1974 cyg & fe & ba & 17 & 37 & 0.125 & @xmath555 & @xmath556 & onemg + hr del & fe & d & 172 & 230 & 0.012 & @xmath557 & @xmath558 & + dq her & fe & ca & 39 & 86 & 0.051 & @xmath559 & @xmath560 & + v446 her & he & a & 7 & 11.5 & 0.426 & @xmath561 & @xmath254 & pg + v533 her & fe & ba & 22 & 46 & 0.091 & @xmath562 & @xmath563 & + cp lac & hy & a & 5.3 & 9.8 & 0.377 & @xmath564 & @xmath565 & + dk lac & fe & ao ? & 11 & 24 & 0.182 & @xmath566 & @xmath567 & + gk per & he & ao & 7 & 13 & 0.286 & @xmath568 & @xmath549 & + rr pic & fe & d & 20 & 127 & 0.100 & @xmath569 & @xmath570 & + cp pup & he & a & 6 & 8 & 0.333 & @xmath552 & @xmath563 & + v351 pup & fe & a & 10 & 26 & 0.200 & @xmath571 & @xmath563 & onemg + fh ser & fe & cb & 42 & 59 & 0.048 & @xmath572 & @xmath573 & + xx tau & fe & cb & 24 & 43 & 0.083 & @xmath574 & @xmath575 & pg + rw umi & ? & ba & 48 & 88 & 0.042 & @xmath576 & @xmath577 & pg + lv vul & fe & ba & 21 & 43 & 0.095 & @xmath578 & @xmath579 & + nq vul & fe & bb & 23 & 53 & 0.087 & @xmath580 & @xmath581 & + pw vul & fe & bb & 82 & 126 & 0.024 & @xmath582 & @xmath583 & + qu vul & fe & ba & 22 & 49 & 0.091 & @xmath584 & @xmath585 & onemg + qv vul & fe & cb & 50 & 53 & 0.040 & @xmath586 & @xmath587 & + v500 aql & 0.1452 & 19.25 & 5.9 & @xmath588 & 4.50 & & 4.85 & novae + v603 aql & 0.1381 & 11.70 & 0.33 & @xmath589 & 3.90 & 17 . & 4.80 & at minimum + v1229 aql & & & 2.1 & @xmath590 & & & & + t aur & 0.2044 & 15.03 & 0.96 & @xmath591 & 4.50 & 57 . & 4.45 & + v842 cen & & 15.82 & 1.15 & @xmath592 & & & & + v450 cyg & & 19 . & 3.5 & @xmath593 & 5 . : & & 5.4 & + v476 cyg & & 17.70 & 1.62 & @xmath594 & 5.85 & & 6.20 & + v1500 cyg & 0.1396 & 17.42 & 1.5 & @xmath595 & 4.95 & ? & 5.35 & + v1819 cyg & & 20.33 & 7.4 & @xmath596 & 4.90 & & 5.25 & + hr del & 0.2142 & 12.16 & 0.76 & @xmath597 & 2.30 & 40 . & 2.80 & + dq her & 0.1936 & 14.38 & 0.48 & @xmath598 & 5.30 & 86.5 & 2.70 & + v446 her & 0.2071 & 17.61 & 1.35 & @xmath599 & 5.85 & & 6.20 & + v533 her & 0.1471 & 15.06 & 1.25 & @xmath600 & 4.60 & ? & 4.95 & + cp lac & & 15.80 & 1.35 & @xmath591 & 4.50 & & 4.90 & + dk lac & & 16.75 & 3.9 & @xmath601 & 2.40 & & 2.75 & + bt mon & 0.3338 & 15.59 & 1.8 & @xmath602 & 3.85 & 82 & 1.90 & + gk per & 1.9968 & 13.01 & 0.455 & @xmath603 & 3.75 & @xmath604&4.15 & + rr pic & 0.1450 & 12.50 & 0.58 & @xmath605 & 3.55 & 65 & 3.45 & + cp pup & 0.0614 & 14.99 & 1.7 & @xmath606 & 3.90 & 40 : & 4.45 & + v351 pup & & 19.55 & & @xmath607 & & & & + v3888 sgr & & 20.96 & 2.5 & @xmath608 & 5.75 & & 6.15 & + ct ser & 0.1950 & 16.14 & 1.4 & @xmath609 & 5.40 & & 5.75 & + fh ser & & 16.97 & 0.95 & @xmath610 & 5.05 & & 5.45 & + xx tau & & 18.58 & 3.5 & @xmath611 & 4.60 & & 4.95 & + rw umi & 0.079 & 18.95 & 5 & @xmath612 & 5.15 & & 5.55 & + lv vul & & 15.90 & 0.92 & @xmath613 & 4.35 & & 4.70 & + nq vul & & 17.70 & 1.16 & @xmath614 & 5.30 & & 5.65 & + pw vul & 0.2137 & 17.51 & 1.8 & @xmath615 & 4.50 & & 4.85 & + qu vul & 0.1118 & 18.31 & 1.75 & @xmath616 & 5.20 & & 5.55 & + qv vul & & 18.61 & 2.7 & @xmath617 & 5.20 & & 5.55 & + + ss aur & 0.1828 & 10.5 & 0.20 & @xmath618 & 3.80 & 38 & 4.40 & dwarf novae + ss cyg & 0.2751 & 8.7 & 0.166 & @xmath619 & 2.47 & 38 & 3.05 & at maximum + u gem & 0.1729 & 9.8 & 0.096 & @xmath620 & 4.62 & 70 & 3.90 & + + ae aqr & 0.4117 & 11.5 & 0.102 & @xmath621 & 6.26 & 58 & 6.20 & novalike + v3885 sgr & 0.2163 & 10.3 & 0.11 & @xmath622 & 5.01 & @xmath623 & 5.40&variables + rw sex & 0.2451 & 10.6 & 0.29 & @xmath624 & 3.24 & 34 & 3.90 & + ix vel & 0.1939 & 9.6 & 0.096 & @xmath625 & 4.40 & 60 & 4.25 & +
binary stars matter in astronomy in many different ways : as calibrators of various stellar properties , as tracers of star formation , and as hosts to diverse astrophysical phenomena such as mass transfer , circumstellar and circumbinary discs , or dynamical resonances , to name a few . knowledge of orbital elements is needed in these applications . however , only a small fraction of known visual binaries have known orbits , mostly because of orbital periods being much longer than the time covered by observations . even known orbits are not always reliable for the same reason : lack of sufficient data . we report here a large set of binary - star measurements made at the 4.1-m southern astrophysical research telescope ( soar ) with the speckle camera . this paper continues the previous series published by ( * ? ? ? * hereafter tmh10 ) , @xcite , @xcite , @xcite , and ( * ? ? ? * hereafter tmh14 ) . our primary goal is improvement of known orbital elements and determination of new orbits . the emphasis is placed on close and nearby pairs resolvable at soar , where orbital periods are measured in years rather than decades or centuries . the orbits of these fast binaries can be computed or improved only after a few years of speckle - interferometry monitoring . many orbits were already computed for the first time or improved using data obtained during previous speckle runs . an `` orbit optimizer '' was used now to select binaries where observations in the coming three years could potentially yield noticeable orbit improvement . the second , overlapping program is the characterization of multiplicity of solar - type stars within 67pc ( the fg-67 sample , * ? ? ? * ) . by resolving known binaries with astrometric acceleration @xcite or variable radial velocity ( rv ) , we obtain estimates of their periods and mass ratios and lay a foundation for future orbit determination . the spectroscopic discoveries are mostly furnished by the geneva - copenhagen survey of ( * ? ? ? * hereafter gcs ) . on the other hand , by observing relatively wide binaries with separations up to 3 we constrain the existence of subsystems around their components . several such subsystems are discovered here . in addition , we have now specially targeted distant physical secondary components to the main fg-67 stars , looking for subsystems . this work @xcite complements the large survey of the northern sky done with robo - ao @xcite and shows that subsystems in the secondary components are as frequent as in the main components . in this paper we also observed _ hipparcos _ binaries within 200pc with southern declinations , so far largely neglected . these observations allow us to evaluate orbital motion and to find a subset of `` fast movers '' for further monitoring . some nearby tight binaries from _ hipparcos _ were monitored from the outset of the soar speckle program . this strategy of focusing on fast binaries has resulted in several new orbits , with more still to come . our approach for this program was therefore similar and complementary to that of the speckle program active at the wiyn 3.5-m telescope at kitt peak in recent years ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? to date , that program has resolved over 150 close companions from the _ hipparcos _ list . we re - observed some close ( hopefully , fast ) pairs resolved at the blanco telescope in 2008 . some of them moved substantially in 6 years , and may yield first orbits . as a `` filler '' , several neglected pairs were also measured or found unresolved . this paper is structured as follows . section [ sec : obs ] describes briefly the instrument and data processing and gives an estimate of data consistency for the whole series of measurements at soar . the results are presented in section [ sec : res ] in the form of data tables with comments on the newly resolved pairs and multiple systems . in section [ sec : orb ] , the new and revised orbits are given , and section [ sec : concl ] concludes the paper . the observations reported here were obtained with the _ high - resolution camera _ ( hrcam ) a fast imager designed to work at the 4.1-m soar telescope @xcite . for practical reasons , the camera was mounted on the soar adaptive module ( sam , * ? ? ? however , the laser guide star of sam was not used ; the deformable mirror of sam was passively flattened and the images are seeing - limited . the sam module corrects for atmospheric dispersion and helps to calibrate the pixel scale and orientation of hrcam , as explained in tmh14 . the transmission curves of hrcam filters are given in the instrument manual . we used mostly the strmgren @xmath0 filter ( 543/22 nm ) and the near - infrared @xmath1 filter ( 788/132 nm ) . the response curve of the latter was re - defined as a product of the filter transmission and detector response furnished by the respective manufacturers . the electron multiplication ccd ( em ccd ) luca - s used in hrcam failed in 2014 july . it was sent to its manufacturer ( andor ) for repair and received back in 2014 december . for the two runs in the fall of 2014 , we used the em ccd luca - r , kindly loaned by g. cecil . it has a larger format of 1004@xmath21002 pixels , with smaller pixels of 8@xmath3 m compared to luca - s ( 658@xmath2496 , 10@xmath3 m pixels ) . the second lens in hrcam was replaced by the achromatic doublet of 75-mm focal length to get the pixel scale of 14.33mas . the detector software driver was also upgraded on this occasion . although the two luca cameras look exactly the same and come from the same manufacturer , their ccds are radically different . luca - s uses a line - transfer ccd , while luca - r has a frame - transfer ccd . we found that luca - r had poor charge transfer efficiency ( cte ) in the vertical direction . weak signals such as cosmic - ray hits and dark current from hot pixels are smeared vertically over @xmath4 5 pixels , while strong signals are transferred with less spread . this results in the signal - dependent loss of resolution in the vertical direction ( along ccd columns ) , which was accounted for in the data processing by an additional adjustable parameter . this problem was revealed unexpectedly during the 2014 october run . in the next run , we placed stars closer to the line register , improving slightly the vertical cte by reducing the number of charge transfers . we studied the distribution of the bias signal of both cameras . it is very well modeled by the sum of a gaussian component ( readout noise ) and a negative exponent that corresponds to the single - electron events amplified stochastically by the gain register . most events are generated by the ccd clocking ( clock - induced charge , cic ) typical for em ccds . the width ( decrement ) of the exponential term depends on the em gain . the rate of cic events is found as the ratio of the areas below these two terms . for luca - r , the readout noise is 4.5 analog - to - digital units ( adu ) and the cic rate is 0.03 . the cic events are not smeared vertically because no charge transfer is involved . for luca - s , the readout noise is 9 adu , the exponential decrement is 30 adu ( for the gain setting of 200 used here ) , and the rate of cic events is 0.13 per pixel . about half of the cic events are removed by the threshold of 17 adu applied in the power - spectrum calculation . the soar telescope suffers from 50-hz vibration ( see tmh14 ) . the vibrations are non - stationary , causing variable loss of resolution if the 20-ms exposure time is used . most data were taken with an exposure of 5ms , sampling 1/4 of the vibration period . the resolution is then recovered and the effect of vibrations becomes less dramatic . it disappears completely at an exposure time of 2ms , used on the brightest targets . with such short exposures , the power spectra extend to the cutoff frequency and are very symmetric . the luca - r detector suffering from the cte problem was used mostly with a 10-ms exposure as a compromise between vibrations and poor cte at low signal levels . the observations consist of taking a cube of 400 images of 200@xmath2200 pixels each . for binaries wider than 15 , the 400@xmath2400 format was used . each object and filter combination is normally recorded twice , these data are processed independently , and the result is averaged . we used 2@xmath22 detector binning on the faintest targets observed in the @xmath1 filter , with a minor under - sampling and loss of resolution . measurements of binaries made with and without binning agree well mutually . faint red dwarfs with @xmath5 still produced useful data . the observing time for this program was allocated through noao ( proposals 2013b-0172 , 2014a-0038 , 2014b-0019 , 8 nights total ) and by the chilean national tac ( proposal cn2014b-27 , 4 nights ) . l l r r r 1 & 2014 jan 15 - 16,22 & 1.26 & 15.23 & 309 + 2 & 2014 mar 7 - 8 & @xmath6 & 15.23 & 201 + 3 & 2014 apr 19 - 20 & @xmath7 & 15.23 & 385 + 4 & 2014 oct 5 - 7 & @xmath8 & 14.33 & 553 + 5 & 2014 nov 7 - 8 & 0.66 & 14.33 & 253 + 6 & 2015 jan 10 - 11 & @xmath90.20 & 15.23 & 248 + table [ tab : runs ] lists the observing runs , the calibration parameters ( position angle offset @xmath10 and pixel scale in mas ) , and the number of objects covered in each run . the calibration of angle and scale was done by referencing to the sam imager , as described in tmh14 . run 1 was affected by transparent clouds . two hours on 2014 jan 22 were added as a backup program . during this run , the nasmyth rotator of soar had a large offset of + 2.8@xmath11 . the seeing was mediocre to poor . the instrument was not dismounted between runs 1 and 2 , but the nasmyth rotator was re - initialized , explaining the difference in @xmath10 . the sky was clear , and two hours were added on march 7 as a backup from other program . the sky was also clear on the two nights of run 3 , with seeing good to excellent . runs 4 and 5 used the substitute detector luca - r and were affected by the cte problem . for this reason the observations were performed mostly in the @xmath1 filter with 10-ms exposure . all three nights of run 4 were clear , and mostly with good seeing ( the median full width at half maximum , fwhm , of re - centered average images was 060 , the best fwhm was around 045 ) . the detector was removed between runs 4 and 5 in an attempt to improve the cte by tuning the electronic parameters ; however , this turned out to be impossible . during run 5 , the seeing varied from average to poor . in the worst case , the star did not fit in the 3 field , illuminating all pixels . in such cases , the image is truncated , the power spectrum contains bright vertical and horizontal lines , and the detection limits are poor . under poor seeing , data obtained with the 400-pixel field and 2@xmath22 binning are of better quality , without image truncation . the two nights of run 6 were clear , with seeing average to poor . the hrcam was returned to its original configuration with the luca - s detector . measurements with the sam internal light source confirmed that the pixel scale did not change compared to runs 13 . the data processing is described in tmh10 . as a first step , power spectra and average re - centered images are calculated from the data cubes . the auto - correlation functions ( acfs ) are computed from the power spectra . they are used to detect companions and to evaluate the detection limits . the parameters of binary and triple stars are determined by fitting the power spectrum to its model , which is a product of the binary ( or triple ) star spectrum and the reference spectrum . we used as a reference the azimuthally - averaged spectrum of the target itself in the case of binaries wider than 01 . for closer pairs , the `` synthetic '' reference was used ( see tmh10 ) . the signal - dependent vertical smearing of the image caused by the cte problem required a modification of the reference spectrum . we model the smearing by an additional gaussian term in the reference spectrum @xmath12 : @xmath13 where @xmath14 and @xmath15 are spatial frequency coordinates along ccd lines and columns respectively , @xmath16 is the diffraction - limited transfer function , @xmath17 is the cutoff frequency , @xmath18 , @xmath19 and @xmath20 describe the speckle signal attenuation ( both are negative , @xmath19 is related to seeing ) , and the new parameter @xmath21 is the fwhm of the cte smear in the column direction in pixels . this `` elliptical '' synthetic reference was used for all binaries , both wide and close , observed in runs 4 and 5 . the parameters of the reference @xmath22 were fitted to the power spectrum jointly with the binary parameters @xmath23 . in the case of close binaries , the parameters of the elliptical reference and of the binary itself are mutually correlated , so the resulting measures could be biased . the close binary star b 430 ( _ 19155@xmath92515 _ ) was observed with position angles of the instrument differing by @xmath24 . the `` fringes '' in the power spectrum had different orientation relative to the ccd . however , the binary parameters @xmath25 fitted to these power spectra are mutually consistent , ( 2837 , 00673 , 0.785 ) and ( 2850 , 00690 , 0.722 ) . therefore , the cte effect is , to first order , accounted for by the modified reference model . still , measurements of close binaries from runs 4 and 5 ( 2014.77 and 2014.86 ) should be taken with some caution . wide binaries resolvable in the re - centered long - exposure images are processed by another code that fits only the magnitude difference @xmath26 , using the binary position from speckle processing . this procedure corrects the bias on @xmath27 caused by speckle anisoplanatism and establishes the correct quadrant ( flag * in the data table ) . for run 5 , we also calculated shift - and - add ( saa or `` lucky '' ) images , centered on the brightest pixel in each frame and weighted proportionally to the intensity of that pixel . all frames without rejection were co - added . binary companions , except the closest and the faintest ones , are detectable in these saa images , helping to identify the correct quadrant . such cases are marked by the flag q in the data table . quadrants of the remaining binary stars are guessed based on prior data or orbits , not measured directly . calibration of scale and angular offset in speckle interferometry is challenging because the accuracy of modern measurements with 4-m telescopes exceeds the accuracy of even the best orbits . comparison with ephemerides is useful only as a sanity check . recognizing this problem , we observed several wide binaries during each run . their motions are slow and can be modeled by linear functions of time . these models can then be used to check the calibration of the archival data . this approach tests only the mutual consistency of speckle runs , rather than absolute calibration of the whole data set . we selected 41 binaries wider than 05 that do not contain resolved subsystems and were observed during at least 4 runs each . the angle and separation of each binary is approximated by linear functions of time . if the fitted slope is less than twice its rms error , a zero slope is assumed . deviations from these models are interpreted as calibration corrections . for each run , they are median - averaged . we then iterate by fitting new linear models corrected for the run s systematics , determining new calibration parameters of the runs , etc . figure [ fig : cal ] shows the offsets in angle and scale determined by this procedure for those of 26 runs where at least @xmath28 calibrators were observed . run 0 refers to observations at the blanco telescope in 2008.5 ( tmh10 ) . runs 16 in this paper correspond to numbers 2126 in the plot . vertical bars show the rms scatter between calibrators ( not errors of the mean , which are smaller by @xmath29 ) . a typical rms scatter is 01 in angle and 0.3% in scale . this analysis reveals good mutual consistency of @xmath45000 speckle measurements produced by hrcam to date . the largest systematics are found in runs 10 and 11 ( 2011.9 and 2012.1 ) . the published data can be corrected by subtracting the angular offsets found here and dividing the separations by the new scale factors . we plan to observe more calibrators and publish improved systematic corrections in the future . the data of this paper rely only on the original instrument calibrations and _ are not corrected _ for the offsets found so far . this study gives an estimate of the accuracy of speckle measurements . after correction for run systematics , the median rms deviations from the models for the 41 calibrator binaries are 01 in angle and 1.9mas in separation . the actual errors should be a factor of @xmath41.5 larger ( @xmath43mas ) , considering that the number of independent measurements is about twice the number of fitted parameters . this study will be repeated in the future with additional speckle runs and calibrators . the data tables have almost the same format as in the previous papers of this series . they are available in full only electronically . table 2 lists 1636 measures of 1218 resolved binary stars and subsystems , including 56 newly resolved pairs . the columns of table 2 contain ( 1 ) the wds @xcite designation , ( 2 ) the `` discoverer designation '' as adopted in the wds , ( 3 ) an alternative name , mostly from the _ hipparcos _ catalog , ( 4 ) besselian epoch of observation , ( 5 ) filter , ( 6 ) number of averaged individual data cubes , ( 7,8 ) position angle @xmath30 in degrees and internal measurement error in tangential direction @xmath31 in mas , ( 9,10 ) separation @xmath32 in arcseconds and its internal error @xmath33 in mas , and ( 11 ) magnitude difference @xmath26 . an asterisk follows if @xmath27 and the true quadrant are determined from the resolved long - exposure image ; a colon indicates that the data are noisy and @xmath26 is likely over - estimated ( see tmh10 for details ) ; the flag q " means the quadrant is determined from the saa image . note that in the cases of multiple stars , the positions and photometry refer to the pairings between individual stars , not the photo - centers of subsystems . for stars with known orbital elements , columns ( 1214 ) of table 2 list the residuals to the ephemeris position and code of reference to the orbit adopted in the sixth catalog ( * ? ? ? * hereafter vb6 ) . new and revised orbits computed in this work ( section [ sec : orb ] ) are referenced as `` this work '' . we did not use image reconstruction and measured the position angles modulo @xmath34 ( except the saa images in run 5 ) . plausible quadrants are assigned on the basis of orbits or prior observations , but they can be changed if required by orbit calculation . for triple stars , however , _ both _ quadrants of inner and outer binaries have to be changed simultaneously ; usually the slowly - moving outer pair defines the quadrant of the inner subsystem without ambiguity . table 3 contains the data on 441 unresolved stars , some of which are listed as binaries in the wds or resolved here in other filters . columns ( 1 ) through ( 6 ) are the same as in table 2 , although column ( 2 ) also includes other names for objects without discoverer designations . for stars that do not have entries in the wds , fictitious wds - style codes based on the position are listed in column ( 1 ) . column ( 8) is the estimated resolution limit , the largest of the diffraction radius @xmath35 and the vertical cte smear @xmath21 ( applicable to runs 4 and 5 only ) . columns ( 8,9 ) give the @xmath36 detection limits @xmath37 at @xmath38 and @xmath39 separations determined by the procedure described in tmh10 ( please note that this is not the resolution of the observations ) . when two or more data cubes are processed , the largest @xmath27 value is listed . the last column marks by colons noisy data mostly associated with faint stars . in such cases , the quoted @xmath26 might be too large ( optimistic ) ; however , the information that these stars were observed and no companions were found is still useful for statistics @xcite . table 4 lists 56 newly resolved pairs . the last two columns of table 4 contain the spectral type ( as given in simbad or estimated from absolute magnitude ) and the _ hipparcos _ parallax ( * ? ? ? * hereafter hip2 ) . fragments of acfs of newly resolved triple systems are shown in figure [ fig : acf ] . we comment on the newly resolved binaries below . the following abbreviations are used : pm proper motion , cpm common proper motion , rv radial velocity , sb1 and sb2 single- and double - lined spectroscopic binaries , int4 - 4th interferometric catalog @xcite.4040 ( tok 412 ac ) was independently discovered by sana et al . ( 2014 , apjs , 215 , 15 ) and is designated in the wds as sna 60 ac . by error , the newly discovered close companion in 04308@xmath95727 ( tok 429 aa , ab ) was omitted from table 4 , figure 2 and section 3.2 , but its measurements are found in table 2 . ] _ 01379@xmath98259 . _ hip 7601 is a nearby ( 27pc ) dwarf also known as gj 67.1 or hr 512 . according to @xcite , it is a young spectroscopic triple detected in x - rays ( 1rxs j013755.4@xmath9825838 ) . gcs also recognized the star as sb2 . spectroscopic monitoring ( tokovinin , in preparation ) shows that all three components are similar stars of approximately one solar mass . here we resolved the outer subsystem ab and observed its fast motion . in fact it was already resolved at soar on 2011.036 at 2210 and 0044 , but this low - quality observation has not been published . the available data indicate that the period of ab is 1.35years ; the pair completed nearly two revolutions since its first resolution in 2011 . _ _ hip 10096 is sb2 according to gcs . the separation corresponds to a period of @xmath45yr . _ 03046@xmath95119 . _ hip 14307 and 14313 form the 38 pair dun 10 ab belonging to the fg-67 sample . the component b is resolved here at 019 . the estimated period of ba , bb is @xmath425yr . no motion is seen in 2 months . the subsystem ba , bb is also manifested by asymmetric line profiles ( tokovinin , in preparation ) . the a component has never been observed at high angular resolution . _ 04386 - 0921 . _ hip 21265 is an x - ray source and an sb2 according to gcs ( one observation only ) . it is resolved here at 56mas , which corresponds to a period of @xmath45yr . _ 04469 - 6036 . _ first resolution of hip 22229 at 0043 , @xmath40 . the measure is uncertain , but the elongation is confirmed with 2-ms exposure and is not seen in other observed stars , so it is not caused by vibration . this is a triple system containing an eclipsing binary al dor of algol type with an eccentric orbit @xcite . the separation implies an orbital period of the outer system on the order of 5yr . _ _ this is hip 26237 , hd 37018 , hr 1892 , 42 ori , a young star in orion which has not been observed at high angular resolution so far , according to int4 . we resolved the known binary ab = da 4 and discovered the spectacular subsystem aa , ab at 016 ( figure [ fig : acf ] ) . _ _ hip 30995 is an sb2 according to gcs . the new companion at 02 with @xmath41 mag is unlikely to correspond to the sb2 , so the system is probably triple . _ 06497 - 7433 . _ hip 32735 has an acceleration detected in hip2 , but no rv data . the resolved 03 pair implies a period of @xmath480yr . the spectral type k0iv given in simbad may be inaccurate because the luminosity and color of the components correspond to main sequence stars with masses of 1.0 and 0.6 @xmath42 . _ _ the 02 pair hds 947 had not been observed since its discovery by _ it is revealed as a triple system ( figure [ fig : acf ] ) with comparable separations between components . hds 947 probably corresponds to ab , while the fainter c component is new . _ 07038 - 4334 . _ hip 34052 = hd 53680 = hip 34065c = gj 264 is a spectroscopic and astrometric binary for which @xcite give an orbit with @xmath43d = 4.62yr and an estimated semi - major axis of 015 . there is also an astrometric orbit @xcite with @xmath44yr that predicts @xmath45 for the moment of our first observation . the binary is resolved at 023 , @xmath46mag and shows some orbital motion . the binary makes a quadruple system together with components a and b ( gj 264.1 and 264 ) that form a 205 pair at 185 from c. the b component was also observed here and found unresolved . _ 07294 - 1500 _ is another nearby multiple system . the main component hip 36395 is a visual binary with a known 728-yr orbit , also measured here . the c component ( nltt 17952 ) at 204 is physical , and yet another cpm component f is found at 1072 @xcite , while the wds components d and e are optical . we observed c and resolved it at 009 . the orbital period of ca , cb is on the order of 6yr , estimated masses are 0.6 and 0.4 @xmath42 . we also targeted f and did not resolve it . the magnitudes and colors of c and f are quite similar . _ 07304 + 1352 _ is a quadruple system . the 77 pair ab ( stf 1102 ) is hip 36485 , the cpm component d = hip 36497 = hd 59450 is located at 112 from it , while the wds components c and e are optical . the physical nature of ad is established by common pm , distance , and rv . d is a known sb1 with @xmath47d=7.4yr @xcite and an estimated semi - major axis of 93mas , also an acceleration binary . we resolved the da , db pair at 011 , @xmath48=2.6 . the minimum mass of db derived from its sb orbit is 0.27 @xmath42 , while we estimate the masses of da and db as 1.05 and 0.6 @xmath49 from their luminosity . a previous non - resolution of d is reported in int4 ; it was also unresolved with robo - ao @xcite . _ 07312 + 0210 , _ hip 36557 = hd 59688 . according to observations by d. latham ( 2012 , private communication ) , this is a spectroscopic triple with an inner period of 70d ( double - lined , also detected by gcs , mass ratio 0.7 ) and an outer period of 2007d or 5.5yr . the outer system is also detected by astrometric acceleration @xcite . we resolve it here at 0057 , @xmath48=2.0 , @xmath21=2.5 mag , and see the orbital motion . the estimated mass of ab is 0.88 @xmath42 . the semi - major axis of the 70-d inner binary aa , ab is 7mas , so accurate measurements of ab can detect the sub - motion to determine the orientation of the inner orbit . _ 08021 - 1710 . _ this is the high - pm m - dwarf lp 784 - 12 ( hip 39293 ) at 30pc from the sun . a new distant component c was found at 18 in addition to the known pair hds 1140 which closed from 04 in 1991.25 to 033 now ( figure [ fig : acf ] ) . it is confirmed as physical by its fixed position during one year , the quadrant of the triple was determined in run 5 . _ 08447 - 2126 . _ like the previous object , the late - type nearby binary hds 1260 was discovered by _ hipparcos _ and is expected to move rapidly ( hip 42910 , bd@xmath50 2665 ) . it was targeted at soar for the first time . to our surprise , the object turned out to be a resolved triple , with the secondary being a 016 pair of equal stars ( figure [ fig : acf ] ) . _ hipparcos _ failed to recognize the triple nature of this star . the estimated period of bc is 15yr . the outer pair ab has closed from 08 to 05 and moved in position angle since its discovery . the separations between components are comparable , so this triple system may be interesting dynamically . _ _ hip 46572 is called a `` high proper motion star '' in simbad , although its pm and rv are actually quite moderate . the binary ab has moved little in the 24 years since its resolution by _ we discover the subsystem bc ( figure [ fig : acf ] ) with an estimated period of @xmath4100yr . _ _ hip 48906 is a double - lined binary according to the gcs , first resolved here at 64mas . the period should be @xmath420yr . _ 10056 - 8405 . _ hip 49442 = hd 88948 is a nearby dwarf in the 39 visual binary hj 4310 ab . according to gcs , the rv of the main component a varies by 3.5 km s@xmath51 . here it is resolved into a 018 aa , ab pair with an estimated orbital period of 25yr . no astrometric acceleration was detected , however . the visual secondary b was targeted separately and found unresolved . _ 10070 - 7129 . _ hip 49546 is an astrometric binary of 1.5yr period @xcite with variable rv . the period corresponds to a semi - major axis of 25mas . the star is resolved here tentatively at 26mas ( with 2-ms exposure ) . this resolution is below the diffraction limit and needs confirmation . the measured position angle of @xmath52 is close to @xmath53 predicted by the astrometric orbit . _ 10223 - 1032 . _ hip 50796 is a single - lined and astrometric binary according to @xcite , with period 570.98d ( 1.56yr ) , @xmath54 km s@xmath51 , @xmath55 . the _ hipparcos _ parallax corrected for the binary motion is 20.6@xmath561.9mas . the spectroscopic secondary companion is over - massive , most likely a close pair of m - dwarfs . if so , the new speckle companion at 166 with a period on the order of 500yr makes the system quadruple . the speckle companion might contribute to the ir excess found by torres . the system is an x - ray source , and is possibly young . _ 10530 + 0458 . _ latham ( 2012 , private communication ) identified hip 5212 = hd 94292 as sb2 with a period of 8.3yr and a highly eccentric orbit . it is resolved here securely at 50mas . _ 12176 + 1427 . _ hip 59933 has a variable rv according to gcs . the 32mas separation corresponds to an orbital period of @xmath42yr . however , the separations in the @xmath0 and @xmath1 filters are somewhat discordant ; further confirmation is needed . _ 12250 - 0414 . _ hip 60574 is a spectroscopic triple with periods 14d and 22yr ( latham , 2012 , private communication ) , also an acceleration binary . we resolved the outer pair at 022 separation , matching the spectroscopic period . the lines of the visual secondary ab could potentially be detected in the spectrum by cross - correlation , leading eventually to a full 3-d orbit . _ 12528 + 1225 . _ hip 62933 ( 41 vir ) was observed on request by f. fekel who studies its spectroscopic orbit . apparently it is resolved for the first time . _ 13132 - 0501 . _ hip 64499 has a variable rv , with a preliminary spectroscopic orbit of 17yr period ( latham 2012 , private communication ) . it is resolved at 01 and shows no motion in one month . _ 13321 - 1115 . _ no previous indication of binarity was available for hip 66018 , apart from the cpm companion b at 84 . we discovered another faint component ab at 089 , @xmath48=4.6 , likely to be physical ( low background density ) . the b component ( @xmath57 ) was targeted , but its speckle signal was weak and no obvious close companions to b were found . _ 13344 - 5931 . _ hip 66230 has a variable rv in the gcs and an astrometric acceleration . we resolved it at 01 , @xmath48=2.5 , estimated period @xmath410yr . the pair moved by @xmath58 in one month . _ 13382 - 2341 . _ hip 66530 has a variable rv according to the gcs . it is resolved at 016 in the @xmath1 band only , estimated period @xmath420yr . this is a triple system , considering the cpm companion b at 28 ( lds 4385 ab ) . _ 13401 - 6033 . _ hip 66676 ( a ) and hd 118735 ( b , g6v , @xmath59 ) at 77 share common pm ( although it is small , 58mas yr@xmath51 ) which , together with photometry , indicates with high probability that it is a physical pair ab @xcite . we targeted the secondary component b and found it to be a resolved triple . the faint star c , at 092 from b , is itself a close 016 pair ca , cb ( figure [ fig : acf ] ) . note that this is a region of the sky with very high stellar density , raising suspicion that the ca , cb pair might be a random background object . re - observation of the triple in 2015 ( to be published ) shows , however , that it is physical because the center of c=(ca , cb ) moves relative to b with a speed of 11mas yr@xmath51 ( or 3 km s@xmath51 ) , compatible with the expected orbital motion of bc and much less than the system s pm of 58mas yr@xmath51 . the estimated period of ca , cb is @xmath430 years , the period of bc is @xmath4300 years . _ 13495 - 2621 . _ hip 67458 is a double - lined chromospherically active binary with orbital period of 7.2d ( latham , 2012 , private communication ) . we found a faint tertiary companion at 073 with an estimated orbital period on the order of 100yr . the speckle survey of chromospherically active stars by @xcite did not detect this tertiary , lacking the dynamic range of hrcam . _ 14014 - 3137 . _ hip 68507 is an acceleration binary with a variable rv resolved here at 006 . the period of aa , ab is on the order of 5yr . there is a faint physical companion b at 67 found in 2mass @xcite . another visual companion at 14 , see 195 , is optical , as revealed by its fast relative motion . _ 14382 + 1402 . _ hip 71572 is an acceleration binary without rv data . it is found to be a tight 90-mas pair with a small @xmath26 and an estimated period under 10yr ( some motion is seen in one month ) . very likely it can be studied as a double - lined sb . _ 14464 - 3346 . _ hip 72235b is located at 9 from the primary star and shares its proper motion ( ab = hds 2082 ) . pre - discovery measurements of this _ hipparcos _ pair were published by @xcite . the rv of a may be variable ( latham , 2012 , private communication ) . the star b turns out to be a 04 binary ba , bb with masses of 0.70 and 0.17 @xmath42 estimated from the luminosity . its period is on the order of 100yr . the whole system could thus be quadruple . _ 15362 - 0623 . _ hip 76400 is identified by the gcs as an sb2 with a mass ratio @xmath60 , but there is no spectroscopic orbit available . we resolved the 019 pair with @xmath61=3.9 , indicating a mass ratio of @xmath40.5 and an orbital period of @xmath430yr . very likely the resolved binary does not match the spectroscopic double - lined system . considering the cpm component b at 80 @xcite , this system could be a quadruple with a 3-tier `` 3 + 1 '' hierarchy . _ 15367 - 4208 . _ hip 76435 is a g5v star from the fg-67 sample . its companion c ( ac = fal 78 ) at 135 is physical , while the _ hipparcos _ companion b at 43 is not seen in the 2mass images and has not been confirmed otherwise . we targeted c and resolved it into a close binary . estimated masses of ca and cb are 0.70 and 0.66 @xmath42 , period @xmath44yr . _ 16142 - 5047 . _ hip 79576 has a variable rv ( gcs ) . the 79mas separation implies an orbital period of @xmath45yr . this is the high - pm star ltt 6467 with a low metallicity [ fe / h]=@xmath90.78 . @xcite found no cpm companions . _ 16195 - 3054 . _ hip 79980 and hip 79979 form a 23 cpm pair ab where the f6iii primary is slightly evolved , while the g1/g2 secondary is closer to the main sequence , but still above it . the _ hipparcos _ parallax of b , @xmath62mas , is obviously wrong , so we assume the parallax of the primary , 20.7mas . the rv(b ) is variable ( gcs ) , and we resolve it into a 40mas pair with an estimated period on the order of 1yr . the pair moved by 10@xmath11 in a month . binary motion is the likely cause of the incorrect _ hipparcos _ parallax . _ 16454 - 7150 . _ hip 82032 is located in a crowded field , so the new faint ab companion found here at 13 could be optical . the star was observed because of its suspected variable rv ( gcs ) , but the newly found companion , even if physical , is too distant to explain this variability . another visual companion b at 115 ( ab = b 2392 ) is optical , as evidenced by its fast relative motion . the star is on the harps exo - planet program . _ 16563 - 4040 . _ hip 82876 is a distant o7v star . the 026 pair ab ( hds 2394 ) was measured among other neglected binaries . we found another faint companion c at 146 ( figure [ fig : acf ] ) . the star has an extensive literature , including multiplicity surveys with speckle interferometry and rv @xcite . owing to the large distance , no detectable orbital motion is expected . indeed , the ab pair was measured with hrcam in 2008.5 at the same position as it is now . those observations in the @xmath0 band did not detect the companion c owing to a lower signal - to - noise ratio . _ 17054 - 3346 . _ the rv of hip 83612 varies by 52 km s@xmath51 ( gcs ) . it is a very close pair with an estimated period of @xmath41.5yr , and the measure near the diffraction limit derived from the elongated power spectrum is tentative . the _ hipparcos _ parallax is likely biased by the binary . the visual component b = hip 83609 ( ab = wno 5 ) at 25 is optical . _ 17098 - 1031 . _ hip 83962 = hr 6375 has a variable rv according to the gcs , while gorynya ( 2013 , private communication ) detected double lines . it is resolved tentatively at 33mas with @xmath63 mag ( the 5-ms exposure makes it unlikely that the asymmetry is caused by vibrations ) . the separation corresponds to an orbital period on the order of 1yr , which could bias the _ hipparcos _ parallax . however , @xcite consider the star as single . the new pair was not resolved in 2014.3 ; presumably it became closer . _ _ hip 85342 and hip 85326 form a physical pair ab at 127 separation ( common pm , rv , and parallax ) . the b component = hip 85326 has a variable rv and an astrometric acceleration which could hardly be produced by the 1 speckle companion bb found here , owing to its long estimated period of @xmath4300yr . it seems that b is triple and the whole system is quadruple . this companion bb was not detected in the previous speckle observations because it is red : @xmath64 , @xmath65mag ; its color matches a dwarf star at the same distance as the system . however , the field is crowded and the newly found companion could still be optical . _ 17266 - 3258 . _ hip 85360 is an acceleration and spectroscopic binary . its preliminary spectroscopic period ( latham 2012 , private communication ) corresponds to a semi - major axis of 80mas . the star is chromospherically active and possibly young . the faint companion found here at 116 is most likely optical , as the field is extremely crowded . re - observation within a year will resolve its status . _ 17341 - 0303 . _ hip 85963 has a variable rv and is an acceleration binary . the 91mas separation implies an orbital period of @xmath410yr ; the estimated masses are 1.37 and 0.81 @xmath42 . despite extensive literature ( 51 references in simbad ) , there is no published spectroscopic orbit , while several high - resolution spectroscopic studies addressed the abundance . _ 17342 - 1910 . _ b 1863 is a known close binary which has been unexpectedly found to be a triple ( figure [ fig : acf ] ) . the new distant component c is detectable also in the @xmath0 filter , but we measured only the inner binary in @xmath0 . the star was observed at the blanco telescope in 2008.5397 , and the pair actually measured then was ac , at 1338 , 0217 , @xmath66 ( same as now , see table [ tab : new ] ) . the inner pair ab with a smaller @xmath67 was unresolved in 2008.5 , while it is clearly resolved now . owing to the large distance from the sun , we expect only a slow motion , so even the inner pair observed since 1929 may not yet be ready for computing its first orbit . _ 17342 - 5454 . _ hip 85969 has a variable rv according to the gcs and confirmed by @xcite . the 055 separation implies a period on the order of 80yr . the star is on the exo - planet program at the anglo - australian telescope . _ 18243 - 0405 . _ the neglected pair ysc 67 turned out to be a new triple ( figure [ fig : acf ] ) . the outer 035 binary ab was known previously , now we detect elongation that implies an inner subsystem aa , ab . we compared with stars in the same area of the sky observed before and after to assure that the elongation is not of instrumental origin ; it is seen in two filters . _ 18267 - 3024 . _ hip 90397 is an acceleration binary with a variable rv ( gcs ) , resolved here at 69mas ( estimated period @xmath44yr ) . the star is targeted by exo - planet programs . _ 18346 - 2734 . _ hip 91075 was noted as a double - lined binary by the gcs ( one observation only ) . the separation of 86mas implies a period of @xmath410yr , while the double - lined observation matches the moderate magnitude difference ( estimated masses 1.04 and 0.77 @xmath49 ) . the star is an x - ray source , so a future combined orbit could determine masses for testing evolutionary models of young stars . the sky around the object is quite crowded ; the 9 companion i 1026 is optical ( it moves too fast ) . _ 19206@xmath90645 . _ hip 95068 is the neglected _ hipparcos _ binary hds 2735 ab , a distant k - type giant . we did not resolve this 01 binary , which remains unconfirmed , but detected instead another faint star at 1 . the stellar background is crowded , the pm is small , and the status of the new companion remains uncertain . _ 19209@xmath93303 . _ hip 95106 and 95110 form the 137 pair hj 5107 ab . the rv variability of a was suspected by the gcs , it is now resolved as a 027 binary with estimated period of @xmath435yr . the component b was also observed and found unresolved . it contains a spectroscopic pair ( tokovinin , in preparation ) , the whole system is quadruple . _ 19221 - 2931 . _ hip 95203 is another acceleration binary with variable rv resolved here . the relatively large separation of 077 corresponds to a period of @xmath4180yr . the actual period can be as short as 60yr if the pair is seen now near its maximum separation ( it would then have been closer at the time of the _ hipparcos _ mission ) . most likely , however , the faint visual companion found here and the spectroscopic / acceleration pair make a triple system . there is another companion hip 95164 at 435 . the status of this wide pair ( is it a real binary or just two members of a moving group ? ) is not clear , but the association of those stars leaves no doubt ( common rv , pm , and parallax ) . _ 19409 - 0152 . _ hip 96834 has a spectroscopic orbit with a period of 1yr and expected semi - major axis of 27mas ( latham 2012 , private communication ) . we resolved this pair , although the measurement near the diffraction limit is uncertain . there are some unsolved questions , however . why , despite the small magnitude difference @xmath68 mag , were double lines not seen ? why , despite the 1-yr period , was the _ parallax not strongly affected and the star appears to be on the main sequence ? _ 19453 - 6823 . _ we resolved the secondary component of hds 2806 ab into a close pair ba , bb ( figure [ fig : acf ] ) . this is a k3 dwarf within 50pc from the sun . the pair ba , bb should be fast and turn around in about 10yrs . _ 22259@xmath97501 . _ hip 110712 and 110719 form the 206 pair dun 238 ab . we observed both components . the a component has a variable rv according to the gcs , but not confirmed by @xcite ; it was unresolved . the newly found pair ba , bb has a period on the order of 10yr . considering the distant companion c found by @xcite , the system contains at least 4 stars . the measurements reported here served to improve or compute anew orbits of 194 binary stars . the orbital elements in standard notation and their formal errors are given in table 5 . its last column lists the vb6 code to a prior orbit if it exists ( astrometric orbits in brackets ) . errors of the elements ( except provisional orbits of grade 5 ) are given in the following line . for the elements that were fixed ( from , for example , a spectroscopic orbit ) , an asterisk replaces the error . as discussed in tmh14 , orbit improvements range from minor `` cosmetic '' upgrades to some quite drastic revisions . blanks in the last column indicate the 43 first - time orbit solutions ( plus three with prior astrometric orbits ) . the grades are assigned according to the general grading rules adopted in vb6 @xcite . most orbital elements are derived using the usno orbit code . we do not publish figures for all new orbits here , as they will be available online in vb6 . figure [ fig : rev ] illustrates orbit revisions ranging from dramatic to minor . three first - time orbits are presented in figure [ fig : first ] . below we comment on some pairs . _ _ wsi 97 is a single - lined nearby binary . using the radial velocities measured by d. latham ( 2012 , private communication ) , we computed a combined orbit ( the previous visual orbit reported by @xcite had a wrong period ) . the inclination is close to @xmath34 and had to be fixed in order to match the rv amplitude . the complete orbit including rvs will be published later . _ _ chr 20 is a hyades binary for which @xcite published an sb2 orbit . we combined his rvs with the speckle data , resulting in a very accurate period . the combined orbit corresponds to a mass sum of 2.1 @xmath70 and an orbital parallax of 22mas , in good agreement with the hip2 parallax of @xmath71mas . _ _ mca 18 aa , ab has an sb1 orbit with @xmath72yr . the orbit given here uses only the speckle data , however . _ _ the period of tok 194 matches the astrometric orbit of @xcite . the mass sum in this pair composed of a giant primary and possibly an a - type secondary is 5.9 @xmath42 . the measure on 2011.93 was ignored as spurious ( it was affected by vibrations ) . _ _ tok 195 is the bright star a pup ( hd 64440 , hr 3080 ) known as a spectroscopic binary . however , examination of the rv data reveals that the orbit by @xcite is only approximate . the binary is difficult to measure , always close to the diffraction limit and with @xmath73 . instead of the spectroscopic period of 6.99yr , our orbit has @xmath74yr and is still preliminary . more rv coverage is obviously needed . _ _ hu 1443 a , bc is a triple system . we provide the first very tentative orbit for the outer binary , but note its large residuals from the recent measures of ab . strictly speaking , the orbit should describe the motion of the center - of - gravity of bc around a , rather than the measures of ab . such refinement was made for the orbit of a 2592 ab _ ( 17156@xmath90949 ) _ , but it is not warranted for this preliminary orbit . _ _ tok 199 is marked as an sb2 in the gcs , while latham ( 2012 , private communication ) derived an orbital period of 916d , now independently confirmed by our orbit ( figure [ fig : first ] ) . _ _ this is the triple system hip 84430 . we computed the first orbit of the secondary subsystem ba , bb which was discovered at soar in 2009 and has just completed one full revolution since . its separation is always close to the diffraction limit . adopting a mass sum of 2.6 @xmath42 for ba , bb , the resulting dynamical parallax is 7.6@xmath561.5mas , while the hip2 parallax is 4.9@xmath560.9mas . the latest orbit of the outer pair a 2592 ab published in tmh14 does not account for the fact that the speckle measurements at soar refer to a , ba and not to ab . here we give a more accurate solution that uses the positions of ab computed from the measures of a , ba under the assumption that ba and bb have equal masses . after this correction and orbit adjustment , the weighted residuals are 4.3mas in separation and 13 in angle . interestingly , there were a considerable number of speckle - interferometry measures of this pair obtained in the 1980-s and 1990-s at 4-m telescopes , but none of them recognized the subsystem ba , bb , despite its small @xmath26 . ignoring the multiplicity , the spectroscopic survey of @xcite determined a moderate axial rotation @xmath75 km s@xmath51 and detected the lithium line of 52.8m equivalent width which , together with the x - ray detection by rosat ( rastyc 1715 - 0948 ) , normally indicates youth . these authors do not mention this star in particular , but discuss a group of active lithium - rich giants in their sample , to which this system apparently belongs . even with the larger dynamical parallax ( instead of the hip2 parallax ) , all three resolved components of hip 84430 are located above the main sequence in the color - magnitude diagram . this multiple system is peculiar and merits further study . we present here a large set of new speckle interferometry measurements of close binary stars , mostly with southern declinations . the total number of measurements made with hrcam since 2008 now exceeds 5000 . this unique data set is used for calculation of 46 new orbits and for improvement of 148 known orbits . for comparison , the data in tmh14 resulted in 13 new orbits and in the improvement ( sometimes drastic ) of 45 previously known orbits . we demonstrate the good internal consistency of speckle astrometry with hrcam by repeated measurements of relatively wide binaries . typical errors are on the order of 3mas even for these wide pairs . the high angular resolution and dynamic range of hrcam give access to close binaries never resolved before . some of those objects had prior indication of binarity from variable rv or astrometric acceleration . in such cases , direct resolution allows us to estimate statistically orbital periods ( which are typically short ) and to evaluate masses . this clarifies the statistics of binary and multiple stars in the solar neighborhood . we also resolved a number of components in previously known nearby wide binaries , converting them into triple or higher - order hierarchies . a total of 56 newly - resolved pairs are reported here , ten of those being inner or outer subsystems in visual binaries ( figure [ fig : acf ] ) . most of those subsystems are totally unexpected . some of the newly resolved binaries or subsystems are interesting for various reasons , such as being young ( e.g. x - ray sources ) , having comparable separations and approaching the dynamical stability limit , such as hip 9497 with periods of 138 and 13.9 years ( tmh14 ) , or being targets of exo - planet programs . we thank the operators of soar d. maturana , p. ugarte , s. pizarro , and j. espinoza for efficient support of our program . g. cecil has kindly loaned us his luca - r detector which was used for five nights instead of our own broken camera . r.a.m . acknowledges support from the chilean centro de excelencia en astrofsica y tecnologas afines ( cata ) basal pfb/06 , and the project ic120009 millennium institute of astrophysics ( mas ) of the iniciativa cientfica milenio del ministerio de economa , fomento y turismo de chile . r.a.m also acknowledges eso / chile for hosting him during his sabbatical leave throughout 2014 . this work used the simbad service operated by centre des donnes stellaires ( strasbourg , france ) , bibliographic references from the astrophysics data system maintained by sao / nasa , and the washington double star catalog maintained at usno . 00026@xmath90829 & a 428 & hip 210 & 14.7632 & i & 2 & 317.6 & 0.3 & 0.1197 & 0.9 & 0.4 : & @xmath91.0 & 0.003 & this work + & & & 14.8535 & i & 2 & 315.6 & 1.4 & 0.1165 & 0.8 & 0.3 & @xmath92.5 & 0.000 & this work + 00028 + 0208 & bu 281 ab & hip 223 & 14.8561 & i & 2 & 160.3 & 0.2 & 1.5938 & 0.2 & 1.4 * & & & + 00058@xmath96833 & hds 4 & hip 488 & 14.7661 & i & 2 & 229.3 & 4.8 & 0.1971 & 1.9 & 1.9 & & & + 00115@xmath95545 & hds 25 & hip 927 & 14.7661 & i & 3 & 75.3 & 0.7 & 0.1546 & 0.6 & 0.8 & & & + 00135@xmath93650 & hds 32 & hip 1083 & 14.7662 & i & 2 & 226.3 & 1.0 & 0.2565 & 0.6 & 0.8 & & & + 00143@xmath92732 & hds 33 & hip 1144 & 14.8534 & i & 2 & 55.5 & 0.3 & 0.1117 & 0.3 & 0.9 & 0.6 & @xmath90.002 & this work + & & & 14.8534 & y & 1 & 55.4 & 0.6 & 0.1115 & 0.6 & 1.0 & 0.4 & @xmath90.002 & this work l l l c c c c c c c 00006@xmath96641 & gli 289 & hip 55 & 14.7661 & i & 2 & 0.039 & 1.95 & 3.07 + 00052@xmath96251 & hip 425 & hip 425 & 14.7661 & i & 2 & 0.039 & 2.78 & 4.10 + 00174@xmath95131 & hds 40 & hip 1393 & 14.7662 & i & 2 & 0.039 & 2.56 & 3.97 + 00250@xmath93042 & hip 1976 & hip 1976 & 14.8534 & i & 2 & 0.039 & 3.42 & 4.82 + 00301 + 0452 & hip 2358 & hip 2358 & 14.8561 & i & 2 & 0.039 & 2.28 & 3.69 + 00310@xmath93138 & hds 69 & hip 2433 & 14.7662 & i & 2 & 0.039 & 1.69 & 4.01 + 00324 + 0657 & mca 1 aa , ab & hip 2548 & 14.7632 & i & 2 & 0.039 & 3.01 & 4.25 l l l cc r r l lr 01379@xmath98259 & tok 426 & hip 7601 & 14.7661 & i & 112.8 & 0.0640 & 0.5 & g1v & 36.5 + 02098@xmath94052 & tok 427 & hip 10096 & 14.7689 & i & 218.3 & 0.0397 & 1.4 & f3v & 14.1 + 03046@xmath95119 & tok 428 ba , bb & hip 14313 & 14.7635 & i & 107.9 & 0.1911 & 0.5 : & g6v ? & 18.7 + 04386@xmath90921 & tok 387 & hip 21625 & 14.0457 & i & 47.0 & 0.0557 & 0.6 & f8 ? & 16.1 + 04469@xmath96036 & tok 388 & hip 22229 & 14.0456 & i & 266.2 & 0.0431 & 1.2 & f8v & 16.1 + 05354@xmath90450 & tok 430 aa , ab & hip 26237 & 14.8567 & i & 6.2 & 0.1589 & 0.2 & b1v & 3.7 + 06303@xmath95252 & tok 435 aa , ab & hip 30995 & 15.0288 & i & 176.1 & 0.2134 & 4.1 q & g4v & 14.0 + 06497@xmath97433 & tok 389 & hip 32735 & 14.0597 & i & 330.7 & 0.3343 & 3.1 & k0iv & 15.7 + 06499@xmath92806 & hds 947 ac & hip 32767 & 15.0289 & i & 231.3 & 0.4237 & 3.2 q & k1v & 14.6 + 07040@xmath94337 & tok 390 ca , cb & hip 34052 & 14.0431 & i & 10.8 & 0.2268 & 4.4 & k6v & 57.4 + 07294@xmath91500 & tok 391 ca , cb & hip 36395 & 14.1853 & i & 121.2 & 0.0902 & 1.2 & m0v & 28.5 + 07304 + 1352 & tok 392 da , db & hip 36497 & 14.0596 & i & 67.6 & 0.1143 & 2.5 & f8 & 22.0 + 07312 + 0210 & tok 393 & hip 36557 & 14.0433 & i & 168.6 & 0.0572 & 2.0 & g0 & 15.7 + 08021@xmath91710 & tok 394 ac & hip 39293 & 14.0460 & i & 332.5 & 1.8351 & 3.0 & m1.5 & 33.3 + 08447@xmath92126 & tok 395 bc & hip 42910 & 14.0460 & i & 184.6 & 0.1645 & 0.1 & k7v & 27.5 + 09299@xmath93629 & tok 440 bc & hip 46572 & 15.0317 & i & 90.4 & 0.1126 & 1.4 q & sdf2 & 5.0 + 09586@xmath92420 & tok 437 & hip 48906 & 15.0317 & i & 292.6 & 0.0640 & 1.3 q & f3v & 8.9 + 10056@xmath98405 & tok 396 aa , ab & hip 49442 & 14.0436 & i & 192.6 & 0.1760 & 1.4 & f8v & 15.6 + 10070@xmath97129 & tok 397 & hip 49546 & 14.3027 & y & 345.5 & 0.0263 & 1.7 & f5v & 16.7 + 10223@xmath91032 & tok 398 & hip 50796 & 14.3002 & i & 147.7 & 1.6626 & 2.8 & k4v & 25.8 + 10530 + 0458 & tok 438 & hip 53212 & 15.0292 & y & 194.9 & 0.0486 & 0.8 q & g5 & 11.9 + 12176 + 1427 & tok 399 aa , ab & hip 59933 & 14.3004 & i & 95.2 & 0.0429 & 1.0 & f8 & 17.5 + 12250@xmath90414 & tok 400 & hip 60574 & 14.1855 & i & 71.2 & 0.2194 & 2.6 & g5 & 21.5 + 12528 + 1225 & tok 401 & hip 62933 & 14.3004 & y & 141.3 & 0.0598 & 1.6 & a7iii & 16.4 + 13132@xmath90501 & tok 402 & hip 64499 & 14.1858 & i & 314.8 & 0.1062 & 1.8 & g5 & 19.2 + 13321@xmath91115 & tok 291 aa , ab & hip 66018 & 14.1858 & i & 159.4 & 0.8947 & 4.6 & g0 & 18.0 + 13344@xmath95931 & tok 403 & hip 66230 & 14.1857 & i & 314.3 & 0.1037 & 2.5 & g0v & 18.6 + 13382@xmath92341 & tok 404 aa , ab & hip 66530 & 14.1857 & i & 63.4 & 0.1570 & 3.9 & g4v & 20.6 + 13401@xmath96033 & tok 292 b , ca & hip 66676b & 14.1857 & i & 293.3 & 0.9178 & 3.7 & g6v & 16.9 + 13401@xmath96033 & tok 292 ca , cb & hip 66676b & 14.1857 & i & 5.2 & 0.1614 & 0.0 & g6v & 16.9 + 13495@xmath92621 & tok 405 & hip 67458 & 14.1858 & i & 15.0 & 0.7306 & 4.4 & g9v & 24.3 + 14014@xmath93137 & tok 159 aa , ab & hip 68507 & 14.3031 & i & 101.6 & 0.0652 & 2.1 & f5 & 16.2 + 14382 + 1402 & tok 406 & hip 71572 & 14.1859 & i & 153.9 & 0.0900 & 1.5 & g5 & 21.2 + 14464@xmath93346 & tok 407 ba , bb & hip 72235b & 14.3031 & i & 41.3 & 0.4116 & 4.1 & k4 & 24.0 + 15362@xmath90623 & tok 301 aa , ab & hip 76400 & 14.1860 & i & 78.1 & 0.1908 & 3.9 & g5 & 15.7 + 15367@xmath94208 & tok 408 ca , cb & hip 76435c & 14.3007 & i & 97.2 & 0.0562 & 0.4 & k4 & 20.6 + 16142@xmath95047 & tok 409 & hip 79576 & 14.1833 & i & 145.4 & 0.0791 & 3.0 & g8v & 24.2 + 16195@xmath93054 & tok 410 ba , bb & hip 79979 & 14.1833 & i & 85.0 & 0.0396 & 1.4 & g1/g2 & 20.7 + 16454@xmath97150 & tok 411 aa , ab & hip 82032 & 14.3034 & i & 267.2 & 1.3373 & 4.1 & f8v & 18.3 + 16563@xmath94040 & tok 412 ac & hip 82876 & 14.3034 & y & 144.0 & 1.4585 & 5.1 & o7v & @xmath90.7 + 17054@xmath93346 & tok 413 aa , ab & hip 83612 & 14.3035 & y & 82.3 & 0.0303 & 1.7 : & g1v & 18.4 + 17098@xmath91031 & tok 414 & hip 83962 & 14.1833 & y & 59.1 & 0.0325 & 1.8 : & f5iv & 24.2 + 17264@xmath94837 & tok 415 ba , bb & hip 85326 & 14.3035 & i & 118.6 & 1.0059 & 2.4 * & k1v & 20.7 + 17266@xmath93258 & tok 416 & hip 85360 & 14.3035 & i & 26.9 & 1.1597 & 5.9 * & g3v & 26.4 + 17341@xmath90303 & tok 417 & hip 85963 & 14.3010 & i & 88.9 & 0.0907 & 2.6 & f8 & 15.3 + 17342@xmath91910 & tok 418 ac & hip 85965 & 14.3036 & i & 133.6 & 0.2163 & 3.3 & f2v & 2.5 + 17342@xmath95454 & tok 419 & hip 85969 & 14.1833 & i & 72.1 & 0.5544 & 3.4 & g4v & 23.7 + 18243@xmath90405 & tok 420 aa , ab & hip 90198 & 14.3037 & i & 83.6 & 0.0477 & 0.8 & a3 & 8.7 + 18267@xmath93024 & tok 421 & hip 90397 & 14.3036 & i & 100.4 & 0.0690 & 2.5 & g0v & 21.4 + 18346@xmath92734 & tok 422 aa , ab & hip 91075 & 14.3037 & i & 101.7 & 0.0864 & 1.7 & g5v & 15.0 + 19206@xmath90645 & tok 432 ac & hip 95068 & 14.3011 & i & 56.4 & 0.9751 & 6.4 * & k0 & 3.0 + 19209@xmath93303 & tok 433 aa , ab & hip 95106 & 14.7683 & i & 160.0 & 0.2721 & 2.5 & g0.5v & 21.2 + 19221@xmath92931 & tok 423 aa , ab & hip 95203 & 14.3010 & y & 36.4 & 0.7672 & 5.8 * & g0iv & 19.6 + 19409@xmath90152 & tok 424 & hip 96834 & 14.3011 & i & 102.2 & 0.0355 & 1.3 : & f8 & 20.8 + 19453@xmath96823 & tok 425 ba , bb & hip 97196 & 14.3038 & i & 312.1 & 0.0829 & 1.4 & k3 & 21.3 + 22259@xmath97501 & tok 434 ba , bb & hip 110719 & 14.7657 & i & 58.1 & 0.1917 & 1.4 : & k5 & 43.4 l l rrr rrr r ccll 00026@xmath90829 & a 428 & 414.95 & 2005.35 & 0.503 & 192.2 & 0.505 & 111.9 & 226.8 & 4 & zul1984b + & & @xmath5620.96 & @xmath562.47 & @xmath560.013 & @xmath562.0 & @xmath560.021 & @xmath561.2 & @xmath564.2 & & + 00143@xmath92732 & hds 33 & 10.22 & 2013.47 & 0.609 & 215.3 & 0.124 & 28.4 & 80.0 & 2 & cve2012 + & & @xmath560.07 & @xmath560.04 & @xmath560.009 & @xmath565.3 & @xmath560.002 & @xmath562.5 & @xmath564.3 & & + 00155@xmath91608 & hei 299 & 4.550 & 1995.363 & 0.360 & 60.8 & 0.306 & 145.6 & 344.5 & 2 & hry1998 + & & @xmath560.002 & @xmath560.016 & @xmath560.005 & @xmath561.4 & @xmath560.002 & @xmath561.3 & @xmath562.2 & & + 00315@xmath96257 & i 260 cd & 44.73 & 2011.62 & 0.810 & 241.6 & 0.495 & 124.7 & 112.2 & 3 & msn2001c + & & @xmath560.34 & @xmath560.01 & @xmath560.001 & @xmath560.2 & @xmath560.001 & @xmath560.3 & @xmath560.2 & & + 00321@xmath90511 & a 111 ab & 10.64 & 2015.19 & 0.538 & 206.5 & 0.122 & 155.7 & 101.6 & 2 & sta1978b + & & @xmath560.02 & @xmath560.19 & @xmath560.014 & @xmath5618.5 & @xmath560.003 & @xmath564.2 & @xmath5614.9 & & + 00533@xmath690405 & a 2307 & 57.78 & 2012.51 & 0.419 & 222.4 & 0.233 & 74.3 & 357.9 & 2 & usn1999b + & & @xmath563.87 & @xmath561.25 & @xmath560.029 & @xmath561.4 & @xmath560.005 & @xmath561.6 & @xmath5613.2 & & + 01024@xmath690504 & hds 135 ab & 28.23 & 2002.69 & 0.671 & 268.6 & 0.461 & 144.8 & 22.1 & 2 & bag2006 + & & @xmath560.17 & @xmath560.01 & @xmath560.002 & @xmath561.0 & @xmath560.002 & @xmath560.7 & @xmath561.2 & & + 01028@xmath690214 & a 2308 & 141.80 & 1958.95 & 0.124 & 306.6 & 0.367 & 62.9 & 10.5 & 3 & baz1984a + & & @xmath565.52 & @xmath564.96 & @xmath560.025 & @xmath561.7 & @xmath560.006 & @xmath562.2 & @xmath5613.9 & & + 01104@xmath96727 & gki 3 & 1.778 & 1986.193 & 0.112 & 68.2 & 0.147 & 133.3 & 73.0 & 3 & ( gln2007 ) + & & @xmath560.006 & @xmath560.176 & @xmath560.033 & @xmath569.8 & @xmath560.009 & @xmath567.0 & @xmath5630.8 & & + 01196@xmath90520 & a 313 & 128.51 & 2014.72 & 0.161 & 171.5 & 0.274 & 133.7 & 245.5 & 2 & usn1999a + & & @xmath562.24 & @xmath561.42 & @xmath560.009 & @xmath562.4 & @xmath560.004 & @xmath560.9 & @xmath564.5 & & + 01220@xmath96943 & i 263 & 272.15 & 1936.56 & 0.374 & 231.3 & 0.791 & 68.6 & 297.3 & 4 & msn1999a + & & @xmath5647.48 & @xmath565.30 & @xmath560.059 & @xmath562.8 & @xmath560.075 & @xmath561.9 & @xmath5614.1 & & + 01308@xmath95940 & tok 183 & 4.66 & 2012.83 & 0.462 & 349.5 & 0.0458 & 75.0 & 13.9 & 4 & + & & @xmath560.23 & @xmath560.23 & @xmath560.067 & @xmath563.4 & @xmath560.0030 & @xmath560.0 & @xmath5616.2 & & + 01350@xmath92955 & daw 31 ab & 4.562 & 1932.600 & 0.317 & 224.3 & 0.174 & 19.5 & 244.6 & 1 & msn1999c + & & @xmath560.002 & @xmath560.034 & @xmath560.007 & @xmath569.9 & @xmath560.003 & @xmath562.9 & @xmath569.4 & & + 01424@xmath90645 & a 1 & 637.62 & 1888.34 & 0.454 & 232.1 & 0.844 & 42.0 & 264.7 & 5 & sca2008c + 01528@xmath90447 & rst 4188 & 618.71 & 2619.26 & 0.553 & 22.7 & 1.048 & 72.1 & 10.9 & 5 & hei1996a + 02158@xmath91814 & htg 1 & 295.93 & 1681.32 & 0.151 & 187.1 & 2.184 & 35.5 & 270.2 & 5 & sod1999 + 02166@xmath95026 & tok 185 & 16.00 & 2011.32 & 0.467 & 243.6 & 0.132 & 63.2 & 312.4 & 5 & + 02374@xmath95233 & tok 186 & 3.22 & 2011.31 & 0.083 & 52.0 & 0.0785 & 115.5 & 17.0 & 3 & + & & @xmath560.06 & @xmath560.23 & @xmath560.020 & @xmath563.0 & @xmath560.0031 & @xmath562.5 & @xmath5625.4 & & + 02405@xmath92408 & see 19 & 310.26 & 2016.80 & 0.839 & 225.6 & 0.393 & 147.2 & 73.2 & 4 & lin2010c + & & @xmath56146.13 & @xmath560.32 & @xmath560.009 & @xmath5639.0 & @xmath560.098 & @xmath563.3 & @xmath5622.9 & & + 02415@xmath97128 & b 1923 & 101.50 & 2011.42 & 0.376 & 222.9 & 0.522 & 116.3 & 252.3 & 4 & hrt2012a + & & @xmath562.77 & @xmath560.33 & @xmath560.015 & @xmath562.8 & @xmath560.010 & @xmath560.9 & @xmath562.0 & & + 02517@xmath95234 & hu 1562 & 65.00 & 2021.92 & 0.876 & 237.4 & 0.264 & 111.7 & 6.1 & 4 & hei1979b + & & @xmath566.18 & @xmath560.35 & @xmath560.029 & @xmath563.1 & @xmath560.005 & @xmath565.2 & @xmath5610.4 & & + 02572@xmath92458 & beu 4 ca , cb & 1.518 & 2007.088 & 0.540 & 2.0 & 0.062 & 160.4 & 341.8 & 2 & tok2014a + & & @xmath560.001 & @xmath560.021 & @xmath560.024 & @xmath569.1 & @xmath560.002 & @xmath5612.0 & @xmath5611.0 & & + 03014@xmath690615 & hds 385 & 15.27 & 2012.90 & 0.397 & 347.4 & 0.117 & 52.5 & 2.7 & 2 & bag2005 + & & @xmath560.09 & @xmath560.07 & @xmath560.004 & @xmath561.0 & @xmath560.001 & @xmath560.7 & @xmath562.4 & & + 03236@xmath94005 & i 468 & 237.68 & 1971.45 & 0.505 & 319.5 & 2.548 & 41.6 & 30.9 & 5 & sod1999 + 03272@xmath690944 & hds 433 & 50.81 & 2008.43 & 0.571 & 109.4 & 0.439 & 26.7 & 8.0 & 3 & cve2010c + & & @xmath560.62 & @xmath560.04 & @xmath560.004 & @xmath562.9 & @xmath560.004 & @xmath561.2 & @xmath562.9 & & + 03339@xmath93105 & b 52 & 19.36 & 1997.19 & 0.363 & 140.8 & 0.224 & 85.7 & 11.1 & 2 & hei1996c + & & @xmath560.05 & @xmath560.17 & @xmath560.011 & @xmath560.7 & @xmath560.003 & @xmath560.8 & @xmath564.1 & & + 03544@xmath94021 & fin 344 ab & 14.03 & 2008.05 & 0.579 & 247.4 & 0.0614 & 32.4 & 49.7 & 2 & hrt2012a + & & @xmath560.04 & @xmath560.04 & @xmath560.008 & @xmath562.0 & @xmath560.0005 & @xmath561.5 & @xmath562.3 & & + 03545@xmath690510 & a 1831 bc & 216.02 & 1986.20 & 0.099 & 231.8 & 0.187 & 56.0 & 300.1 & 4 & ole1998a + & & @xmath5642.23 & @xmath5625.43 & @xmath560.144 & @xmath565.3 & @xmath560.013 & @xmath565.4 & @xmath5647.3 & & + 04008@xmath690505 & a 1937 & 42.93 & 2014.94 & 0.550 & 25.9 & 0.1000 & 47.1 & 14.7 & 2 & tok2014a + & & @xmath561.13 & @xmath560.15 & @xmath560.019 & @xmath564.4 & @xmath560.0012 & @xmath563.1 & @xmath568.0 & & + 04070@xmath91000 & hds 521 ab & 21.42 & 1996.78 & 0.687 & 219.0 & 0.225 & 122.9 & 76.4 & 3 & msn2011a + & & @xmath560.37 & @xmath560.06 & @xmath560.009 & @xmath561.8 & @xmath560.005 & @xmath561.0 & @xmath560.5 & & + 04163@xmath690710 & wsi 97 & 5.6805 & 1997.088 & 0.808 & 358.1 & 0.1011 & 148.0 & 47.6 & 3 & rao2014 + & & @xmath560.0058 & @xmath560.011 & @xmath560.005 & @xmath561.7 & @xmath560.0009 & * & @xmath561.7 & & + 04368@xmath91733 & a 2915 & 109.99 & 2012.35 & 0.470 & 151.8 & 0.305 & 62.0 & 94.1 & 3 & baz1986a + & & @xmath567.99 & @xmath561.14 & @xmath560.053 & @xmath564.1 & @xmath560.025 & @xmath565.0 & @xmath5612.9 & & + 04374@xmath90951 & rst 3401 & 126.18 & 1967.68 & 0.366 & 98.9 & 0.325 & 146.2 & 52.9 & 3 & nov2007d + & & @xmath566.11 & @xmath561.47 & @xmath560.026 & @xmath5610.9 & @xmath560.011 & @xmath565.2 & @xmath5614.5 & & + 04389@xmath91207 & hds 599 & 48.48 & 2003.67 & 0.828 & 152.5 & 0.333 & 78.4 & 281.2 & 4 & cve2014 + & & @xmath564.70 & @xmath560.22 & @xmath560.032 & @xmath560.9 & @xmath560.006 & @xmath561.0 & @xmath560.8 & & + 04422@xmath690259 & a 2424 & 55.79 & 2011.46 & 0.348 & 228.8 & 0.147 & 85.0 & 90.8 & 3 & wrh1976a + & & @xmath564.07 & @xmath560.64 & @xmath560.012 & @xmath560.6 & @xmath560.009 & @xmath560.5 & @xmath569.0 & & + 04505@xmath690103 & a 2622 & 403.34 & 1956.27 & 0.477 & 259.9 & 0.345 & 127.1 & 270.1 & 4 & sca2003a + & & @xmath5662.54 & @xmath565.59 & @xmath560.046 & @xmath5610.2 & @xmath560.044 & @xmath564.9 & @xmath5623.0 & & + 04506@xmath691505 & chr 20 & 5.734 & 2003.829 & 0.058 & 130.1 & 0.0904 & 114.4 & 92.8 & 3 & sod1999 + & & @xmath560.005 & @xmath560.096 & @xmath560.007 & @xmath560.7 & @xmath560.0010 & @xmath561.7 & @xmath566.0 & & + 04515@xmath93454 & fin 320 & 42.92 & 2006.48 & 0.826 & 198.4 & 0.227 & 111.6 & 292.1 & 3 & doc2013d + & & @xmath560.39 & @xmath560.13 & @xmath560.004 & @xmath560.6 & @xmath560.001 & @xmath560.3 & @xmath560.2 & & + 04545@xmath90314 & rst 5501 & 84.66 & 2004.32 & 0.515 & 219.1 & 0.243 & 120.8 & 310.3 & 3 & msn2011c + & & @xmath561.42 & @xmath560.39 & @xmath560.011 & @xmath561.0 & @xmath560.003 & @xmath561.4 & @xmath561.9 & & + 05025@xmath92115 & don 91 ab & 43.55 & 1997.46 & 0.720 & 246.5 & 1.062 & 60.7 & 280.6 & 4 & sod1999 + & & @xmath560.27 & @xmath560.86 & @xmath560.028 & @xmath563.9 & @xmath560.049 & @xmath562.1 & @xmath561.2 & & + 05073@xmath98352 & hds 669 & 26.45 & 2014.41 & 0.669 & 6.1 & 0.272 & 61.0 & 71.3 & 4 & + & & @xmath561.01 & @xmath560.03 & @xmath560.016 & @xmath562.8 & @xmath560.009 & @xmath561.4 & @xmath562.8 & & + 05245@xmath90224 & mca 18 aa , ab & 9.399 & 2011.613 & 0.435 & 129.0 & 0.0462 & 103.5 & 34.7 & 3 & bag1999b + & & @xmath560.040 & @xmath560.000 & @xmath560.432 & @xmath566.4 & @xmath560.0195 & @xmath5615.3 & @xmath5627.1 & & + 05289@xmath90318 & da 6 & 1491.21 & 1995.70 & 0.846 & 247.1 & 1.021 & 48.3 & 349.7 & 4 & lin2010c + & & @xmath56672.53 & @xmath561.14 & @xmath560.045 & @xmath564.3 & @xmath560.287 & @xmath562.3 & @xmath561.1 & & + 05484@xmath692052 & stt 118 ab & 227.52 & 1986.06 & 0.869 & 137.9 & 0.792 & 89.7 & 290.8 & 4 & pal2005b + & & @xmath5615.48 & @xmath560.09 & @xmath560.026 & @xmath562.0 & @xmath560.037 & @xmath560.8 & @xmath561.4 & & + 05532@xmath96150 & slr 15 & 1251.74 & 1970.32 & 0.824 & 138.4 & 1.303 & 122.1 & 292.2 & 4 & hei1993d + & & @xmath5678.90 & @xmath560.75 & @xmath560.008 & @xmath561.7 & @xmath560.061 & @xmath562.2 & @xmath563.6 & & + 05542@xmath92909 & fin 382 & 20.16 & 2014.80 & 0.473 & 169.2 & 0.154 & 129.5 & 110.8 & 3 & hrt2012a + & & @xmath560.03 & @xmath560.08 & @xmath560.018 & @xmath564.0 & @xmath560.003 & @xmath562.4 & @xmath565.1 & & + 06293@xmath90248 & b 2601 ab & 16.53 & 1999.33 & 0.387 & 209.6 & 1.072 & 53.3 & 42.0 & 3 & sgr2000 + & & @xmath560.01 & @xmath560.01 & @xmath560.002 & @xmath560.3 & @xmath560.002 & @xmath560.2 & @xmath560.4 & & + 06314@xmath690749 & a 2817 & 31.55 & 2015.34 & 0.281 & 57.3 & 0.192 & 37.3 & 320.8 & 2 & pop1969b + & & @xmath561.41 & @xmath560.28 & @xmath560.015 & @xmath564.4 & @xmath560.007 & @xmath561.9 & @xmath5611.7 & & + 06359@xmath93605 & rst 4816 ba , bb & 14.03 & 2004.09 & 0.577 & 288.8 & 0.182 & 111.9 & 296.2 & 2 & tok2012b + & & @xmath560.02 & @xmath560.03 & @xmath560.005 & @xmath560.3 & @xmath560.001 & @xmath560.2 & @xmath560.4 & & + 06439@xmath95434 & hds 934 & 12.37 & 2014.21 & 0.216 & 93.4 & 0.132 & 18.5 & 70.3 & 3 & hrt2012a + & & @xmath560.02 & @xmath560.04 & @xmath560.004 & @xmath564.6 & @xmath560.001 & @xmath561.9 & @xmath564.2 & & + 06454@xmath93148 & ehr 9 ba , bb & 6.865 & 2014.728 & 0.224 & 189.7 & 0.126 & 138.9 & 5.6 & 4 & + & & @xmath560.137 & @xmath560.318 & @xmath560.038 & @xmath5613.6 & @xmath560.010 & @xmath568.3 & @xmath5615.5 & & + 06481@xmath90948 & a 1056 & 616.86 & 1962.17 & 0.522 & 256.0 & 0.501 & 56.7 & 280.6 & 4 & sca1983f + & & @xmath56513.30 & @xmath5610.30 & @xmath560.282 & @xmath5613.7 & @xmath560.255 & @xmath567.0 & @xmath5632.6 & & + 06490@xmath91509 & ac 4 & 581.32 & 2031.70 & 0.761 & 157.5 & 0.957 & 52.6 & 280.6 & 5 & + 06493@xmath90216 & fin 322 & 58.69 & 1969.56 & 0.293 & 241.6 & 0.148 & 112.2 & 68.0 & 3 & hrt2011d + & & @xmath561.61 & @xmath560.78 & @xmath560.015 & @xmath560.9 & @xmath560.006 & @xmath561.2 & @xmath566.0 & & + 07026@xmath691558 & a 2462 ab & 44.45 & 2012.79 & 0.661 & 67.3 & 0.204 & 134.4 & 45.5 & 3 & baz1976 + & & @xmath560.44 & @xmath560.02 & @xmath560.002 & @xmath560.7 & @xmath560.001 & @xmath560.5 & @xmath560.8 & & + 07043@xmath90303 & a 519 ab & 90.40 & 2012.50 & 0.260 & 93.6 & 0.397 & 96.6 & 283.0 & 3 & tok2014a + & & @xmath561.44 & @xmath560.42 & @xmath560.026 & @xmath560.7 & @xmath560.006 & @xmath560.7 & @xmath563.4 & & + 07113@xmath91032 & a 2122 & 169.68 & 2021.25 & 0.712 & 150.3 & 0.203 & 120.2 & 228.6 & 4 & usn2002 + & & @xmath5613.75 & @xmath562.39 & @xmath560.021 & @xmath565.1 & @xmath560.007 & @xmath563.7 & @xmath562.5 & & + 07143@xmath92621 & fin 323 & 118.54 & 1971.01 & 0.747 & 131.3 & 0.178 & 80.2 & 85.0 & 4 & ole2004b + & & @xmath5611.14 & @xmath560.74 & @xmath560.094 & @xmath563.0 & @xmath560.012 & @xmath561.0 & @xmath563.6 & & + 07155@xmath97552 & i 312 & 847.09 & 2012.33 & 0.859 & 152.9 & 2.308 & 97.5 & 86.1 & 5 & + 07175@xmath94659 & i 7 & 85.01 & 1958.70 & 0.955 & 236.0 & 0.891 & 104.4 & 244.8 & 3 & msn2011c + & & @xmath561.22 & @xmath560.44 & @xmath560.014 & @xmath565.1 & @xmath560.048 & @xmath562.3 & @xmath562.5 & & + 07374@xmath93458 & fin 324 ac & 80.68 & 2016.58 & 0.648 & 262.1 & 0.323 & 158.6 & 32.7 & 3 & hrt2012a + & & @xmath561.63 & @xmath560.37 & @xmath560.014 & @xmath568.4 & @xmath560.004 & @xmath564.9 & @xmath568.2 & & + 07435@xmath92711 & b 737 & 290.20 & 1967.59 & 0.566 & 86.2 & 0.335 & 105.7 & 324.4 & 4 & + & & @xmath5614.74 & @xmath561.93 & @xmath560.014 & @xmath562.1 & @xmath560.016 & @xmath561.2 & @xmath564.3 & & + 07456@xmath93410 & tok 193 & 15.00 & 2010.83 & 0.305 & 58.4 & 0.473 & 67.7 & 342.5 & 4 & + & & * & @xmath560.15 & @xmath560.032 & @xmath560.8 & @xmath560.016 & @xmath561.4 & @xmath566.8 & & + 07490@xmath92455 & tok 194 & 2.422 & 2011.557 & 0.408 & 212 . & 0.0374 & 29.5 & 295.3 & 3 & + & & @xmath560.075 & @xmath560.084 & @xmath560.072 & @xmath5632.&@xmath560.0031 & @xmath5613.9 & @xmath5628.8 & & + 07522@xmath94035 & tok 195 & 7.388 & 2012.067 & 0.380 & 90.3 & 0.0624 & 86.2 & 170.0 & 4 & ( jnc2005 ) + & & @xmath560.657 & @xmath560.206 & @xmath560.000 & @xmath566.8 & @xmath560.0071 & @xmath567.0 & @xmath560.0 & & + 08061@xmath90047 & a 1971 & 572.52 & 1910.43 & 0.409 & 202.8 & 1.070 & 125.6 & 99.1 & 5 & ole1993 + 08125@xmath94616 & chr 143 aa , ab & 31.42 & 2017.49 & 0.339 & 173.6 & 0.0738 & 71.7 & 257.6 & 3 & msn2010c + & & @xmath560.33 & @xmath560.21 & @xmath560.020 & @xmath561.0 & @xmath560.0012 & @xmath561.7 & @xmath562.0 & & + 08144@xmath94550 & fin 113 ab & 385.48 & 1973.09 & 0.700 & 103.4 & 0.566 & 81.6 & 346.0 & 5 & cve2010e + 08251@xmath94910 & rst 321 & 25.78 & 2000.85 & 0.219 & 128.6 & 0.310 & 40.9 & 232.1 & 2 & hrt2012a + & & @xmath560.41 & @xmath560.15 & @xmath560.005 & @xmath560.8 & @xmath560.004 & @xmath560.7 & @xmath564.1 & & + 08270@xmath95242 & b 1606 & 14.66 & 2011.84 & 0.353 & 277.0 & 0.152 & 61.7 & 11.9 & 2 & tok2012b + & & @xmath560.08 & @xmath560.01 & @xmath560.002 & @xmath560.2 & @xmath560.000 & @xmath560.1 & @xmath560.2 & & + 08317@xmath92601 & i 807 & 720.66 & 1936.16 & 0.370 & 134.2 & 0.631 & 129.4 & 244.5 & 5 & + 08345@xmath93236 & fin 335 & 17.36 & 2014.08 & 0.572 & 268.9 & 0.139 & 30.6 & 43.8 & 2 & doc2013d + & & @xmath560.03 & @xmath560.03 & @xmath560.004 & @xmath561.9 & @xmath560.001 & @xmath561.3 & @xmath562.0 & & + 08380@xmath90844 & hds 1242 & 41.66 & 2014.55 & 0.392 & 198.2 & 0.249 & 24.9 & 254.0 & 4 & hrt2012a + & & @xmath564.50 & @xmath560.53 & @xmath560.011 & @xmath5614.8 & @xmath560.020 & @xmath566.3 & @xmath565.1 & & + 08391@xmath95557 & hu 1443 ab & 876.85 & 1860.89 & 0.644 & 104.3 & 1.400 & 128.1 & 314.0 & 5 & + 08421@xmath95245 & b 1624 & 74.05 & 1995.69 & 0.240 & 89.7 & 0.474 & 71.4 & 0.1 & 3 & hrt2012a + & & @xmath560.43 & @xmath560.31 & @xmath560.003 & @xmath560.2 & @xmath560.002 & @xmath560.2 & @xmath562.1 & & + 08431@xmath91225 & rst 3603 & 98.92 & 1964.70 & 0.393 & 155.5 & 0.268 & 154.5 & 14.1 & 3 & hrt2010a + & & @xmath563.48 & @xmath561.39 & @xmath560.031 & @xmath5613.1 & @xmath560.007 & @xmath5610.0 & @xmath5617.8 & & + 08447@xmath94238 & chr 238 & 2.257 & 2012.557 & 0.615 & 42.4 & 0.0785 & 145.8 & 332.8 & 3 & hrt2012a + & & @xmath560.004 & @xmath560.056 & @xmath560.042 & @xmath567.1 & @xmath560.0046 & @xmath5614.0 & @xmath567.9 & & + 08474@xmath91703 & bu 586 & 442.24 & 2009.74 & 0.529 & 239.7 & 0.538 & 63.9 & 317.3 & 5 & msn2009 + 08538@xmath94731 & fin 316 & 7.209 & 2007.601 & 0.253 & 108.5 & 0.0781 & 11.1 & 106.8 & 2 & hrt2012a + & & @xmath560.004 & @xmath560.009 & @xmath560.002 & @xmath568.8 & @xmath560.0003 & @xmath561.6 & @xmath568.7 & & + 08539@xmath690149 & a 2554 & 43.86 & 2021.18 & 0.502 & 112.4 & 0.210 & 162.4 & 339.7 & 3 & zir2007 + & & @xmath560.80 & @xmath560.30 & @xmath560.030 & @xmath5616.7 & @xmath560.005 & @xmath5611.4 & @xmath5617.3 & & + 08589@xmath690829 & del 2 & 5.566 & 2006.445 & 0.778 & 281.8 & 0.398 & 122.2 & 23.7 & 3 & hrt2012a + & & @xmath560.024 & @xmath560.039 & @xmath560.019 & @xmath561.7 & @xmath560.008 & @xmath563.1 & @xmath563.2 & & + 09149@xmath690427 & hei 350 & 198.56 & 2000.40 & 0.574 & 84.6 & 1.856 & 122.2 & 344.7 & 5 & ( hei1994a ) + 09173@xmath96841 & fin 363 ab & 3.445 & 2013.376 & 0.455 & 154.6 & 0.0888 & 140.7 & 117.2 & 2 & doc2013d + & & @xmath560.002 & @xmath560.007 & @xmath560.003 & @xmath560.8 & @xmath560.0004 & @xmath560.6 & @xmath560.7 & & + 09194@xmath97739 & koh 83 aa , ab & 19.84 & 1999.00 & 0.795 & 20.2 & 0.146 & 65.0 & 276.2 & 5 & + 09228@xmath90950 & a 1342 ab & 52.98 & 2020.20 & 0.074 & 202.4 & 0.169 & 68.5 & 58.7 & 2 & msn2011a + & & @xmath560.37 & @xmath560.45 & @xmath560.008 & @xmath560.4 & @xmath560.001 & @xmath560.6 & @xmath563.3 & & + 09252@xmath91258 & wsi 73 & 27.3 & 2014.80 & 0.860 & 101.1 & 0.235 & 86.0 & 283.4 & 5 & + 09275@xmath95806 & chr 240 & 2.621 & 2014.453 & 0.173 & 267.4 & 0.0478 & 111.4 & 125.6 & 3 & + & & @xmath560.025 & @xmath560.049 & @xmath560.029 & @xmath562.3 & @xmath560.0018 & @xmath562.3 & @xmath566.6 & & + 09278@xmath90604 & b 2530 & 34.80 & 2010.11 & 0.270 & 330.8 & 0.428 & 85.4 & 341.9 & 2 & sod1999 + & & @xmath560.06 & @xmath560.10 & @xmath560.003 & @xmath560.1 & @xmath560.001 & @xmath560.1 & @xmath561.1 & & + 09313@xmath91329 & kui 41 & 18.38 & 2002.17 & 0.335 & 230.8 & 0.647 & 139.9 & 105.3 & 3 & sod1999 + & & @xmath560.11 & @xmath560.03 & @xmath560.004 & @xmath560.9 & @xmath560.002 & @xmath560.2 & @xmath560.6 & & + 09327@xmath690152 & fin 349 & 40.57 & 1972.12 & 0.420 & 138.6 & 0.154 & 54.7 & 72.6 & 2 & msn2011a + & & @xmath560.19 & @xmath560.16 & @xmath560.009 & @xmath560.6 & @xmath560.001 & @xmath560.6 & @xmath560.8 & & + 09387@xmath93937 & i 202 & 163.69 & 2005.49 & 0.414 & 178.6 & 0.932 & 111.4 & 203.4 & 4 & hrt2012a + & & @xmath563.53 & @xmath560.67 & @xmath560.012 & @xmath561.1 & @xmath560.017 & @xmath560.9 & @xmath562.2 & & + 09407@xmath95759 & b 780 & 10.63 & 2005.86 & 0.338 & 86.1 & 0.123 & 128.6 & 17.6 & 2 & drs2012 + & & @xmath560.01 & @xmath560.02 & @xmath560.002 & @xmath560.3 & @xmath560.000 & @xmath560.3 & @xmath560.8 & & + 09495@xmath91033 & a 1344 & 260.75 & 1991.89 & 0.781 & 119.5 & 0.355 & 116.2 & 90.9 & 4 & hei1986b + & & @xmath5625.54 & @xmath561.16 & @xmath560.030 & @xmath562.4 & @xmath560.009 & @xmath562.6 & @xmath561.8 & & + 10050@xmath95119 & hu 1594 & 134.51 & 2028.97 & 0.629 & 87.3 & 0.281 & 64.4 & 21.0 & 3 & usn2002 + & & @xmath563.76 & @xmath562.28 & @xmath560.033 & @xmath561.6 & @xmath560.009 & @xmath563.3 & @xmath564.6 & & + 10120@xmath92836 & b 194 & 91.89 & 2003.58 & 0.676 & 198.3 & 0.226 & 76.0 & 59.7 & 3 & usn2002 + & & @xmath560.85 & @xmath560.22 & @xmath560.012 & @xmath560.5 & @xmath560.003 & @xmath560.7 & @xmath561.1 & & + 10161@xmath92837 & tok 199 & 2.507 & 2014.734 & 0.353 & 278.5 & 0.0450 & 114.5 & 51.8 & 4 & + & & @xmath560.022 & @xmath560.020 & @xmath560.017 & @xmath561.7 & @xmath560.0013 & @xmath561.8 & @xmath563.0 & & + 10282@xmath92548 & fin 308 ab & 32.76 & 2018.13 & 0.739 & 155.4 & 0.144 & 48.6 & 270.6 & 3 & hrt2012a + & & @xmath562.67 & @xmath561.28 & @xmath560.135 & @xmath564.5 & @xmath560.033 & @xmath5616.2 & @xmath562.7 & & + 10375@xmath90932 & rst 3708 & 154.81 & 1971.26 & 0.457 & 183.9 & 0.416 & 54.6 & 24.4 & 4 & hei1991 + & & @xmath5655.88 & @xmath5611.57 & @xmath560.197 & @xmath5617.0 & @xmath560.026 & @xmath5619.7 & @xmath5637.9 & & + 10419@xmath97811 & hds 1530 & 43.48 & 2007.33 & 0.637 & 114.9 & 0.277 & 55.2 & 121.2 & 5 & + 10426@xmath690335 & a 2768 & 81.66 & 1976.39 & 0.550 & 57.8 & 0.401 & 140.1 & 354.9 & 2 & hrt2010a + & & @xmath562.05 & @xmath560.21 & @xmath560.005 & @xmath561.0 & @xmath560.002 & @xmath561.7 & @xmath562.9 & & + 10465@xmath96416 & fin 364 ab & 13.25 & 2020.60 & 0.025 & 316.7 & 0.0805 & 115.2 & 17.6 & 2 & hrt2012a + & & @xmath560.20 & @xmath562.49 & @xmath560.022 & @xmath561.7 & @xmath560.0014 & @xmath562.2 & @xmath5664.8 & & + 10529@xmath91717 & hds 1556 & 14.93 & 2002.74 & 0.544 & 110.5 & 0.173 & 99.2 & 51.4 & 3 & hrt2012a + & & @xmath560.26 & @xmath561.39 & @xmath560.056 & @xmath562.1 & @xmath560.028 & @xmath563.6 & @xmath5617.3 & & + 10557@xmath690044 & bu 1076 & 138.90 & 1921.62 & 0.714 & 219.3 & 0.750 & 125.7 & 323.4 & 4 & hrt2010a + & & @xmath563.02 & @xmath565.07 & @xmath560.036 & @xmath564.0 & @xmath560.069 & @xmath569.5 & @xmath5610.8 & & + 11009@xmath94030 & fin 365 & 26.86 & 1990.21 & 0.087 & 107.4 & 0.168 & 100.5 & 50.5 & 3 & tok2012b + & & @xmath560.23 & @xmath561.14 & @xmath560.029 & @xmath561.1 & @xmath560.004 & @xmath560.9 & @xmath5616.3 & & + 11014@xmath91204 & hds 1572 & 18.64 & 2013.69 & 0.686 & 142.4 & 0.173 & 98.7 & 133.8 & 3 & + & & @xmath560.72 & @xmath560.03 & @xmath560.013 & @xmath560.4 & @xmath560.003 & @xmath560.8 & @xmath561.5 & & + 11102@xmath91122 & hds 1590 & 20.63 & 1995.82 & 0.698 & 255.8 & 0.136 & 129.4 & 72.2 & 3 & hrt2012a + & & @xmath561.62 & @xmath561.68 & @xmath560.011 & @xmath565.4 & @xmath560.006 & @xmath560.9 & @xmath561.5 & & + 11125@xmath91830 & bu 220 & 365.68 & 1983.93 & 0.434 & 325.3 & 0.553 & 99.8 & 337.3 & 4 & hei1995 + & & @xmath568.02 & @xmath563.59 & @xmath560.014 & @xmath560.6 & @xmath560.022 & @xmath560.8 & @xmath566.7 & & + 11272@xmath91539 & hu 462 & 48.40 & 2008.46 & 0.087 & 127.6 & 0.445 & 166.5 & 347.7 & 2 & wsi2006b + & & @xmath560.11 & @xmath560.49 & @xmath560.004 & @xmath5611.7 & @xmath560.003 & @xmath562.9 & @xmath5610.5 & & + 11297@xmath90619 & a 7 & 204.32 & 2050.54 & 0.656 & 93.3 & 0.447 & 105.1 & 59.8 & 4 & usn2002 + & & @xmath5611.20 & @xmath5619.37 & @xmath560.017 & @xmath564.9 & @xmath560.077 & @xmath565.8 & @xmath5613.4 & & + 11544@xmath93745 & hwe 71 & 248.80 & 2008.48 & 0.720 & 276.3 & 1.004 & 99.6 & 162.0 & 5 & + 12396@xmath93717 & daw 63 & 56.85 & 2021.02 & 0.287 & 194.5 & 0.387 & 21.6 & 3.7 & 3 & hrt2010a + & & @xmath560.32 & @xmath560.55 & @xmath560.015 & @xmath5613.8 & @xmath560.004 & @xmath565.3 & @xmath5616.4 & & + 13145@xmath92417 & fin 297 ab & 60.81 & 1957.92 & 0.691 & 188.3 & 0.233 & 66.0 & 295.0 & 2 & msn2010c + & & @xmath563.27 & @xmath563.17 & @xmath560.041 & @xmath562.7 & @xmath560.010 & @xmath561.7 & @xmath562.2 & & + 13169@xmath93436 & i 1567 & 40.88 & 2006.40 & 0.457 & 145.5 & 0.326 & 121.6 & 274.0 & 2 & tok2012b + & & @xmath560.18 & @xmath560.01 & @xmath560.002 & @xmath560.2 & @xmath560.000 & @xmath560.1 & @xmath560.4 & & + 13310@xmath93924 & see 179 & 83.14 & 1956.12 & 0.521 & 146.5 & 0.161 & 145.2 & 244.4 & 2 & fin1964b + & & @xmath561.26 & @xmath561.91 & @xmath560.013 & @xmath568.6 & @xmath560.004 & @xmath564.0 & @xmath563.6 & & + 13342@xmath91623 & rst 3844 & 121.85 & 2010.94 & 0.432 & 264.5 & 0.255 & 117.6 & 19.4 & 3 & usn2002 + & & @xmath562.70 & @xmath560.49 & @xmath560.007 & @xmath561.1 & @xmath560.003 & @xmath560.7 & @xmath562.4 & & + 13472@xmath90943 & kui 65 & 155.21 & 2014.17 & 0.854 & 99.0 & 0.268 & 144.2 & 22.7 & 3 & zir2012 + & & @xmath5620.85 & @xmath560.11 & @xmath560.006 & @xmath564.5 & @xmath560.012 & @xmath566.1 & @xmath567.6 & & + 13571@xmath92731 & i 234 & 200.72 & 2057.91 & 0.725 & 165.9 & 0.773 & 117.3 & 83.9 & 4 & mro1966b + & & @xmath5617.44 & @xmath5610.04 & @xmath560.023 & @xmath568.6 & @xmath560.043 & @xmath561.3 & @xmath561.1 & & + 14020@xmath92108 & wsi 79 & 30.00 & 2015.16 & 0.746 & 123.3 & 0.239 & 135.3 & 119.4 & 4 & + & & * & @xmath560.17 & @xmath560.037 & @xmath569.9 & @xmath560.011 & @xmath562.7 & @xmath569.2 & & + 14025@xmath92440 & b 263 & 161.43 & 2014.56 & 0.499 & 20.4 & 0.480 & 38.8 & 81.7 & 4 & + & & @xmath5628.68 & @xmath561.92 & @xmath560.019 & @xmath569.9 & @xmath560.072 & @xmath569.0 & @xmath5610.0 & & + 14190@xmath90636 & hds2016 ab & 18.67 & 2003.74 & 0.302 & 285.0 & 0.348 & 107.9 & 71.0 & 3 & + & & @xmath560.33 & @xmath560.03 & @xmath560.011 & @xmath560.8 & @xmath560.001 & @xmath560.5 & @xmath562.2 & & + 14295@xmath93702 & hds 2045 aa , ab & 18.39 & 2008.92 & 0.468 & 160.0 & 0.137 & 126.9 & 137.3 & 3 & + & & @xmath560.74 & @xmath560.06 & @xmath560.018 & @xmath562.6 & @xmath560.006 & @xmath562.0 & @xmath562.2 & & + 14375@xmath690217 & chr 42 aa , ab & 21.54 & 1994.64 & 0.844 & 312.4 & 0.143 & 53.7 & 41.9 & 3 & hrt2000a + & & @xmath560.10 & @xmath560.20 & @xmath560.016 & @xmath563.6 & @xmath560.007 & @xmath563.0 & @xmath565.6 & & + 14581@xmath94852 & wsi 80 & 23.40 & 1998.44 & 0.724 & 267.3 & 0.342 & 113.1 & 297.0 & 4 & + & & @xmath560.94 & @xmath562.07 & @xmath560.020 & @xmath564.9 & @xmath560.050 & @xmath566.6 & @xmath568.6 & & + 15047@xmath95625 & rst 2937 & 146.05 & 2006.50 & 0.809 & 173.6 & 0.214 & 105.9 & 113.4 & 4 & + & & @xmath5624.74 & @xmath560.03 & @xmath560.054 & @xmath562.0 & @xmath560.028 & @xmath564.3 & @xmath566.0 & & + 15226@xmath94755 & slr 20 & 429.60 & 1972.88 & 0.469 & 56.0 & 1.899 & 109.7 & 344.6 & 4 & hei1993d + & & @xmath564.57 & @xmath561.56 & @xmath560.004 & @xmath560.9 & @xmath560.019 & @xmath560.5 & @xmath562.0 & & + 15227@xmath94441 & cop 2 ab & 736.69 & 2009.83 & 0.808 & 125.1 & 1.409 & 117.6 & 19.6 & 5 & zir2007 + 15246@xmath94835 & b 1288 ab & 449.82 & 2013.70 & 0.742 & 143.4 & 0.326 & 67.5 & 284.9 & 4 & usn2002 + & & @xmath5650.91 & @xmath560.60 & @xmath560.019 & @xmath561.7 & @xmath560.028 & @xmath562.3 & @xmath566.1 & & + 15252@xmath94659 & rst 767 & 249.27 & 1921.63 & 0.504 & 116.6 & 0.525 & 136.0 & 335.4 & 5 & csa1975c + 15262@xmath92819 & rst 769 & 201.47 & 1977.28 & 0.379 & 163.2 & 0.392 & 141.9 & 238.0 & 5 & doc1996b + 15355@xmath94751 & hds 2191 & 62.52 & 2005.25 & 0.315 & 201.9 & 0.302 & 88.6 & 267.1 & 4 & + & & @xmath5614.96 & @xmath561.10 & @xmath560.243 & @xmath561.1 & @xmath560.037 & @xmath560.9 & @xmath5614.6 & & + 15420@xmath690027 & a 2176 & 52.68 & 1987.16 & 0.653 & 97.7 & 0.148 & 27.2 & 279.3 & 2 & usn2006b + & & @xmath562.04 & @xmath560.38 & @xmath560.027 & @xmath5612.9 & @xmath560.008 & @xmath567.4 & @xmath569.9 & & + 16035@xmath95747 & see 258 ab & 26.84 & 1990.87 & 0.515 & 42.0 & 0.328 & 168.7 & 320.0 & 2 & sod1999 + & & @xmath560.09 & @xmath560.21 & @xmath560.012 & @xmath5643.2 & @xmath560.006 & @xmath569.9 & @xmath5644.1 & & + 16065@xmath94027 & rst 1876 & 96.95 & 2037.31 & 0.006 & 97.5 & 0.303 & 97.5 & 338.1 & 4 & + & & @xmath562.07 & @xmath561.54 & @xmath560.027 & @xmath560.8 & @xmath560.011 & @xmath560.9 & @xmath566.2 & & + 16085@xmath91006 & bu 949 & 56.33 & 2018.34 & 0.774 & 198.1 & 0.289 & 82.3 & 157.8 & 2 & hrt2009 + & & @xmath560.20 & @xmath560.35 & @xmath560.016 & @xmath560.5 & @xmath560.003 & @xmath560.5 & @xmath562.9 & & + 16170@xmath95342 & i 987 & 102.47 & 1952.08 & 0.262 & 59.3 & 0.384 & 15.9 & 12.2 & 3 & hei1986a + & & @xmath561.77 & @xmath561.72 & @xmath560.021 & @xmath5639.9 & @xmath560.012 & @xmath5610.8 & @xmath5643.4 & & + 16224@xmath93220 & jsp 691 & 264.96 & 1958.53 & 0.590 & 101.9 & 0.870 & 122.0 & 307.2 & 4 & hei1981a + & & @xmath5643.30 & @xmath560.81 & @xmath560.046 & @xmath562.9 & @xmath560.102 & @xmath562.2 & @xmath564.7 & & + 16229@xmath91701 & chr 54 & 39.89 & 2011.76 & 0.634 & 236.6 & 0.183 & 101.9 & 306.0 & 3 & lin2011c + & & @xmath561.21 & @xmath561.14 & @xmath560.049 & @xmath562.7 & @xmath560.008 & @xmath562.1 & @xmath562.8 & & + 16318@xmath90216 & a 693 & 96.18 & 1991.33 & 0.278 & 220.9 & 0.195 & 129.3 & 344.2 & 2 & hrt2010a + & & @xmath560.89 & @xmath560.68 & @xmath560.011 & @xmath562.2 & @xmath560.003 & @xmath561.7 & @xmath563.8 & & + 16391@xmath93713 & fin 340 ab & 23.99 & 2013.87 & 0.621 & 149.6 & 0.102 & 55.4 & 143.3 & 2 & hrt2012a + & & @xmath560.06 & @xmath560.02 & @xmath560.004 & @xmath560.4 & @xmath560.000 & @xmath560.4 & @xmath560.8 & & + 16544@xmath93806 & hds 2392 & 37.25 & 2010.49 & 0.325 & 166.5 & 0.187 & 60.2 & 321.9 & 5 & + 16589@xmath93737 & see 315 & 38.73 & 1984.63 & 0.154 & 243.9 & 0.186 & 39.4 & 102.4 & 2 & sta1981a + & & @xmath564.48 & @xmath561.11 & @xmath560.106 & @xmath561.7 & @xmath560.006 & @xmath568.6 & @xmath5611.2 & & + 17156@xmath91018 & bu 957 & 90.33 & 1934.61 & 0.554 & 23.0 & 0.296 & 102.0 & 7.0 & 3 & hei1984b + & & @xmath561.52 & @xmath562.41 & @xmath560.067 & @xmath562.1 & @xmath560.013 & @xmath564.6 & @xmath5611.3 & & + 17156@xmath93836 & fin 355 & 14.23 & 1985.97 & 0.476 & 191.9 & 0.249 & 115.3 & 137.2 & 2 & msn2010c + & & @xmath560.03 & @xmath560.06 & @xmath560.010 & @xmath560.6 & @xmath560.002 & @xmath560.6 & @xmath561.0 & & + 17157@xmath90949 & tok 53 ba , bb & 5.10 & 2010.25 & 0.416 & 349.0 & 0.0305 & 150.0 & 326.3 & 2 & + & & @xmath560.18 & @xmath560.52 & @xmath560.076 & @xmath56556 & @xmath560.0056 & @xmath5635.0 & @xmath5678.0 & & + 17157@xmath90949 & a 2592 ab & 157.60 & 2013.20 & 0.330 & 31.7 & 0.381 & 133.6 & 267.0 & 3 & tok2014a + & & @xmath569.0 & @xmath561.56 & @xmath560.017 & @xmath561.4 & @xmath560.017 & @xmath561.6 & @xmath567.5 & & + 17166@xmath90027 & a 2984 & 140.76 & 1890.53 & 0.866 & 220.5 & 0.959 & 65.1 & 287.5 & 4 & ole1993 + & & @xmath560.23 & @xmath560.36 & @xmath560.003 & @xmath560.0 & @xmath560.008 & @xmath560.0 & @xmath560.0 & & + 17181@xmath93810 & see 324 & 300.00 & 1997.49 & 0.484 & 249.6 & 0.439 & 106.5 & 42.2 & 5 & + 17283@xmath92058 & a 2244 ab & 44.52 & 2015.23 & 0.578 & 265.5 & 0.168 & 40.2 & 25.0 & 2 & msn1999c + & & @xmath562.18 & @xmath560.26 & @xmath560.044 & @xmath569.7 & @xmath560.003 & @xmath563.9 & @xmath5615.3 & & + 17305@xmath91006 & rst 3978 & 49.24 & 1981.10 & 0.343 & 96.8 & 0.547 & 83.0 & 79.5 & 3 & + & & @xmath561.19 & @xmath561.52 & @xmath560.090 & @xmath561.8 & @xmath560.044 & @xmath562.2 & @xmath567.4 & & + 17471@xmath93807 & i 1336 & 36.06 & 2017.29 & 0.578 & 15.8 & 0.132 & 45.0 & 11.3 & 3 & hei1986b + & & @xmath562.08 & @xmath560.19 & @xmath560.050 & @xmath568.4 & @xmath560.001 & @xmath560.1 & @xmath5614.2 & & + 18018@xmath690118 & bu 1125 ab & 213.97 & 2028.73 & 0.601 & 161.3 & 0.787 & 55.1 & 66.0 & 4 & hrt2009 + & & @xmath5623.83 & @xmath561.62 & @xmath560.074 & @xmath564.4 & @xmath560.033 & @xmath565.1 & @xmath565.1 & & + 18044@xmath95953 & rst 5099 & 45.64 & 2009.10 & 0.513 & 114.4 & 0.259 & 68.2 & 107.1 & 4 & hrt2010a + & & @xmath562.10 & @xmath560.10 & @xmath560.010 & @xmath561.0 & @xmath560.004 & @xmath561.0 & @xmath562.7 & & + 18092@xmath92211 & rst 3157 & 9.325 & 2015.286 & 0.423 & 250.8 & 0.154 & 49.6 & 54.5 & 3 & hei1990c + & & @xmath560.021 & @xmath560.175 & @xmath560.042 & @xmath566.5 & @xmath560.005 & @xmath565.2 & @xmath566.3 & & + 18171@xmath94336 & hds 2583 & 153.90 & 2016.88 & 0.724 & 58.5 & 0.153 & 129.9 & 24.2 & 4 & + & & @xmath56308.47 & @xmath562.38 & @xmath560.025 & @xmath5615.5 & @xmath560.082 & @xmath5644.0 & @xmath5681.0 & & + 18281@xmath92645 & hds 2615 & 150.28 & 1990.48 & 0.609 & 166.6 & 0.897 & 96.4 & 328.7 & 4 & + & & @xmath56211.19 & @xmath566.64 & @xmath560.004 & @xmath563.2 & @xmath560.366 & @xmath564.2 & @xmath5676.5 & & + 18305@xmath92848 & hds 2624 & 42.93 & 2007.53 & 0.531 & 42.3 & 0.0979 & 144.5 & 14.0 & 3 & + & & @xmath5610.94 & @xmath561.05 & @xmath560.104 & @xmath5622.1 & @xmath560.0211 & @xmath5633.2 & @xmath5624.7 & & + 18323@xmath91439 & chr 73 & 2.774 & 2012.600 & 0.602 & 213.9 & 0.0365 & 135.6 & 2.0 & 3 & ole2005d + & & @xmath560.002 & @xmath560.189 & @xmath560.166 & @xmath5610.6 & @xmath560.0053 & @xmath5623.2 & @xmath5622.7 & & + 18439@xmath90649 & ysc 133 & 13.52 & 2002.36 & 0.124 & 97.6 & 0.0787 & 112.1 & 290.2 & 3 & + & & @xmath562.60 & @xmath564.77 & @xmath560.023 & @xmath562.4 & @xmath560.0117 & @xmath563.2 & @xmath5677.4 & & + 18465@xmath90058 & mca 53 aa , ab & 42.48 & 2012.42 & 0.665 & 163.7 & 0.102 & 131.1 & 355.5 & 3 & msn2010c + & & @xmath562.20 & @xmath561.12 & @xmath560.057 & @xmath5621.8 & @xmath560.008 & @xmath5611.3 & @xmath5639.0 & & + 18516@xmath96054 & rst 5126 & 54.16 & 2011.97 & 0.492 & 195.0 & 0.130 & 21.7 & 12.8 & 3 & + & & @xmath563.17 & @xmath560.36 & @xmath560.020 & @xmath5634.5 & @xmath560.005 & @xmath5613.8 & @xmath5636.1 & & usn2002 + 19029@xmath95413 & i 1390 & 47.36 & 2009.65 & 0.682 & 241.2 & 0.198 & 56.0 & 51.0 & 3 & + & & @xmath560.80 & @xmath560.60 & @xmath560.052 & @xmath567.8 & @xmath560.012 & @xmath564.2 & @xmath567.5 & & + 19040@xmath93804 & i 1391 & 47.38 & 2006.14 & 0.641 & 271.3 & 0.178 & 35.8 & 56.4 & 4 & hei1973b + & & @xmath560.22 & @xmath560.23 & @xmath560.019 & @xmath565.2 & @xmath560.003 & @xmath561.3 & @xmath565.7 & & + 19105@xmath95813 & b 2468 & 182.29 & 2013.45 & 0.891 & 168.5 & 0.655 & 140.0 & 5.0 & 5 & + 19155@xmath92515 & b 430 & 19.95 & 2014.98 & 0.506 & 285.1 & 0.132 & 82.1 & 7.5 & 2 & hrt2001b + & & @xmath560.09 & @xmath560.41 & @xmath560.042 & @xmath560.5 & @xmath560.004 & @xmath561.0 & @xmath565.7 & & + 19164@xmath691433 & chr 85 aa , ab & 13.673 & 2008.572 & 0.022 & 242.3 & 0.0583 & 131.3 & 262.7 & 2 & mca1993 + & & @xmath560.069 & @xmath561.101 & @xmath560.010 & @xmath564.5 & @xmath560.0014 & @xmath563.7 & @xmath5628.7 & & + 19194@xmath90136 & hds 2734 aa , ab & 41.19 & 2021.53 & 0.506 & 27.8 & 0.321 & 30.4 & 348.2 & 4 & + & & @xmath565.46 & @xmath562.07 & @xmath560.068 & @xmath5639.7 & @xmath560.018 & @xmath567.2 & @xmath5657.6 & & + 19296@xmath91239 & hu 75 & 129.67 & 2009.47 & 0.652 & 183.1 & 0.496 & 31.5 & 245.3 & 3 & sca2003d + & & @xmath562.70 & @xmath560.29 & @xmath560.012 & @xmath565.9 & @xmath560.013 & @xmath563.4 & @xmath564.7 & & + 19398@xmath92326 & see 389 & 46.20 & 1976.63 & 0.088 & 327.9 & 0.221 & 93.5 & 0.4 & 2 & doc1994a + & & @xmath560.10 & @xmath560.60 & @xmath560.007 & @xmath560.3 & @xmath560.001 & @xmath560.2 & @xmath564.7 & & + 19407@xmath90037 & chr 88 aa , ab & 20.56 & 1996.01 & 0.058 & 185.7 & 0.0854 & 150.8 & 328.1 & 3 & msn2010c + & & @xmath560.14 & @xmath561.04 & @xmath560.030 & @xmath5612.6 & @xmath560.0024 & @xmath567.0 & @xmath5627.5 & & + 19573@xmath690513 & a 604 & 228.15 & 1935.43 & 0.268 & 94.7 & 0.291 & 108.0 & 306.0 & 3 & hei1991 + & & @xmath5617.63 & @xmath565.63 & @xmath560.019 & @xmath561.9 & @xmath560.016 & @xmath561.2 & @xmath5612.2 & & + 19581@xmath94808 & hds 2842 & 32.04 & 2015.65 & 0.807 & 254.7 & 0.273 & 74.1 & 67.2 & 4 & + & & @xmath567.08 & @xmath560.23 & @xmath560.061 & @xmath567.9 & @xmath560.031 & @xmath560.9 & @xmath566.0 & & + 20210@xmath91447 & bla 7 aa , ab & 3.765 & 2015.704 & 0.043 & 34.3 & 0.0665 & 80.9 & 120.0 & 3 & msn1994 + & & * & @xmath560.035 & * & @xmath562.1 & @xmath560.0023 & @xmath563.2 & * & & + 20347@xmath96319 & hu 1615 & 357.17 & 1999.08 & 0.851 & 257.5 & 0.387 & 46.0 & 263.0 & 5 & usn2002 + 20507@xmath93116 & b 997 & 63.67 & 2014.34 & 0.179 & 281.1 & 0.207 & 41.5 & 306.4 & 3 & hrt2010a + & & @xmath5634.66 & @xmath5614.29 & @xmath560.142 & @xmath5627.2 & @xmath560.053 & @xmath568.1 & @xmath56141.1 & & + 20562@xmath93146 & b 1001 & 265.93 & 1969.51 & 0.369 & 187.0 & 0.322 & 131.2 & 219.0 & 4 & usn2002 + & & @xmath5610.98 & @xmath562.93 & @xmath560.021 & @xmath565.1 & @xmath560.016 & @xmath563.7 & @xmath569.4 & & + 21051@xmath690757 & hds 3004 ab & 80.42 & 2016.78 & 0.691 & 88.1 & 0.463 & 58.0 & 48.7 & 4 & lin2012a + & & @xmath56163.73 & @xmath561.39 & @xmath560.062 & @xmath5632.2 & @xmath560.326 & @xmath5621.2 & @xmath5663.5 & & + 21058@xmath95744 & hu 1625 & 560.04 & 2072.29 & 0.000 & 222.1 & 0.657 & 81.6 & 118.5 & 5 & + 21074@xmath90814 & bu 368 ab & 345.28 & 1954.64 & 0.665 & 270.1 & 0.576 & 87.6 & 304.2 & 3 & pal2005b + & & @xmath5615.75 & @xmath561.11 & @xmath560.020 & @xmath560.5 & @xmath560.013 & @xmath560.6 & @xmath561.9 & & + 21158@xmath95316 & fin 329 & 36.04 & 1997.91 & 0.115 & 260.5 & 0.179 & 107.1 & 284.8 & 3 & doc2013d + & & @xmath560.59 & @xmath563.77 & @xmath560.035 & @xmath563.3 & @xmath560.006 & @xmath560.7 & @xmath5629.5 & & + 21243@xmath690343 & a 2288 & 122.72 & 1950.41 & 0.659 & 135.4 & 0.300 & 139.4 & 86.1 & 3 & sca2003e + & & @xmath562.29 & @xmath560.50 & @xmath560.011 & @xmath564.3 & @xmath560.007 & @xmath561.8 & @xmath563.3 & & + 21255@xmath690203 & a 2289 ab & 159.45 & 2008.15 & 0.652 & 141.1 & 0.207 & 111.0 & 51.1 & 4 & usn2002 + & & @xmath5621.43 & @xmath561.85 & @xmath560.025 & @xmath564.2 & @xmath560.013 & @xmath563.1 & @xmath567.1 & & + 21274@xmath90701 & hds 3053 & 20.93 & 2015.86 & 0.349 & 152.8 & 0.166 & 51.1 & 148.5 & 2 & msn2010c + & & @xmath560.84 & @xmath560.35 & @xmath560.043 & @xmath562.2 & @xmath560.005 & @xmath561.7 & @xmath563.3 & & + 21436@xmath91108 & lv 10 & 1219.17 & 2038.43 & 0.494 & 9.1 & 2.039 & 36.2 & 4.2 & 5 & usn2002 + 21477@xmath91813 & chr 223 & 113.86 & 2018.85 & 0.748 & 284.3 & 0.365 & 99.8 & 71.5 & 5 & + 21536@xmath91019 & fin 358 & 49.38 & 1976.32 & 0.770 & 289.4 & 0.0791 & 30.5 & 9.2 & 3 & msn2001c + & & @xmath561.98 & @xmath561.86 & @xmath560.054 & @xmath5640.9 & @xmath560.0025 & @xmath5611.4 & @xmath5649.2 & & + 21552@xmath96153 & hdo 296 ab & 27.94 & 1997.59 & 0.420 & 282.0 & 0.276 & 83.2 & 41.4 & 2 & doc2011f + & & @xmath560.32 & @xmath560.41 & @xmath560.031 & @xmath560.4 & @xmath560.009 & @xmath560.5 & @xmath565.0 & & + 22029@xmath691547 & hds 3129 & 20.57 & 2005.43 & 0.777 & 230.1 & 0.100 & 124.9 & 273.1 & 3 & hor2010 + & & @xmath560.07 & @xmath560.02 & @xmath560.002 & @xmath560.5 & @xmath560.000 & @xmath560.2 & @xmath560.3 & & + 22156@xmath94121 & chr 187 & 25.87 & 2014.92 & 0.386 & 91.6 & 0.201 & 68.8 & 66.7 & 3 & + & & @xmath560.13 & @xmath560.05 & @xmath560.006 & @xmath560.4 & @xmath560.004 & @xmath560.7 & @xmath560.6 & & + 22241@xmath90450 & bu 172 ab & 145.07 & 1987.66 & 0.702 & 113.5 & 0.402 & 161.4 & 296.9 & 2 & doc2007d + & & @xmath561.85 & @xmath560.05 & @xmath560.003 & @xmath563.0 & @xmath560.003 & @xmath560.7 & @xmath563.0 & & + 22384@xmath90754 & a 2695 & 132.45 & 2027.46 & 0.576 & 285.5 & 0.252 & 76.8 & 47.9 & 3 & hrt2010a + & & @xmath564.91 & @xmath566.74 & @xmath560.039 & @xmath561.4 & @xmath560.022 & @xmath562.8 & @xmath567.8 & & + 22500@xmath93248 & hdo 301 & 26.46 & 2015.07 & 0.516 & 194.6 & 0.184 & 162.9 & 255.8 & 2 & hrt2010a + & & @xmath560.04 & @xmath560.04 & @xmath560.003 & @xmath565.2 & @xmath560.001 & @xmath561.5 & @xmath564.9 & & + 22546@xmath691054 & hds 3257 & 249.14 & 2007.00 & 0.415 & 51.2 & 0.344 & 67.1 & 20.9 & 4 & + & & @xmath5699.01 & @xmath5645.27 & @xmath560.361 & @xmath5618.4 & @xmath560.255 & @xmath5626.0 & @xmath56122.6 & & + 23227@xmath91502 & hu 295 & 64.33 & 2006.96 & 0.148 & 276.3 & 0.406 & 77.8 & 1.6 & 2 & usn1999b + & & @xmath560.31 & @xmath560.45 & @xmath560.003 & @xmath560.3 & @xmath560.002 & @xmath560.5 & @xmath562.9 & & + 23374@xmath690737 & fox 102 ab & 127.12 & 2006.75 & 0.292 & 139.1 & 0.263 & 22.2 & 259.4 & 3 & hrt2014b + & & @xmath563.20 & @xmath560.64 & @xmath560.010 & @xmath566.6 & @xmath560.005 & @xmath562.8 & @xmath567.8 & & +
the description of strong electron correlation effects is still nowadays a challenging problem for both density - functional theory ( dft ) and wavefunction theory ( wft ) communities . even though kohn sham dft ( ks - dft ) is in principle exact , standard approximate exchange correlation functionals usually do not enable an adequate description of multi - configurational systems . on the other hand , the `` gold standard '' single - reference _ coupled cluster _ ( cc ) method should also be able to model near - degeneracies but high - order excitations would then be necessary . in practice multi - reference ( mr ) perturbation theories such as the _ second order complete active space perturbation theory _ ( caspt2 ) @xcite or the _ n - electron valence state perturbation theory _ ( nevpt2 ) @xcite are usually employed . despite their success , these methods may suffer from the perturbative description of the short - range dynamical correlation which is sometimes not accurate enough . for these reasons multi - configurational extensions of both dft and cc have been investigated for many years and despite significant efforts and encouraging results , it is still unclear how the incorporation of a _ complete active space self - consistent field _ ( casscf ) into dft and cc should be performed . a recent review article by bartlett and coworkers @xcite gives an extensive picture of the developments in mrcc . we shall focus here on dft and restrict ourselves to the description of the ground state . + the rigorous combination of casscf with dft is already difficult at the formal level due to the infamous double counting problem . specific functionals must indeed be developed for complementing the casscf energy that already contains some correlation effects @xcite . the latter are usually referred to as static correlation effects . only the complementary correlation , known as dynamical correlation , should therefore be assigned to the density functional . turning such a scheme into a practical computational method is not an easy task since approximations used for standard functionals can not be applied straightfowardly in this context . one way to overcome the double counting problem consists in separating correlation effects in the coordinate space , as initially proposed by savin @xcite . the resulting range - separated dft methods combine rigorously long - range wft with short - range dft . even though range separation allows for a multi - configurational description of the electron density , it can not completely isolate static correlation from dynamical correlation simply because the former is usually not a purely long - range effect , even in a dissociated molecule . this point will be discussed further in the following . + let us mention that savin earlier combined configuration interaction ( ci ) with dft by separating correlation effects in the natural orbital space @xcite . orbitals with occupation numbers larger than a given threshold @xmath0 were correlated explicitly , at the ci level , while the remaining correlation energy was described with a complementary @xmath0-dependent density functional . more recently , gutl and savin @xcite proposed alternative ci - dft schemes where the correlation energy is still split in the orbital space but through the introduction of gap shift or cutoff parameters . in all these hybrid ci - dft schemes , a ci energy is complemented by a functional of the electron density . in this work we propose to revisit at the formal level the separation of correlation effects in the orbital space . as a major difference with the approaches discussed previously , orbitals occupation will be used as basic variable rather than the electron density . the idea originates from condensed matter physics where strongly correlated electrons are usually described with model hamiltonians ( such as the hubbard hamiltonian ) rather than the true physical one @xcite . as discussed further in the following , adapting such an approach to quantum chemistry is appealing since static correlation is usually defined in the orbital space . molecular orbitals can then be viewed as sites with an occupation to be determined . as shown in this work , a wft description of static correlation can be combined rigorously with an orbital - occupation functional modeling of dynamical correlation . a potential drawback of such a scheme lies in the fact that , in contrast to range - separated dft , the complementary functional loses its universality since it depends on the molecular orbitals . the theory is derived for a particular choice of orbitals , namely those obtained by diagonalizing the non - interacting hamiltonian ( kinetic and nuclear potential energy operators only are considered ) . its generalization is left for future work . + the paper is organized as follows : after a short introduction to the double counting problem in multi - configuration dft ( sec . [ sec:2blecounting ] ) , the exact multi - determinant extension of standard dft based on range separation is briefly reviewed ( sec . [ subsec : srdft ] ) and its application to the dissociated h@xmath1 molecule is discussed in sec . [ subsec : h2_min_basis ] . we then leave the framework of dft by using orbitals occupation as basic variable rather than the electron density . this approach is first introduced for model hamiltonians such as the hubbard hamiltonian ( sec . [ sec : soft ] ) . by isolating an impurity site and separating its on - site repulsion from the interactions on the remaining sites ( referred to as the bath ) , an exact _ site - occupation embedding theory _ ( soet ) is derived and compared with the _ density matrix embedding theory _ ( dmet ) of knizia and chan @xcite . illustrative results are then given for the simple two - site hubbard model in sec . [ sec : two_site_hubb ] . finally , we show in sec . [ sec : soft_in_qc ] that an exact combination of ci with orbital occupation functional theory is obtained when applying soet to a quantum chemical hamiltonian . conclusions and perspectives are given in sec . [ sec : conclusions ] . in wft the exact ground - state energy of an electronic system can be obtained variationally as follows , @xmath2 where @xmath3 is the kinetic energy operator , @xmath4 denotes the two - electron repulsion operator and @xmath5 is the nuclear potential operator . in a regular casscf calculation , the minimization in eq . ( [ eq : exactvp_wft ] ) is restricted to linear combinations of slater determinants that belong to a given _ complete active space _ ( cas ) . the orbitals are also optimized variationally . the cas is obtained by distributing a given number of active electrons in selected active orbitals that become consequently partially occupied . the selection of active electrons and orbitals is usually based on chemical intuition . doubly occupied orbitals are referred to as inactive orbitals . note that the latter are not frozen in a casscf calculation . the remaining orbitals , that are not occupied , are the virtuals . for convenience we will denote @xmath6 the space of trial casscf wavefunctions . the superscript @xmath7 specifies all the restrictions in the minimization ( number of active electrons and active orbitals for example ) . the casscf energy can therefore be written as @xmath8 note that , in practice , a trial casscf wavefunction @xmath9 will be parametrized in second quantization as follows @xcite , @xmath10 where @xmath11 is the basis of determinants for the cas . the singlet excitation operator @xmath12 that allows for orbital rotations is defined as @xmath13 where @xmath14 and @xmath15 can be inactive , active or virtual orbitals . consequently , the minimization in eq . ( [ casscf_ener_exp ] ) is performed over both @xmath16 and @xmath17 parameters . + the energy difference @xmath18 is usually referred to as dynamical correlation energy . the latter is routinely described within multi - reference perturbation theories such as caspt2 or nevpt2 . a combined casscf - dft approach should ideally be exact and variational . dynamical correlation would be described with a density functional @xmath19 $ ] that would complement the casscf energy . this can be formulated rigorously when considering the universal levy lieb functional @xcite , @xmath20 = \underset{\psi\rightarrow n}{\rm min}\langle \psi\vert \hat{t}+{\hat{w}_{\rm ee}}\vert\psi\rangle,\end{aligned}\ ] ] where the minimization is performed over all wavefunctions with density @xmath21 , and the following partitioning @xcite , @xmath22 = f^{m}[n]+\overline{e}^{{{\rm dyn}}}_{\rm c}[n ] . \end{aligned}\ ] ] the ll functional associated with the casscf space @xmath6 is defined as @xmath23 = \underset{\psi\in\mathcal{s}^{m}\rightarrow n}{\rm min}\langle \psi\vert \hat{t}+{\hat{w}_{\rm ee}}\vert\psi\rangle,\end{aligned}\ ] ] where the minimization is restricted to casscf wavefunctions in @xmath6 with density @xmath21 . according to the hohenberg kohn ( hk ) theorem @xcite , the exact ground - state energy can be obtained variationally as follows , @xmath24 + ( v_{\rm ne}\vert n ) \big\},\end{aligned}\ ] ] where the minimization is performed over electron densities @xmath25 that integrate to a fixed number @xmath26 of electrons . the notation @xmath27 has been used . combining eq . ( [ ll_cas_fun ] ) with eq . ( [ eq : hk2_theo ] ) leads to @xmath28+\overline{e}^{{{\rm dyn}}}_{\rm c}[n]+ ( v_{\rm ne}\vert n ) \big\}.\end{aligned}\ ] ] consequently , for any trial casscf wavefunction @xmath9 in @xmath6 , @xmath29}}&\geq & f^{{{m}}}[n_{\psi}]+{\overline{e}^{{{\rm dyn}}}_{\rm c}[n_{\psi}{]}}+(v_{\rm ne}\vert n_\psi ) \nonumber\\ & \geq & e,\end{aligned}\ ] ] where @xmath30 is the electron density obtained from the trial casscf wavefunction @xmath9 . thus we conclude that , provided that the exact ground - state density can be reproduced by a casscf wavefunction , the exact ground - state energy can be written as @xmath31 } } \big\}.\end{aligned}\ ] ] in order to turn eq . ( [ eq : var_casdft_ener ] ) into a practical computational method , approximate complementary correlation functionals should be developed . obviously standard correlation functionals can not be used otherwise correlation effects would be double counted . this double counting problem arises also when the size of the cas varies . in the exact theory , the functional should vary with the cas so that correlation effects are transferred from the casscf to the dft part of the energy and the total energy remains constant and equal to the exact ground - state energy . deriving cas - dependent density functionals is a complicated task simply because the electron density is defined in the coordinate space while a cas is defined in the orbital space . in order to rigorously overcome such difficulties , savin @xcite proposed to separate correlation effects in the coordinate space , thus leading to the so - called range - separated dft methods . this approach is briefly reviewed in sec . [ sec : mc - srdft ] . we then propose an alternative approach where correlation effects are separated in the orbital space , like in a regular casscf calculation . in this context , the orbitals occupation will be used as basic variable rather than the electron density . in the standard ks - dft scheme @xcite , the universal ll functional in eq . ( [ eq : universal_ll_fun ] ) is decomposed into the non - interacting kinetic energy functional @xmath32= \underset{\psi\rightarrow n}{\rm min}\langle \psi\vert \hat{t}\vert\psi\rangle $ ] and the hartree - exchange - correlation ( hxc ) energy functional , @xmath33=t_{\rm s}[n]+e_{\rm hxc}[n].\end{aligned}\ ] ] consequently , the hk variational principle can be reformulated as follows , @xmath34 \big\},\end{aligned}\ ] ] where the minimizing ks determinant @xmath35 fulfils the self - consistent equation @xmath36\,\hat{n}({\bf r})\bigg ) \vert\phi^{\rm ks}\rangle=\mathcal{e}^{\rm ks}\vert\phi^{\rm ks}\rangle.\end{aligned}\ ] ] therefore , within ks - dft , the two - electron interaction is fully described by a density functional . as originally shown by savin @xcite , it is in fact possible to describe only a part of the two - electron repulsion within dft and leave the remaining part to wft . this is achieved by separating the interaction in the coordinate space into two complementary contributions . the range separation based on the error function , @xmath37 has for example been used extensively in the last decade ( see ref . @xcite and the references therein ) , but any separation like the simpler linear one @xcite can be considered . range separation , that is controlled by the @xmath38 parameter , is appealing as it enables to isolate the coulomb hole and assign it to a density functional while long - range correlation can be described in wft . this hybrid range - separated wft - dft approach can be derived rigorously from the alternative partitioning of the ll functional , @xmath39= f^{\rm lr,\mu}[n]+\overline{e}^{\rm sr,\mu}_{\rm hxc}[n],\end{aligned}\ ] ] where the long - range ll functional equals @xmath40= \underset{\psi\rightarrow n}{\rm min}\langle \psi\vert \hat{t}+{\hat{w}^{\rm lr,\mu}_{\rm ee}}\vert\psi\rangle , \end{aligned}\ ] ] and @xmath41 $ ] is the complementary @xmath38-dependent short - range hxc density functional . let us stress that , for any @xmath38 value , this functional is universal since it depends on the electron density only . combining eq . ( [ eq : ll_range_separation ] ) with eq . ( [ eq : hk2_theo ] ) leads to the exact ground - state energy expression @xmath42 \right\},\end{aligned}\ ] ] where the minimizing wavefunction @xmath43 fulfils the self - consistent equation @xmath44 \hat{n}(\mathbf{r}).\end{aligned}\ ] ] note that ks - dft and pure wft are recovered when @xmath45 and @xmath46 , respectively . + since the long - range interaction is now described explicitly , the auxiliary wavefunction @xmath43 , that has exactly the same density as the physical system , is multi - determinantal . casscf can therefore be applied in this context in conjunction with local or semi - local short - range functionals @xcite . even though encouraging results were obtained with such a range - separated casdft scheme , better functionals are still needed for the method to be reliable @xcite . as discussed further in sec . [ subsec : h2_min_basis ] , one major problem is that static correlation is not a purely long - range correlation effect . + returning to the exact theory , an adiabatic connection ( ac ) @xcite expression for the short - range functional can be obtained from the auxiliary equations , @xmath47 and the density constraint @xmath48 for @xmath49 . indeed , according to eqs . ( [ eq : ll_range_separation ] ) and ( [ eq : ll_longrange ] ) , @xmath50&=&f[n]-f^{\rm lr,\mu}[n ] \nonumber\\ & = & \int^{+\infty}_\mu \dfrac{\ddroit\mathcal{e}^\nu}{\ddroit \nu}\ddroit\nu\ + ( v^\mu - v^{+\infty}\vert n),\end{aligned}\ ] ] thus leading , according to the hellmann feynman theorem , to @xmath51 & = & \int^{+\infty}_\mu \langle { \psi}^{\nu}\vert\partial \hat{w}^{\rm lr,\nu}_{\rm ee}/\partial \nu \vert{\psi}^{\nu } \rangle . \end{aligned}\ ] ] we obtain for @xmath45 a range - separated ac expression for the conventional hxc functional , @xmath52 & = & \int^{+\infty}_0 \langle { \psi}^{\nu}\vert\partial \hat{w}^{\rm lr,\nu}_{\rm ee}/\partial \nu \vert{\psi}^{\nu } \rangle . \end{aligned}\ ] ] it is readily seen from eqs . ( [ eq : ac_srfun ] ) and ( [ eq : ac_rs_convhxc ] ) that conventional functionals can not be used straightfowardly in multi - determinant range - separated dft otherwise there would be double counting of long - range correlation effects through the wft treatment . by connecting the non - interacting ks system ( @xmath53 ) to the long - range interacting one ( @xmath54 ) as follows , @xmath55-t_{\rm s}[n ] \nonumber\\ & = & { e}^{\rm lr,\mu}_{\rm hxc}[n],\end{aligned}\ ] ] where @xmath56 $ ] is the purely long - range hxc functional , we obtain from eqs . ( [ eq : ac_srfun ] ) and ( [ eq : ac_rs_convhxc ] ) the following expression , @xmath50={e}_{\rm hxc}[n]-{e}^{\rm lr,\mu}_{\rm hxc}[n],\end{aligned}\ ] ] that has been used by toulouse @xcite for developing approximate local and semi - local short - range functionals . we should stress that all the formalism briefly reviewed in this section is in fact general and can be applied in a different context , for example in dft for model hamiltonians , as proposed in sec . [ subsec : soet ] . we consider in this section the h@xmath1 molecule in a slater minimal basis consisting of the @xmath57 and @xmath58 atomic orbitals localized on the left and right hydrogen atoms , respectively @xcite . the basis functions are identical with @xmath59 . for large bond distances the bonding and anti - bonding molecular orbitals are equal to @xmath60 and @xmath61 , respectively . the ground state will then be written in the basis the two slater determinants @xmath62 and @xmath63 . since the latter differ by a double excitation , they are not coupled by one - electron operators . therefore , when approaching the dissociation limit , the matrix representation of the auxiliary long - range hamiltonian in eq . ( [ timeindepsrdfteq ] ) reduces to @xmath64=\left [ \begin{array}{c c } { e}^\mu & k^{{\mu } } \\ { k^{{\mu } } } & { e}^\mu \end{array } \right],\end{aligned}\ ] ] where diagonal elements are identical since atomic orbitals do not overlap , and the coupling term equals @xmath65 while the second long - range two - electron integral on the right - hand side of eq . ( [ eq : kmu_exp ] ) reduces to the regular one @xmath66 and becomes zero in the dissociation limit , the first term is an `` on - site '' integral computed with the long - range interaction . thus we conclude that the coupling between @xmath62 and @xmath63 determinants is determined by the contribution at short range of the long - range interaction . in this case static correlation , that is also referred to as left - right correlation , can obviously not be interpreted as a purely long - range correlation effect . nevertheless , if the error function is used , we see from its taylor expansion for small @xmath67 , @xmath68 that the coupling term can be expanded as follows @xmath69 consequently , as already pointed out by gori - giorgi and savin @xcite , even an infinitesimal @xmath38 value ensures that @xmath70 is not strictly equal to zero , thus providing the correct multi - configurational description of the dissociated h@xmath1 molecule in the ground state : @xmath71 this simple example illustrates how difficult it is to describe static correlation in the coordinate space , in contrast to short - range dynamical correlation that is connected with the coulomb hole . we propose in the rest of this paper an alternative approach where correlations are separated in the orbital space . in order to overcome the double counting problem , the orbitals occupation will be used rather than the electron density . for clarity , this approach will be introduced first for model hamiltonians . as mentioned previously , it is convenient to work in the orbital space rather than the coordinate space when it comes to separate static and dynamical correlation effects . a change of paradigm is then necessary in order to avoid the double counting problem . the basic variable in dft is the electron density @xmath72 that is defined in the coordinate space . as we want to reformulate dft in the orbital space , orbitals occupation @xmath73 seems to be the variable of choice . this is known in condensed matter physics as _ site - occupation functional theory _ ( soft ) @xcite . the latter is nothing but the formulation of dft for model hamiltonians such as the hubbard hamiltonian . after a short introduction to the ks - soft scheme in sec . [ subsec : ks - soft ] , we will show in sec . [ subsec : soet ] how formal analogies with range - separated dft can lead in this context to an exact embedding theory for model hamiltonians . let us consider the hubbard hamiltonian with an external potential @xmath74 : @xmath75 where @xmath76 is the hopping integral , @xmath77 denotes the on - site two - electron repulsion , @xmath78 and @xmath79 . the site - occupation operator equals @xmath80 and , for a given wavefunction @xmath9 , the occupation of site @xmath81 is defined as @xmath82 . for simplicity we will consider a fixed number of electrons @xmath26 and only discuss ground - state properties . as shown by gunnarsson and schnhammer @xcite , the hk theorem can be adapted to the hamiltonian in eq . ( [ eq : hubb_hamil_ext_pot ] ) . there is indeed a one - to - one correspondence between the external potential @xmath83 and the ground - state sites occupancy @xmath84 . consequently , for fixed @xmath76 and @xmath77 parameters , the exact ground - state energy can be obtained from the following variational principle , @xmath85 where @xmath86 . the analog of the universal hk functional is a function of the sites occupation that can be written within the ll constrained - search formalism as @xmath87 since the hk theorem still holds when @xmath88 , a ks formulation of soft ( ks - soft ) is obtained from the following partitioning , @xmath89 where the non - interacting kinetic energy functional is defined in analogy with ks - dft as @xmath90 note that , in this context , the non - interacting ks system is _ not _ described by sites that are strictly singly , doubly or not occupied , @xmath91 in other words @xmath92 is not a single slater determinant in the basis of the site orbitals . indeed , @xmath92 is the ground state of the non - interacting hamiltonian @xmath93 where @xmath94 is _ non - local _ in the orbital space . let us stress that there is no hk theorem without a non - local term like @xmath94 included into the hamiltonian . for example , for two electrons , the two hamiltonians that differ by more than a constant , @xmath95 and @xmath96 where @xmath97 and @xmath98 , yield the same ground - state occupancies @xmath99 and @xmath100 . ( 1,0 ) ; ( -0.2,0 ) node [ left]@xmath101 ; ( 0.4,-0.5 ) ( 0.4,0.5 ) ; ( 0.6,-0.5 ) ( 0.6,0.5 ) ; ( 0,1.4 ) ( 1,1.4 ) ; ( -0.2,1.4 ) node [ left]@xmath102 ; ( 0,0 ) ( 1,0 ) ; ( -0.2,0 ) node [ left]@xmath101 ; ( 0.4,-0.5 ) ( 0.4,0.5 ) ; ( 0.6,-0.5 ) ( 0.6,0.5 ) ; ( 0,2.0 ) ( 1,2.0 ) ; ( -0.2,2.0 ) node [ left]@xmath103 ; this is an important difference with ks - dft where the quantity to reproduce from singly or doubly occupied ks orbitals is the exact electron density @xmath72 . in ks - soft , the quantity to reproduce is the exact sites occupation @xmath21 that may be fractional due to the non - zero hopping integral @xmath104 . + by analogy with ks - dft , an exact ac expression for the hxc site - occupation functional in eq . ( [ eq : hxcsofun_def ] ) can be obtained from the following auxiliary equations , @xmath105 with the site - occupation constraint @xmath106 fulfilled for @xmath107 , thus leading to @xmath108 according to the hellmann feynman theorem . note that within the mean - field approximation and in the particular case of a singlet ground state , the hxc functional is simplified as follows , @xmath109 the latter expression is usually referred to as the hartree energy in condensed matter physics @xcite . the exchange energy is then considered to be zero . in the spirit of _ dynamical mean - field theory _ @xcite and dmet @xcite , we propose to isolate one site , referred to as impurity and for which correlation effects will be described explicitly , while the remaining sites ( the bath ) are treated within soft . such an approach , that will be referred to as _ site - occupation embedding theory _ ( soet ) in the following , is formally very similar to range - separated dft . the main difference is that the interactions are separated here in the orbital space rather than in the coordinate space . for convenience the impurity and bath sites will be labelled as @xmath110 and @xmath111 , respectively . by analogy with eqs . ( [ eq : ll_range_separation ] ) and ( [ eq : ll_longrange ] ) , soet can be derived from the following partitioning of the ll functional , @xmath112 where the impurity ll functional equals @xmath113 and @xmath114 denotes the complementary hxc bath functional that describes the bath as well as the coupling between the bath and the impurity . note that , in contrast to range - separated dft , the separation of correlation effects is , in soet , not controlled by a single continuous parameter . it rather relies on the selection of an impurity site . as discussed in the following more impurity sites can actually be considered . + we now return to the single - impurity case . since , according to eqs . ( [ eq : vp_soft ] ) , ( [ eq : fimp_plus_ebath ] ) and ( [ eq : llimpfun ] ) , any normalized trial wavefunction @xmath9 fulfils @xmath115 the exact ground - state energy can be expressed as @xmath116 where the minimizing wavefunction @xmath117 fulfils the self - consistent equation @xmath118\hat{n}_i\bigg)\vert{\psi}^{\rm imp}\rangle \nonumber \\ & & = { \mathcal{e}}^{\rm imp } \vert{\psi}^{\rm imp}\rangle . \end{aligned}\ ] ] thus we obtain an exact embedding scheme where the bath is described by a site - occupation functional . its connection with dmet will be discussed in the following . let us first focus on the complementary bath functional . from the ks decomposition @xmath119 where @xmath120 describes the repulsion on the impurity that is embedded into a non - interacting bath , we obtain with eqs . ( [ eq : hxcsofun_def ] ) and ( [ eq : fimp_plus_ebath ] ) , @xmath121 this expression is convenient for developing approximate functionals . for example , within the mean - field approximation , the hxc functional for the impurity is simplified as follows @xmath122 thus leading , when combined with eq . ( [ mf_approx_hxc ] ) , to @xmath123 better approximate functionals could be developed from the local density approximation for a luttinger liquid @xcite , in complete analogy with the development of short - range functionals from the uniform electron gas @xcite . work is currently in progress in this direction . + returning to the exact theory , a more explicit expression for the complementary bath functional can be obtained within the ac formalism when considering , in analogy with eq . ( [ eq : ac_full_scale ] ) , the following auxiliary equations @xmath124 where @xmath125 for @xmath126 , @xmath127 , and @xmath128 . in such an ac the on - site repulsion is switched on continuously site after site . a graphical representation is given in fig . [ fig : ac_graph_rep ] . c the embedded impurity system is recovered when @xmath129 and @xmath130 or , equivalently , when @xmath131 and @xmath132 . from eqs . ( [ eq : llimpfun ] ) and ( [ eq : ll_imp_ksdecomp ] ) , the site - occupation constraint @xmath133 and the hellmann feynman theorem , we obtain @xmath134 by introducing the multiple - impurity ll functional @xmath135 the ( fully - interacting ) hxc functional becomes @xmath136 since @xmath137 , thus leading to the alternative expression ( see eq . ( [ eq : hxc_fun_acexp ] ) ) , @xmath138 combining eqs . ( [ eq : bathfun_as_diff ] ) , ( [ eq : impfun_ac_formula ] ) and ( [ eq : hxc_fun_ac_alternative ] ) finally gives the exact expression @xmath139 if the exact complementary bath functional was known , its derivative could be computed and , by solving eq . ( [ eq : sc_eq_imp ] ) self - consistently , we would obtain the exact embedded impurity wavefunction @xmath117 that has the same sites occupation @xmath21 as the physical system described by the hubbard hamiltonian in eq . ( [ eq : hubb_hamil_ext_pot ] ) : @xmath140 if @xmath21 is known , the exact embedding potential @xmath141 can in fact be reached without knowing the complementary bath functional . indeed , according to the variational principle , for any local potential @xmath142 , the embedded impurity wavefunction fulfils @xmath143 where @xmath144 is the ground - state energy of @xmath145 . thus we obtain from eqs . ( [ eq : llimpfun ] ) , ( [ eq : sc_eq_imp ] ) and ( [ eq : nimp_equals_n ] ) , the legendre fenchel transform expression @xcite @xmath146 where @xmath21 is the input sites occupation and @xmath147 the embedding potential to be optimized . in practice approximations to @xmath21 could be used for example by performing a mean - field description of the physical system . in this case , the ( approximate ) embedding potential would enable to reproduce the mean - field occupation in both the impurity and the bath . + thus a connection with dmet can be established ( see eq . ( 5 ) in ref . a substantial difference though is that , in dmet , the potential to be optimized is the mean - field interaction operator @xmath147 that defines the mean - field approximation to the hubbard hamiltonian , @xmath148 and the sites occupation to be reproduced by the ground state of @xmath149 is the one obtained from the ground state of the dmet impurity hamiltonian , @xmath150 where @xmath151 . the terms @xmath152 and @xmath153 are defined in ref . @xcite and depend indirectly on @xmath147 . note that for simplicity dmet equations are written here for a uniform external potential . + at a given iteration @xmath154 of a dmet calculation , the mean - field potential is set to @xmath155 and the impurity hamiltonian generates the ground - state sites occupation @xmath156 . if it exists a local potential @xmath157 such that the ground state @xmath158 of @xmath159 has the same sites occupation then , according to the variational principle , any trial potential @xmath147 fulfils @xmath160 where @xmath161 is the ground - state energy of @xmath149 , thus leading to the legendre fenchel transform expression : @xmath162 if a maximum exists , it then corresponds to @xmath163 . the updated occupation @xmath164 is then obtained from @xmath165 , thus generating a new potential until convergence is reached . in this respect , dmet can be considered as a ks _ optimized effective potential _ ( oep ) scheme since the mean - field hamiltonian is in this context nothing but a non - interacting hamiltonian . the oep is used to model the interaction on the impurity . a formal analogy can actually be made with the range - separated ks - oep approach proposed in ref . @xcite , where the long - range interaction is described with an oep while the short - range interaction is modeled with a density functional . the oep is then obtained from a density constraint , exactly like in dmet . + let us finally mention that an alternative and exact formulation of dmet is obtained when introducing the following partitioning of the ll functional @xmath166 where @xmath167 in connect to connect dmet with soet one may fix @xmath147 and let @xmath153 vary independently as a potential to be optimized . if it exists a potential @xmath168 such that the ground state @xmath169 of @xmath170 has its site occupation equal equal to the one generated by @xmath149 , that we denote @xmath171 , then we have according to the variational principle , @xmath172 at the first iteration of a dmet calculation , the latter will be fixed and for any new trial potential @xmath147 , the variational principle leads to @xmath173 let us finally mention that , in practical soet calculations , the embedding potential would be obtained self - consistently by using an approximate ( local ) hxc functional for the bath and solving eq . ( [ eq : sc_eq_imp ] ) . the computational implementation of soet is left for future work . an illustration is given in sec . [ sec : two_site_hubb ] with the simple two - site hubbard hamiltonian . + returning to soet , let us finally mention that the theory can be extended to more impurity sites simply by using the following partitioning of the ll functional , @xmath174 where @xmath175 . in this case , the ac formula for the complementary bath functional becomes @xmath176 soet is applied in this section to the simple two - site hubbard model with a uniform external potential . the analytical construction of the ac is presented in sec . [ subsec : ac_for_two_sites ] . the resulting integrand expressions are then analyzed in sec . [ subsec : results ] . let us consider a two - electron system described by the two - site hubbard hamiltonian @xmath177 since we are interested here in the singlet ground state only , the matrix representation of the hamiltonian can be reduced to the basis of the `` doubly - occupied site '' states @xmath178 with @xmath110 or @xmath179 , and @xmath180 that corresponds to singly - occupied sites , thus leading to @xmath181 & = & \left [ \begin{array}{c c c } u_0+v_0-v_1 & 0 & -\sqrt{2}t\\ 0 & u_1+v_1-v_0 & -\sqrt{2}t\\ -\sqrt{2}t & -\sqrt{2}t & 0 \end{array } \right ] + ( v_0+v_1 ) .\end{aligned}\ ] ] for simplicity we choose for the physical hamiltonian @xmath182 and @xmath183 . consequently the site occupation is uniform : @xmath184 we obtain by diagonalization the well - known expressions for the ground - state energy @xmath185 and the corresponding wavefunction @xmath186 where @xmath187 since @xmath188 , thus leading to the more explicit expression @xmath189 note the hellmann feynman theorem , @xmath190 that will be used in the following . the double occupancy is the same for both sites and equal to @xmath191 along the ac described in eq . ( [ eq : ac_impurities ] ) and fig . [ fig : ac_graph_rep ] , symmetry in the on - site repulsions is broken in the auxiliary hamiltonian operator . this is clear for the embbeded impurity where @xmath192 and @xmath193 . nevertheless , symmetry can be restored in the hamiltonian matrix simply by adjusting the local potential components @xmath194 and @xmath195 such that @xmath196 thus leading to @xmath197 the latter condition defines uniquely ( up to a constant ) the local embedding potential that gives a uniform site occupation . in this case the effective on - site repulsion is the same on each site and is simply expressed as @xmath198 therefore the ac can be constructed analytically as follows for the impurity ( @xmath129 ) , @xmath199 and for the bath ( @xmath131 ) , @xmath200 consequently the wavefunctions along the ac for the bath and the impurity are @xmath201 and @xmath202 respectively . the corresponding double occupancies , that are nothing but hxc integrands per unit of @xmath77 for the impurity and the bath ( see eqs . ( [ eq : impfun_ac_formula ] ) and ( [ eq : ac_comp_bath_exp ] ) ) , respectively , can then be expressed , according to eqs . ( [ eq : explicit_exp_2bleocc ] ) and ( [ eq : same_2bleocc ] ) , as @xmath203 in order to derive the expression for the bath integrand in eq . ( [ eq : ac_comp_bath_exp ] ) , we express the embedded wavefunction along the ac as @xmath204 where the coefficient @xmath205 is obtained by diagonalizing the hamiltonian matrix in eq . ( [ hmatrix ] ) and using eqs . ( [ eq : hubb_gs_ener ] ) and ( [ eq : ac_p1 ] ) , thus leading to @xmath206 since @xmath207 we obtain the alternative expression @xmath208 and @xmath209 note that the embedded impurity wavefunction is obtained when @xmath130 along the ac for the impurity , and @xmath132 along the ac for the bath : @xmath210 since , according to eqs . ( [ hmatrix ] ) and ( [ eq : auxhamil_ac_bath ] ) , @xmath211 it comes from the site occupation constraint in eq . ( [ eq : so_constraint_two_sites ] ) that @xmath212 moreover , according to eqs . ( [ hell feyn_th_2sites ] ) and ( [ eq : wf_ac_bath ] ) , the double occupancy can be rewritten along the ac for the bath as @xmath213 thus giving with eq . ( [ eq : ac_comp_bath_exp ] ) , @xmath214 or , more explicitly , @xmath215 as expected , the physical energy @xmath216 is recovered when adding contributions from the impurity and the bath : @xmath217 let us finally mention that , in the conventional ac ( see eq . ( [ eq : ac_full_scale ] ) ) , the on - site repulsion is scaled on both the impurity and the bath sites . in this case the interaction on the impurity is not separated from the repulsions in the bath . the corresponding ac will therefore be defined as @xmath218 and @xmath219 as a result the ac integrand can be expressed as @xmath220 thus leading to the explicit expression @xmath221 note that , according to eq . ( [ hell feyn_th_2sites ] ) , the hxc integrand can also be rewritten as @xmath222 finally by rewriting the double occupancy along the ac for the impurity ( see eq . ( [ eq:2ble_occ_imp_explicit_exp ] ) ) as follows , @xmath223 we see from eqs . ( [ eq : hxc_fun_acexp ] ) , ( [ eq : hxc_fun_ac_alternative ] ) , ( [ eq:2ble_occ_ac_bath ] ) and ( [ eq : convac_hellth_int ] ) that , as expected , both the conventional hxc integrand and the sum of bath and impurity integrands lead to the same hxc energy after integration over [ 0,1 ] : @xmath224 or , more explicitly , @xmath225 * * * * * * * * * * * * * * * * * * * * * * * * + the embedded impurity wavefunction is actually obtained for @xmath132 , @xmath226 so that since sites are both singly occupied , it comes @xmath227 when defining the hartree exchange energy as the mean field one , that is @xmath228 for a singly occupied site , the bath correlation energy can be written as @xmath229 where the bath correlation integrand equals @xmath230 ac for the impurity : @xmath231 c double occupancies along the ac for the impurity and the bath ( see eqs . ( [ eq:2ble_occ_imp_explicit_exp ] ) and ( [ eq:2ble_occ_bath_explicit_exp ] ) ) , that can be identified as hxc integrands per unit of @xmath77 , have been plotted for various correlation regimes . results are shown in figs . [ fig:2site_hubb_imp_ac ] and [ fig:2site_hubb_bath_ac ] . for analysis purposes , comparison is made between the sum of these two integrands and the conventional hxc integrand in eq . ( [ eq : conv_ac_int_explicit_exp ] ) . results are shown in fig . [ fig:2site_hubb_bath_comparacs ] . + let us first focus on the impurity . as expected from accurate quantum chemical calculations of the ac for the h@xmath1 molecule at equilibrium @xcite , the integrand varies linearly with the interaction strength @xmath232 in the weakly correlated regime ( @xmath233 ) . the integrand gains curvature as @xmath234 increases . in the strongly correlated regime ( @xmath235 ) , the double occupancy of the impurity becomes zero for large interaction strengths ( see the bottom panel in fig . [ fig:2site_hubb_imp_ac ] ) . it only varies in the vicinity of the non - interacting case ( @xmath132 ) with a large negative slope . this pattern was found by teale for the stretched h@xmath1 molecule when computing the ac at the _ full configuration interaction _ ( fci ) level in a large basis set ( see fig . 5 ( a ) in ref . one should mention that , along the ac for the impurity , the on - site repulsion in the bath is set to zero . comparison with ref . @xcite is then more relevant when considering the conventional ac integrand for the two sites . as shown in fig . [ fig:2site_hubb_bath_comparacs ] , all correlation regimes observed numerically for h@xmath1 along the ac and the bond breaking coordinate are qualitatively well reproduced by applying soft to the two - site hubbard hamiltonian . + c turning to the ac for the bath , we first notice in the top panel of fig . [ fig:2site_hubb_bath_ac ] that the double occupancy of the bath remains essentially linear in @xmath232 for all correlation regimes , in contrast to the ac for the impurity . this is due to the fact that the on - site repulsion is already switched on for the impurity . scaling is only used for the repulsion on the bath site so that the embedded impurity ( @xmath132 ) can be connected with the physical system ( @xmath130 ) . as expected , the two systems exhibit different double occupancies even though they have the same sites occupation ( see the bottom panel in fig . [ fig:2site_hubb_bath_ac ] ) . variations with the correlation strength are consistent with those obtained for the 1d - hubbard model ( see case @xmath236 in fig . 2 of ref . + c let us finally discuss the separation of interactions between the impurity and the bath . as mentioned previously and illustrated in fig . [ fig:2site_hubb_bath_comparacs ] , both the conventional and bath+impurity ac integrands give the same total hxc energy after integration over @xmath237 $ ] but they of course differ along the ac . in the weakly correlated regime , both the bath and the impurity contribute significantly to the hxc energy . note that , in practical calculations , we want to describe the embedded impurity only , meaning that the contributions of the bath to the hxc energy will be described by a site - occupation functional . it becomes clear from this simple example that the latter can contribute significantly and that approximations beyong the mean field ( which is exact only when @xmath238 ) will be needed . returning to the ac , in the strongly correlated regime , the contribution of the bath becomes negligible for all interaction strengths . note that , for @xmath132 , the impurity and conventional hxc integrands are equal to @xmath239 and @xmath240 , respectively . since both are non - zero only for small interaction strengths and behave similarly ( with a large negative slope ) , they both integrate to the same hxc energy simply because the effective on - site repulsion along the ac is smaller for the impurity alone ( @xmath241 ) than for the two sites ( @xmath242 ) . we propose in this section to apply soft to quantum chemical hamiltonians . the formulation of a ks scheme in this context is discussed in sec . [ subsec : ks - soft_qc ] . we then show in sec . [ subsec : cassoft ] that soet can lead to an alternative casdft method . in order to adapt soft to quantum chemistry , let us first consider the orthonormal basis of molecular orbitals @xmath243 obtained for non - interacting electrons , so that @xmath244 with @xmath245 this is obviously not a good starting point for conventional quantum chemical calculations but it is convenient for deriving a ks - soft scheme in this context . as explained further in the following , the theory can in principle be adapted to different choices of orbitals . using eq . ( [ eq : orbitals_soft_qc ] ) leads to the following second - quantized expression for the molecular hamiltonian @xmath246 , @xmath247 where @xmath248 denotes the fluctuation potential . the hartree fock ( hf ) potential operator can be decomposed as follows , @xmath249 where @xmath250 denotes a regular two - electron integral . the first term on the right - hand side of eq . ( [ eq : uhfdecomp ] ) is local in the orbital space while the second one is non - local and thus enables hopping between occupied and unoccupied orbitals , hence the notation @xmath251 in analogy with the kinetic energy operator in the hubbard hamiltonian . therefore the molecular hamiltonian becomes @xmath252 with @xmath253 . this expression enables a direct comparison with the hubbard hamiltonian in eq . ( [ eq : hubb_hamil_ext_pot ] ) : the sites are now molecular orbitals and their energies @xmath254 play the role of the external potential . + following sec . [ subsec : ks - soft ] , we introduce the ll functional @xmath255 so that the exact ground - state energy , that is the fci energy in this context , can be written as @xmath256 let us consider the ks partitioning , @xmath257 where @xmath258 and , in contrast to eq . ( [ eq : hxcsofun_def ] ) , no hartree and exchange energy contributions have been introduced . indeed , the fluctuation potential induces correlation effects only since , when written with normal ordering @xcite , it generates double excitations only . consequently , within ks - soft , the fci energy is obtained variationally as follows , @xmath259 where the minimizing wavefunction @xmath260 fulfils the self - consistent equation @xmath261\hat{n}_p\bigg)\vert{\psi^{\rm ks}}\rangle = { \mathcal{e}^{\rm ks}}\vert{\psi^{\rm ks } } \rangle.\end{aligned}\ ] ] eqs . ( [ eq : fciener_var_soft ] ) and ( [ eq : ks - soft_eq_qc ] ) should in principle enable to recover the fci energy without treating electron correlation explicitly . orbital rotations would therefore be sufficient . one should of course investigative potential @xmath262-representability problems for example by computing legendre fenchel transforms . approximate functionals may also be developed from model hamiltonians such as the hubbard hamiltonian . let us finally note that different orbitals could be used in eq . ( [ eq : ks - soft_eq_qc ] ) . one would need to adapt the correlation functional in order to recover the correct fci energy . work is in progress in these directions . following knizia and chan @xcite , we propose in this section to apply soet to the molecular hamiltonian in eq . ( [ eq : mol_hamil_like_hubb ] ) . the impurity sites will be the active orbitals ( @xmath263 ) whose selection is usually based on chemical intuition , while the inactive and virtual orbitals correspond to the bath . a graphical representation is given in fig . [ fig : active_space_graph_rep ] . as already mentioned for soet , the separation of correlation effects is not controlled here by a single continuous parameter like in range - separated dft . in the spirit of a regular casscf calculation , it rather relies on the selection of active orbitals . this ensures that only correlation effects within the active space are described explicitly in wft while the remaining correlation , including core correlation , is modeled by an orbital - occupation functional . @xmath264 separating correlation effects in the active orbital space @xmath265 from the remaining ones leads to the alternative partitioning of the ll functional , @xmath266 where @xmath267 and @xmath268 is the reduction of the fluctuation potential operator to the active orbital space : @xmath269 note that the hf potential in the second term on the right - hand side of eq . ( [ eq : fluctu_pot_active ] ) is here calculated with the active orbitals that are doubly occupied , by analogy with eq . ( 31 ) in ref . @xcite . combining eqs . ( [ eq : fciener_vp_n ] ) , ( [ eq : part_f_imp_dyn ] ) and ( [ eq : fimp_ll_qc ] ) leads to the exact variational expression @xmath270 where the minimizing wavefunction @xmath271 fulfils the self - consistent equation @xmath272\hat{n}_p\bigg)\vert{\psi_{\mathcal a}^{\rm imp}}\rangle = { \mathcal{e } _ { \mathcal a}^{\rm imp}}\vert{\psi_{\mathcal a}^{\rm imp } } \rangle.\end{aligned}\ ] ] the method will be referred to as cassoft . interestingly the impurity hamiltonian is very similar to the one proposed by zgid and chan in ref . a major difference is that the embedding potential @xmath273 should in principle enable to recover the exact orbitals occupancy in both the bath and the impurity and , through the functional , the fci energy . the impurity hamiltonian can also be viewed as an embedded version of dyall s hamiltonian @xcite . @xmath274k . g. dyall , j. chem . 102 , 4909 ( 1995 ) . + @xmath275c . angeli , r. cimiraglia , and j .- malrieu , j. chem . 117 , 9138 ( 2002 ) . a numerical validation of eq . ( [ eq : sc_cassoft ] ) for small molecules can be achieved when computing from the input fci orbitals occupation @xmath21 the legendre fenchel transform @xmath276 where @xmath277 is the ground - state energy of @xmath278 . work is in progress in this direction . + let us stress that , in contrast to the short - range density functional used in range - separated dft , the complementary bath orbital - occupation functional @xmath279 is not a universal functional . it will _ a priori _ depend on the molecular orbital basis in which cassoft is formulated . it also depends on the active orbital space . even though range - separated dft is much simpler in that respect , the choice of an optimal range - separation parameter is , in terms of accuracy , not universal partly because static correlation is not a purely long - range effect . local and semi - local approximations to the short - range density functional are usually not accurate enough for modeling strongly multi - configurational systems ( see ref . @xcite and the references therein ) . working in the orbital space will at least allow for a proper separation of static and dynamical correlation effects . one may consider the development of approximate functionals to be a cumbersome task in this context . a similar challenge is to some extent encountered in _ natural orbital functional theory _ ( noft ) @xcite . recent advances in the field might be useful for such developments . this is left for future work . + returning to exact cassoft , an ac formula can actually be derived for the complementary bath correlation functional @xmath279 by introducing orbital - dependent active spaces @xmath280 where the indices @xmath281 and @xmath282 refer to the lowest and highest active orbitals in energy , respectively . let @xmath283 denote the active space of interest @xmath265 . we then consider the auxiliary equations @xmath284 where the orbitals occupation constraint @xmath285 is fulfilled for @xmath107 , @xmath286 and @xmath287 . the virtual and inactive increment operators are defined as @xmath288 respectively . the explicit expression for @xmath289 is deduced from eq . ( [ eq : fluctu_pot_active ] ) . the superscripts `` @xmath290 '' and `` @xmath291 '' refer to the incorporation into the active space @xmath292 of the virtual @xmath293 and inactive @xmath294 orbitals , respectively . a graphical representation is given in fig . [ fig : active_space_graph_rep ] . c the embedded active electrons are recovered along the ac when @xmath295 , @xmath296 and @xmath132 . according to eqs . ( [ eq : part_f_imp_dyn ] ) and ( [ eq : fimp_ll_qc ] ) , the complementary bath correlation functional can be expressed as @xmath297 thus leading , by analogy with eq . ( [ eq : hxc_fun_ac_alternative ] ) , to @xmath298 it is readily seen from eq . ( [ eq : ac_bath_cassoft ] ) that the complementary bath correlation functional describes correlation effects that ( i ) couple inactive or active orbitals with the virtual orbitals ( first term on the right - hand side ) and that ( ii ) couple inactive orbitals with active orbitals ( second term ) . the correlation effects within the active space are treated explicitly . note that cassoft is in principle exact ( in a given basis set ) and it is free from double counting problems . the reason is that it relies on an exact separation of the ll functional in the orbital space . the exact formulation of multi - configuration density - functional theory has been discussed . the infamous double counting problem can be avoided when separating correlation effects either in the coordinate space or in the orbital space . in the latter case , orbitals occupation should be used as basic variable rather than the electron density . this approach has been applied to the hubbard hamiltonian , thus leading to an exact _ site occupation embedding theory _ ( soet ) . the connection with _ density matrix embedding theory _ ( dmet ) has been discussed and an adiabatic connection ( ac ) formula has been derived for the complementary bath functional . the ac has been constructed analytically for the simple two - site hubbard model . the computational implementation of soet as well as the development of approximate local bath functionals is left for future work . + we then proposed to apply soet to a quantum chemical hamiltonian , thus showing that multi - configuration methods can be merged rigorously with orbital - occupation functionals . the method is referred to as cas _ site occupation functional theory _ ( cassoft ) . in this context , impurity sites correspond to the active orbitals while inactive and virtual orbitals are the bath . an ac formalism has also been derived . the latter should be useful for developing approximate functionals . a connection with multi - reference perturbation theory may be achieved from a perturbation expansion of the ac integrand . note also that , in regular dft , the uniform electron gas played a crucial role in the development of electron density functionals . by analogy , model hamiltonians such as the hubbard hamiltonian could be used for developing local and semi - local orbital - occupation functionals . finally , following knizia and chan @xcite , it would be interesting to explore connections between cassoft and _ density matrix renormalization group _ ( dmrg ) methods @xcite . work is in progress in these directions . the author would like to thank the editorial board for the kind invitation to submit a `` new views '' paper to mol . ef is grateful to markus reiher and stefan knecht for their kind invitation to give a seminar at eth in august 2014 and their helpful comments on this work . ef thanks masahisa tsuchiizu , vincent robert , laurent mazouin , andreas savin , lucia reining and bernard amadon for stimulating discussions on strongly correlated electrons . the author finally acknowledges financial support from the phc program sakura 2969uk , the labex `` chemistry of complex systems '' and the anr ( mcfunex project ) .
in recent years , a considerable amount of research has focused on the study of graph structures arising from technological , biological and sociological systems . graphs are the tool of choice in modeling such systems since the latter are typically described as a set of pairwise interactions . important examples of such datasets are the internet graph ( vertices are routers , edges correspond to physical links ) , the web graph ( vertices are web pages , edges correspond to hyperlinks ) , social networks ( vertices are humans , edges correspond to friendships ) , information networks like facebook and linkedin ( vertices are accounts , edges correspond to online friendships ) , biological networks ( vertices are proteins , edges correspond to protein interactions ) , math collaboration network ( vertices are mathematicians , edges correspond to collaborations ) and many more . towards the end of the 90s , a series of papers observed that the classic models of random graphs introduced by erds and rnyi @xcite and gilbert @xcite did not explain the empirical properties of real - world networks @xcite . typical properties of such networks include skewed degree distributions @xcite , large clustering coefficients @xcite and small average distances @xcite , a phenomenon typically referred to as `` small worlds '' . skewed degree distributions have widely been modeled as power laws : the proportion of vertices of a given degree follows an approximate inverse power law , i.e. , the proportion of vertices of degree k is approximately @xmath14 . understanding the properties of real world networks has attracted considerable research interest in the recent years @xcite . a large body of work has focused on finding models which generate graphs which mimick real world networks . most of the existing work on network formation models falls into the stochastic and the strategic categories . kronecker graphs @xcite , the cooper - frieze model for the web graph @xcite , the aiello - chung - lu model @xcite , protean graphs @xcite and numerous other models belong to the former category . the strategic approach has its origins in the work of boorman @xcite , aumann @xcite and aumann and myerson @xcite and a variety of models fall into this category , e.g. , the fabrikant - koutsoupias - papadimitriou model @xcite . despite the large amount of work on models which generate power law graphs , a considerably smaller amount of work has focused on generative models for planar graphs . planar graphs have been studied mainly in the context of transportation networks and occur in numerous real - world problems : modelling city streets @xcite , crowd simulations @xcite , river networks , railway and road maps printed circuits , chemical molecules , see also @xcite and references therein . in this work , we prove fundamental properties of random apollonian networks ( rans ) @xcite , a popular model of planar graphs with power law properties @xcite . we use the symbol @xmath15 to denote the random graph at time @xmath2 . the details of the model are exposed in section [ sec : ran ] . specifically our main results are the following theorems : let @xmath16 be the @xmath0 highest degrees of the random apollonian network @xmath15 at time @xmath2 where @xmath0 is a fixed positive integer . also , let @xmath17 be a function such that @xmath4 as @xmath5 . then _ whp _ @xmath18 and for @xmath19 @xmath20 [ thrm : thrm1 ] the growing function @xmath17 can not be removed , see @xcite . using theorem [ thrm : thrm1 ] we relate the highest @xmath0 degrees and eigevalues . let @xmath21 be the largest @xmath0 eigenvalues of the adjacency matrix of @xmath15 . then _ whp _ @xmath22 [ thrm : thrm2 ] also , we prove the following theorem for the diameter . the diameter @xmath10 of @xmath15 satisfies in probability @xmath23 where @xmath12 is the unique solution greater than 1 of the equation @xmath13 . [ thrm : ternary ] the outline of the paper is as follows : in section [ sec : ran ] we describe the model and discuss existing work on rans . in sections [ sec : degree ] and [ sec : eigen ] we prove theorems [ thrm : thrm1 ] and [ thrm : thrm2 ] respectively . in section [ sec : diameter ] we give a simple proof for the asymptotic growth of the diameter and prove theorem [ thrm : ternary ] . finally in section [ sec : waiting ] we investigate another property of the model . finally , in section [ sec : concl ] we conclude by proposing several problems for future work . apollonius of perga was a greek geometer and astronomer noted for his writings on conic sections . apollonius introduced the problem of space filling packing of spheres which in its classical solution , the apollonian packing see @xcite for remarkable properties of apollonian circles- , exhibits a power law behavior . specifically , the circle size distribution follows a power law with exponent of about 1.3 @xcite . apollonian networks ( ans ) were introduced in @xcite and independently in @xcite . zhou et al . @xcite introduced the random apollonian networks ( rans ) and gave formulae for their order , size , degree distribution ( using heuristic arguments ) , clustering coefficient and diameter . high dimensional rans were introduced in @xcite . the degree distribution of rans was shown to follow a power law in @xcite . other properties have also been analyzed , e.g. , average distance of two vertices @xcite , properties of the connectivity profile @xcite . zhou et al . @xcite proposed a simple rule that generates a random two - dimensional apollonian networks with very large clustering coefficient . rans are planar 3-trees , a special case of random @xmath0-trees @xcite . the general result of the degree distribution of random @xmath0-trees was proved by cooper & uehara and gao @xcite . in rans in contrast to the general model of random @xmath0 trees the random @xmath0 clique chosen at each step has never previously been selected . for example in the two dimensional case any chosen triangular face is being subdivided into three new triangular faces by connecting the incoming vertex to the vertices of the boundary . random @xmath0-trees due to their power law properties have been proposed as a model for complex networks , see , e.g. , @xcite and references therein . recently , a variant of @xmath0-trees , namely ordered increasing @xmath0-trees has been proposed and analyzed in @xcite . an example of a two dimensional ran is shown in figure [ fig : fig1 ] . the ran generator takes as input a parameter @xmath24 equal to the number of iterations the algorithm will perform and runs as follows : * let @xmath25 be a triangle @xmath26 embedded on the plane , e.g. , as an equilateral triangle . * _ for _ @xmath27 to @xmath24 : * * sample a face @xmath28 of the planar graph @xmath29 uniformly at random . * * insert the vertex @xmath30 inside this face ( e.g. , in the barycenter of the corresponding triangle ) and draw the three edges @xmath31 , @xmath32,@xmath33 , e.g. , as straight lines . it s worth pointing out why we expect this model to yield a power law degree distribution : consider any vertex @xmath34 at some time @xmath2 . vertex @xmath34 sees a wheel graph around it ( except for vertices 1,2,3 who see a wheel modulo one edge ) as figure [ fig : fig2 ] shows . since we pick a face uniformly at random the probability that @xmath34 increases its degree by 1 is ( roughly ) proportional to its degree . therefore , we expect this process to result in a graph , which is obviously planar , and which _ whp _ occurs with high probability _ whp _ if @xmath35}}=1 $ ] . ] has properties similar to those generated by a preferential attachment process @xcite . therefore , it should not come as a surprise that existing techniques developed in the context of preferential attachment models can be adapted to solve problems concerning the structure of random apollonian networks . finally , we shall make use of the following formulae for the number of vertices ( @xmath36 ) , edges ( @xmath37 ) and faces ( @xmath38 ) at time @xmath2 in a ran @xmath15 : @xmath39 note that a ran is a maximal planar graph since for any planar graph @xmath40 . we decompose the proof of theorem [ thrm : thrm1 ] into several lemmas which we prove in the following . the proof of theorem [ thrm : thrm1 ] follows from lemmas [ lem : lemma2 ] , [ lem : lemma3 ] , [ lem : lemma4 ] , [ lem : lemma5 ] , [ lem : lemma6 ] . we partition the vertices into three sets : those added before @xmath41 , before @xmath42 and after @xmath42 where @xmath43 and @xmath44 . we define a supernode to be a collection of vertices and the degree of the supernode the sum of the degrees of its vertices . let @xmath45 denote the degree of vertex @xmath46 at time @xmath2 . and let @xmath47denote the rising factorial function . then , for any positive integer @xmath0 @xmath48 } \leq \frac { ( k+2)!}{2 } \big ( \frac{2t}{s } \big)^{\frac{k}{2}}. \label{eq : rising}\ ] ] [ lem : lemma1 ] we distinguish two types of vertices . this analysis will be omitted in the subsequent lemmas . ] since as it can also be seen in figure [ fig : fig2] the three initial vertices @xmath49 have one face less than their degree whereas all other vertices have degree equal to the number of faces . + note that @xmath50 and that @xmath51 . by conditioning successively we obtain @xmath52 } & = { { \mathbb e}\left [ { { { \mathbb e}\left [ { d_t(s)^{(k ) } | d_{t-1}(s)}\right ] } } \right]}\\ & = { { \mathbb e}\left [ { ( d_{t-1}(s))^{(k ) } \big ( 1-\frac{d_{t-1}(s)}{2t-1 } \big ) + ( d_{t-1}(s)+1)^{(k ) } \frac{d_{t-1}(s)}{2t-1}}\right ] } \\ & = { { \mathbb e}\left [ { ( d_{t-1}(s))^{(k ) } \big ( 1-\frac{d_{t-1}(s)}{2t-1 } \big ) + ( d_{t-1}(s))^{(k ) } \frac{d_{t-1}(s)+k}{d_{t-1}(s ) } \frac{d_{t-1}(s)}{2t-1}}\right]}\\ & = { { \mathbb e}\left [ { ( d_{t-1}(s))^{(k)}}\right]}\big ( 1+\frac{k}{2t-1 } \big ) = ... = 3^{(k ) } \prod_{t'=s+1}^t ( 1+\frac{k}{2t'-1 } ) \\ & \leq 3^{(k ) } \exp{\big ( \sum_{t'=s+1}^t \frac{k}{2t'-1 } \big ) } \leq 3^{(k ) } \exp { \big ( k \int_{s}^{t } \ ! \frac{\mathrm{d}x}{2x-1 } \ , \big ) } \\ & \leq \frac{(k+2)!}{2 } \exp { \big ( k \log { \frac{2t-1}{2s-1 } } \big ) } \leq \frac { ( k+2)!}{2 } \big ( \frac{2t}{s } \big)^{\frac{k}{2}}. \end{aligned}\ ] ] + note that initially the degree of any such vertex is 2 . for any @xmath53 @xmath52 } & = { { \mathbb e}\left [ { { { \mathbb e}\left [ { d_t(s)^{(k ) } | d_{t-1}(s)}\right ] } } \right]}\\ & = { { \mathbb e}\left [ { ( d_{t-1}(s))^{(k ) } \big ( 1-\frac{d_{t-1}(s)-1}{2t-1 } \big ) + ( d_{t-1}(s)+1)^{(k ) } \frac{d_{t-1}(s)-1}{2t-1 } } \right]}\\ & = { { \mathbb e}\left [ { ( d_{t-1}(s))^{(k ) } \big ( 1+\frac{k}{2t-1 } \big ) - ( d_{t-1}(s))^{(k ) } \frac{k}{(2t-1)d_{t-1}(s ) } } \right]}\\ & \leq { { \mathbb e}\left [ { ( d_{t-1}(s))^{(k)}}\right ] } \big ( 1+\frac{k}{2t-1 } \big ) \leq \ldots \leq \frac { ( k+2)!}{2 } \big ( \frac{2t}{s } \big)^{\frac{k}{2}}.\end{aligned}\ ] ] the degree @xmath54 of the supernode @xmath55 of vertices added before time @xmath41 is at least @xmath56 _ whp _ . [ lem : lemma2 ] we consider a modified process @xmath57 coupled with the ran process , see also figure [ fig : fig3 ] . specifically , let @xmath58 be the modified degree of the supernode in the modified process @xmath57 which is defined as follows : for any type of insertion in the original ran process note there exist three types of insertions with respect to how the degree @xmath54 of the supernode ( black circle ) gets affected , see also figure [ fig : fig3] @xmath58 increases by 1 . we also define @xmath59 . note that @xmath60 for all @xmath61 . let @xmath62 and @xmath63}}$ ] . the following technical claim is proved in the appendix [ sec : appendix ] . @xmath64 [ claim : lemma2 ] let @xmath65 denote the event that the supernode consisting of the first @xmath41 vertices has degree @xmath58 in the modified process @xmath57 less than @xmath56 . note that since @xmath66 it suffices to prove that @xmath67 } } = o(1)$ ] . using claim [ claim : lemma2 ] we obtain @xmath68 } } & \leq \sum_{r=0}^ { t_0^{1/4}\sqrt{t } -(6t_0 + 6 ) } { r+6t_0 + 5 \choose 6t_0 + 5 } \big(\frac{2t_0 + 3}{2t+1}\big)^{3t_0 + 3 } e^{-\frac{3}{2}-2t_0+\frac{2t_0^{1/4}}{3 } } \\ & \leq t_0^{1/4}t^{1/2 } \frac{\big ( t_0^{1/4}t^{1/2 } \big)^{6t_0 + 5}}{(6t_0 + 5 ) ! } \big(\frac{2t_0 + 3}{2t+1}\big)^{3t_0 + 3 } e^{-\frac{3}{2}-2t_0+\frac{2t_0^{1/4}}{3 } } \\ & \leq \big ( \frac{t}{2t+1 } \big)^{3t_0 + 3 } \frac{t_0^{3t_0/2 + 3/2 } ( 2t_0 + 3)^{3t_0 + 3}}{(6t_0 + 5)^{6t_0 + 5 } } e^{4t_0 + 7/2 + 2/3 t_0^{1/4}}\\ & \leq 2^{-(3t_0 + 3 ) } \frac{e^{4t_0 + 7/2 + 2/3 t_0^{1/4}}}{(6t_0 + 5)^{\tfrac{3}{2}t_0+\tfrac{1}{2 } } } = o(1).\end{aligned}\ ] ] no vertex added after @xmath42 has degree exceeding @xmath69 _ whp _ . [ lem : lemma3 ] let @xmath70 denote the event that some vertex added after @xmath42 has degree exceeding @xmath69 . we use a union bound , a third moment argument and lemma [ lem : lemma1 ] to prove that @xmath71}}=o(1)$ ] . specifically @xmath72 } } & \leq \sum_{s = t_1}^t { { { \bf{pr}}\left[{d_t(s ) \geq t_0^{-2}t^{1/2}}\right ] } } = \sum_{s = t_1}^t { { { \bf{pr}}\left[{d_t(s)^{(3 ) } \geq ( t_0^{-2}t^{1/2})^{(3)}}\right ] } } \\ & \leq t_0^{6}t^{-3/2 } \sum_{s = t_1}^t { { \mathbb e}\left[{d_t(s)^{(3)}}\right ] } \leq 5!\sqrt{2 } t_0 ^ 6 \sum_{s = t_1}^t s^{-3/2 } \leq 5!2\sqrt{2 } t_0 ^ 6 t_1^{-1/2 } = o(1 ) . \end{aligned}\ ] ] no vertex added before @xmath42 has degree exceeding @xmath73 _ whp _ . [ lem : lemma4 ] let @xmath74 denote the event that some vertex added before @xmath42 has degree exceeding @xmath73 . we use again a third moment argument and lemma [ lem : lemma1 ] to prove that @xmath75}}=o(1)$ ] . @xmath76 } } & \leq \sum_{s=1}^{t_1 } { { { \bf{pr}}\left [ { d_t(s ) \geq t_0^{1/6}t^{1/2}}\right ] } } = \sum_{s=1}^{t_1 } { { { \bf{pr}}\left [ { d_t(s)^{(3 ) } \geq ( t_0^{1/6}t^{1/2})^{(3 ) } } \right ] } } \\ & \leq t_0^{-1/2}t^{-3/2 } \sum_{s=1}^{t_1 } { { \mathbb e}\left [ { d_t(s)^{(3 ) } } \right ] } \leq t_0^{-1/2}t^{-3/2 } \sum_{s=1}^{t_1 } 5!\sqrt{2 } \frac{t^{3/2}}{s^{3/2 } } \\ & \leq 5!\sqrt{2}\zeta(3/2)t_0^{-1/2 } = o(1)\end{aligned}\ ] ] where @xmath77 . the @xmath0 highest degrees are added before @xmath42 and have degree @xmath78 bounded by @xmath79 _ whp _ . [ lem : lemma5 ] for the upper bound it suffices to show that @xmath80 . this follows immediately by lemmas [ lem : lemma3 ] and [ lem : lemma4 ] . the lower bound follows directly from lemmas [ lem : lemma2 ] , [ lem : lemma3 ] and [ lem : lemma4 ] . assume that at most @xmath81 vertices added before @xmath42 have degree exceeding the lower bound @xmath82 . then the total degree of the supernode formed by the first @xmath41 vertices is @xmath83 . this contradicts lemma [ lem : lemma2 ] . finally , since each vertex @xmath84 has degree at most @xmath85 the @xmath0 highest degree vertices are added before @xmath42 _ whp _ . the @xmath0 highest degrees satisfy @xmath86 _ whp _ . [ lem : lemma6 ] let @xmath87 denote the event that there are two vertices among the first @xmath42 with degree @xmath82 and within @xmath88 of each other . by the definition of conditional probability and lemma [ lem : lemma3 ] @xmath89 } } & = { { { \bf{pr}}\left[{\mathcal{a}_4|\bar{\mathcal{a}_3}}\right ] } } { { { \bf{pr}}\left[{\bar{\mathcal{a}_3}}\right ] } } + { { { \bf{pr}}\left[{\mathcal{a}_4|\mathcal{a}_3}\right]}}{{{\bf{pr}}\left[{\mathcal{a}_3}\right ] } } \leq { { { \bf{pr}}\left[{\mathcal{a}_4|\bar{\mathcal{a}_3}}\right ] } } + o(1 ) \end{aligned}\ ] ] it suffices to show that @xmath90}}=o(1)$ ] . note that by a simple union bound @xmath89 } } & \leq \sum_{1 \leq s_1 < s_2 \leq t_1 } \sum_{l=-\frac{\sqrt{t}}{f(t)}}^{\frac{\sqrt{t}}{f(t ) } } p_{l , s_1,s_2 } = o\big(t_1 ^ 2 \frac{\sqrt{t}}{f(t ) } \max { p_{l , s_1,s_2 } } \big)\end{aligned}\ ] ] where @xmath91}}$ ] . we consider two cases and we show that in both cases @xmath92 . + note that at time @xmath42 there exist @xmath93 edges in @xmath94 . @xmath95 } } \tag{2}\\ & \leq t_0^{1/6}t^{1/2 } \sum_{d_1,d_2=3}^{4t_1 } { 2t_0^{1/6}t^{1/2 } \choose d_1 - 1 } { 2t_0^{1/6}t^{1/2 } \choose d_2 - 1 } \big(\frac{2t_0 + 3}{2t+1}\big)^{(d_1+d_2)/2 } e^{\frac{3}{2}+t_1+\frac{2t_0^{1/6}}{3 } } \tag{3}\\ & \leq t_0^{1/6}t^{1/2 } \sum_{d_1,d_2=3}^{4t_1 } ( 2t_0^{1/6}t^{1/2})^{d_1+d_2 - 2 } \big(\frac{2t_0 + 3}{2t+1}\big)^{(d_1+d_2)/2 } e^{2t_1 } \\ & \leq t_0^{1/6}t^{1/2 } e^{2t_1 } t_1 ^ 2 ( 2t_0^{1/6}t^{1/2})^{8t_1 - 2 } \big(\frac{2t_0 + 3}{2t+1}\big)^{4t_1 } \\ & = t_0^{4t_1/3 + 1/6}t^{-1/2 } e^{2t_1 } t_1 ^ 2 2^{8t_1 } ( 2t_0 + 3)^{4t_1 } \big ( \frac{t}{2t+1 } \big)^{4t_1 } \\ & = o\big ( \frac{f(t)}{t_1 ^ 2\sqrt{t } } \big)\end{aligned}\ ] ] note that we omitted the tedious calculation justifying the transition from ( 2 ) to ( 3 ) since calculating the upper bound of the joint probability distribution is very similar to the calculation of lemma [ lem : lemma2 ] . + + notice that in any case @xmath96 share at most two faces ( which may change over time ) . note that the two connected vertices @xmath97 share a common face only if @xmath98 . the other case is treated in the same manner . ] . consider the following modified process @xmath99 : whenever an incoming vertex `` picks '' one of the two common faces we do nt insert it . we choose two other faces which are not common to @xmath97 and add one vertex in each of those . notice that the number of faces increases by 1 for both @xmath97 as in the original process and the difference of the degrees remains the same . an algebraic manipulation similar to case 1 gives the desired result . in complex network analysis the spectrum of the adjacency matrix is an important aspect of the network or more generally of the network model with several applications , e.g. , triangle counting @xcite . to the best of our knowledge the spectrum of random apollonian networks has been studied only experimentally @xcite . here , we prove theorem [ thrm : thrm2 ] . we decompose the proof of the main theorem in lemmas [ lem : lemma7 ] , [ lem : lemma8 ] , [ lem : lemma9 ] , [ lem : lemma10 ] . having computed the highest degrees of a ran in section [ sec : degree ] , eigenvalues are computed by adapting existing techniques @xcite , see also @xcite for a closely related analysis , and taking into account the special properties of the model . we decompose the proof of theorem [ thrm : thrm2 ] in four lemmas . specifically , in lemmas [ lem : lemma7 ] , [ lem : lemma8 ] we bound the degrees and co - degrees respectively . having these bounds , we decompose the graph into a star forest and show in lemmas [ lem : lemma9 ] and [ lem : lemma10 ] that its largest eigenvalues , which are @xmath100 , dominate the eigenvalues of the remaining graph . we partition the vertices into three set @xmath101 . specifically , let @xmath102 be the set of vertices added after time @xmath103 and at or before time @xmath104 where @xmath105 in the following we use the recursive variational characterization of eigenvalues @xcite . specifically , let @xmath106 denote the adjacency matrix of a simple , undirected graph @xmath107 and let @xmath108 denote the @xmath109-th largest eigenvalue of @xmath106 . then @xmath110 where @xmath111 ranges over all @xmath112 dimensional subspaces of @xmath113 . we shall use the following lemma in our proof , specifically in the proof of lemma [ lem : lemma9 ] . for any @xmath114 and any @xmath17 with @xmath4 as @xmath5 the following holds _ whp _ : for all @xmath46 with @xmath115 , for all vertices @xmath116 , then @xmath117 . [ lem : lemma7 ] set @xmath118 . we use lemma [ lem : lemma1 ] , a simple union bound and markov s inequality to obtain : @xmath119 } } & \leq \sum_{s = f(t)}^t \sum_{r=1}^s { { { \bf{pr}}\left [ { d_s(r)^{(q ) } \geq ( s^{1/2+\epsilon } r^{-1/2})^{(q ) } } \right]}}\\ & \leq \sum_{s = f(t)}^t \sum_{r=1}^s { { { \bf{pr}}\left [ { d_s(r)^{(q ) } \geq ( s^{-(q/2+q\epsilon ) } r^{q/2 } ) } \right ] } } \\ & \leq \sum_{s = f(t)}^t \sum_{r=1}^s \frac{(q+2)!}{2 } \big ( \frac{2s}{r } \big)^{q/2 } s^{-q/2 } s^{-q\epsilon } r^{q/2 } \\ & = \frac{(q+2)!}{2 } 2^{q/2 } \sum_{s = f(t)}^t s^{1-q\epsilon } \\ & \leq \frac{(q+2)!}{2 } 2^{q/2 } \int_{f(t)-1}^t \ ! x^{1-q\epsilon } \ , \mathrm{d}x \\ & \leq \frac{(q+2)!}{2(q\epsilon-2 ) } 2^{q/2 } ( f(t)-1)^{2-q\epsilon } = o(1).\end{aligned}\ ] ] let @xmath120 be the set of vertices in @xmath121 which are adjacent to more than one vertex of @xmath122 . then @xmath123 _ whp _ . [ lem : lemma8 ] first , observe that when vertex @xmath46 is inserted it becomes adjacent to more than one vertex of @xmath122 if the face chosen by @xmath46 has at least two vertices in @xmath122 . we call the latter property @xmath124 and we write @xmath125 when @xmath46 satisfies it . at time @xmath42 there exist @xmath126 faces total , which consist of faces whose three vertices are all from @xmath122 . at time @xmath127 there can be at most @xmath128 faces with at least two vertices in @xmath122 since each of the original @xmath126 faces can give rise to at most 3 new faces with at least two vertices in @xmath129 . consider a vertex @xmath130 , i.e. , @xmath127 . by the above argument , @xmath131 } } \leq \frac{6t_1 + 3}{2t+1}$ ] . writing @xmath132 as a sum of indicator variables , i.e. , @xmath133 and taking the expectation we obtain @xmath134 } & \leq \sum_{s = t_2}^t \frac{6t_1 + 3}{2t+1 } \leq ( 6t_1 + 3 ) \int_{t_2}^t \ ! ( 2x+1)^{-1 } \ , \mathrm{d}x \\ & \leq ( 3t^{\tfrac{1}{8}}+\tfrac{3}{2 } ) \ln{\frac{2t+1}{2t_2 + 1}}= o(t^{1/7})\end{aligned}\ ] ] by markov s inequality : @xmath135 } } \leq \frac { { { \mathbb e}\left[{|s_3'|}\right ] } } { t^{1/6 } } = o(1).\ ] ] therefore , we conclude that @xmath136 _ whp _ . let @xmath137 be the star forest consisting of edges between @xmath122 and @xmath138 . let @xmath16 denote the @xmath0 highest degrees of @xmath107 . then @xmath139 _ whp _ . [ lem : lemma9 ] it suffices to show that @xmath140 for @xmath141 . note that since the @xmath0 highest vertices are inserted before @xmath42 _ whp _ , the edges they lose are the edges between @xmath122 and the ones incident to @xmath120 and @xmath142 and we know how to bound the cardinalities of all these sets . specifically by lemma [ lem : lemma8 ] @xmath136 _ whp _ and by theorem [ thrm : thrm1 ] the maximum degree in @xmath143 is less than @xmath144 , @xmath145 for @xmath146 respectively _ whp _ . also by theorem [ thrm : thrm1 ] @xmath147 . hence , we obtain @xmath148 to complete the proof of theorem [ thrm : thrm2 ] it suffices to prove that @xmath149 is @xmath150 where @xmath151 . we prove this in the following lemma . the proof is based on bounding maximum degree of appropriately defined subgraphs using lemma [ lem : lemma7 ] and then using standard inequalities from spectral graph theory @xcite . @xmath152 _ whp _ . [ lem : lemma10 ] from gershgorin s theorem @xcite the maximum eigenvalue of any graph is bounded by the maximum degree . we bound the eigenvalues of @xmath153 by bounding the maximum eigenvalues of six different induced subgraphs . specifically , let @xmath154 $ ] , @xmath155 where @xmath156 $ ] is the subgraph induced by the vertex set @xmath111 and @xmath157 is the subgraph containing only edges with one vertex is @xmath111 and other in @xmath158 . we use lemma [ lem : lemma9 ] to bound @xmath159 and lemma [ lem : lemma8 ] for the other eigenvalues . we set @xmath160 . @xmath161 @xmath162 @xmath163 @xmath164 @xmath165 @xmath166 therefore _ whp _ we obtain @xmath167 as shown in appendix b of @xcite using physicist s methodology the diameter of a ran is asymptotically upper bounded by a logarithmic factor @xmath168 . similarly , zhang et al . @xcite show that the diameter scales logarithmically . in the following we give a simple proof that the diameter @xmath10 of the graph @xmath15 created @xmath2 steps is @xmath169 _ whp _ and in theorem [ thrm : ternary ] we give a refined upper bound for the diameter . . each face is labelled with its depth.,scaledwidth=30.0% ] we begin with a necessary definition for the proof of claim [ claim : diameter ] . we define the _ depth of a face _ recursively . initially we have one face @xmath170 its depth is @xmath171 . for each new face @xmath172 created by picking a face @xmath173 , we have @xmath174 . an example is shown in figure [ fig : depth ] , where each face is labelled with its corresponding depth . the diameter @xmath10 of the graph @xmath15 created @xmath2 steps is @xmath169 _ whp _ . [ claim : diameter ] note that if @xmath175 is the maximum depth of a face then @xmath176 . hence , we need to upper bound the depth of a given face after @xmath2 rounds . let @xmath177 be the number of faces of depth @xmath0 at time @xmath2 , then : @xmath178 } & = \sum_{1 \leq t_1<t_2<\ldots < t_k\leq t } \prod_{j=1}^k \frac{1}{2t_j+1 } \leq \frac{1}{k ! } ( \sum_{j=1}^t \frac{1}{2j+1 } ) ^k \leq \frac{1}{k ! } ( \frac{1}{2 } \log{t})^k \leq ( \frac{e\log{t}}{2k})^{k+1}\end{aligned}\ ] ] by the first moment method we obtain @xmath179 _ whp _ and hence @xmath180 _ whp _ . the depth of a face can be formalized via a bijection between random ternary trees and rans . using this bijection we prove theorem [ thrm : ternary ] which gives a refined upper bound on the asymptotic growth of the diameter . consider the random process which starts with a single vertex tree and at every step picks a random leaf and adds three children to it . let @xmath158 be the resulting tree after @xmath2 steps . there exists a natural bijection between the ran process and this process , see @xcite and also figure [ fig : ternary ] . the depth of @xmath158 in probability is @xmath181 where @xmath12 is the unique solution greater than 1 of the equation @xmath13 , see broutin and devroye @xcite , pp . 284 - 285 . note that the diameter @xmath10 is at most twice the height of the tree and hence the result follows . the above observation , i.e. , the bijection between rans and random ternary trees can not be used to lower bound the diameter . a counterexample is shown in figure [ fig : counter ] where the height of the random ternary tree can be made arbitrarily large but the diameter is 2 . albenque and marckert proved in @xcite that if @xmath182 are two i.i.d . uniformly random internal vertices , i.e. , @xmath183 , then the distance @xmath184 tends to @xmath185 with probability 1 as the number of vertices @xmath186 of the ran grows to infinity . however , an exact expression of the asymptotic growth of the diameter to the best of our knowledge remains an open problem . finally , it is worth mentioning that the diameter of the ran grows faster asymptotically than the diameter of the classic preferential attachment model @xcite which _ whp _ grows as @xmath187 , see bollobs and riordan @xcite . consider the three faces created after the insertion of the first point . let s call the face which gets the first vertex @xmath188 and the other two faces @xmath189 , see also figure [ fig : fig1](b ) . also for simplicity of the notation let the configuration of figure [ fig : fig1](b ) correspond to time @xmath190 . let @xmath191 equal the number of steps until a new vertex picks face @xmath192 or @xmath193 . clearly , @xmath194 . what is the expectation @xmath195}$ ] ? for any @xmath196 @xmath197 } } & = \prod_{j=1}^t \frac{3 + 2(j-1)}{5 + 2(j-1 ) } = \frac{3}{2t+3}.\end{aligned}\ ] ] using now the fact that @xmath195 } = \sum_{t=1}^{+\infty } { { { \bf{pr}}\left[{x \geq t}\right]}}= 1 + \sum_{t=1}^{+\infty } { { { \bf{pr}}\left[{x > t}\right]}}$ ] and substituting we obtain that @xmath195 } = + \infty$ ] . in this work we studied several aspects of random apollonian networks , namely the highest degree vertices , the spectrum , the diameter and the `` waiting times '' . there are various research directions which we plan to address in future work . indicatively , we report the following : ( a ) in contrast to the preferential attachment model , ran have small separators . specifically , by the lipton - tarjan theorem @xcite we know the existence of a good separator of size @xmath198 . this is in accordance with the empirical observation that real world networks have good separators @xcite . we plan to investigative the minimum size separator in future work . ( b ) consider the natural process where an adversary with probability @xmath170 at every step deletes a vertex from the network , e.g. , uniformly at random or in a maliscious way . under which conditions do we obtain a connected ran ? ( c ) what is the exact asymptotic expression for the diameter of @xmath15 ? we would like to thank luc devroye and alexis darrasse for pointing out references @xcite and @xcite respectively . 99 aiello , w. , chung , f. , lu , l. : _ a random graph model for power law graphs _ experimental mathemathics , 10(1 ) , 53 - 66 , 2001 albenque , m. , marckert , j.f . : _ some families of increasing planar maps _ electronic journal of probability , 13 , pp . 1624 - 1671 albert , r. , barabsi , a. , jeong , h. : _ diameter of the world wide web . _ nature 401 , pp . 103131 , 1999 andrade , j.s . , herrmann , h.j . , andrade , r.f.s . , da silva , l.r . : _ apollonian networks : simultaneously scale - free , small world euclidean , space filling , and with matching graphs _ phys . 94 ( 2005 ) 018702 . andrade , r.f.s . , miranda , j.g.v . : _ spectral properties of the apollonian network _ physica a , 356 , ( 2005 ) aumann , r. : _ acceptable points in general cooperative @xmath186-person games _ contributions to the theory of games iv , pp . 287 - 324 , princeton university press , 1959 aumann , r. , myerson , r. : _ endogenous information of links between players and coalition : an application to the shapley value _ shapley value , pp . 175 - 191 , cambridge university press , 1988 bala , v. , goyal , s. : _ a noncooperative model of network formation _ econometrica , econometric society , vol . 68(5 ) , pp . 1181 - 1230 , 2000 barabsi , a. , albert , r. : _ emergence of scaling in random networks _ science 286 , pp . 509512 , 1999 barthlemy , m. , flammini , a. : _ modeling urban street patterns _ physical review letters , 100(13 ) , 138702(4 ) , 2008 blandford , d. , blelloch , g. , kash , i. : _ compact representations of separable graphs _ proceedings of the fourteenth annual acm - siam symposium on discrete algorithms ( soda 03 ) , pp . 679 - 688 , 2003 bodini , o. , darrasse , a. , soria , m. : _ distances in random apollonian network structures _ available at arxiv http://arxiv.org/abs/0712.2129 bollobs , b. , riordan , o. : _ the diameter of a scale - free random graph _ combinatorica , 2002 bonato , a. , hadi , n , horn , p. , pralat , p. : _ models of online social networks _ internet mathematics , to appear ( 2011 ) boorman , s. : _ a combinatorial optimization model of transmission of job information through contact networks _ bell journal of economics , 6:216 - 249 , 1975 boyd , d.w . : _ the sequence of radii of the apollonian packing _ mathematics of computation , vol . 249 - 254 , 1982 broder , a. , kumar , r. , maghoul , f. , raghavan , p. , rajagopalan , s. , stata , r. , tomkins , a. , wiener , j. : _ graph structure in the web _ in proc . of the 9th intl . world wide web conference , pp . amsterdam : north - holland publishing company , 2000 chung graham , f. : _ spectral graph theory _ american mathematical society 1997 chung graham , f. , lu , l. : _ complex graphs and networks _ american mathematical society , no . 107 ( 2006 ) chung , f. , lu , l , vu . v.h . : _ spectra of random graphs with given expected degrees _ proceedings of the national academy of sciences of the united states of america , 100 , 63136318 cooper , c. , frieze , a. : _ a general model of web graphs _ random structures & algorithms , volume 22 issue 3,pp . 311335 , 2003 cooper , c. , uehara , r. : _ scale free properties of random @xmath0-trees _ mathematics in computer science , 3(4 ) , pp . 489496 , 2010 darrasse , a. , soria , m. : _ degree distribution of random apollonian network structures and boltzmann sampling _ 2007 conference on analysis of algorithms , aofa 07 , dmtcs proceedings . darrasse , a. , hwang , h .- k . , bodini , o. , soria , m. : _ the connectivity - profile of random increasing k - trees _ available at arxiv : http://arxiv.org/abs/0910.3639 broutin , n. , devroye , l. : _ large deviations for the weighted height of an extended class of trees _ algorithmica , 46 , pp . 271 - 297 , 2006 doye , j.p.k , massen , c.p . : _ self - similar disk packings as model spatial scale - free networks _ e 71 ( 2005 ) 016128 . eisenstat , d. : _ random road networks : the quadtree model _ available at arxiv erds , p. , rnyi , a. : _ on random graphs i _ publicationes mathematicae debrecen 6 , pp . 290 - 297 , 1959 erds , p. , rnyi , a. : _ on the evolution of random graphs _ magyar tud . akad . mat . kutat int . kzl 5 , pp . 17 - 61 , 1960 fabrikant , a. , koutsoupias , e. , papadimitriou , c. : _ heuristically optimized trade - offs _ proceedings of the 29th international colloquium on automata , languages , and programming ( icalp ) , pp . 110122 , 2002 faloutsos , m. , faloutsos , p. , faloutsos , c. : _ on power - law relationships of the internet topology _ proceedings of the conference on applications , technologies , architectures , and protocols for computer communication ( sigcomm ) , 1999 flaxman , a. , frieze , a. , fenner , t. : _ high degree vertices and eigenvalues in the preferential attachment graph _ internet mathematics , 2(1 ) , 2005 gao , y. : _ the degree distribution of random k - trees _ theoretical computer science , 410(8 - 10 ) , 2009 . gao , y. , hobson , c. : _ . random k - tree as a model for complex networks _ in the fourth workshop on algorithms and models for the web - graph ( waw2006 ) , 2006 gilbert , e.n . : _ random graphs _ annals of mathematical statistics , 30 , pp . 1141 - 1144 , 1959 graham , r.l . , lagarias , j.c . , mallows , c.l . , wilks , a.r . , yan , c.h . : _ apollonian circle packings : number theory _ j. number theory 100 , no . 1 , 1 - 45 , mr1971245 guy , s. , chhugani , j. , curtis , s. , dubey , p. , lin , m. , manocha , d. : _ pledestrians : a least - effort approach to crowd simulation _ eurographics / acm siggraph symposium on computer animation , 2010 . hayashi , y. , matsukubo , j. : _ a review of recent studies of geographical scale - free networks _ available at arxiv : cond / physics/0512011 kloks , t. : _ treewidth : computations and approximations _ springer - verlag , 1994 kolountzakis , m.n , miller , g.l . , peng , r. , tsourakakis , c.e . : _ efficient triangle counting in large graphs via degree - based vertex partitioning _ 7th workshop on algorithms and models for the web graph ( waw 10 ) , 2010 lipton , r.j . , tarjan , r. : _ a separator theorem for planar graphs _ siam journal on applied mathematics 36 : 177189 , 1979 leskovec , j. , faloutsos , c. : _ scalable modeling of real graphs using kronecker multiplication _ machine learning proceedings of the twenty - fourth international conference ( icml 2007 ) , corvalis , oregon , usa , june 20 - 24 , 2007 masucci , a.p . , smith , d. , crooks , a. , batty , m. : _ random planar graphs and the london street network _ european physics journal b ( 2009 ) mihail , m. , papadimitriou , c. : _ on the eigenvalue power law _ random 02 , pp . 254 - 262 ( 2002 ) panholzer , a. , seitz , g. : _ ordered increasing @xmath0-trees : introduction and analysis of a preferential attachment network model _ dmtcs proc . , aofa10 , pp . 549 - 564 , 2010 pralat , p. , wormald , n. : _ growing protean graphs _ internet mathematics , 4(1 ) , pp . 1 - 16 , 2007 strang , g. : _ linear algebra and its applications _ brooks cole , 2005 tsourakakis , c.e . : _ fast counting of triangles in large real networks without counting : algorithms and laws _ eighth ieee international conference on data mining ( icdm 08 ) , 2008 watts , d. , strogatz , s. : _ collective dynamics of small - world networks _ nature , 1998 , * 393 * , 440 - 442 wu , z .- x . , xu , x .- j . , wang , y .- h . : _ comment on `` maximal planar networks with large clustering coefficient and power - law degree distribution '' _ physical review , e 73 , 058101 ( 2006 ) zhang , z.z . , comellas , f. , fertin , g. , rong , l.l . : _ high dimensional apollonian networks _ available at arxiv http://arxiv.org/abs/cond-mat/0503316 zhou , t. , yan , g. , wang , b.h . : _ maximal planar networks with large clustering coefficient and power - law degree distribution _ phys . e 71 ( 2005 ) 046141 . let @xmath199 be a vector denoting that @xmath58 increases by 1 at @xmath200 for @xmath201 . we upper bound the probability @xmath202 of this event in the following.note that we consider the case where the vertices have same degree as the number of faces around them , figure [ fig : fig2](b ) . the other case is analyzed in exactly the same way , modulo a negligible error term . @xmath203 \bigg [ \prod_{k=0}^r \prod_{j=\tau_k+1}^{\tau_{k+1}-1 } \big(1-\frac{d_0+k}{2j+1}\big ) \bigg ] \\ & = d_0 ( d_0 + 1 ) \ldots ( d_0+r-1)\bigg [ \prod_{k=1}^r \frac{1}{2\tau_k+1 } \bigg ] \exp{\bigg ( \sum_{k=0}^r \sum_{j=\tau_k+1}^{\tau_{k+1}-1 } \log{\big(1-\frac{d_0+k}{2j+1}\big ) } \bigg ) } \\ & = \frac{(d_0+r-1)!}{(d_0 - 1)!}\bigg [ \prod_{k=1}^r \frac{1}{2\tau_k+1 } \bigg ] \exp{\bigg ( \sum_{k=0}^r \sum_{j=\tau_k+1}^{\tau_{k+1}-1 } \log{\big(1-\frac{d_0+k}{2j+1}\big ) } \bigg ) } \\\end{aligned}\ ] ] @xmath225 \\ & \leq c(r , d_0,t_0,t ) \sum_{t_0-\lceil \tfrac{d_0}{2 } \rceil + 1 \leq \tau_1 ' \leq .. \leq \tau_r ' \leq t-\lceil \tfrac{d_0+r}{2 } \rceil } \prod_{k=1}^r \big [ ( 2\tau_k'+3)^{-1/2 } \big ( 1+\frac{1}{\tau_k'+\frac{d_0+k}{2 } + 1/2}\big)\big ] \\ & \leq \frac{c(r , d_0,t_0,t)}{r ! } \bigg ( \sum_{t_0-\lceil \tfrac{d_0}{2 } \rceil}^{t-\lceil \tfrac{d_0+r}{2 } \rceil } ( 2\tau_k'+3)^{-1/2}+\tfrac{1}{\sqrt{2 } } ( \tau_k'+3/2)^{-3/2 } \bigg)^r \\ & \leq \frac{c(r , d_0,t_0,t)}{r ! } \bigg ( \int_{0}^{t-\frac{d+r}{2 } } \ ! \big [ ( 2x+3)^{-1/2 } + \tfrac{1}{\sqrt{2 } } ( x+3/2)^{-3/2 } \big ] \ , \mathrm{d}x\bigg)^r \\ & \leq \frac{c(r , d_0,t_0,t)}{r ! } \big(\sqrt{2t+3-(d_0+r)}+2/3\big)^{r } \\ & \leq \frac{c(r , d_0,t_0,t)}{r ! } ( 2t)^{r/2 } e^{-\tfrac{r}{2 } \tfrac{d_0+r-3}{2 t } } e^{\frac{2r}{3\sqrt{2t-(d_0+r)+3 } } } \\ & \leq { d_0+r-1 \choose d_0 - 1 } \big(\frac{2t_0 + 3}{2t+1}\big)^{d_0/2 } \big [ \big ( 1-\frac{d_0}{2t_0 + 3}\big)^{-(1-\frac{d_0}{2t_0 + 3 } ) } \big]^{t_0 + 3/2 } \times \\ & \big(\frac{2t}{2t+1}\big)^{r/2 } \exp{\big ( -\frac{d_0}{2 } + \frac{(d_0+r)^2}{8t+4}+ \frac{(d_0+r)^3}{4(2t+1)^2 } - \frac{r(d_0+r-3)}{4t}+\frac{2r}{3\sqrt{2t+3-(d_0+r ) } } \big ) } \\ \end{aligned}\ ] ]
with the advent of wide field surveys our knowledge of galaxy properties has significantly improved . the large amount of high quality data of nearby and high redshift objects obtained so far is allowing astronomers to shed light on how galaxies formed and evolved at different redshifts and in different environments ( e.g. @xcite ) . however our picture of the universe still remains limited since it is mostly based on the study of rest - frame optical selected objects . first of all optically - selected galaxies already contain , by definition , a large amount of stars which may not be the case in younger or less evolved galaxies . secondly , surface brightness selection is very dramatic and nearly all catalogued objects lie in an extremely narrow range of surface brightnesses @xcite . + since each wavelength has its own selection effects and traces different baryonic components , the combination of blind surveys at various frequencies is of vital importance . alternative methods of baryon selection will allow us to have a less biased view of our universe and enable us to correctly reconstruct the evolutionary history of galaxies ( e.g. @xcite ) . in particular , the most interesting results can come from the comparison of optical surveys with blind surveys tracing components not directly related to stellar emission like , for example , neutral and molecular hydrogen . for molecular hydrogen we will have to wait a couple of years for the first results from alma , but with the advent of multi - beam instruments on large single dish radio telescopes it has become possible during the last decade to carry out fully sampled blind hi surveys of the sky ( e.g. @xcite ) . neutral hydrogen is extremely important because it represents the fuel for the future star formation activity of a galaxy . it is also one of the galactic components that can most easily be affected by environmental mechanism ( e.g. ram pressure , tidal interaction ) influencing its star formation history . therefore hi line observations of galaxies can provide us with some of the most powerful diagnostics on the role of the environment in regulating the evolution of galaxies @xcite . what we lack is deep and high spatial resolution hi blind surveys of cluster of galaxies and their outskirts . the hi parkes all sky survey ( hipass ) has represented a real breakthrough for hi blind surveys , but its low spatial resolution ( @xmath115 arcmin ) has made it very difficult to carry out detailed studies of high density regions . fortunately , with the refurbishing of the 305 m arecibo radio telescope , blind hi surveys with much improved sensitivity , spatial ( 3.5 arcmin @xmath2 ) and velocity ( 5 @xmath3 ) resolution have become possible thanks to the installation of the alfa multi beam instrument . the abell cluster 1367 , due to its proximity ( z@xmath10.0216 , v@xmath16500 ) and to the fact that it is currently forming at the intersection of two filaments in the great wall @xcite , represents an ideal target for hi surveys ( e.g. @xcite ) . it is ideal for the study of the properties of hi selected galaxies and environmental effects . for these reasons , abell 1367 is one of the region observed as a part of the arecibo galaxy environment survey ( ages , @xcite ) , one of the new hi blind survey carried out with the alfa multi - beam system . ages aims to study the atomic hydrogen properties of different galactic environments to faint sensitivity limits ; low hi masses ( 6@xmath4 , assuming a 200 velocity width at 92.8 mpc , the distance of the a1367 cluster ) and column densities of @xmath1 3@xmath5 ( for a source that fills the beam ) . the environments ages will survey range from apparent voids in the large - scale structure of galaxies , via isolated spiral galaxies and their haloes , to galaxy - rich regions associated with galaxy clusters and filamentary structures . ages plans to cover @xmath6 square degrees centered on abell 1367 . here we report on the early results obtained from observations of a subset of the whole ages - a1367 region : a @xmath7 square degree area centered on the cluster core . as part of the ages , we observed the abell 1367 region in may 2006 and april 2007 using the alfa feed array at the 305-m arecibo telescope . a total of @xmath163 hours observing time were allocated with an average observing time of @xmath12.3 hours per night . since the allocated time was insufficient to cover all the ages - a1367 region ( 5@xmath04 square degrees ) we decided to give the highest priority to the cluster center and its immediate outskirts covering a strip of @xmath15@xmath01 square degrees ( 11:34:00@xmath8r.a.(j.2000)@xmath811:54:15 , 19:15@xmath8dec(j.2000)@xmath820:20 ) centered on the cluster core . the observation and data reduction techniques adopted by ages are extensively described in @xcite ; here we will only provide a brief summary of them . + ages uses the 7-feed alfa multi - beam receiver and a spectral line backend capable of instantaneously recording spectra from the two linear polarizations of each beam and covering a bandwidth of 100 mhz . the angular resolution is given by the size of the alfa beams ( @xmath13.3@xmath03.8 arcmin ) and the velocity resolution is @xmath125khz , corresponding to @xmath1 5.5 . observations are performed in _ drift scan mode _ ( e.g. @xcite ) : the array is kept at a fixed azimuth and elevation while the sky drifts overhead . the earth s rotation rate governs the on source time in a drift scan . for arecibo this means that each point in the sky takes @xmath112 s to cross the beam . twenty - five separate scans are then required to reach the 300 s / beam integration time . in order to compensate for the change in parallactic angle , and thus achieve uniform sky coverage , alfa must be rotated before every scan and , in order to attain fully sampled sky coverage , it is necessary to stagger the declination ( dec . ) of individual scans by @xmath11 arcmin ( half the beam separation ) . we record data every second and total power values for each of the seven beams , both polarizations and 4096 channels , are recorded as four - bit floating point numbers . calibration is performed at the beginning of every scan using a high - temperature noise diode that is injected into each beam for a duration of 1 s. a uniform final depth of @xmath1 300 sec was achieved for all of the area observed . data reduction was performed using the aips++ packages livedata and gridzilla @xcite , developed by the australian telescope national facility ( atnf ) . livedata performs bandpass estimation and removal , doppler tracking and calibrates the residual spectrum . gridzilla is a gridding package that co - adds all the spectra using a suitable algorithm , to produce 3d data cubes . the abell 1367 5@xmath01 square degrees region was gridded using a median gridding technique into 1@xmath01 square arcmin pixels , each of which contains a 4096 channel spectrum , and was smoothed at a velocity resolution of 10 . the ages observing strategy is highly successful at reducing sidelobe variations by making a nyquist - sampled map with every alfa beam individually . thus , when the observations are median combined , the variations across the beams are removed , leaving a circular beam with symmetrical sidelobes at a @xmath15 - 10% level @xcite . the output data cube has two spatial dimensions and the spectral dimension can be chosen by the user to be frequency , wavelength or velocity . the noise distribution is fairly gaussian with a standard deviation of 0.8439@xmath90.0003 mjy , consistent with the expected value @xcite for minimal smoothing of the beam . the cube is publicly available and can be downloaded at the following link @xmath10 . radio frequency interference ( rfi ) represents a constant source of contamination in certain regions of the observed spectral window . in particular two main rfi sources were present in all the scans : the san juan faa radar at 1350 mhz and its 3rd harmonic at 1387.8 mhz . in addition an intermittent source of rfi ( gps satellite l3 ) appeared only in some of the scans at approximately 1381 mhz . the strong constant sources at 1350 mhz ( v@xmath114620@xmath9140 ) and 1387.8 mhz ( v@xmath17009@xmath919 ) completely obscure a part of the cosmic volume : in particular the 1387.8 mhz rfi lies approximately at the velocity of abell 1367 making part of the cluster volume inaccessible . the intermittent gps source at 1381 mhz ( v@xmath18500 ) reduces our sensitivity limit and could significantly influence the number of sources detected in a frequency range @xmath1 300 wide . in addition to the rfi another two sources of contamination affect our data . the first one is the galaxy : livedata is not able to correctly subtract the strong emission from the milky way , resulting in a blindness of our survey in about the velocity range -50@xmath11 + 50 . moreover the abell 1367 region includes the 3c264 radio galaxy ( r.a.=11:45:05.5 , dec=+19:36:23 ) . this strong radio source has a 1.4ghz continuum flux of @xmath15 jy , producing strong spectral standing waves and making a @xmath14 arcmin radius region around the galaxy inaccessible . in every blind survey a crucial step in the data reduction is represented by the source extraction . the objective is to obtain a catalogue of confirmed sources to well defined selection limits : as the signal - to - noise ratio ( s / n ) decreases , the contamination by spurious noise signals quickly rises . detection techniques need to be optimized to discriminate real from spurious signals as well as possible . currently source extraction is carried out using three different methods : the cube is visually inspected independently by two of us ( in this case rra and lc ) and by an automatic extractor ( polyfind , written by r. f. minchin ) based on a peak flux threshold . the automatic extractor is essentially based on the cross - correlation with templates method described by @xcite . the program initially looks for peak values above a pre - defined value ( in this case 4@xmath12 ) and then cross - correlates with templates accepting the best fit as long as the correlation coefficient is above 0.75 and the total signal - to - noise ratio of the detection is above 4 @xcite . automatically extracted candidates are then checked by eye , visually inspecting the full ages spectrum summed from a 5@xmath13 box around the position of the detection . the aim of this process is to quickly remove obvious spurious detections while not rejecting any real source . these methods provided a list of 104 candidate hi sources 210 . these systems are described in the appendix of this paper . ] , of which 78 were detected by all the three different signal extraction techniques . in the following we will refer to these 78 objects as `` sure '' sources and to the remaining 26 as `` possible '' sources . the distribution of the total 21 cm flux as a function of the line width at the 50% level of peak maximum for our 104 candidate sources is shown in fig.[ftotw ] . filled circles indicate `` sure '' sources while empty circles are `` possible '' sources . the void in the bottom right part of the plot ( i.e. large velocity widths and small fluxes ) is due to a selection effect present in the source extraction technique : at fixed low total flux it is easier to detect a narrow hi source , since the survey is not flux but surface brightness limited . in order to compare the ages s / n threshold with the recent results presented by alfalfa @xcite we plotted in fig.[ftotw ] a s / n limit of @xmath14=6.5 for our survey , where @xcite : @xmath15 @xmath16 is the total flux in jy , @xmath17 is the velocity width measured at the 50% level , @xmath18 is either @xmath19 for @xmath20 or 400/@xmath21 for @xmath22 , @xmath23 is the spectral velocity resolution and @xmath24 is the r.m.s . noise across the spectrum measured in mjy at @xmath25 spectral resolution . in our case @xmath25=10 . the great majority of our `` sure '' sources lie well above @xmath14=6.5 while the `` possible '' sources have @xmath266.5 , in agreement with @xcite . this suggests that also for our sample @xmath14=6.5 is a reliable threshold for discriminating between `` sure '' and `` possible '' sources . in order to test the reliability of our extraction technique and the real nature of the 26 `` possible '' sources , 10 hours of telescope time where scheduled with the arecibo single pixel l - band wide ( lbw ) receiver in april 2007 . the lbw observations were made covering a spectral band of 12.5 mhz ( 2048 spectral channels ) centered at the expected source recessional velocity . the integration time varied between 5 and 10 minutes depending on the intensity of the source , reaching an rms in the range @xmath10.5 - 0.8 mjy beam@xmath27 at a velocity resolution of 10 km s@xmath27 . all observations were taken using the position - switching technique , with each blank sky ( or off ) position observed for the same duration , and over the same portion of the telescope dish ( az and el ) as the on - source ( on ) observation . each 5 min + 5 min on + off pair was followed by a 10 s on + off observation of a calibrated noise diode . none of the 26 observed sources lies within the sidelobes of a `` sure '' hi detection , making unlikely a sidelobe contamination of the l - wide follow - up observations . of the 26 `` possible '' sources , only 4 ( indicated with crosses in fig.[ftotw ] ) were not confirmed by arecibo follow - up observations . we therefore have a sample of 100 sure hi sources detected in the 5@xmath01 square degrees observed by ages in the abell 1367 region . all the 100 sources in the sample were inspected and parametrized using the miriad task mbspect @xcite . the source position was estimated from a gaussian fit to the moment map within the galaxy velocity range and the spectrum was then extracted from the data cubes . this spectrum is a weighted average , with the weight depending on the distance of each pixel from the fitted position . velocity widths were then measured at the 20% and 50% levels ( @xmath28 and @xmath17 ) relative to the peak signal , using a width - maximizing algorithm @xcite . the uncertainties in each derived quantity are computed following the recipes given by @xcite . in table [ suretab ] we present the main parameters of all the 100 hi sources detected in the cube , namely : + col.1 : hi source i d + col.2 - 3 : hi r.a . ( j.2000 ) coordinate and relative error + col.4 - 5 : hi dec . ( j.2000 ) coordinate and relative error + col.6 : heliocentric velocity and relative error ( @xmath29 , i.e. optical reference frame ) , measured as the midpoint between the velocities where the flux density reaches 50% of peak maximum level . + col.7 : velocity width and relative error of the source line profile measured at the 50% level of peak maximum . + the velocity widths are not corrected for instrumental broadening , turbulent motions , disk inclination or cosmological effects . + col.8 : velocity width and relative error of the source line profile measured at the 20% level . + col.9 : peak flux density and relative error in mjy . + col.10 : total flux and relative error in jy . + col.11 : object flag , defined as follows : flag 0 indicates sources detected by all the three independent methods used for source extraction making them reliable sources . flag 1 indicates objects confirmed only after lwide follow - up observations . flag 2 indicates objects which are contaminated by rfi : they are in general sure detections but the hi parameters ( e.g. flux , velocity width , etc . ) can be strongly affected and should not be used . + the ages source catalogue and the spectra for all the hi sources are publicly available at the following link @xmath10 . the results of the lbw follow - up observations are presented in table [ lwidetab ] , which lists the following : + col.1 : hi source i d + col.2 : heliocentric velocity and relative error of the source ( @xmath29 ) , measured as the midpoint between the channels at which the flux density drops to 50% . + col.3 : velocity width and relative error of the source line profile measured at the 50% level . + the velocity widths are not corrected for instrumental broadening , turbulent motions , disk inclination or cosmological effects . + col.4 : velocity width and relative error of the source line profile measured at the 20% level . + col.5 : total flux and relative error in jy . + col.6 : the rms noise per channel at 10 km s@xmath27 velocity resolution . + in fig.[lwide ] we compare the measurements obtained using the alfa and l - wide receivers for the 22 sources confirmed during follow - up observations ( filled circles ) . although these are the lowest s / n sources within our sample , the agreement between the dataset is quite good with an average dispersion of 19 in recessional velocity , 35 in velocity width and @xmath114% in total flux . the coma - a1367 supercluster is one of the regions most intensively investigated by optically selected hi surveys ( i.e. @xcite ) . however only 21 out of the 100 sources detected by ages had a previous hi measurement . the high number of new hi detections in ages ( @xmath179% ) confirms once more the importance of hi blind surveys for having an unbiased view of the neutral hydrogen distribution in the local universe . among the 21 sources previously known @xcite , three of these objects ( namely cgcg97 - 079 , cgcg97 - 087 and cgcg127 - 052 ) are strongly contaminated by rfis in the ages cube , and can not be used for a comparison with the literature . our measurements for the remaining 18 objects are in satisfactory agreement with past works as illustrated in fig.[lwide ] ( empty circles ) . there is quite a good agreement in both fluxes and velocity width measurements and the average scatter is @xmath110% in flux and @xmath1 28 in velocity width . in one case ( cgcg97 - 138 ) our velocity width estimate differs considerably from literature values : @xcite gives a velocity width of 160whereas we measure only 62 ) . however our new l - wide measurement is consistent with the value obtained from the ages cube ( w=58 ) . the abell 1367 region has been covered in the optical by the sloan digital sky survey ( sdss ) as part of data release 5 . we have used this homogenous dataset to look for optical counterparts of our hi sources . we cross - correlated our list of hi detections with the sdss galaxy catalogue using a search radius of 4 arcmin . in the few cases in which no candidate optical counterpart was found , we visually inspected the sdss plates to look for low surface brightness galaxies not included in the sdss catalogue . once a candidate optical counterpart was found , we looked into the literature and ned for an estimate of its optical recessional velocity . in addition for 5 of the 46 candidates without an available redshift in the literature we obtained new optical recessional velocity estimates in january - febrary 2007 using the imaging spectrograph bfosc attached to the cassini 1.5 m telescope at loiano ( italy ) . observations and data reduction techniques applied are described in @xcite . confirmed optical counterparts are defined as those optical galaxies lying within 4 arcmin and having a recessional velocity within @xmath9200 from the radio velocity . candidate optical counterparts are galaxies which lie within 4 arcmin but do not have a published optical recessional velocity . of the 100 hi sources in our sample , 55 have confirmed optical counterparts , 34 have a unique candidate ( i.e. no optical recessional velocity available ) optical counterpart , 7 have two or more candidate counterparts and 4 sources appear to have no optical counterparts . three of these objects are associated with the interacting groups in the outskirts of abell 1367 described in the appendix of this paper . a list of the optical counterparts ( confirmed and candidate ) associated with the hi sources is presented in table [ optcounttab ] . in fig.[centroid ] we show the distribution of the difference between the optical and hi positions and recessional velocities for the 55 sources having confirmed optical counterparts . optical and radio measurements show a very good agreement with a median offset of only 18 arcsec in position ( consistent with the typical error of 16 arcsec in the estimate of the centroid for radio sources ) and of @xmath14@xmath916 in recessional velocity . for all the hi sources having 1 optical counterpart ( confirmed or candidate ) we downloaded sdss - dr5 @xmath30 images @xcite and performed aperture photometry and surface brightness profile decomposition using the task _ ellipse _ within iraf . the radial profiles have been constructed by integrating the available images within elliptical , concentric annuli . the ellipticity and position angles have been determined and then fixed using the @xmath31 band image . total magnitudes , effective radii ( @xmath32 ) and effective surface brightness ( @xmath33 , i.e. average surface brightness within @xmath32 ) have been determined following the procedure of @xcite . fig.[distrall ] illustrates the sky ( upper panel ) and the hi mass vs distance distributions ( lower panel ) for all the 100 sources detected in the a1367 cube . over half of the sources ( 54 ) belongs to the abell 1367 cluster and its outskirts , lying in the velocity range 4000@xmath119000 . this roughly explains why in this cube we detect twice the number of sources than in other ages fields : excluding the abell 1367 region ( 4000@xmath119000 ) we are left with 46 sources consistent with the number found in ngc628 @xcite , ngc1156 ( auld et al . in prep . ) and ngc7332 @xcite . the bottom panel of fig.[distrall ] allows a direct comparison of ages with alfalfa @xcite showing that , in the same area of sky , alfalfa would miss @xmath1 half of the sources detected here . this is quite expected since alfalfa has an integration time @xmath1 6.3 times shorter ( @xmath148 sec instead of @xmath1300 sec ) than ages . more impressive is the difference between ages and the northern hipass extension @xcite : whereas ages finds 100 sources , hipass does not detect a single galaxy in the 5@xmath01 square degrees here studied . we use the hi data from ages and the optical data from sdss to derive the physical properties of our sample . the hi mass ( @xmath34 ) is defined as : @xmath35 where @xmath16 is the integrated hi flux in jy and d is the galaxy distance in mpc . for those galaxies belonging to abell 1367 ( 4000@xmath119000 ) we assume @xmath36=92.8 mpc @xcite while we assume @xmath37 with @xmath38 70 @xmath39 for the rest of the sample . the stellar mass is obtained from the optical colors and luminosity as defined in @xcite : @xmath40 where @xmath41 is the absolute magnitude in the @xmath31 band . therefore the total amount of baryons is @xmath42=@xmath43+@xmath34 ( e.g. @xcite ) . this quantity does not include the contribution of helium and metals ( usually assumed equal to 1.4@xmath0m(hi ) ) , the molecular hydrogen and any warm or hot gas component . therefore @xmath42 must be considered as a lower limit of the real value . finally we define the galaxy dynamical mass as : @xmath44 where @xmath45 is the gravitational constant , @xmath46 is the radius containing 75% of the @xmath31 band flux , @xmath47 is the hi velocity width ( here we use the average between @xmath17 and @xmath28 hi velocity widths ) corrected for inclination and @xmath48 is the ratio between @xmath46 and the extension of the hi emission ( i.e. @xmath49 ) . @xmath48 should lie between 2 - 3 @xcite , but for our galaxies radio interferometric data ( necessary to estimate @xmath50 ) is not available , so this remains unknown and we leave it as a free parameter . the galaxy inclination is obtained from the ellipticity determined in the @xmath31 band via : @xmath51 where @xmath52 is the galaxy ellipticity and @xmath53 is the intrinsic axial ratio of the galaxy here assumed equal to 0.13 @xcite . we excluded from this analysis face - on galaxies having @xmath54 ( 20 galaxies ) , for which it is not possible to accurately estimate the dynamical mass . the correlations between different optical and hi quantities and their distributions for our sample are presented in fig.[paramall ] . from left to right : the ratio of the gas to stellar mass , the baryon fraction ( @xmath55 ) , the total hi mass , the effective surface brightness in @xmath31 band , the @xmath56 color and the dynamical mass are presented . only galaxies with one optical counterpart and @xmath54 ( 69 galaxies ) are shown : filled dots are objects with confirmed optical counterparts while empty dots indicate objects with optical counterpart candidates . our sample covers almost three orders of magnitude in both hi and dynamical mass . the optical surface brightness distribution shows a significant tail at low surface brightness ( @xmath5722.5 - 24 mag @xmath58 ) as usually observed in hi selected sample ( e.g. @xcite ) . hi selected samples are often considered as composed of only blue , gas - rich galaxies . this is not the case for our sample which spans over 1 magnitude in the @xmath56 color , as is typically observed in optically selected samples @xcite , and covers a wide range of gas to stars ratios ( @xmath59 ) : from gas poor objects ( @xmath60 0.1 ) to extremely gas rich galaxies ( @xmath60 10 ) . more massive galaxies appear to be redder , have higher surface brightness and lower gas content than dwarf systems . this is also found to be the case in optically selected samples @xcite . well known relations like color - mass , surface brightness - mass and hydrogen - dynamical mass are here rediscovered . unfortunately the small number of objects in our sample does not allow a more detailed comparison between the slopes , scatters and distributions of the relations followed by hi and optically selected galaxies . we postpone this kind of analysis to the moment when we will have a significantly large ages database at our disposition . + the most intriguing result presented in fig.[paramall ] is the absence of any correlation involving the baryon fraction . on the contrary all galaxies in our sample appear to have almost the same baryon fraction @xmath61=(-0.2-log(@xmath48))@xmath90.13 independently from their mass , surface brightness or gas content . the little dispersion in the baryon fraction distribution is in fact well within the observational errors . assuming 0.1 dex error in the hi mass ( see sec.4.1 ) , in the stellar mass @xcite and in the dynamical mass , the uncertainty on the estimate of the baryon fraction is @xmath10.17 dex . the gas and stellar mass fraction cover a larger dynamical range ( i.e. have a larger dispersion : @xmath62=(-0.4-log(@xmath48))@xmath90.24 and @xmath63=(-0.6-log(@xmath48))@xmath90.3 ) ) than the baryon fraction , as shown in fig.[massratios ] and accordingly to the kolmogorov - smirnov test the baryon fraction distribution is not compatible with the gas and star fraction distributions at a 99.99% confidence level . we can therefore conclude that our sample shows approximately the same baryon fraction . a constant baryonic mass fraction in galaxies implies a direct correlation between dynamical mass and baryonic mass : exactly what is usually required for the tully - fisher relation @xcite . this result is also consistent with the recent study of an optically selected sample of extremely low mass dwarf galaxies carried out by @xcite . to our knowledge , this is the first confirmation of a constant baryonic fraction coming from an hi selected sample . this observational evidence is also consistent with models of galaxy formation @xcite but in contradiction with recent models of galaxy evolution which invoke supernova feedback and other physical mechanisms to remove baryons preferentially from low mass systems and to reproduce the observed scaling relations @xcite . even if our sample includes low surface brightness galaxies it appears that physical mechanisms that preferentially remove baryons from low mass galaxies are not effective in the mass range and selection criteria of our sample . we remark that the only way to obtain a correlation between the baryon fraction and the dynamical mass in our sample would be to assume that @xmath48 ( i.e. the ratio of the hi to optical radius ) is a strong function of the mass and/or the contribution of molecular hydrogen , hot and warm gas to the total baryon mass varies with the mass of the galaxy . we can not a priori exclude the first possibility , even if unlikely since @xcite found that the ratio of the hi to optical radius varies by only 20% along the hubble sequence . conversely , the molecular hydrogen represents only @xmath115% of the total gas reservoir in normal , late - type galaxies @xcite and the contribution of hot and warm gas components is a few percents ( @xmath641 - 2% ) of the total gas mass ( e.g. @xcite ) making the second scenario extremely unlikely . more than half of the hi sources detected in the cube belong to the abell 1367 volume . this includes not only the cluster core but also a part of the great wall : a large scale filament of galaxies connecting abell 1367 to the coma cluster @xcite . all galaxies in the filament lie approximately at the same distance from us as abell 1367 , implying that the system a1367+great wall constitutes a volume limited sample . at the distance of abell 1367 ( @xmath192.8 mpc ) the surveyed sky area has a physical size of 8.1@xmath01.6 @xmath65 and our sensitivity limit for a source with @xmath17=200 is @xmath16@xmath4 . since the virial radius of abell 1367 is @xmath1 2.3 mpc @xcite , our region covers the cluster virialized region and its immediate neighborhood . in order to compare the hi and optical properties of the abell 1367 region we extracted an optically selected sample of galaxies from sdss - dr5 . we select all galaxies in the ages region having @xmath66 mag ( corresponding to @xmath67 at the abell 1367 distance ) : 259 galaxies in total , of which 208 have optical redshift available and 155 are confirmed cluster members ( 4000@xmath119000 ) . we note that the time necessary to carry out a hi targeted survey of this optically selected sample ( reaching the same ages noise level ) would be similar to the observing time needed by ages to cover all the 5@xmath01 square degrees in abell 1367 . the sky distribution of the 54 hi sources belonging to the a1367 region ( 4000@xmath119000 ) is shown in fig.[skya1367 ] . it is interesting to note that no significant over - density of sources is observed corresponding to the cluster center : the only hi galaxy lying well within the x - ray cluster contours belongs to a small group of galaxies infalling for the first time into the cluster , suggesting that it is just entering the cluster and it is only projected on the center of abell 1367 @xcite . this appears much more evident when we compare the sky distribution of our sample with the typical sky distribution obtained for the optically selected sample . in fig.[skya1367 ] we show the distribution of the 155 confirmed cluster members ( empty circles ) and of the 51 galaxies without redshift available mag , suggesting that a not negligible fraction is composed by background galaxies @xcite . ] ( empty stars ) extracted from the optically selected sample . the optical selected sample is strongly clustered on the cluster center and almost half of the sources detected lie within the x - ray emitting region . moreover the `` finger of god '' feature , typical of clusters of galaxies , is not observed in the hi selected sample ( see fig.[wedge ] ) . conversely the hi - selected galaxies show a different pattern in the wedge diagram shown in fig.[wedge ] : galaxies on the east side of the cluster have a lower recessional velocity than galaxies in the west side . this probably suggests the presence of two different infalling directions into the cluster core as proposed by @xcite . the differences observed between the optically and hi selected samples are somehow expected since clusters of galaxies are made hi deficient structures through the combination of the morphology - density relation @xcite and of the `` gas - density '' relation @xcite . this is also consistent with recent determinations of the correlation function for hi selected samples @xcite which show that hi galaxies are among the least clustered objects in the universe . the influence of the harsh cluster environment is barely visible when we look for radial variations of the gas to star mass ratio and of the hi - deficiency ( see fig.[defradius ] ) . the hi deficiency is defined as the difference , in logarithmic units , between the observed hi mass and the value expected from an isolated galaxy with the same morphological type t and optical linear diameter d : hi def = @xmath68 @xcite . we used the equations in @xcite to calculate the expected hi mass from the optical diameter . the hi deficiency does not significant vary with the cluster - centric distance ( fig.[defradius ] , bottom panel ) , contrary to what usually observed in optically selected samples ( i.e. @xcite ) . a slight decrease on the average value of the gas to star mass ratio is observed in correspondence of the cluster center , however also in this case no strong gradients have been found . we can therefore conclude that when observed in hi the abell cluster 1367 almost completely disappears and the only evidence of its hidden presence is a slightly higher number of galaxies with a low gas to star mass ratio near the cluster core . the abell 1367 local volume can be used for a more detailed analysis of selection effects representing a unique dataset for which complete samples independently selected at optical and radio wavelengths are available . we are particularly interested in the comparison of the hi mass distribution and optical luminosity function estimated from the two samples . in the past , different authors have used optically selected samples to determine the hi mass distributions of galaxies in the local universe obtaining results significantly different from radio selected samples @xcite . we therefore used the optically selected sample extracted from sdss - dr5 , complete to @xmath66 mag , in order to compare the hi distribution and the @xmath69 band luminosity function for optically and hi selected galaxies . contrary to the hi selected sample , the sdss sample is not complete in redshift , but the redshift completeness rapidly drops below 50% for @xmath70 mag ( @xmath71 ) . this complicates the estimate of the luminosity function at low luminosities . in order to overcome the redshift incompleteness of cluster samples we used the _ completeness corrected _ method proposed by @xcite . this method is based on the assumption that the spectroscopic sample ( i.e. galaxies with redshift available ) is representative of the entire cluster , i.e. the fraction of galaxies that are cluster members is the same in the ( incomplete ) spectroscopic sample as in the ( complete ) photometric one . the hi mass distributions obtained from the optically and hi selected samples are shown in fig.[lfhi ] . as expected the two samples produce hi distributions significantly different . the optical selected sample drops quicker at low hi masses and does not include nearly half of the hi sources in the a1367 volume . the two samples start to differ at @xmath722@xmath73 , well before our sensitivity limit . low luminosity , low surface brightness , gas rich objects are in fact not included in the optical sample , but they represent a significant fraction of the total hi budget in abell 1367 . this result is consistent with the recent work by @xcite who showed that the faint end slope of the hi mass function for an optically selected sample is less steep than the one obtained from a 21 cm selected sample . our analysis suggests that this difference is mainly due to selection effects and that hi distributions obtained from optically selected samples can not be generally considered as good proxies of the hi mass function in the local universe . the hi and optically selected samples also differ in the estimate of the optical luminosity functions ( fig.[lfopt ] , lower panel ) . the luminosity function ( lf ) for the hi selected galaxies is flatter and includes fewer objects ( i.e. lower @xmath74 ) than the one obtained from the optically selected sample . a great part of this difference is due to the overabundance of red ( @xmath750.9 ) , gas poor galaxies in the center of abell 1367 : more than 80% of bright red galaxies ( @xmath76 ) in the a1367 region are not detected at 21 cm ( see fig.[lfopt ] , upper panel ) . the difference between the two samples is less evident if we compare the luminosity functions obtained only for blue galaxies ( fig.[lfopt ] , middle panel ) , but even in this case the hi sample misses @xmath130% of blue cluster galaxies . this result is quite interesting since we generally associate hi deficiency to red colors . in order to further investigate the properties of these blue galaxies not detected in hi , in fig.[colmag ] we compare the color @xmath56 , @xmath31 magnitude relation for the confirmed cluster members in the optically ( empty circles ) and hi selected ( triangles ) samples . almost all the hi galaxies lie on the blue sequence , but the opposite is not true : not all the galaxies in the blue sequence are detected in hi . this could be due to a number of different reasons . first of all we have to take into account our sensitivity limit ( @xmath726@xmath4 for w@xmath77 @xmath1 200 ) which we need to relate to optical color and magnitude . to do so we have used the correlation between the hi mass and @xmath31 band luminosity ratio and the @xmath56 color observed for our sample . the best bisector linear fitting gives the following relation : @xmath78 with a dispersion of 0.37 dex . we can use this equation to estimate our sensitivity limit , which is indicated by the solid line in fig.[colmag ] . we consider undetected blue galaxies those objects not detected by ages , lying outside the one @xmath12 region and with a @xmath56 color bluer than 0.9 mag : 10 galaxies in total . some of these sources lie within ( in both space and velocity ) strong hi galaxies and can be missed , due to the size of the arecibo beam ( fwhm@xmath13.5 arcmin ) and to the fact that in these cases the hi emission is totally assigned to the bright source . moreover some galaxies have a recessional velocity within the rfis frequencies and can not be detected . excluding these cases ( indicated with crosses in fig.[colmag ] , left ) we are left with eight blue galaxies which are not detected in hi . these objects have a @xmath69 band luminosity in the range 2@xmath010@xmath79 10@xmath80 l@xmath81 ( i.e. -19.4@xmath82 - 17.9 mag ) , implying that we are dealing with intermediate mass galaxies and not dwarf systems . moreover their broad - band morphology and structural parameters are indistinguishable from the other blue hi detected galaxies . the limits obtained from eq.[ml ] imply that the undetected galaxies have a hi deficiency @xmath830.37 , suggesting that we are dealing with blue , deficient objects , consistent with the marginal detection of two of these galaxies ( cgcg97 - 092 , cgcg97 - 093 ) by deeper surveys @xcite implying an hi deficiency 0.31 and 0.57 respectively . this suggests that we are dealing with the same population of hi poor , star forming spirals originally observed in nearby clusters by @xcite . only when observed in h@xmath84 emission of a galaxy is due to the hydrogen ionized in hii regions by massive ( @xmath8510 m@xmath81 ) , young ( @xmath8 2 @xmath0 10@xmath86 yr ) stars . ] these galaxies appear different from healthy spirals , showing a truncation of the star forming disk . in particular , in all but one object for which h@xmath84 data are available ( 8 galaxies , @xcite ) the h@xmath84 radius ( e.g. the isophotal radius within @xmath87 ) is @xmath12 - 4 times less extended than the @xmath88 band 24th mag @xmath58 isophotal radius ( see table [ nodettab ] ) . this is consistent with various independent studies @xcite which have shown that the hi deficiency is associated with a truncation of the star forming disk . all the deficient objects lie at a projected distance of less than @xmath11 degree from the center of abell 1367 , suggesting that the absence of hi and the truncation of the star formation disk is likely due to their interaction with the cluster environment . the fact that these objects have lost almost 40% of their original gas content but their broadband morphology and optical colours are still those typical of normal star forming spirals implies that they are recent arrivals within the cluster environment . it appears that the neutral hydrogen has been already stripped well within the optical radius quenching the star formation in the galaxy outskirts , but a type stars have not yet died and the galaxies still appear as blue as they were before losing their gas . given that h@xmath84 traces stars younger than @xmath89 yr and the typical lifetime of a stars is @xmath11 gyr , we can speculate that our galaxies started to loose their gas @xmath1100myr ago , in agreement with the typical timescale for gas stripping observed in virgo cluster galaxies ( e.g. @xcite ) and predicted by numerical models ( e.g. @xcite ) . in less than 1 gyr , when all the a stars are dead , these objects will lie in the @xmath56 red sequence and will not be any different from the bulk of the gas - poor ( red ) galaxy cluster population . only future detailed spectroscopic investigation will allow us the better estimate the time scale of this transformation . finally , it is interesting to note that this population of blue gas - poor galaxies corresponds to @xmath1 one third of the whole population of blue galaxies in the a1367-volume , supporting the idea that a1367 is a cluster still young and a considerable number of objects are infalling into the cluster core for the first time @xcite . we have presented the results of a hi blind survey of 5@xmath01 square degrees centered on the core of abell 1367 as part of the arecibo galaxy environment survey . one hundred hi galaxies have been detected ( 79 new measurements and 50 new redshifts ) , half of which belong to the abell 1367 volume . comparing the properties of the hi selected galaxies with an optically selected sample extracted from sdss - dr5 we have shown that the large scale distribution of galaxies is strongly wavelength dependent : in hi the cluster core almost completely disappears and hi selected galaxies are homogeneously distributed in the cluster volume . the difference between the two samples is even more evident when we compare their luminosity and hi distribution . our optically selected sample misses a considerable number of low luminosity , low surface brightness galaxies underestimating by almost a factor 2 the number of hi rich galaxies present in the cluster volume . on the contrary our hi sample contains a factor 3 less bright galaxies than the optical one . these are mainly ellipticals or early type , red spirals but also a not negligible fraction of the blue sequence cluster spirals is missed . all these blue gas - poor objects are located near the edge of the x - ray cluster contours suggesting that their gas has been recently ( @xmath1 100 myr ) stripped by ram pressure so that the galaxies are already gas poor whereas their stellar populations are still young . + although the spatial and number distribution of samples selected at different wavelengths are significantly different , the internal properties of galaxies seem not to be significantly wavelengths dependent . hi galaxies seem to follow the same scaling relations observed in optical and bigger galaxies are redder , have a higher surface brightness and a lower gas content than lower luminosity galaxies . the most interesting result of our analysis is that the baryon fraction of our sample appears to be almost constant , suggesting the difference in the star to gas ratio observed between bright and faint galaxies is only due to a different star formation efficiency ( i.e. the time scale necessary to convert gas into stars ) . this is the first time that such a result is obtained from an hi selected sample and even if consistent with models of galaxy formation it suggests that during their evolutionary history less massive galaxies do not lose a larger fraction of their baryons than bigger objects . the analysis presented in this paper points out the importance of hi blind surveys like ages in order to have an unbiased view of the properties of galaxies in the local universe . once the survey is at a more advanced stage and the size of our sample is significantly bigger than the one presented here it will be possible to carry out more detailed quantitative studies and comparisons between optically and hi selected samples , gaining more insight into the evolutionary history of nearby galaxies . lc wishes to thank all the ages consortium and in particular elias brinks & noah brosh for useful discussions and comments on this manuscript . lc is supported by the uk particle physics and astronomy research council . this work is based on observations collected at arecibo observatory . the arecibo observatory is part of the national astronomy and ionosphere center , which is operated by cornell university under a cooperative agreement with the national science foundation . we wish to thank the arecibo observatory for granting us the time to complete this part of the project and its staff for the help during the observations and data reduction . the tac of the loiano telescope is acknowledged for the time allocation to this project . this research has made use of the nasa / ipac extragalactic database , which is operated by the jet propulsion laboratory , california institute of technology , under contract to nasa and of the goldmine database @xcite . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the sdss web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , university of cambridge , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . 83 natexlab#1#1 , m. g. , moore , b. , & bower , r. g. 1999 , , 308 , 947 , j. k. , ageros , m. a. , allam , s. s. , et al . 2007 , , 172 , 634 , r. , minchin , r. f. , davies , j. i. , et al . 2006 , , 371 , 1617 , i. k. , glazebrook , k. , brinkmann , j. , et al . 2004 , , 600 , 681 , d. g. , staveley - smith , l. , de blok , w. j. g. , et al . 2001 , , 322 , 486 , s. , plionis , m. , kova , k. , & voglis , n. 2007 , , 370 , e. f. , mcintosh , d. h. , katz , n. , & weinberg , m. d. 2003 , , 585 , l117 , e. f. , mcintosh , d. h. , katz , n. , & weinberg , m. d. 2003 , , 149 , 289 , a. , boissier , s. , cortese , l. , et al . 2006 , , 651 , 811 , a. & gavazzi , g. 2006 , , 118 , 517 , a. , lequeux , j. , & gavazzi , g. 2002 , , 384 , 33 , f. h. & rao , s. 1993 , , 417 , 494 , v. , takeuchi , t. t. , iglesias - paramo , j. , et al . 2007 , apjs in press ( astro - ph/0609738 ) , j. s. , dekel , a. , kolatt , t. s. , et al . 2001 , , 555 , 240 , b. , haynes , m. p. , & giovanelli , r. 2005 , , 130 , 1037 , v. , kotanyi , c. , balkowski , c. , & van gorkom , j. h. 1994 , , 107 , 1003 , g. l. , giovanelli , r. , & haynes , m. p. 1983 , , 269 , 13 , s. , lacey , c. g. , baugh , c. m. , & frenk , c. s. 2000 , , 319 , 168 , l. , boselli , a. , gavazzi , g. , et al . 2005 , , 623 , l17 , l. , gavazzi , g. , boselli , a. , et al . 2006 , , 453 , 847 , l. , gavazzi , g. , boselli , a. , iglesias - paramo , j. , & carrasco , l. 2004 , , 425 , 429 , l. , gavazzi , g. , iglesias - paramo , j. , boselli , a. , & carrasco , l. 2003 , , 401 , 471 , l. , marcillac , d. , richard , j. , et al . 2007 , , 376 , 157 , l. l. , songaila , a. , hu , e. m. , & cohen , j. g. 1996 , , 112 , 839 , r. a. , eke , v. r. , frenk , c. s. , et al . 2007 , , 377 , 41 , j. , minchin , r. , sabatini , s. , et al . 2004 , , 349 , 922 , j. i. , de blok , w. j. g. , smith , r. m. , et al . 2001 , , 328 , 1151 , r. , colless , m. , driver , s. p. , et al . 2003 , , 342 , 725 , a. 1980 , , 236 , 351 , k. c. 1970 , , 160 , 811 , w. , staveley - smith , l. , calabretta , m. , et al . 2005 , in bulletin of the american astronomical society , vol . 37 , bulletin of the american astronomical society , 1316+ , g. 1987 , , 320 , 96 , g. 1989 , , 346 , 59 , g. , boselli , a. , donati , a. , franzetti , p. , & scodeggio , m. 2003 , , 400 , 451 , g. , boselli , a. , van driel , w. , & oneil , k. 2005 , , 429 , 439 , g. , franzetti , p. , scodeggio , m. , boselli , a. , & pierini , d. 2000 , , 361 , 863 , g. , oneil , k. , boselli , a. , & van driel , w. 2006 , , 449 , 929 , g. , pierini , d. , & boselli , a. 1996 , , 312 , 397 , m. , blanton , m. r. , masjedi , m. , & west , a. a. 2006 , , 653 , 240 , r. & haynes , m. p. 1985 , , 292 , 404 , r. , haynes , m. p. , herter , t. , et al . 1997 , , 113 , 22 , r. , haynes , m. p. , kent , b. r. , et al . 2005 , , 130 , 2598 , r. , haynes , m. p. , kent , b. r. , et al . 2007 , , 133 , 2569 , m. , giuricin , g. , mardirossian , f. , mezzetti , m. , & boschin , w. 1998 , , 505 , 74 , f. , willman , b. , mayer , l. , et al . 2007 , , 374 , 1479 , m. p. & giovanelli , r. 1984 , , 89 , 758 , m. p. , giovanelli , r. , herter , t. , et al . 1997 , , 113 , 1197 , p. a. , springob , c. m. , catinella , b. , et al . 2006 , in american astronomical society meeting abstracts , vol . 208 , american astronomical society meeting abstracts , 53.04+ , p. a. , staveley - smith , l. , ekers , r. d. , et al . 2000 , , 119 , 2686 , s. j. u. , armus , l. , higdon , j. l. , soifer , b. t. , & spoon , h. w. w. 2006 , , 648 , 323 , j. , boselli , a. , cortese , l. , vlchez , j. m. , & gavazzi , g. 2002 , , 384 , 383 , g. , heckman , t. m. , white , s. d. m. , et al . 2003 , , 341 , 33 , jr . , r. c. , bothun , g. d. , & schommer , r. a. 1984 , , 89 , 1279 , r. a. , haynes , m. p. , & catinella , b. 2006 , , 131 , 716 , b. s. , staveley - smith , l. , kilborn , v. a. , et al . 2004 , , 128 , 16 , r. h. , boyce , p. j. , kilborn , v. a. , et al . 2003 , , 342 , 738 , b. m. 1983 , , 88 , 962 , s. & de blok , e. 1998 , in astronomical society of the pacific conference series , vol . 136 , galactic halos , ed . d. zaritsky , 210+ , s. s. , schombert , j. m. , bothun , g. d. , & de blok , w. j. g. 2000 , , 533 , l99 , m. j. , zwaan , m. a. , webster , r. l. , brown , m. j. i. , & staveley - smith , l. 2007 , , 654 , 702 , r. f. , auld , r. , davies , j. i. , et al . 2007 , in iau symposium , vol . 235 , iau symposium , ed . f. combes & j. palous , 227229 , r. f. , disney , m. j. , boyce , p. j. , et al . 2003 , , 346 , 787 , r. f. , momjian , e. , cortese , l. , et al . 2006 , in american astronomical society meeting abstracts , 95.03+ , e. & brggen , m. 2007 , , 765 , j. l. & schneider , s. e. 2000 , , 130 , 177 , a. 2007 , , 133 , 2087 , s. , mould , j. r. , hughes , s. m. g. , et al . 2000 , , 529 , 698 , e. e. & hoffman , g. l. 1996 , , 465 , 595 , r. j. , teuben , p. j. , & wright , m. c. h. 1995 , in asp conf . 77 : astronomical data analysis software and systems iv , ed . r. a. shaw , h. e. payne , & j. j. e. hayes , 433+ , y. , bekki , k. , couch , w. j. , & de propris , r. 2002 , , 565 , 223 , j. m. , giovanelli , r. , & haynes , m. p. 1996 , , 461 , 609 , j. m. , manrique , a. , garca - gmez , c. , et al . 2001 , , 548 , 97 , j. g. & schneider , s. e. 1998 , , 119 , 159 , c. m. , haynes , m. p. , & giovanelli , r. 2005 , , 621 , 215 , c. m. , haynes , m. p. , giovanelli , r. , & kent , b. r. 2005 , , 160 , 149 , iii , w. t. , bates , b. , bothun , g. d. , & schommer , r. a. 1981 , , 86 , 919 , r. a. , van albada , t. s. , van der hulst , j. m. , & sancisi , r. 2002 , , 390 , 829 , d. , bomans , d. j. , hensler , g. , & junkes , n. 2001 , , 380 , 40 , b. , balkowski , c. , cayatte , v. , van driel , w. , & huchtmeier , w. 2004 , , 419 , 35 , b. , cayatte , v. , balkowski , c. , & duschl , w. j. 2001 , , 561 , 708 , b. c. , gilmore , d. m. , & jones , c. 1993 , , 407 , 489 , o. i. , ryan - weber , e. v. , garcia - appadoo , d. a. , et al . 2006 , , 371 , 1855 , a. i. , geller , m. j. , huchra , j. p. , & ramella , m. 1993 , , 106 , 1301 [ [ high - velocity - clouds ] ] high velocity clouds + + + + + + + + + + + + + + + + + + + + two extended features well separated from the milky way emission and probably associated with high velocity clouds ( hvcs ) have been detected in the velocity range 140@xmath11210 ( fig.[hvc ] ) . the first lies in the range 11:50@xmath9011:55 and probably extends outside the region observed . the brightest region is in the south - east corner of the cube ( 11:54:00 , + 19:20:00 , 160@xmath11210 ) and a low column density stream seems to extend to the north - west . the second and strongest hvc occupies great part of the east part of the cube and extends over 1.5 degrees ( 11:40@xmath9011:46 , 19:25@xmath9120:15 ) with a velocity in the range 140@xmath11175 . the cgcg97027-group is located at a projected distance of @xmath12 degrees ( @xmath13.2 mpc ) from the centre of abell 1367 . it is composed of two star forming spirals cgcg97027 ( v@xmath16630 ) and cgcg97026 ( v@xmath16220 ) and one lenticular cgcg97023 ( v@xmath16320 ) . at a projected distance of @xmath113 arcmin ( @xmath10.7 mpc ) sw from the group lies the elliptical galaxy cgcg97 - 021 ( v@xmath16648 ) . our data show the presence of diffuse hi emission all over the region occupied by the four bright zwicky galaxies ( see fig.[97027 ] ) . in particular and high velocity 6500@xmath116800 streams extends from cgcg97 - 021 to the north ( fig.[97027 ] , left ) and two of the hi sources without optical counterparts belong to this stream ( ages j113614 + 195910 and ages j113626 + 195102 ) . extended emission is also detected in the north part of the cgcg97 - 027 group in the velocity range 6000@xmath116500 ( fig.[97027 ] , right ) . : the cgcg97041-group is located at a projected distance of 1 degrees ( 1.6 mpc ) from the centre of abell 1367 and , as the cgcg97027-group , it is composed of two spirals cgcg97036 ( v@xmath16595 ) and agc210559 ( v@xmath16825 ) and one lenticular cgcg97041 ( v@xmath16778 ) . we detect hi emission extended over @xmath1200 kpc north of the three group members , apparently not associated with either of them ( see fig.[97041 ] ) . this source ( @xmath92 7382 ) has no optical counterpart catalogued in the sdss database . however in the sdss plates a low surface brightness ( @xmath93 24 @xmath94 galaxy is present in correspondence of the hi coordinates . we considered this object as candidate optical counterpart and labeled it as sdsslsb in table [ optcounttab ] . this source lies at a projected distance of @xmath15.8 arcmin from the starburst galaxy cgcg97 - 068 . no galaxies or extended features are found in deep cfht b - band @xcite and galex nuv - fuv @xcite data in correspondence with the peak of hi emission , suggesting that it could be a stream of gas stripped from cgcg97 - 068 . unfortunately the distance of the source from cgcg97 - 068 is consistent with the region in which the first sidelobe is expected and the fact that ages j114239 + 201150 lies at exactly the same recessional velocity of the starburst galaxy makes impossible to discard the possibility that this source is sidelobe contaminated . vla d - configuration observations has been recently obtained for the interacting groups described above and for cgcg97 - 068 and will be used to study in detail the properties of these sources . the results of these observations will be presented in a forthcoming work . band luminosity function for galaxies in the a1367 volume as obtained from the optically ( empty circles ) and hi ( histogram ) selected sample . all galaxies in the two samples are shown in the bottom panel , while blue ( @xmath970.9 ) and red galaxies ( @xmath750.9 ) are shown in the middle and upper panel respectively.,width=302 ] color @xmath41 magnitude relation for confirmed cluster members in the abell 1367 region . triangles indicate galaxies detected in the ages cube . crosses indicate those galaxies lying in the rfis velocity range or within the beam occupied by brighter sources . the solid ( dashed ) line indicate our sensitivity limit ( @xmath91 @xmath12 ) obtained converting the hi sensitivity into color limit using the @xmath56 versus @xmath98 relation given by eq.[ml ] . , width=302 ] @xmath101j114017 + 194718 & 11:40:17.26 & 19:47:19.1 & -- \\ ages~j114027 + 192429 & kug1137 + 196 & 11:40:26.15 & 19:24:51.9 & 3421 \\ ages~j114039 + 195455 & [ ibc2002]j114038 + 195437 & 11:40:38.99 & 19:54:38.5 & 7784 \\ ages~j114113 + 193240 & kug1138 + 198 & 11:41:13.43 & 19:32:21.7 & 10011 \\ ages~j114118 + 200829 & kug 1138 + 204 & 11:41:17.45 & 20:08:33.4 & 14500 \\ ages~j114129 + 200550 & sdssj114130.18 + 200629.8 & 11:41:30.18 & 20:06:29.8 & -- \\ & sdssj114129.01 + 200550.4 & 11:41:29.01 & 20:05:50.4 & -- \\ & sdssj114129.74 + 200506.0 & 11:41:29.74 & 20:05:06.0 & -- \\ ages~j114131 + 201912 & sdssj114134.75 + 201917.9 & 11:41:34.75 & 20:19:17.9 & -- \\ & sdssj114128.58 + 201803.8 & 11:41:28.58 & 20:18:03.8 & -- \\ ages~j114143 + 193536 & maps - ngpo\_434\_0037838 & 11:41:43.22 & 19:35:26.2 & -- \\ ages~j114148 + 194538 & [ ibc2002]j114149 + 194605 & 11:41:49.83 & 19:46:04.4 & 7789 \\ ages~j114204 + 194536 & sdssj114205.74 + 194611.7 & 11:42:05.74 & 19:46:11.7 & -- \\ ages~j114209 + 201924 & ngc3821 & 11:42:09.09 & 20:18:56.6 & 5788 \\ ages~j114211 + 195827 & kug1139 + 202 & 11:42:14.77 & 19:58:35.2 & 7815 \\ ages~j114216 + 200302 & kug1139 + 203 & 11:42:15.66 & 20:02:55.5 & 6102 \\ ages~j114224 + 200710 & cgcg097 - 068 & 11:42:24.48 & 20:07:09.9 & 5974 \\ ages~j114239 + 201150 & -- & -- & -- & -- \\ ages~j114243 + 200143 & cgcg097 - 072 & 11:42:45.21 & 20:01:56.6 & 6332 \\ ages~j114255 + 195734 & cgcg097 - 073 & 11:42:56.46 & 19:57:58.4 & 7275 \\ ages~j114308 + 194210 & kug1140 + 199 & 11:43:07.96 & 19:41:57.4 & 12988 \\ ages~j114311 + 200032 & cgcg097 - 079 & 11:43:13.32 & 20:00:17.5 & 7000 \\ ages~j114312 + 193655 & [ ibc2002]j114313 + 193645 & 11:43:13.02 & 19:36:46.6 & 6121 \\ ages~j114331 + 193747 & cgcg097 - 083 & 11:43:31.22 & 19:37:40.6 & 13126 \\ ages~j114347 + 195820 & ugc06697 & 11:43:49.11 & 19:58:06.2 & 6725 \\ ages~j114358 + 200430 & ngc3840 & 11:43:59.01 & 20:04:37.3 & 7368 \\ ages~j114402 + 202313 & 2masxj11435706 + 2022383 & 11:44:01.92 & 20:22:51.6 & 11962 \\ ages~j114416 + 201309 & cgcg097 - 102 & 11:44:17.22 & 20:13:23.9 & 6368 \\ ages~j114435 + 192841 & maps - ngpo\_434\_0047286 & 11:44:36.39 & 19:28:40.1 & -- \\ ages~j114447 + 200737 & ugc06719 & 11:44:47.12 & 20:07:30.2 & 6571 \\ ages~j114451 + 192723 & ngc3859 & 11:44:52.24 & 19:27:15.2 & 5468 \\ ages~j114452 + 194635 & cgcg097 - 125 & 11:44:54.85 & 19:46:34.9 & 8271 \\ ages~j114503 + 195832 & ngc3861 & 11:45:03.87 & 19:58:25.1 & 5085 \\ ages~j114508 + 200837 & [ ibg2003 ] j114506 + 200849 & 11:45:06.40 & 20:08:49.2 & 3822 \\ ages~j114543 + 200147 & cgcg097 - 138 & 11:45:44.67 & 20:01:51.6 & 5313 \\ ages~j114545 + 194210 & [ ibg2003]j114545 + 194130 & 11:45:45.27 & 19:41:30.7 & 6123 \\ ages~j114612 + 202403 & ngc3884 & 11:46:12.19 & 20:23:29.9 & 6946 \\ ages~j114625 + 193024 & sdssj114626.45 + 193011.9 & 11:46:26.45 & 19:30:11.9 & -- \\ ages~j114638 + 194854 & sdssj114636.47 + 194834.1 & 11:46:36.47 & 19:48:34.1 & -- \\ & sdssj114640.32 + 194820.8 & 11:46:40.32 & 19:48:20.8 & -- \\ ages~j114646 + 191906 & 2masxj11464700 + 1919462 & 11:46:47.03 & 19:19:46.2 & -- \\ ages~j114724 + 192324 & maps - ngpo\_434\_0048296 & 11:47:25.08 & 19:23:31.4 & -- \\ ages~j114727 + 194906 & sdssj114730.39 + 194859.0 & 11:47:30.39 & 19:48:59.0 & -- \\ ages~j114730 + 195353 & 2masxj11473099 + 1952201 & 11:47:30.99 & 19:52:20.8 & 5778 \\ ages~j114739 + 195621 & cgcg097 - 152 & 11:47:39.35 & 19:56:22.0 & 6166 \\ ages~j114739 + 200924 & sdssj114737.57 + 200900.6 & 11:47:37.57 & 20:09:00.6 & -- \\ ages~j114809 + 192109 & -- & -- & -- & -- \\ ages~j114823 + 194244 & sdssj114825.21 + 194217.0 & 11:48:25.21 & 19:42:17.0 & -- \\ ages~j114824 + 202331 & lsbcd571 - 02 & 11:48:26.40 & 20:23:19.0 & 6796 \\ ages~j114829 + 194529 & maps - ngpo\_434\_0031444 & 11:48:28.43 & 19:45:36.0 & -- \\ ages~j114956 + 192840 & sdssj114956.75 + 192840.4 & 11:49:56.76 & 19:28:40.4 & -- \\ & sdssj114956.93 + 192830.7 & 11:49:56.93 & 19:28:30.7 & -- \\ ages~j115030 + 200505 & maps - ngpo\_434\_0021239 & 11:50:29.99 & 20:05:08.2 & -- \\ ages~j115034 + 194301 & sdssj115033.61 + 194250.1 & 11:50:33.61 & 19:42:50.1 & -- \\ ages~j115052 + 193102 & kug1148 + 197a & 11:50:52.48 & 19:31:10.8 & -- \\ ages~j115101 + 193728 & 2masxj11510128 + 1937311 & 11:51:01.29 & 19:37:31.3 & -- \\ ages~j115101 + 202355 & ugc06821 & 11:51:01.12 & 20:23:57.3 & 6438 \\ ages~j115120 + 192931 & kug1148 + 197 & 11:51:21.28 & 19:29:33.9 & -- \\ ages~j115147 + 192057 & kug1149 + 196 & 11:51:48.37 & 19:21:29.3 & 6932 \\ \noalign{\smallskip } \hline \end{array}\ ] ] * : new optical redshift obtained in winter 2007 using the loiano 1.5 meter telescope .
perhaps the most exciting astronomical observation leading to our current understanding of black holes has been the discovery of quasars . these celestial objects , originally found in the early sixties as point - like radio emitters , were identified with apparently stellar sources , possessing somewhat unusual spectra , with prominent emission lines . the identification by @xcite of these lines as redshifted systems implied that quasars are distant and extremely luminous , commonly producing @xmath2 erg s@xmath3 ; this is a hundred times or more in excess of the total luminosity of all the stars in a galaxy . sensitive imaging of the nebulosities which often surround them implied that quasars are nuclei of galaxies , and thus are higher - luminosity counterparts of the compact nuclei of seyferts , studied some twenty years before the discovery of quasars as unusual emission line objects : hereafter , we assume that they are respectively the lower and higher luminosity end of the same population . a variety of scenarios were advanced to explain their nature , and this included multiple supernovae or massive spinning stars , but the proposal that quasars are powered by an accretion of surrounding matter onto a black hole , advanced in the mid-60s by @xcite and @xcite , became the paradigm that we are developing and testing today . while this is a viable and very attractive paradigm , only the last few years brought a solid evidence for it , allowing also to measure the mass of the central object . it is important to note here that quasars are much more numerous at a redshift @xmath4 2 than they are locally , meaning that a substantial fraction of galaxies must have undergone the quasar phase . it is thus likely that many otherwise normal local galaxies harbor supermassive black holes , `` dead quasars . '' in fact , as we discuss below , there is a number of relatively anonymous galaxies that show no signs of nuclear activity , but _ do _ show evidence for such black holes . we present the observational evidence for black holes in agn in section 2 ; in section 3 , we discuss the effects of the orientation of the accretion disk surrounding the back hole on the appearance of the nucleus . in section 4 , we discuss the jet - dominated agn known as blazars . in section 5 , we review the observations of agn in x rays and @xmath1rays , the bands that sample the regions closest to the black hole , and in section 6 , we review the radiation processes proposed to explain the emission in these bands . there are two general lines of argument that are used to `` prove '' the existence of black holes in agn . the first attempts to measure the total mass within a volume , and argues that no other form besides a black hole can have these parameters . this is done either via estimation of the volume from variability data ( via the light travel time arguments ) and mass from the luminosity ( via the eddington limit ) ; alternatively , this can be determined by a measurement of velocity of matter at a specified distance from the central object , essentially using kepler s laws . the second method , discussed in more detail in chapter 5.3 ( by a. fabian ) , relies on the distortion of the emission line shapes caused by strong gravity resulting from the presence of a black hole , and we cover it here only briefly . the variability of active galaxies generally shows the highest amplitude and the shortest time scales in the x ray and @xmath1ray bands , which happen to be clearly separated from the optical / uv bands by the strong absorption of the interstellar medium in our own galaxy . this rapid variability as well as other lines of argument indicate that the x ray / @xmath1ray radiation arises the closest to the central source , and in many cases , is the primary source of energy in active galaxies . while the total bolometric luminosity of quasars is often dominated by the optical and uv flux ( see , e. g. , laor et al . 1997 ; for a recent review , see ulrich , maraschi , & urry 1997 ) , the bulk of this flux probably arises in more distant regions from the central source than the x rays and @xmath1rays . the optical and uv radiation arising in the innermost regions of the nucleus , on the other hand , is most likely a result of reprocessing of x ray / @xmath1ray photons . this as well as the author s personal interest in x rays is the reason why this chapter focuses primarily on the high energy emission from quasars . in general , the x ray variability of quasars is aperiodic . while a measurement of periodic variability would give us a clue to the circumnuclear environment and thus the nature of the black hole , besides the ill - fated ngc 6814 ( cf . madejski et al . 1993 ) , no strict periodicity was reported for any agn . however , there were two reports of quasi - periodic variation of flux of active galaxies : ngc 5548 ( papadakis & lawrence 1993 ) and ngc 4051 ( papadakis & lawrence 1995 ) inferred from the exosat data , but the quality of the data is only modest , and these still need to be confirmed . more statistically significant is the quasi - periodic variability of of iras 18325 - 5926 by @xcite , but this is inferred from only a few ( @xmath5 10 ) cycles , and requires confirmation via further monitoring before drawing any detailed conclusions . nonetheless , we understand relatively little about the details of variability of active galaxies , although the recent light curves are sufficiently good to discriminate if the time series are linear or non - linear ; this is discussed later in this chapter . in any case , this rapid variability implies a compact source size . this is of course the standard causality argument : no stationary source of isotropic radiation can vary faster than the time it takes for light to cross it . in the x ray band , the power spectrum of variability generally is rather flat at long time scales , and above some characteristic frequency , it shows a power - law behavior , such that the variable power drops with decreasing time scale ( see , e.g. , mchardy 1989 ) ; the fourier phases of these light curves show no coherence ( cf . krolik , done , & madejski 1993 ) . as it was pointed by many authors , for this form of variability , the doubling time scale has no definitive meaning , but for the lack of better data , it suffices for the illustrative purposes : it is certainly valid as an order - of - magnitude relationship between the source radius @xmath6 and the time scale for doubling of the source flux @xmath7 such that @xmath8 @xmath7 . again , for quasars , this is particularly true in the x ray band , where the variability is most rapid : even the early x ray data gave us a clue that quasars are very compact . for example , an x ray light curve for the seyfert galaxy ngc 5506 ( by no means an extreme object ) as observed by the me detector onboard the exosat satellite shown on fig . 1 illustrates it well . this , and other observations of it , with redshift @xmath9 and a 2 - 10 kev x ray flux of @xmath10 erg @xmath11 s@xmath3 , imply a luminosity @xmath12 of @xmath13 erg s@xmath3 , corresponding to an eddington mass of at least @xmath14 @xmath15 . this corresponds to a schwarzschild radius @xmath16 of @xmath17 cm . the x ray data show a doubling time of @xmath18 s , corresponding to @xmath19 cm , which , for a @xmath20 @xmath15 black hole , would imply that the x ray emission arises from a region of radius @xmath21 @xmath16 . this example is by no means extreme we made a number of assumptions that are probably even too conservative : more realistic assumptions imply a mass of @xmath22 @xmath15 , and @xmath23 @xmath16 . a number of more extreme cases including very luminous objects were reported recently on the basis of the rosat data by @xcite . these are generally for the so - called `` narrow - line seyfert 1s , '' and we will return to those objects later . in brief , a doubling time scale of @xmath24 s for a source with @xmath25 erg s@xmath3 is not uncommon , implying that the bulk of the x ray emission arises around 10 - 100 @xmath16 . however , the use of the observed variability time scale does not provide an `` airtight '' argument for the size of the emitting region , since the observed emission may well be anisotropic , yielding an underestimate of the emitting volume , as is almost certainly the case for blazars . we discuss this in more detail later on . the other line of evidence for the presence of black holes in galaxies ( both active and `` normal '' ) is the velocity field of the matter emitting closely to the nucleus . this kind of work has been recently reviewed by @xcite , and it dates back to the ground - based observations made in the late 70s , when w. sargent and collaborators showed that the stellar velocity dispersion in the radio galaxy m 87 increases to 350 km s@xmath3 in the innermost 1.5@xmath26 from the nucleus . m 87 was in fact observed by the planetary camera by @xcite and the faint object spectrograph by @xcite onboard the repaired hubble space telescope , and the images showed the presence of a disk - like structure of ionized gas in the innermost few arc seconds . the spectroscopy provided a measure of the velocity of the gas at an angular distance from the nucleus of 0.25@xmath26 ( corresponding to @xmath27 pc , or @xmath28 cm ) , showing that in the reference frame of the object , it recedes from us on one side , and approaches us on the other , with a velocity difference of @xmath29 km s@xmath3 ( see fig . 2 ) . this implies a mass of the central object of @xmath30 @xmath31 , and besides a black hole , we know of no other form of mass concentration that can `` fit '' inside this region . as an aside , it is worth noting that m 87 is known to have a relativistic jet perpendicular to the disk structure mentioned above , expanding with the bulk lorentz factor @xmath32 of 4 . as such , this object is probably just a blazar , with the jet oriented at an angle @xmath33 to the line of sight , and thus is probably the closest to a hard evidence that blazars ( which we discuss in more detail below ) indeed _ do _ harbor black holes . besides m 87 , similar spectroscopy observations done with hubble space telescope revealed high stellar velocities in the central regions of a number of normal galaxies which otherwise show no evidence for an active nucleus . these were summarized recently by @xcite , and include well - publicized observations ( by l. ferrarese , h. ford , j. kormendy and others ) of ngc 6251 , ngc 4261 , ngc 4594 , ngc 3115 , but doubtless by now probably there are several new objects . such high velocities in the innermost regions of these galaxies can not be explained in any other way besides invoking the presence of massive ( @xmath34 @xmath15 ) black holes in their centers . ( interestingly , several of these galaxies also show weak radio jets ! ) even for our own milky way galaxy , the infrared data and velocity measurements readily performed from the ground ( cf . eckart & genzel 1997 ) , at much higher resolution than the hst data for external galaxies reveal a `` modest '' nuclear black hole with a mass of @xmath35 @xmath15 . the evidence is building that supermassive black holes are quite common ; recent estimates by @xcite as well as @xcite imply that they inhabit perhaps as many as half of all galaxies , and may well be the `` dead quasars '' of the past . active galactic nuclei are thus probably only the `` tip of the iceberg '' of their population . perhaps the most elegant observation showing the presence of a keplerian disk around a black hole and thus capable of measuring the mass of the hole independently of the otherwise uncertain estimates of its distance was the very large baseline interferometry observation of megamasers in the vicinity of the nucleus of seyfert 2 galaxy ngc 4258 reported by @xcite . the masing activity can only be observed along a line of sight where the velocity gradient is zero , meaning that these can be seen at locations with masers either between us and their source of energy , or locations at @xmath4 90@xmath36 to the line of sight . indeed , this is the spatial distribution of maser spots around the nucleus of ngc 4258 , illustrated in fig . 3 . specifically , these observations reveal individual masing spots revolving at distances ranging from @xmath37 pc to about twice that around the central object presumably , again , a black hole with a mass of @xmath38 @xmath15 . what is truly remarkable about these data is the near - perfect keplerian velocity distribution , in a slightly warped disk - like formation ; this implies that almost all the mass is located well within the inner radius where the megamasers reside . it is not possible to have a cluster of distinct , dark , massive objects responsible for such gravitational potential ; at least some of the objects would escape on a relatively short time scale , and form a potential well with a different shape , which would now force a departure of the megamaser emitting material from pure keplerian motion ( cf . maoz 1995 ) . since we know the central mass quite precisely , ngc 4258 has been a terrific laboratory to study the details of the accretion disk . in particular , this is a relatively low luminosity ( @xmath39 erg s@xmath3 ) object , making it quite sub - eddington , with @xmath40 . such sub - eddington sources are likely to obey unique solutions of accretion disk structure ( see , e.g. , ichimaru 1977 ; narayan & yi 1994 ; abramowicz et al . 1995 ) , where the accreting gas is optically thin and radiates inefficiently , and the accretion energy that is dissipated viscously , is advected with the accretion flow . with this , as was argued by @xcite , the accretion disk in ngc 4258 can well be advection - dominated ( but this does not _ have _ to be the case ; see , e.g. , neufeld & maloney 1995 ) . however , it is important to note that such low eddington rate can not be universal among quasars ; if most of them radiated at such low @xmath41 , the black hole masses of the most luminous sources would be much larger than expected on other grounds . however , an intriguing possibility ( cf . yi 1996 ) is that quasars in their `` youth , '' when the black hole masses were more modest , had standard , `` cold '' ( shakura - sunyaev ) accretion disks , and this is why they were so luminous in the past . the masses of the black holes grew with time , and even if the mass rate supplied for accretion remained constant , @xmath41 actually decreased , and thus the the inner accretion disks switched from `` cold '' ( bright , shakura - sunyaev ) phase to `` hot '' ( fainter , advection - dominated ) phase even if the rate of mass supply _ did not _ decrease . this is perhaps why some of the yesterday s bright quasars are today s dormant , `` dark '' black holes , revealed only via the kinematical studies mentioned above . nonetheless , ngc 4258 is probably _ not _ a unique object . we know of a large class of `` low activity '' active galaxies , known collectively as `` low ionization nuclear emission region '' objects , or liners ; they generally have low luminosity , coupled with the absence of the luminous inner disk as evidenced by the emission line ratios . however , unlike the _ bona fide _ seyferts with low luminosity , which seem to vary relatively rapidly in x rays , implying they are `` scaled down '' quasars with relatively low mass black holes liners are known _ not _ to vary rapidly in any band , suggesting that the low activity is not due to a low black hole mass , but rather due to a low accretion rate ( cf . ptak 1997 ) . as the nuclei of these objects are not very bright , the data are sparse , and thus the details of the radiative processes are poorly known ; while workable models exist ( see , e.g. , lasota et al . 1996 ) , they still require more work on the details of the transition between the `` standard '' and advection - dominated regions of the disk . perhaps the most convincing evidence that a strong gravitational field is present in active galactic nuclei comes from the recent measurements of the shape of the fe k@xmath0 fluorescence line , arising in a geometrically thin , but optically thick accretion disk . this is discussed in more detail by a. fabian ( chapter 5.3 ) , so what follows is a brief summary . the inner part of the disk is illuminated by x rays . because of the relative cosmic abundances and the fluorescence yields of various elements , the strongest discrete spectral feature predicted from the disk is the 6.4 kev fluorescent fe k@xmath0 line ; the strength ( equivalent width ) of the line of @xmath4 150 ev , as measured by @xcite , is in fact roughly consistent with predictions of @xcite . since this line arises from matter in motion , its profile is a tracer of the velocity field of the accreting matter . the asca observations of the x ray bright seyfert galaxy mcg-6 - 30 - 15 by tanaka et al . ( 1995 ) indeed showed that the line ( see fig . 4 ) has a characteristic two - pronged shape expected to arise from matter flowing in a disk - like structure . the matter approaching us is responsible for the blue wing of the line , while that receding produces the red wing , with an additional redshift , since the photons are emitted in a strong gravitational field ; the exact shape also depends on the inclination of the disk . a detailed spectral fitting of the line shape indicates that the emitted energy of the line is indeed 6.4 kev , while the bulk of its flux arises at @xmath5 10 @xmath16 , implying in turn the presence of nearly neutral material very closely to the black hole . the analysis of a number of seyfert spectra from the asca archives by @xcite suggests that many seyferts indeed show the fe k@xmath0 line profiles that require an emission close to the black hole , but the quality of data is only modest . fortunately , a number of more sensitive observatories such as axaf , astro - e , xmm , and constellation - x , will be launched in the next few years , providing ample opportunities for x ray observations of effects of strong gravity . the megamaser source ngc 4258 is only the first of three active galaxies showing a spatial distribution of masing spots from which it is possible to measure the mass of the central object . the other two are the well - known ngc 1068 ( greenhill 1998 ) , and ngc 4945 ( greenhill , moran , & herrnstein 1997 ) . both are also classified as seyfert 2s , which from the observational side means that they show narrow emission lines , implying velocities on the order of 1,000 km s@xmath3 ; these lines show no variability , and thus it is generally accepted that they originate in a relatively large regions , on the order of 100 pc or more . seyfert 1s as well as luminous quasars , on the other hand , exhibit generally very different spectra , with permitted emission lines , and these lines are usually broad , with velocities upwards of 1,500 km s@xmath3 , often reaching 30,000 km s@xmath3 . this , together with the variability of the lines that is commonly observed on time scales of weeks or months , implies that the broad line region is located much closer to the nucleus than the narrow line region . the likely relationship between the two classes of active galaxies was revealed by the seminal observation of the well - known seyfert 2 galaxy ngc 1068 by antonucci & miller ( 1985 ) . the spectropolarimetric study of the h@xmath42 line revealed that when observed in polarized light , the line is broad . they interpreted the polarization as due to electron scattering by material that is distributed preferentially along the symmetry axis of the system , and advanced the widely accepted scenario explaining the differences between the two types of seyferts as an orientation effect . this is illustrated in fig . 5 ; all seyfert galaxies are surrounded by a geometrically and optically thick torus , with an inner radius of a fraction of a parsec . such a torus can be , for instance , the outer regions of a severely warped disk , as recently suggested by m. begelman and j. pringle ( see chapter 8.1 ) . when an object is viewed along the axis of the torus , it is a seyfert 1 , revealing all the ingredients of the nucleus : the broad line region and the unobscured x ray source , both commonly varying on a short time scale . when viewed in or close to the plane of the torus , the broad line region is completely obscured , and the soft end of the x ray spectrum is absorbed due to the photoelectric absorption by the material in the torus . the opening of the torus contains a `` mirror '' of ionized gas with free electrons , and these are responsible for the scattering of the broad line light into the line of sight , and hence the broad lines are seen only in polarized light . in the context of this `` unification '' model , it is thus not surprising that all three megamasers with measured rotation curves are seyfert 2 objects : to even observe a megamaser , we have to be located in the plane of material which has large column density and presumably is roughly co - axial with the torus . likewise , the x rays traveling closely to the plane of the torus encounter a larger column density , resulting in the greater photoelectric absorption . this is in fact , observed ; even early x ray spectra of seyferts showed preferentially larger absorption in sources showing only narrow lines ( seyfert 2s ) , with the broad line region obscured by the same material that absorbs soft x rays . note that this is different than the case of only narrow emission lines present in some low - luminosity radio galaxies , such as m 87 ( collectively known as the fr - i objects ) : there , the broad lines can not be obscured , as we see no evidence of x ray absorption . furthermore , while seyfert 2s sometimes do show highly ionized , narrow permitted lines ( in addition to the commonly detected forbidden lines ) , fr - i objects _ do not _ , implying an absence of the very strong isotropic ionizing uv continuum . we will return to this point below . the use of the observed variability time scale does not , however , provide an `` airtight '' argument for the size of the emitting region , since the observed emission may well be anisotropic , yielding an underestimate of the emitting volume . as pointed out by @xcite , if the radiating matter moves at a relativistic speed towards the observer , the variability time scale appears shortened by the doppler effect . furthermore , the flux of radiation is boosted in the direction of motion of the emitting matter , and thus the total luminosity inferred under an assumption of isotropy is an overestimate . we are nearly certain that this is the case for blazars . this sub - class of quasars consists of sources that generally show strong radio emission and are highly variable and polarized in all observable wavelengths , and also includes members with very weak or absent emission lines , known as bl lacertae type objects . high angular resolution radio observations , performed since the late 60s using very long baseline interferometry implied compact radio emission regions , often associated with a rapid change of their structure . with the assumption that the redshifts are indeed cosmological , these structures appeared to expand at transverse speeds exceeding @xmath43 , and in many cases , they had shapes of jets . if the speed of the emitting matter is very relativistic , pointing closely to our line of sight , the apparent superluminal motion is just a projection effect , providing further support for the doppler - boosting scenario for blazars . in a number of aspects , blazars are the most extremely active galactic nuclei . observations of many blazars by the egret instrument onboard the compton gamma - ray observatory ( cgro ) indicated that usually they are strong emitters in the gev @xmath1ray band ; we know of at least 50 such objects . the gev emission can dominate the overall observed electromagnetic output of these sources , and this is illustrated in fig . 6 for 3c 279 , the first of the erget - discovered gev blazars . in a few cases , the @xmath1ray emission has been observed with cerenkov radiation telescopes to extend as far as the tev range . this is particularly exciting , as tev radiation can be observed from the ground , allowing a study of quite exotic phenomena occurring in quasars without the expense of space - borne platforms . interestingly , we know of no `` radio - quiet '' mev or gev @xmath1ray emitting quasars , implying an association of the @xmath1rays with the compact radio source and thus a jet . in many cases , the @xmath1ray emission from blazars is variable on a time scale of a day , and recent , simultaneous multi - wavelength monitoring observations indicate that the flux swings are reasonably well correlated between various bands , with the amplitude of variability increasing with the increase of the energy of the observing band . if the emission in the lower energy bands arises from a region co - spatial with that where the @xmath1rays originate and the tracking of light curves in various bands supports this then the opacity to the @xmath1-@xmath1 pair production would imply large optical thickness . similarly , even the early calculations indicated that if the radio sources are as luminous and as compact as the variability data imply , the cooling should be predominantly via the compton process , with the radiating electrons losing all their energy via upscattering of the just produced photons , regardless of the radiation mechanism . even the large @xmath1ray fluxes detected by egret are often orders of magnitude lower than the predictions of this self - compton process . both these difficulties , however , go away if the radio as well as the gev emission are both doppler - boosted , meaning that the true luminosity is lower , and source sizes are greater than inferred under an assumption of isotropy . in fact , we now believe that the entire broad - band continuum is produced in the relativistic jet ( cf . knigl 1981 ; ulrich et al . 1997 ) , and statistical considerations imply that the bulk lorentz factors @xmath32 of these jets are on the order of 10 ( cf . vermeulen & cohen 1994 ) . for the case of bl lac type blazars , the doppler enhancement ( going roughly as a square of @xmath32 ; see , e.g. , appendix a of sikora et al . 1997 ) ) is so strong that the continuum completely outshines the emission lines ; in some cases , however , it is possible that the emission lines are absent to begin with . in any case , the mechanism for an acceleration and collimation of these powerful jets to such high relativistic speeds is not known , but general considerations of jet energetics are strongly suggestive that this energy is ultimately tapped from an accretion onto a compact object : the best picture today involves the so - called blandford - znajek mechanism ( blandford & znajek 1977 ) , where the power of the jet derives from the spin of the black hole . the electromagnetic emission from jets in blazars recently reviewed by @xcite and ulrich et al . ( 1997 ) is interesting in its own right , and the comparison of the broad - band spectra of blazars and the `` radio - quiet '' quasars shows it clearly . the strong gev component which clearly dominates the spectra of the former ( cf . 6 ) is entirely absent in the latter , and the relative strength of the gev emission as compared to that in kev range is often different by a factor of @xmath44 . the radio and optical emission shows polarization , and spectra are always non - thermal . all these facts argue that the radio - through - uv emission is produced by the synchrotron process , with ultrarelativistic electrons ( with @xmath45 or more ) radiating in magnetic field on the order of a gauss . the compton upscattering of lower energy photons mentioned above is indeed present - and most current models suggest that it is responsible for the radiation observed above a few kev . in the lineless bl lac type objects , we have no evidence for any strong external radiation field , and the dominant `` seed '' photons for compton upscattering are likely to be produced by the synchrotron process internally to the jet . as was shown by @xcite as well as by @xcite and @xcite , in blazars with emission lines , the external , diffuse radiation dominates ; since this radiation is isotropic in the stationary frame , it appears doppler - boosted in the frame of the jet . this scenario predicts that there should be many objects with jets pointing farther away , up to the right angle to our line of sight . we now believe that for the quasar - type blazars , with strong emission lines , these are the giant , powerful radio galaxies . the difference between the quasar - type and bl lac type blazars may well be related to the presence of the inner , extremely luminous accretion disk in the former . as suggested by m. begelman , and , more recently @xcite , some radio galaxies ( as , for instance , m 87 ) have massive black holes , yet relatively modest luminosity , and essentially no broad emission lines ; this might argue for a very sub - eddington accretion rate , with a likelihood of an advection - dominated inner accretion disk . an intriguing possibility is that these low - luminosity radio galaxies are the `` misdirected '' bl - lac - type blazars : the absence of the inner , luminous disk would then be responsible for the absence of the strong ionizing radiation and thus broad emission lines . this in turn would mean that in bl lac - type blazars , there are no external photons to act as `` seeds '' to be comptonized by the relativistic electrons in the jet , and the only `` seed '' photons are internal , produced by the synchrotron process . we are nearly certain that there is a clear _ intrinsic _ distinction between the jet - dominated blazars , and the more common , radio - quiet quasars . however , the reason for the different behavior of these two subclasses is far from certain , and this is primarily because we do not have a good understanding of the formation and acceleration of relativistic jets . even though we can not use the causality arguments to infer the presence of black holes in blazars solely from their variability since we do not know for sure as to what extent the true variability time scales are shortened by doppler boosting more indirect evidence implies that ultimately , accretion onto a black hole powers blazars as well . as to the difference between the two categories , perhaps the most appealing scenario , advanced by @xcite , as well as by @xcite , is where the two classes differ by strong or weak spin of the black hole ; again , the jet is formed and powered by tapping the rotational energy of the hole . this is somewhat similar to the distinction between the galactic binary `` microquasars '' suggested by @xcite , discussed in chapter 2.1 ( by p. charles ) . nonetheless , this is an area of very active research where no clear conclusions have been reached . in particular , if the shape of the fluorescence fe k line seen in the seyfert galaxy mcg-6 - 30 - 15 ( see above ) and possibly in other seyferts ( nandra et al . 1997a ) indeed requires that the cold material powering the nucleus flows via a disk - like structure , this implies that the fluorescence occurs at at a distance @xmath46 @xmath16 . since the last stable orbit in a non - rotating ( schwarzschild ) black hole is at 3 @xmath16 and beyond this , the matter is in a free fall , this would imply that the black hole is spinning ( kerr ) , where the last stable orbit can be substantially closer to the black hole , depending on its spin . however , we note that an alternative scenario , advanced by @xcite does _ not _ require a disk at @xmath46 @xmath16 ; nonetheless , this still requires x ray emission at least at or just beyond 3 @xmath16 , merely eliminating the requirement of _ spinning _ black hole , but it still implies that at least a non - rotating black hole is present . as we mentioned above , active galaxies are generally variable , and the most rapid variability is observed in the x ray and @xmath1ray bands ; since we are interested in understanding the regions closest to the black hole , these bands deserve the most detailed study . a recent , excellent article by mushotzky , done , & pounds ( 1993 ) reviewed the x ray spectra of agn ; however , this covered the status of observations before the results from cgro became available , while we do include these results here . the most conclusive results are gleaned by a study of the non - blazar active galaxies , to avoid any potential contamination by the jet . since no radio - quiet (= jet - less ) active galaxy was detected above several hundred kev , the most relevant spectral region is the x ray and soft @xmath1ray bands . this covers nearly four decades in energy , from @xmath47 kev to @xmath48 mev , and often requires observations with multiple satellites , which must be made simultaneously , since active galaxies are variable in all bands . the early x ray observations of active galaxies by @xcite , @xcite , and @xcite implied that , to the first order , x ray spectra of active galaxies are power laws with the energy index @xmath0 ( defined such that the flux density @xmath49 ) of about 0.7 , modified by photoelectric absorption at the low energy end . a major advance came from sensitive observations with the rosat and ginga satellites ( covering respectively the bands of 0.1 to 2 , and 2 to @xmath4 30 kev ) , revealing that the spectra are more complex , with a somewhat softer underlying continuum , with @xmath50 . this continuum is modified by photoelectric absorption in the host galaxy of the quasar ; the absorbing material can be either cold , or partially ionized , and this manifests itself as isolated edges , most notably of oxygen , seen in the rosat ( and more recently , also in asca ) data . the ginga spectra , reported by @xcite , showed a strong emission line at the rest energy of @xmath51 kev , presumably due to fluorescence of the k shell of iron , and a hardening above @xmath52 kev . these last two features were interpreted as signatures of reprocessing ( `` reflection '' ) of the primary continuum in the cold material that is accreting onto the nucleus . the fe k line is indeed the strongest expected , via the combination of the relatively high abundance of iron ( decreasing with the atomic number @xmath53 ) , and fluorescence yield ( increasing with @xmath53 ) ; as pointed out by @xcite , the line equivalent width as measured by ginga is too high to be produced in absorbing material of any column . reprocessing by the matter that accretes onto the black hole is thus the most viable alternative . the hardening , predicted earlier in a seminal paper by lightman & white ( 1988 ) , arises as an additional spectral component , due to compton reflection from cold matter . this is produced roughly at one thomson depth , or when the equivalent hydrogen column density is @xmath54 @xmath11 . the cosmic abundances of elements are such that at low energies , the photoelectric absorption dominates , and unless the accreting matter is substantially ionized , this component emerges only above the last significant absorption edge , again from iron . the strengths of the reflection component and the line are in fact in agreement with theoretical predictions by @xcite . with the intensity of the incident rays greatest closely to the central source ( as inferred from the variability of the continuum ) , its kinematic and gravitational doppler shifts are powerful diagnostics of the immediate circumnuclear region . perhaps the brightest _ bona fide _ seyfert 1 on the sky is the luminous object ic 4329a , and it has been observed simultaneously with the rosat ( 0.2 - 2 kev ) and cgro osse ( 50 - 1000 kev ) ( fabian et al . 1993 ; madejski et al . this left a gap between @xmath55 and @xmath56 kev , so the observations were supplemented by non - simultaneous data obtained by ginga a few years earlier ; the ginga data were renormalized to match the rosat flux at 2 kev , so this combined data set provided a good representation of a broad band high energy spectrum of a seyfert 1 , and is illustrated in fig . the data indeed showed a modest photoelectric absorption due to neutral material ( most likely the ism of the host galaxy ) , plus an additional modest column of ionized absorber . beyond @xmath55 kev , the spectrum is the primary continuum , with @xmath50 . at @xmath4 6.4 kev , there is a strong fe k line , which in the case of ic 4329a and many other seyfert galaxies is broad , with @xmath57 of at least 200 ev . the detailed studies of this line in another object , mcg-6 - 30 - 15 @xcite , show the characteristic two - pronged shape , implies relativistic motion in an accretion disk inclined to the line of sight as discussed above . the region beyond the fe k line shows general hardening of the spectrum , accompanied by the pseudo - edge due to fe k ; this is a signature of compton reflection , as discussed above . the intensity of this component is consistent with the reflector being a semi - infinite plane ; it peaks at @xmath58 kev , and beyond this , it steepens gradually , most likely due to the compton recoil as well as the klein - nishina effects . of course the underlying power law also can steepen there we only observe a sum of the primary and reflected spectrum . the detailed fits ( spectral decomposition ) of the data imply that the primary spectrum is consistent with the power law having an exponential cutoff at an e - folding energy of @xmath59 kev . the recent simultaneous rxte and osse observations of this object generally confirm this picture . a very important constraint on any theoretical models for radiation processes in active galaxies is the _ absence _ of strong annihilation line at 511 kev . is this high energy spectrum of ic 4329a unique ? given the variability of active galaxies , simultaneous observations in x rays and soft @xmath1rays are required , but these are sparse ; an alternative approach by @xcite is to co - add many non - simultaneous observations . this in fact produced an average spectrum that is remarkably similar to the above picture , and thus any theoretical interpretation of the data for ic 4329a is probably valid for seyfert nuclei in general . it is important to note that luminous objects similar to ic 4329a are a majority of seyfert 1s , but there is one important sub - class of seyferts that shows decidedly distinct x ray spectra , differing from the above description in soft x rays , below @xmath55 kev . these are the so - called `` ultra - soft '' seyferts . first observed in the heao data by s. pravdo and collaborators , they were suggested to be a possibly distinct class of active galaxies by f. cordova on the basis of the einstein observatory imaging proportional counter data . the exosat data analyzed by @xcite as well as by @xcite showed that in a number of active galaxies , the extrapolation of the hard power law towards low energies ( taking properly into consideration the absorption from our own galaxy ) _ underpredicted _ the soft x ray flux , indicating that there is an additional component of x ray emission . this , the so - called `` soft excess , '' meant that in a given object , the ultra - soft component can co - exist with the hard power law , implying that active galaxies generally have two - component power law spectra , where either or both components are visible . it is important to note that this component appears in both low- and high - luminosity sources ; for instance , it was observed in the einstein observatory data in the quasar pg 1211 + 143 by @xcite . for the sources where both components are present , they intersect at @xmath60 kev ( but this may be an observational artifact ) ; the spectrum of the soft component is very soft ( steep ) , and can be described as a power law with an index @xmath0 @xmath55 or even steeper ( but a power law is usually not a unique model ) , while the hard component has just the `` canonical '' hard seyfert 1 spectrum with @xmath0 @xmath48 . in the very few cases that the variability of both components has been measured , it appears that the two vary independently , showing no correlation between them ( see , e.g. , the case of mkn 335 in turner 1988 ) . on the other hand , such a lack of correlated variability may be the result of a complex spectral deconvolution procedure , since only a tail of the soft excess component is observable due to intervening absorption in the ism of the host , or our own galaxy . in fact , a correlation of uv and euv fluxes was clearly observed in the `` soft excess '' seyfert ngc 5548 ( marshall et al . this implies some coupling of the soft excess component to the uv / x ray reprocessing cycle discussed in more detail in section 6 below . similar correlation has been also detected in that source between the uv flux and soft x ray residuals observed simultaneously in ginga spectra ( cf . magdziarz et al . since the euv observations suggest much higher variability amplitude than the uv data , the lack of apparent correlation in fainter sources may be related to either variation of the cut off energy of the tail of the soft excess , or confusion with spectral index variations in the hard component . `` ultra - soft '' seyferts appear to be quite common in the rosat all - sky survey , which is not surprising , since rosat is a very sensitive instrument below 2 kev . detailed follow - up studies by @xcite showed a very interesting result : the optical / uv emission lines in these objects are generally quite narrow , with the widths on the order of 1,000 - 3,000 km s@xmath3 , as compared to @xmath61 5,000 km s@xmath3 for `` normal '' seyfert 1s . these are _ not _ similar to seyfert 2s at all ; these are widths of permitted lines , while seyfert 2s only rarely show permitted lines . important clues to the nature of these soft components may be in the variability patterns , although only very few well - sampled light curves exist . the exosat data were consistent with the hard x ray variability that is aperiodic but linear , meaning that the time series can be described as uncorrelated noise ( cf . czerny & lehto 1997 ) ; an example of a light curve for this is shown in fig . 1 . the variability of soft x rays , on the other hand , is often `` episodic , '' with large flares ( up to a factor of 100 ! ) ( see fig . 8 for an example ) , and decidedly non - linear ( cf . boller et al . ( in this case a `` non - linear '' behavior manifests itself qualitatively in an episodic , flare - like behavior such as that illustrated in fig . 8 ; see , e.g. , vio et al . quantitatively , a `` non - linear '' time series is a positive , definite one , which has the ratio of its standard deviation to its mean which is larger than unity ; see , e.g. , green 1993 . ) this difference may imply different emission mechanisms in the hard and soft x ray components ; however two observational effects have to be considered before drawing any conclusions . first , the observational noise or / and presence of additional higher frequency variability in the hard component may effectively dissolve apparent signatures of non - linearity ( cf . second , if the variability related to the energy reprocessing does indeed originate from variations of the soft excess component ( e.g. , magdziarz et al . 1998 ) , then the signatures of non - linearity should be suppressed in the uv and the hard continuum . early modeling attempted to describe the `` soft excess '' as the tail end of the thermal , multi - blackbody emission from an accretion disk ; however , this seems _ not _ to be the case , at least for the bright and well - studied ngc 5548 . magdziarz et al . ( 1998 ) have shown that the uv component in that object may be associated with rather cold disk continuum , with a temperature on the order of a few ev , while the soft excess requires a separate spectral component . we will return to this below . as it was mentioned above , the popular `` unification '' picture explains the differences between the spectra of seyfert 1s vs. seyfert 2s as due to the orientation effects . to the first order , the only difference that should be seen in the x ray spectra of the two classes is the amount of photoelectric absorption , while the underlying continuum should be the same . just as in the case of seyfert 1s , this requires simultaneous observations by multiple satellites , only the problem is more acute here , as the large amount of absorption leaves generally fewer soft x ray photons to allow for a sensitive measurement of the continuum . nonetheless , the observations with the ginga satellite by @xcite revealed that x ray spectra of seyfert 2s are in fact equivalent to spectra of seyfert 1s , absorbed by various column densities of cold gas . more detailed studies by @xcite implied that the continua of seyfert 2s may be somewhat harder , but only marginally so ; a more conclusive results should be obtained from observations by the rossi x ray timing explorer , which features a broader bandpass , extending to 50 kev ( or , for brighter sources , even to 100 kev ) . in any case , the inferred column densities in seyfert 2s are @xmath62 @xmath11 or more . in a few cases as , for instance , the well - studied ngc 1068 we only know that no primary x ray continuum is seen , so the absorber must be quite thomson - thick , with the absorbing column greater than @xmath63 @xmath11 . a good , illustrative example of a nearly - extreme seyfert 2 but , with the primary continuum still barely penetrating the absorber is ngc 4945 , a seyfert 2 which also shows megamaser emission , and thus is most likely observed in the plane of the putative torus . in the asca and ginga ranges ( below @xmath64 kev ) , the source is relatively faint , but above @xmath64 kev , the spectrum rises sharply ( @xcite ) . osse observations of it by @xcite revealed that at 50 - 100 kev , this is the second brightest radio - quiet active galaxy in the sky ( see fig . the absorbing column is large , @xmath65 @xmath11 , and while the observations were not simultaneous , a comparison of fig . 7 and fig . 9 reveals that the underlying continuum , to the first order , is consistent with that of a seyfert 1 . another interesting aspect of this source is the fact that the megamaser distribution implies a mass of the black hole of @xmath66 m@xmath67 ( greenhill et al . 1997 ) . with the bolometric luminosity of the nucleus of at least @xmath39 erg s@xmath3 ( iwasawa et al . 1993 ; done et al . 1996 ) , this source radiates at a few percent of eddington luminosity , and thus is unlikely to be advection - dominated , as may be the case for ngc 4258 . so far , we discussed primarily the relatively low - luminosity quasars , and an obvious question to be asked is : does this general picture hold for the higher luminosity counterparts ? in general , the answer is yes , but with some modifications . it is important here to compare `` apples to apples , '' and in the case of quasars , this means selecting radio - quiet objects , as radio - loud quasars tend to have higher x ray to optical flux ratios , which may be due to a contamination by a possible jet ; unfortunately , this paucity of x ray photons in the radio - quiet objects makes spectral studies somewhat more difficult . in general , the more luminous objects show a lower ratio of x ray to optical luminosities ( see , e.g. , kriss & canizares 1985 ; avni & tananbaum 1986 ) . the recent work by laor et al . ( 1997 ) ( using the rosat pspc data ) and @xcite ( using asca data ) indicates that the continuum x ray spectra of higher luminosity objects appear generally similar to those of the lower luminosity counterparts . notable exception is an absence of the fe k line and the compton reflection component in quasars ( see , e.g. , nandra et al . 1997b ) . with the more luminous central source , this may be the result of a nearly - complete ionization of the accreting material , such that the reflection component is present , but can not be distinguished by its tell - tale fe k line and spectral hardening above @xmath52 kev : for an entirely ionized reflector , the incident or emerging photons encounter no photoelectric absorption , and the reflection is ( below @xmath58 kev ) identical to the incident spectrum , with no fe k line present . unfortunately , those quasars are generally too faint to be studied in detail above @xmath64 kev , where the only instruments currently available are non - imaging proportional counters such as the rxte , dominated by uncertainties of the instrumental as well as the cosmic x ray background . as a result , any detailed studies must await x ray reflective optics sensitive beyond 10 kev , already under development ; such telescopes are essential for studies of these faint objects , as they permit subtraction of background from the same image as the source , and thus will yield the best quality data for luminous quasars . the availability of good quality high energy spectra permits us to constrain the possible emission mechanisms that can operate in active galaxies . we briefly discussed the case of blazars above ; observationally , the continuum high energy spectra of radio - quiet objects are also decidedly non - thermal , but these mechanisms are probably somewhat different than the synchrotron + compton model discussed for the blazar jets . the early work by @xcite , using the diffusion approximation , suggested that a power - law spectrum can be produced by a repeated compton - upscattering of soft photons by a compton - thick bath of hot electrons . a more general variant of this model is essentially what is used today to explain the primary , high energy spectrum in these objects as well as in the galactic black hole candidates . in summary , a successful model has to explain a power law spectrum with an energy index @xmath0 of @xmath4 0.9 1 , exponentially cutting off at @xmath68 kev : other spectral features are merely signatures of reprocessing . there are essentially two flavors of the comptonization model that can be applied to the isotropic emission in quasars : the original thermal comptonization version , and a non - thermal variant . the difference between the two is related to the distribution of electron energies . the non - thermal version , developed by a. zdziarski , a. lightman , p. coppi , as well as by c. done , r. svensson , g. ghisellini , and a. fabian ( zdziarski et al . 1990 ; zdziarski & coppi 1991 ) involves acceleration of particles to relativistic energies , and a subsequent pair cascade ; these particles comptonize uv photons , believed to be produced in abundance by the accretion disk . the attractive feature of this model is that the pairs thermalize to relatively low temperatures ( @xmath4 a few kev ) , providing the medium which , again , upscatters the uv photons to form the `` soft excess '' discussed above . one feature , however , predicted by this version of the model , is the annihilation line that should be present at @xmath69 kev . no spectrum of any active galaxy collected so far with the cgro osse detector showed such a feature ( johnson et al . 1997 ) , and thus the model is somewhat out of favor , despite its natural ability to provide the `` soft excess . '' refinements to the thermal comptonization model , primarily by @xcite , allowed the quasi - analytical calculation in the optically thin - to intermediate regime . this is the most viable current model for radio - quiet active galaxies , and is discussed in more detail in chapter 5.3 by j. poutanen . roughly , for regime relevant to the hard x ray spectra of seyferts , the energy of the exponential cutoff @xmath70 determines the temperature of the comptonizing plasma , such that @xmath71 @xmath72 @xmath73 , while the index of the power law , together with the cutoff , determine its optical depth @xmath74 , such that @xmath74 @xmath72 @xmath75 / @xmath76 ( pietrini & krolik 1995 ; poutanen , krolik , & ryde 1997 ) . for seyferts , the comptonizing plasma has to have a temperature of @xmath44 kev , and optical depth @xmath77 1 ; the soft `` seed '' photons are available in abundance from the inner accretion disk , as evidenced by the presence of the reflection component ( zdziarski et al . 1997 ) . however , a simple `` sandwich '' type structure ( with a cold disk covered by a uniform corona ) can not work , as this would produce too many `` seed '' photons for comptonization . instead , an example of a good phenomenological model is a `` patchy corona '' above a surface of the disk , proposed by haardt , maraschi , & ghisellini ( 1994 ) ; however , none of these models address the processes responsible for the particle acceleration . the issue of the `` soft excess '' remains unresolved in the context of the above models . spectral fitting to the data for the well - studied ngc 5548 imply that the soft excess can be produced by a relatively cold ( @xmath78 ev ) but optically thick ( @xmath79 ) plasma , while the hard continuum requires @xmath80 kev , but @xmath81 ( magdziarz et al . 1998 ) . coexistence of such two phases may be related to the disk structure and dynamics of possible multi - phase transition regions ( see , e.g. , magdziarz & blaes 1998 ) . in fact , such multi - phase medium appears naturally in local solutions of the disk corona transition layer ( see , e. g. , @xcite ) . however , without high quality observational data on the spectral shape and temporal correlation between both components , so far , we lack clear clues as to the time evolution of the plasma energetics . the author s prejudice ( based partially on the `` episodic '' nature of the soft light curves ) is that we probably witness some form of a limit cycle operating in the inner disk , and thus the best avenue for this is a development of a more detailed theory for the structure of the inner accretion region , and in particular , the issue of stability of the transition region between the _ bona fide _ disk and the matter free - falling onto the black hole . however , any tests of theories require sensitive observations : especially needed are well - sampled light curves obtained simultaneously over a broad energy range , from the softest energies accessible ( @xmath82 kev ) up to the end of the observable spectrum , in the mev range . the prospects for such observations are very good : with the impending launch of axaf , xmm , astro - e , integral , and , eventually , constellation - x , we should have the data for the more definitive modeling . johnson , w. n. , zdziarski , a. a. , madejski , g. m. , paciesas , w. s. , steinle , h. , & lin , y .- c . 1997 , in _ proceedings of the fourth compton symposium _ , eds . c. d. dermer , m. s. strickman , & j. d. kurfess ( aip : new york ) , aip conference proceedings 410 , p. 283
near - infrared surveys prompted the discovery of a population of objects with very red optical - infrared colors ( extremely red objects , eros hereafter ; e.g. elston et al . 1988 , mccarthy et al . 1992 , hu & ridgway 1994 ; thompson et al . 1999 ; yan et al . 2000 ; scodeggio & silva 2000 ) . in general , objects have been classified as eros when they had redder colors than late type galaxies ( with negligible dust extinction ) at any redshift . however , depending on the depth of the photometry and on the available filters , different selection criteria have been used to select eros . in this paper , eros are defined as objects with @xmath55 ( see sect . 5 for more details on this choice ) . the red colors of eros are consistent with two classes of galaxies : they could be old , passively evolving elliptical galaxies at @xmath10 which are so red because of the large @xmath7correction . eros may also be strongly dust reddened star forming galaxies or agn . the observational results of the last few years showed that both classes of galaxies are indeed present in the ero population : on one hand , a few objects were spectroscopically confirmed to be @xmath11 ellipticals ( dunlop et al . 1996 , spinrad et al . 1997 , liu et al . 2000 , and marginally , soifer et al . 1999 ) , or to have surface brightness profiles consistent with being dynamically relaxed early type galaxies ( e.g. stiavelli et al . 1999 , benitez et al . 1999 ) . on the other hand , other eros have been detected in the sub mm ( cimatti et al . 1998 , dey et al . 1999 , smail et al . 1999 , andreani et . al . 2000 ) , thus providing examples of high redshift starburst galaxies reddened by strong dust extinction and characterized by high star formation rates . the relative contribution of the two classes of objects to the whole ero population is still unknown , but there are preliminary indications , based on near - ir and optical spectroscopy and on surface brightness analysis , that ellipticals may represent the largest fraction of this population ( e.g. cimatti et al . 1999 , 2000 ; liu et al . 2000 ; moriondo et al . 2000 ) . a small fraction of low - mass - stars and brown dwarfs among eros is also expected in case of unresolved objects ( e.g. thompson et al . 2000 , cuby et al . 1999 ) . the importance of studying eros is clear especially for the clues that they could provide on the formation and evolution of elliptical galaxies . for instance , existing realizations of hierarchical models of galaxy formation predict a significant decline in the comoving density of the ellipticals with @xmath9 , as they should form through merging at @xmath12 ( kauffmann 1996 , baugh et al . 1996a ) , so that a measure of a decline of their comoving density would provide a stringent proof of these models . conflicting results have been found so far about such issue : some works claim the detection of a deficit of @xmath11 ellipticals ( e.g. kauffmann et al . 1996 , zepf 1997 , franceschini et al . 1998 , barger et al . 1999 , menanteau et al . 1999 ) , whereas others find that a constant comoving density of ellipticals up to @xmath13 is consistent with the data ( totani & yoshii 1997 , benitez et al . 1999 , broadhurst & bowens 2000 , schade et al . 1999 ) . a potentially serious problem in these studies is suspected to be the influence of the field - to - field variations in the density of eros due to the small fields of view usually covered in the near - infrared . for instance , barger et al . ( 1999 ) found a very low surface density of eros in a 60 arcmin@xmath2 survey to @xmath14 , while on a similar area mccracken et al . ( 1999 ) observed a density three times larger at the same @xmath7 level . similar uncertainties have been found in the attempts of deriving the fraction of high redshift galaxies among ir selected samples , which is believed to be a stringent test for the formation of the massive galaxies ( broadhurst et al . 1992 , kauffmann & charlot 1998 ) . fontana et al . ( 1999 ) , using photometric redshifts , found that the fraction of high@xmath9 galaxies in a collection of small deep fields complete to @xmath15 was low and comparable with the predictions of the cold dark matter ( cdm ) models , but not with passive evolution ( ple ) models . the preliminary results of eisenhardt et al . ( 2000 ) , suggesting a much higher fraction of high@xmath9 galaxies , are instead consistent with both cdm and ple models . the main aim of our survey was to encompass the difficulties induced by the cosmic variance , obtaining a sample of eros on a large area , at moderately deep @xmath7 levels , in order to minimize and possibly to detect the effects of their clustering , and to compare the observed density with that expected in the case of passive evolution of ellipticals . our survey is larger by more than a factor of four than the thompson et al . ( 1999 ) survey , and by more than an order of magnitude than all the other previous surveys for eros , at the same limiting magnitudes . with the large area covered we aimed also to detect a sample , or place limits to the surface density of the very rare class of extreme eros with @xmath16 . in this paper , the observational results of this survey and their main implications are presented . a more detailed interpretation of our findings will be presented in a forthcoming paper ( daddi et al . , in preparation ) . the paper is organized as follows : we first describe the data reduction and analysis , then we present the counts of field galaxies . in sect . [ sec : eros ] the sample of eros is described . [ sec : clustering ] contains the analysis of the clustering of field galaxies and eros . the main implications of our findings are discussed in sect . [ sec : implic ] . km s@xmath18 mpc@xmath18 throughout the paper . the @xmath0 observations were made with the eso ntt 3.5 m telescope in la silla , during the nights of 2730 march 1999 , using the sofi camera ( moorwood et al . 1998 ) with a field of view of about 5@xmath195@xmath20 . sofi is equipped with a hawaii hgcdte 1024x1024 array , with a scale of 0.29@xmath21/pixel . the @xmath0 filter has @xmath22 and @xmath23 and it is slightly bluer than the standard @xmath7 filter in order to reduce the thermal background . the center of the observed field is at @xmath24 = 14@xmath2549@xmath2629@xmath27 and @xmath28 = 09@xmath29 ( j2000 ) . the observed field is one of the fields described in yee et al . ( 2000 ) to which we refer for details about its selection . the main criteria were not to have any apparent nearby clusters and to be at high galactic latitude . the images were taken with a pattern of fixed offsets of 144@xmath21 ( about half of the sofi field of view ) over a grid of 9@xmath3013 pointings . the total area covered by the observations was about 24@xmath3034 arcmin , with a local integration time of 12 minutes in the central deepest region of the field . in the shallower region , the effective integration time is reduced to not less than 6 minutes . the total amount of time required to cover the whole field was about 5.5 hours . the data reduction was carried out using the iraf software . the images were flat - fielded with twilight flats . the sky background was estimated and subtracted for each frame using a clipped average of 68 adjacent frames ( excluding the central frame itself ) . the photometric calibration was achieved each night with the observation of 5 - 7 standard stars taken from persson et al . the zero - points have a scatter of @xmath31 0.015 magnitudes in each night and a night - to - night variation within 0.02 magnitudes . each frame was scaled to the same photometric level correcting for the different zero - points and airmasses . accurate spatial offsets were measured for each frame using the area in common with the adjacent frames . the images were then combined , masking the known bad pixels , in order to obtain the final mosaic . the cosmic rays were detected and replaced by the local median using the task _ cosmicrays _ of the iraf package _ ccdred_. the effective seeing of the final coadded mosaic ranges from 0.9@xmath21 to 1.1@xmath21 . the @xmath1-band data were taken in may 1921 1998 with the 4.2 m william herschel telescope on la palma . the observations were done using the prime focus camera , equipped with a thinned 2048x4096 pixels eev10 chip , with a scale of 0.237@xmath21/pixel . this gives a field of view of about 8.1@xmath32 . a standard johnson @xmath1filter was used . the whole field has been covered by a mosaic of 6 pointings . each pointing consisted of at least 3 exposures of 1200s taken with small offsets . the total integration times per pointing was therefore 3600s , with the exceptions of two pointings with 4800s and 6000s . the photometric calibration was achieved with standard stars taken from landolt ( 1992 ) with a scatter in the zeropoints , from the different stars used , below 0.01 magnitudes . the images were de - biased and then flatfielded , using a master flatfield constructed from the science exposures , scaled to the same zeropoint and then combined . the seeing of the final @xmath1-band mosaic was between 0.7@xmath21 and 0.8@xmath21 . the software sextractor ( bertin & arnouts 1996 ) was run on the @xmath0 mosaic with a background weighted threshold in order to take into account the depth variations across the area , as defined by sextractor . among the detected objects , all those with s / n@xmath33 in a 2@xmath34 circular aperture ( twice the average seeing fwhm ) were selected and added to the catalog . a few spurious detections ( e.g. close to image defects ) have been excluded after a visual inspection of the image . the final catalog includes 4585 objects . in the central deepest region , the 5@xmath35 limiting aperture magnitude is @xmath0(2@xmath34)=19.6 , whereas in the remaining area such limit is @xmath0(2@xmath34)19.2 because of the reduced integration time . isophotal magnitudes were measured with a limiting threshold of about 0.7@xmath36 corresponding in the central area to a surface brightness limit of about @xmath37 mag arcsec@xmath38 . the aperture correction from 2@xmath34 to total magnitudes was estimated throughout the area by measuring the difference between the isophotal and the 2@xmath21 aperture magnitudes for the stars with @xmath3916 . a differential correction , in the range of 0.160.30 magnitudes , was measured for different regions of the mosaic with a typical scatter less than 0.03 magnitudes . for the bright objects the isophotal magnitudes were on average consistent with the kron automatic aperture magnitudes . however , we adopted the isophotal magnitudes because the kron magnitudes are rather unstable at faint flux levels , where the low signal often does not allow to define the correct automatic aperture . the total @xmath0 magnitudes were then defined as the brightest between the isophotal and the corrected aperture magnitude . this allowed to safely assign a total magnitude for both the faint and the bright objects . the typical @xmath0 magnitude where the corrected aperture magnitude begins to be adopted as the total one is @xmath0@xmath3118 in the central deepest region . the completeness of our catalog has been estimated by adding artificial objects to the @xmath0 mosaic in empty positions , using the iraf package _ artdata_. point - like sources as well as objects with de vaucouleurs and exponential profiles ( convolved with the seeing psf ) were simulated , and sextractor was run with the same detection parameters as for the real data . the 85% completeness magnitude for the deepest area is @xmath40 for point - like sources . the completeness decreases to @xmath31 70% for the worst case that we have tested , i.e. for exponential galaxies with 0.7@xmath21 half light radius . in the shallower area the corresponding limiting magnitude is @xmath3 . most of the @xmath41 galaxies are anyway expected to be only barely resolved with the @xmath0 1@xmath21 seeing ( saracco et al . 1997 ) , and this certainly occurs for the distant @xmath10 ellipticals , and thus their completeness limits can be assumed to be similar to those for stars . in order to recover the @xmath1-band counterparts of the @xmath0-selected objects , a coordinate mapping between the @xmath0 and the @xmath1 images was derived . sextractor was then run in assoc mode with a search box of 2@xmath30 fwhm@xmath42 . the regions around bright stars or defects in the @xmath1 and @xmath0 band images were excluded from this analysis . the final effective area is 701 arcmin@xmath2 at @xmath3 and 447.5 arcmin@xmath2 at @xmath43 . whenever an object had s / n@xmath443 in the @xmath1-band image 3@xmath35 limits were assigned . the 3@xmath35 limiting magnitude in a 2@xmath21 diameter aperture is @xmath45 for most of the area , reaching @xmath46 in the deepest pointing . when s / n@xmath473 , 2@xmath21 diameter corrected aperture magnitudes were assigned to each object . the aperture correction was derived in the same way as for the @xmath0-band , with slightly smaller corrections because of the better @xmath1-band seeing . the magnitudes were dereddened for galactic extinction . at the galactic coordinates of the center of our field ( @xmath48 , @xmath49 ) , the extinction coefficients from burstein & heiles ( 1982 ) and from schlegel et al . ( 1998 ) are @xmath50 and @xmath51 respectively . since the two values are derived in different ways , neither of the two can be discarded . the average was therefore adopted , obtaining a correction of 0.052 magnitudes in @xmath1 and negligible in @xmath0 . this introduces an uncertainty of @xmath31 0.03 magnitudes in the dereddened @xmath1 magnitudes . finally , the @xmath8 colors are defined for all the objects as the difference between the @xmath1 and @xmath0 corrected aperture magnitudes . thanks to the depth of the @xmath1-band data , colors as red as @xmath52 could be measured down to the @xmath0 magnitude limits of our survey . because of the long integrations used in the @xmath1-band the objects with @xmath53 were saturated . this effect is obviously more important for stars than for galaxies ; moreover , since we are interested in the extremely red galaxy population , the effects due to saturation have no impact on our results . the star galaxy ( s / g ) separation was done by means of the sextractor class_star parameter in both the @xmath1 and @xmath0 band . this classification was found to be reliable for objects with @xmath5417.5 and @xmath5524 . because of the seeing variations through the area in both bands , a variable class_star threshold was adopted in different subareas . given the better seeing of the @xmath1-band , the classification was based mostly on that band , switching to the @xmath0 class_star for objects close to the saturation level in @xmath1 . objects which are fainter than @xmath0 = 17.5 and @xmath1 = 24 have not been classified and have been considered to be galaxies . this means that a s / g separation can be provided only for objects with colors @xmath565 at @xmath0=19 , @xmath566 at @xmath0=18 , and so on ( see fig . [ fig : zoom ] and the upper panel of fig . [ fig : ccdiag ] ) . from the small number of objects classified as stars which are close to the diagonal straight line which indicates our s / g classification limit in the color - magnitude plane , we can safely conclude that our inability to properly classify very red , faint objects has almost no effect on the total star number counts and , therefore , on the galaxy number counts . in addition , it is possible that near the faint limit of our survey a very small fraction of very compact , blue galaxies , such as for instance agn or compact narrow emission line galaxies ( e.g. koo & kron 1988 ) , could have been incorrectly classified as stars . galaxy number counts in the @xmath0 band can provide more advantages in studying galaxy evolution and cosmological geometry than optical counts because they are much less sensitive to the evolution of stellar population and to the dust extinction . our survey , which covers the magnitude range @xmath57 , represents the widest among the previous deep surveys at levels fainter than @xmath58 . [ tab : xx ] cccc & & & + 11.7 12.2 & 701 & - & 4 + 12.2 12.7 & `` & - & 7 + 12.7 13.2 & '' & - & 9 + 13.2 13.7 & `` & - & 17 + 13.7 14.2 & '' & 4 & 21 + 14.2 14.7 & `` & 6 & 38 + 14.7 15.2 & '' & 16 & 37 + 15.2 15.7 & `` & 30 & 62 + 15.7 16.2 & '' & 74 & 73 + 16.2 16.7 & `` & 100 & 88 + 16.7 17.2 & '' & 178 & 128 + 17.2 17.7 & `` & 372 & 127 + 17.7 18.2 & '' & 633 & 144 + 18.2 18.7 & `` & 892 & 185 + & & & + 18.7 18.8 & '' & 200 & 32 + 18.8 19.2@xmath59}$ ] & 447.5 & 628 & 84 + table [ tab : xx ] summarizes the number of galaxies and stars detected in each @xmath0 bin and fig . [ fig : stars1 ] shows the corresponding differential counts in 0.5 magnitude bins . no correction for incompleteness was applied . galaxies start to dominate over stars at @xmath60 and their surface density is about a factor of 8 higher than the stellar surface density at @xmath61 the slopes of the galaxy number counts were derived over the magnitude range covered by our survey . at bright magnitudes a slope of @xmath62 is found in the range @xmath63 . we confirm that the @xmath7-band galaxy counts show a flattening at @xmath64 , where the best fit slope changes from @xmath65 to @xmath66 ( see fig . [ fig : counts_gala ] ) . the leveling off of the counts below a slope of 0.4 indicates that the differential contribution to the extragalactic background light ( ebl ) in the @xmath7 band peaks at @xmath67 and then starts to decrease at fainter fluxes . the contribution to the ebl over the magnitude range @xmath68 sampled by our survey is about @xmath69 nw / m@xmath2/sr , which constitutes about @xmath70 of the estimated ebl from discrete sources in the @xmath7-band ( cf . pozzetti et al . 1998 , madau & pozzetti 2000 ) . [ fig : counts_gala ] shows the differential galaxy number counts in our survey compared with a compilation of @xmath7-band published surveys . no attempt was made to correct for different filters . as shown in the figure , our counts are in very good agreement with the average counts of previous surveys ( hall & green 1998 ) . [ tab : median ] ccc & & + 16.5 17.0 & 143 & 3.54 + 17.0 17.5 & 277 & 3.80 + 17.5 18.0 & 522 & 3.92 + 18.0 18.5 & 759 & 4.08 + 18.5 18.8 & 613 & 4.11 + 18.8 19.2 & 628 & 4.04 + [ tab : eros ] c|c|ccc|ccc|ccc|ccc & & & & & + & & & & & & & & & & & & & + @xmath7117.0 & 701 & 2 & 0.006 & 0.003 & & & & & & & & & + @xmath7117.5 & 701 & 15 & 0.025 & 0.02 & 5 & 0.008 & 0.007 & & & & & & + @xmath7118.0 & 701 & 58 & 0.051 & 0.08 & 19 & 0.017 & 0.027 & & & & & & + @xmath7118.5 & 701 & 158 & 0.084 & 0.23 & 75 & 0.040 & 0.11 & 7 & 0.004 & 0.01 & & & + @xmath7118.8 & 701 & 279 & 0.111 & 0.40 & 150 & 0.060 & 0.21 & 30 & 0.012 & 0.04 & 2 & 0.0008 & 0.003 + @xmath7119.0 & 447.5 & 220 & 0.116 & 0.49 & 133 & 0.070 & 0.30 & 33 & 0.017 & 0.07 & 4 & 0.0021 & 0.009 + @xmath7119.2 & 447.5 & 281 & 0.126 & 0.63 & 173 & 0.079 & 0.39 & 44 & 0.020 & 0.10 & 5 & 0.0023 & 0.011 + @xmath7119.2 & 701 + 447.5 & 279 + 119 & 0.127 & 0.67 & 150 + 87 & 0.076 & 0.40 & 30 + 27 & 0.018 & 0.10 & 2 + 4 & 0.0019 & 0.012 + in fig . [ fig : ccdiag ] the @xmath8 vs. @xmath0 color - magnitude diagram is plotted for both stars and galaxies in our sample . the diagonal straight line in the upper panel indicates our s / g classification limit in this plane . because objects above this line are not classified , no star can appear in the upper right corner . however , this figure shows that there are very few stars with a color redder than @xmath72 , even in the region of this color - magnitude diagram where our morphological classification is still reliable . this suggests that very few stars should be present in the sample of objects for which no morphological classification was possible . [ fig : histoc ] shows the distribution of colors for galaxies fainter than @xmath73 , from which the median color at different @xmath0 magnitude levels have been calculated ( see table [ tab : median ] ) . in spite of the presence of objects with color upper and lower limits , the use of the median colors ( instead of the mean colors ) allows unbiased estimates in the @xmath0 range we are considering . the faintest bin with 18.8@xmath7419.2 includes only the objects detected in the deeper region . the median @xmath8 color of the galaxies increases by 0.5 magnitudes from @xmath73 to @xmath75 and then it remains almost constant up to the limits of our survey ( @xmath0=19.2 ) . this trend is similar to what is found by saracco et al . ( 1999 ) for the median @xmath76 galaxies color , which also reaches a maximum at @xmath77 1819 and then it becomes bluer , while the median @xmath78 color gets significantly bluer at brighter ( @xmath79 ) magnitudes ( gardner et al . 1993 ) . our wide - field survey allows us to select a statistically significant and complete sample of eros which can provide stringent constraints on the density of high-@xmath9 ellipticals ( see sect . 1 ) . for this reason , eros are defined as objects with @xmath80 because this corresponds to select passively evolving ellipticals at @xmath81 ( see fig . [ fig : color_fig ] ) . table [ tab : eros ] shows the results of the selection for different magnitude and color thresholds . among the others , the threshold @xmath82 is used because it corresponds to the selection of @xmath83 elliptical galaxies ( see fig . [ fig : color_fig ] ) . we detected 279 eros with @xmath3 from the whole area and 119 eros with @xmath43 in the deeper area , yielding a total sample of 398 objects ( see fig . [ fig : zoom ] and table [ tab : eros ] ) . this is by far the largest sample of eros obtained to date . a small but complete sample of eros with @xmath16 has also been selected , and we estimate for the first time their surface density to be @xmath84 arcmin@xmath38 at @xmath85 . a comparison of the surface densities of the eros in our sample with those in the thompson et al . ( 1999 ) survey can be directly done after taking into account the different filters used in the two surveys . for the @xmath86 filter used by thompson et al . ( 1999 ) , @xmath0 @xmath31 @xmath86 - 0.2 ( adopting @xmath87 ) , and therefore their limit at @xmath88 corresponds to @xmath89 which is the shallower limit of our survey , while their @xmath90 color is bluer than our @xmath8 by about 0.1 magnitudes for the redder objects ( thompson , private communication ) . at the level of @xmath3 we find a density of @xmath91 arcmin@xmath38 for eros with @xmath92 , to be compared with the value of 0.039@xmath930.016 that they find ( these errors are poissonian ) . thus , the two surface densities are in excellent agreement with each other . we also verified that the average @xmath8 color of all our objects with @xmath94 ( @xmath95 , determined with a kaplan - meier estimator ) , is in good agreement with the average @xmath96 in @xmath97 by thompson et al . statistical measurements of the clustering of faint galaxies are important for studying the evolution of galaxies and the formation of structures in the universe . in fact , the amplitude of clustering in 2d space is a useful probe of the underlying 3d structure ( e.g. connolly et al . @xcite , efstathiou et al . 1991 , magliocchetti & maddox 1999 ) . the clustering of galaxies on the sky has been studied extensively especially in the optical , but also in the near - infrared ( e.g. roche et al . 1998 and 1999 , postman et al . 1998 , baugh et al . our survey , as noted before , is the widest at the limits of @xmath85 . it is therefore interesting to estimate the clustering of our sample of galaxies . the angular two - point correlation function @xmath98 is defined as the excess probability ( over a poissonian distribution ) of finding galaxies separated by the apparent distance @xmath99 : @xmath100d\omega_{\rm 1}d\omega_{\rm 2 } \label{eq : dp}\ ] ] where n is the mean density per steradian ( groth & peebles , 1977 ) . several methods for estimating @xmath98 from a set of object positions have been proposed and used , but the most bias free and suitable for faint galaxies samples resulted to be the landy & szalay technique ( landy & szalay 1993 , see also kerscher et al this technique ( adopted for the calculations in this paper ) consists in deriving the counts of objects binned in logarithmic distance intervals , for the data - data sample @xmath101 $ ] , the data - random sample @xmath102 $ ] and the random random sample @xmath103 $ ] . these counts have to be normalized , i.e. divided for the total number of couples in each of the 3 samples . from them we can estimate @xmath104 as : @xmath105 - 2[dr]+[rr]}{[rr ] } \label{eq : w_b}\ ] ] which is biased to lower values with respect to the real correlation function @xmath98 : @xmath106 where @xmath107 is the `` integral constraint '' ( groth & peebles , 1977 ) : @xmath108 assuming that the angular correlation function @xmath98 can be described by a power law of the form @xmath109 , then , following roche et al . ( 1999 ) , we can extimate the ratio between @xmath107 and the amplitude @xmath110 using the random random sample : @xmath111 the amplitude of the real two - point correlation function @xmath98 can then be estimated by fitting to the measured @xmath104 the function : @xmath112 the errors can be estimated , following baugh et al . ( 1996b ) , as : @xmath113{(1+w_{\rm b}(\theta))/dd ) } \label{eq : error}\ ] ] where @xmath114 is the non normalized histogram of @xmath101 $ ] . ( [ eq : error ] ) is equivalent to assuming 2@xmath35 poissonian errors for the correlations , and it gives estimates that are comparable to the errors obtained with the bootstrap technique ( ling , frenk & barrow 1986 ) . this is necessary because it is known that , as the counts in the different bins are not completely independent , assuming the 1@xmath35 poissonian errors would result in an underestimate of the true variance of the global parameters of the angular correlation ( see mo et al . 1992 ) . in case of the presence of a randomly distributed spurious component among the analyzed sample of objects ( an example of this case would be a residual stellar component among the galaxy sample ) , the resulting amplitudes are apparently reduced by a factor @xmath115 , where @xmath116 is the fraction of the randomly distributed component ( see e.g. roche et al . 1999 ) , and the corresponding correction should be applied . [ tab : clust_all ] cccccc & & & & & + 18.0 & 701 & 1131 & 1.3@xmath930.5 & 0.8 & 4.55 + 18.5 & 701 & 1890 & 1.6@xmath930.3 & 0.8 & 4.55 + 18.8 & 701 & 2503 & 1.5@xmath930.2 & 0.8 & 4.55 + 19.2 & 447.5 & 2222 & 1.6@xmath930.2 & 0.8 & 5.16 + the random samples used in our analysis were obtained using the pseudo random number generator routine of the c library function _ drand48_. random samples with up to 200000 objects were used . typically the number of objects in the random samples were a factor of 100200 larger than the number of observed objects . the random sample was generated with the same geometrical constraints as the data sample , avoiding for instance to place objects in the regions excluded around the brightest stars . in our analysis a fixed slope of @xmath117 0.8 was assumed , as this is consistent with the typical slopes measured in both faint and bright surveys ( e.g. baugh et al . 1996b , roche et al . 1996 , maddox et al . 1990 ) , and because it gives us the possibility to directly compare our results with the published ones that are typically obtained adopting such a slope . the factor @xmath118 was estimated ( with eq . ( [ eq : c ] ) ) for both the whole and the deeper areas , turning out to be 4.55 and 5.16 respectively ( the angles are expressed in degrees , if not differently stated ) . in fig . [ fig : all_clust ] the observed , bias corrected , two - point correlation functions @xmath98 are shown ; the bins have a constant logarithmic width ( @xmath119 ) , with the bin centers ranging from 3.6@xmath21 to 15@xmath20 . we clearly detect a positive correlation signal for our sample with an angular dependence broadly consistent with the adopted slope @xmath117 0.8 , even if the measurements show some deviations , in particular for the brightest samples . a few cluster candidates are present in our survey . these possible clusters include galaxies with @xmath120 , and are therefore expected to be at @xmath121 . a detailed analysis of the cluster candidates will be given in a forthcoming paper . for the purpose of the present work , we tested that the measured clustering amplitudes are stable in case of removal of the galaxies of the most evident cluster from the sample . however , the presence of such clusters , most of which happen to be in the shallower area , could partly explain the observed deviations from the fitted @xmath122 power laws for the three brightest samples . the derived clustering amplitudes are presented in table [ tab : clust_all ] . the amplitude errors are obtained from the fit assuming eq . ( [ eq : error ] ) . no correction for the stellar contamination was applied . in fig . [ fig : roche ] the clustering amplitudes of our samples are compared with other published measurements . the data are shown together with a number of ple models with different clustering evolutions , which are described in detail in roche et al . ( 1998 , 1999 ) . our measurements are in good agreement with both the models and the previous estimates of roche et al . ( 1999 ) , except for the point with limiting magnitude @xmath123 , which is however the most uncertain of our data points . as a check it was verified that the correlations of the stellar sample are consistent with zero , within the measured errors , at all the scales . this is a confirmation that the stars are homogeneously distributed on the field ( as they should be ) and the seeing variations across the area did not cause a detectable bias in our classification . [ tab : clust_eros ] cccccccc & & & & & + & & & & & & & + 18.0 & 701 & 58 & 24@xmath9310 & & & 0.8 & 4.55 + 18.25 & 701 & 106 & 25@xmath935 & & & 0.8 & 4.55 + 18.5 & 701 & 158 & 22@xmath933 & & & 0.8 & 4.55 + 18.8 & 701 & 279 & 14@xmath932 & 150 & 14@xmath933.4 & 0.8 & 4.55 + 19.2 & 447.5 & 281 & 13@xmath931.5 & 173 & 12@xmath932.3 & 0.8 & 5.16 + the large sample of eros derived from our survey allowed us for the first time to estimate their clustering properties . even a simple visual inspection of the sky distribution of the objects with @xmath80 ( see fig . [ fig : distrib_sky ] ) shows that the eros have a very inhomogeneous distribution . the results of the quantitative analysis of the clustering are shown in fig . [ fig : eros_clust ] , where the observed , bias corrected angular correlations @xmath98 of the objects with @xmath8@xmath1245 are displayed . a strong clustering is indeed present at all the scales that could be measured , and its amplitudes ( table [ tab : clust_eros ] ) are about an order of magnitude higher than the ones of the field population at the same @xmath0 limits . the correlations are very well fitted by a @xmath125 power law . no attempt was made to correct the amplitudes for the stellar contamination ( see sect . [ calculus ] ) , and we stress that such corrections would increase them . adopting the errors derived from the fits , our detections are significant at more than 7@xmath35 level for the samples with @xmath80 and @xmath0 limits equal to or fainter than @xmath126 . the amplitudes shown in fig . 5 suggest a possible trend of decreasing strength of the clustering for fainter eros : the @xmath7119.2 eros are less clustered than the ones with @xmath7118.5 and the difference is significant at @xmath127 , based on the derived errors . the significance of this effect would , however , decrease if the contamination by randomly distributed stars increases towards the limit of our survey . although difficult to be quantified , this is likely to happen because , differently from the brightest eros , only a small fraction of the eros fainter than @xmath0 = 18.5 could be morphologically classified ( see fig . [ fig : zoom ] ) . defining redder thresholds drastically reduces the number of eros and it is not possible to estimate with sufficient accuracy how the amplitudes change for objects with even redder @xmath8 colors . we could only verify that the sample of eros with @xmath82 has clustering amplitudes consistent with those of the @xmath80 samples ( see table [ tab : clust_eros ] ) . to measure the clustering of the @xmath56 eros , an area at least 10 times larger than ours ( i.e. @xmath312 square degrees ) at @xmath128 would be needed , assuming that their clustering amplitudes are similar to those of the eros with @xmath80 . finally , it was studied if and how the clustering amplitude changes as a function of @xmath8 for the @xmath7118.8 sample ( see fig . [ fig : col_var ] ) . a clear increase of @xmath110 with @xmath8 is present for colors @xmath8@xmath1243.5 , while the @xmath53 sample has an amplitude that is consistent with that of the whole sample of field galaxies . the variation of @xmath110 can be described with a power law in the range of @xmath129 . previous efforts to disentangle the clustering properties of the red and blue populations in faint @xmath7selected samples probably failed because the ero population was not sufficiently sampled . for instance , kummel & wagner ( 2000 ) did not find significant differences in the clustering of objects with color bluer or redder than @xmath8=3.49 for their @xmath13017 sample . this is not surprising since at k@xmath4417 the ero population is almost absent ( see table [ tab : eros ] and fig . [ fig : ccdiag ] ) . to check for the stability of these results , possible systematics that could produce a bias in our work were analyzed . first af all , as the clustering of our @xmath0selected galaxies is in good agreement with the literature data ( fig . [ fig : roche ] ) , we can exclude the presence of measurable biases coming from the selection of the sample . regarding the color measuraments , since eros are the tail of objects in the @xmath8 color distribution , systematic variations of the photometric zeropoints across the area could have the effect of creating artificial ero overdensities and voids . to exclude this possibility we verified that the blue tail of the @xmath8 distribution is homogeneously spread across our survey , with a very low clustering amplitude . in case of zeropoints variations these should produce the same effect in both the tails of the color distribution . moreover , to test the reality of the large void of eros clearly seen in the bottom region of our survey ( see fig . [ fig : distrib_sky ] ) , the @xmath8 color distribution of the galaxies inside and outside this large void were compared by means of a kolmogorov - smirnov test , selecting only the galaxies with @xmath1314 in both regions . the probability that the two distributions are extracted from the same population is 43% . thus , the two regions are fully consistent with each other with respect to the color distribution of the blue population . this , together with the fact that no underdensity is present in this region when only the bluer galaxies are considered , shows that the void of eros should be considered a real feature . all these tests strongly suggest that the inhomogeneous ero distribution is a real effect . the strong clustering signal that we find to increase with the @xmath8 threshold and to reach very high values for the eros is potentially capable of giving insight about the nature of these objects . the main possible source of objects that may contribute to the ero population , as discussed in the introduction , are old passively evolving @xmath10 ellipticals , dust - reddened starburst galaxies and , in case of unresolved objects , low - mass stars or brown dwarfs . our field being at high galactic latitude ( @xmath49 ) , stars are expected to have no clustering , and to be homogeneously distributed , and certainly not to give the strong clustering signal detected . as for the starburst galaxies , it must be noted that in such galaxies the red colors are mainly driven by the amount of dust extinction and not by the redshift , as in the case of ellipticals ( see fig . [ fig : color_fig ] ) , and therefore a wide redshift distribution is expected which should dilute their intrinsic clustering . moreover , it is known that the iras selected galaxies ( which are typically star forming galaxies ) have very low intrinsic clustering ( e.g. fisher et al . we can therefore reasonably conclude that the observed signal is due to the clustering of high redshift ellipticals . this is also suggested by studies of the local universe which have shown that early - type galaxies are much more clustered than late - type galaxies ( e.g. guzzo et al . 1997 , willmer et al . 1998 ) . in this regard the results plotted in fig . [ fig : col_var ] could be qualitatively explained by noting that in selecting redder samples the fraction of early type galaxies increases ( the color of local ellipticals is just around @xmath132 ) and by assuming that these galaxies are intrinsically more clustered , while such plot would be difficult to understand if mainly driven by the strongly reddened starburst galaxies . these considerations strongly suggest that eros are mainly composed by @xmath10 ellipticals , confirming the previous indications that had been found on this issue . as the elliptical galaxies are the dominant population of galaxy clusters , we investigated the possibility that the detected clustering of eros could be the result of a few massive clusters at @xmath133 present in our field . for example , in the region inside the circle in fig . [ fig : distrib_sky ] , a large ero overdensity is found , that one could suspect to be due to a high-@xmath9 cluster of galaxies . however , fig . [ fig : eros_cm ] shows that there is no clear color - magnitude sequence among the @xmath134 objects inside that region , suggesting that they do not all belong to a single cluster . in case of a cluster , even at high@xmath9 , a well defined color - magnitude sequence is in fact generally observed ( e.g. stanford et al . 1998 ) . in the last years a few examples of massive @xmath1351 clusters of galaxies have been discovered ( e.g. stanford et al . 1997 , rosati et al . 1999 ) , with x ray luminosity of @xmath136 erg s@xmath18 . a crude estimate of the number of structures of this sort that could be observed in our survey can be derived by calculating the number of high@xmath9 clusters with l@xmath137 erg s@xmath18 expected in the volume we are sampling . from the x ray luminosity function of such structures at @xmath138 ( rosati et al . 2000 ) we estimate that the expected number of massive clusters in our field in the redshift interval @xmath139 is only @xmath140 ( for @xmath141 ) . moreover , the detection of the ero positive correlation , following a @xmath125 power law on all the scales from 10@xmath21 to 15@xmath20 ( corresponding to @xmath318 mpc at @xmath138 ) suggests that the clustering signal does not come from a few possible clusters detected in our field , but rather from the whole large scale structure traced by the elliptical galaxies . our results on the clustering of eros have important consequences on the problem of estimating the density of high-@xmath9 ellipticals ( see sect . 1 ) . the existence of an ero angular correlation with @xmath125 and with a high amplitude implies significant surface density variations around the mean value even for relatively large areas . in the presence of a correlation with amplitude a , the rms fluctuations of the counts around the mean value @xmath142 is ( see for example roche et al . 1999 ) : @xmath143 the factor @xmath118 is the same as in eq . ( [ eq : c ] ) and , by applying eq . ( [ eq : c ] ) for several areas , it was found that it can be approximated as : @xmath144 if the area is expressed in arcmin@xmath2 and @xmath125 . the validity of such an approximation has been tested for square regions and for areas not larger than the ones of our survey . with eq . ( [ eq : sigma ] ) and ( [ eq : c2 ] ) the expected variations of the ero number counts can be calculated , once their clustering amplitude is known . to verify the consistency of this picture , we derived the distribution of the number of eros ( with @xmath8@xmath1245 and @xmath7118.8 , i.e. those in fig . [ fig : distrib_sky ] ) that can be recovered in our area by sampling it with a field of view of 5@xmath1455@xmath20 , which is the typical field of view of a near - infrared imager such as sofi . in fig . [ fig : sofi_f ] the observed frequencies of the number of eros recovered in this counts - in - cell analysis is plotted . as the mean expected number of eros is about 10 , the poissonian fluctuations would be @xmath1463.2 , while fluctuations with @xmath1475.4 are actually observed . applying eq . ( [ eq : sigma ] ) , the measured clustering amplitude @xmath148 implies @xmath149 , in excellent agreement with the measured @xmath35 value . we also note that the distribution of the numbers of eros in fig . [ fig : sofi_f ] is not only asymmetric , but also very broad , ranging from @xmath150=0 to @xmath150=30 . in 29% of the cases the number of eros recovered is @xmath1515 , corresponding to a surface density half of the real one , while only in 19% of the cases the observed number is @xmath15215 . this shows that , on average , it is more probable to underestimate the real surface density of these objects . this is a clear property of the sky distribution that we observe , as the voids extend on a large fraction of the surveyed area . these results show how strong the effects of the field - to - field variations are in the estimate of the sky surface density of eros . in this respect , it should be noted here that all previous estimates of the number density of high@xmath9 ellipticals were based on surveys made with small fields of view , typically ranging from 1 arcmin@xmath2 in the case of the nicmos hdf - s ( benitez et al . 1999 ) to 60 arcmin@xmath2 in the case of barger et al . ( 2000 ) . the selection of galaxies with colors @xmath134 can be used to search for elliptical galaxies at @xmath153 ( see fig . [ fig : color_fig ] ) , and to study their evolution by comparing their observed surface densities with those expected from ple or hierarchical models of massive galaxy evolution . in this respect , very discrepant results have been obtained so far , making the formation of spheroids one of the most controversial problems of galaxy evolution ( see the introduction ) . our results on the ero clustering clearly show that for such a comparison to be reliable , both a wide field survey ( resulting in a large number of eros ) and a consistent estimate of their surface density fluctuations are necessary before reaching solid conclusions on the evolution of elliptical galaxies . in this section , with the main aim to show the effect of the increased uncertanties due to the clustering , a preliminary comparison is presented between the sky density of eros observed in our survey ( table [ tab : eros ] ) and the predictions of an extreme ple model similar to that used by zepf ( 1997 ) . in this model , ellipticals formed at @xmath154=5 and their star formation rate ( @xmath155 ) is characterized by an exponentially decaying burst with @xmath156 , with @xmath157 gyr . adopting the markze et al . ( 1994 ) local luminosity function of ellipticals , and the bruzual & charlot ( 1997 ) models with solar metallicity and salpeter imf , the expected surface densities of passively evolving ellipticals with @xmath92 was calculated for different limiting @xmath0 magnitudes . [ fig : eros_dens ] shows the comparison between the expected and the observed densities of eros with @xmath92 ( such a color threshold should select passively evolving galaxies at @xmath158 ) . for each data point we show three different error bars , which are actually the region of confidence in the poissonian case ( at 1@xmath35 ) and in the true ( i.e. clustering corrected ) case ( at 1@xmath35 and 2@xmath35 ) . such confidence regions have been estimated , following the prescriptions of eq . ( [ eq : sigma ] ) , by finding the range of values for the true average counts @xmath142 for which the observed @xmath150 would represent a deviation of the required number of @xmath35 from the real density . in other words such ranges are defined from the two solutions of the equation : @xmath159 where @xmath24 is the number of @xmath35 considered . the amplitudes of the angular correlation function used for the @xmath92 eros are those derived for the @xmath80 eros , which is likely to be a conservative assumption as the amplitudes of redder samples should be higher , as suggested by fig . [ fig : col_var ] . fig . [ fig : eros_dens ] shows that the observed eros densities are indeed lower than the predictions of this particular ple model . however , even for the most deviant point , this ple model can be rejected at only the 2.5@xmath35 and 2.3@xmath35 level for @xmath141 and @xmath160 respectively , if we use the `` correct '' error bars . note that the different points plotted in that figure are not statistically independent because they are partially obtained with the same objects ( they are cumulative values ) . it should be recalled here that the observed ero densities plotted in fig . 14 are an overestimate of the true density of the ellipticals because of the contamination by dusty starbursts ( see cimatti et al . 1998 , 1999 ; dey et al . 1999 ; smail et al . 1999 ) and by field low - mass stars . the fraction of dusty starbursts in complete ero samples is not known yet , as discussed in the introduction , but our results show that they should not be the dominant population . for instance , assuming that the fraction of dusty starbursts and low - mass stars is 20% and 10% of the ero population respectively , this would decrease the observed densities plotted in fig . [ fig : eros_dens ] accordingly , but it would also increase by a factor of 2 the clustering amplitudes of the high-@xmath9 ellipticals ( see sect . [ calculus ] ) , and hence the error bars related to those points . as a consequence , the statistical significance of the difference between data and model in this case would be only at the 2@xmath35 level . it is relevant to mention that the predictions of the ple models depend very strongly on many parameters that have to be adopted _ a priori _ such as @xmath161 , @xmath162 , the local lf of ellipticals ( uncertain by up to a factor of 2 ) , the redshift of formation @xmath154 , the history of star formation , the metallicity , the imf , the spectral synthesis models . for instance , even just a decrease of @xmath154 , or a small residual star formation at z @xmath31 1.5 ( menanteau et al . 98 , jimenez et al . 99 ) , would decrease the predicted numbers of eros making them more consistent with our data . we therefore conclude that it seems premature to reject even extreme ple models at a high level of statistical significance on the basis of these data . a preliminary comparison of our results can be made with some aspects of the hierarchical models of galaxy formation . first of all , our findings could qualitatively fit into the predictions of such models , where high-@xmath9 ellipticals should be very clustered ( kauffmann et al . 1999 ) because they are expected to be linked to the most massive dark matter haloes which are strongly clustered at high@xmath9 . the indication ( marginally significant at @xmath163 level ) of a decrease of the clustering amplitude of the eros with the @xmath0 magnitude ( see sect . [ sec : clust_eros ] ) , if mainly due to the mass of the galaxies , could also fit well in this framework because smaller objects should be connected to smaller dark matter haloes which are expected to be less correlated . on the other hand , our results seem to conflict with the predictions made by kauffmann & charlot ( 1998 ) on the fraction of @xmath7-selected galaxies with @xmath164 . in fact , the fraction of galaxies observed to have color @xmath82 ( which corresponds to the selection of @xmath83 ellipticals ) is about 7% of the total in our survey ( see table [ tab : eros ] ) , to be compared with the 23% of @xmath11 galaxies with k@xmath16519 expected in the kauffmann & charlot ( 1998 ) hierarchical model . this result on the fraction of @xmath10 galaxies in our @xmath7-selected sample broadly agree with the finding of eisenhardt et al . the main results of this work are : * we have presented a survey which covers 701 arcmin@xmath2 and is 85% complete to @xmath3 over the whole area and to @xmath4 over 447.5 arcmin@xmath2 ; the r - band limit is @xmath166 at the 3@xmath35 level . * the observed galaxy counts are derived over the largest area so far published in the range of @xmath167 . such counts are in excellent agreement with other published data . * the median @xmath8 color of field galaxies increases by 0.5 mags from @xmath73 to @xmath168 , and it remains constant to @xmath40 . * a sample of 398 eros has been selected . this sample is the largest published to date and is characterized by an area larger by about four times than previous surveys . the ero counts and surface densities have been derived for several color thresholds and @xmath0 limiting magnitudes . in particular , we find @xmath169 ( poissonian ) eros arcmin@xmath38 with @xmath80 and @xmath170 eros arcmin@xmath38 with @xmath92 at @xmath4 . * the surface density of eros with @xmath16 has been estimated for the first time to be of the order of @xmath84 arcmin@xmath38 at @xmath85 . * the angular correlation function of field galaxies , fitted with a fixed slope @xmath171 , has an amplitude @xmath172 at @xmath173 , in agreement with previous measurements . * for the first time , we detected the clustering of eros , with an amplitude @xmath174 for the objects with @xmath80 , in the range @xmath173 ) which is about a factor of ten higher than that of field galaxies . the ero two point correlations are very well fitted by a @xmath171 power law . * the clustering amplitude of the galaxies increases with the @xmath8 color threshold following the relation @xmath175 , for @xmath176 at @xmath89 . * the strong clustering of eros is shown to be a direct evidence that a large fraction of these objects are indeed high@xmath9 ellipticals . our result is therefore the first detection of the large scale structure traced by the elliptical galaxies at @xmath138 . * the ero clustering explains the conflicting results obtained so far on the density of high-@xmath9 ellipticals in terms of strong field - to - field variations affecting the surveys based on small fields of view ( e.g. @xmath177 arcmin ) . * taking into account the clustering of eros , even the predictions of extreme ple models for the comoving density of high@xmath9 ellipticals can not be rejected at much more than 2@xmath35 significance level . we would like to thank nathan roche for providing his models in digital form , gustavo bruzual and stephane charlot for their synthetic stellar population models . we also thank leonardo vanzi for his assistance during the ntt observations and the anonymous referee for useful comments . lp acknowledges the support of cnaa during the realization of this project .
the kinematic dynamo effect is generally believed to be facilitated by a separation of spatial scales between the flow and the magnetic field . this idea has its roots in the discovery of the @xmath5-effect ( steenbeck , 1966 ; krause and rdler , 1980 ) for non - parity - invariant flows ( an instance of which are helical flows ) . indeed , recent results indicate that helicity and scale separation can lead to experimental dynamos ( gailitis , 2000 ; busse , 2000 , stieglitz and mller , 2001 ) . lanotte ( 1999 ) showed that scale separation can also be useful for flows possessing parity invariance and thus lacking @xmath5-effect . they found a first instance of a three - dimensional parity - invariant flow , the `` modified taylor green flow '' , which can amplify a large - scale magnetic field by a negative eddy diffusivity mechanism , as conjectured by kraichnan ( 1976 ) . examples of negative eddy diffusivities when the magnetic field depends on all three coordinates , while the flow depends only on two ( two and a half dimensional dynamos ) were actually known since the work of roberts ( 1972 ) . new instances of large - scale dynamos with the roberts flow were found more recently by tilgner ( 1997 ) and plunian and rdler ( 2000 , 2001 ) . emergence of the @xmath5-effect and the negative eddy diffusivity effect can be demonstrated by asymptotic theory involving , in principle , very large scale separation . both the aforementioned experiments and numerical simulations with helical ( galanti , 1992 ) flows indicate that even a moderate scale separation of a factor two can help for obtaining a dynamo ( a physical interpretation for the case of the abc flows has been proposed by archontis ( 2000 ) ) . the following argument due to fauve ( 1999 ) shows , how crucial for experimental dynamos it is to lower the critical magnetic reynolds number @xmath6 . here , @xmath7 is the overall scale of the apparatus , @xmath8 is the typical velocity and @xmath9 is the ( molecular ) magnetic diffusivity . given the typically very small value of the magnetic prandtl number , the flow is expected to be strongly turbulent . hence , the kinetic energy dissipated per unit time in the apparatus is @xmath10 , where @xmath11 is the fluid density and @xmath12 is the typical spatial scale of the velocity . thus , @xmath13 , where @xmath14 is the scale - separation factor . therefore , for a given fluid ( @xmath9 ) , geometry ( @xmath15 ) and size ( @xmath7 ) , the required pump power behaves as the cube of the attainable @xmath16 . it is generally believed that the cost of an experiment scales roughly as its volume , i.e. @xmath17 ; thus , if there are practical limitations on the maximum pump power available , the cost will scale as the ninth power of the attainable @xmath16 . all this makes it desirable to design apparatus with the lowest possible critical magnetic reynolds number . our intention is to begin a systematic investigation of the kinematic dynamo effect for parity - invariant flows with large or moderate scale separation with emphasis on ( i ) the genericity of the phenomenon of negative eddy diffusivity and ( ii ) the lowering of the critical magnetic reynolds number . note , that parity - invariant flows are at least as `` natural '' as helical ones . indeed , the navier - stokes equations are parity - invariant . hence , even if this particular symmetry , like most other symmetries , is broken by the mechanisms generating the flow , we can expect the symmetry to be dynamically restored in a statistical sense at high reynolds numbers ( frisch , 1995 ) . note also that the existence of three - dimesnional flow with negative eddy diffusivity is a recent discovery ; contrary to the alpha effect which can be derived in purely analytic fashion by working , e.g. , at low magnetic reynolds numbers or with a delta - correlated ( in time ) velocity , only ( very cpu - demanding ) numerical solutions of the equations arising from multi - scale techniques can demonstrate the existence of negative eddy - diffusivities and can show that it is not a pathologically unusual effect . we shall work here with time - independent flows , which we may think of as being averages of turbulent flows . we are of course aware that time dependence of the flow and , generally more realistic choices of the basic flow could somewhat modify our conclusions ; such issues are left for future work . the paper is organized as follows . in sections [ s : expansion ] and [ s : abundance ] we consider the case where the spatial scale of magnetic field is much larger than that of the flow . then @xmath15 is a natural small parameter , enabling one to apply asymptotic methods . multiscale expansions in hydrodynamics may be viewed as adaptations of general homogenization methods used to determine effective large - scale transport coefficients of periodic or random materials ( bensoussan , 1978 ; kozlov , 1978 ; zhikov , 1979 ) . usually the effective coefficients can not be obtained without solving suitable auxiliary small - scale problems . in three space dimensions this may require significant computational resources . in lanotte ( 1999 ) equations for the large - scale component of magnetic fields are derived along the same lines as in previous works on negative eddy viscosity in ordinary incompressible hydrodynamics ( dubrulle and frisch , 1991 ; gama , 1994 ; wirth , 1995 ) . here we use a similar but a distinct approach based on the work of vishik ( 1986 , 1987 ) , which yields a complete decomposition of magnetic modes and their eigenvalues ( in arbitrarily high - order terms of such expansions are derived ) . to investigate , how commonly the phenomenon of negative magnetic eddy diffusivity is exhibited by fully three - dimensional flows , we follow in a strategy similar to the one , which gama ( 1994 ) applied to estimate the abundance of two - dimensional flows with negative eddy viscosity . by solving numerically the relevant auxiliary problems we determine the minimum over directions of the ( anisotropic ) magnetic eddy diffusivities for two ensembles of flows with random fourier components this way we determine the fraction of flows such that , when decreasing molecular magnetic diffusivity , a large - scale negative eddy diffusivity instability appears before the flow becomes a small - scale dynamo . an interpretation of `` outliers '' with strongly negative eddy diffusivity , found in computations , is given in . since it is important for our purpose to analyze as many instances as possible , we have used a new efficient algorithm for solving the elliptic auxiliary problems arising from the two - scale expansion . this method is an outgrowth of the chebyshev iterative method developed by podvigina and zheligovsky ( 1997 ) for self - adjoint negative definite linear operators . the operators arising in hydrodynamical problems are not self - adjoint , but they naturally split into a dominant self - adjoint negative definite part ( the molecular diffusion term ) and the remaining transport term . we can then combine chebyshev methods with operator splitting to reduce the harm stemming from complex eigenvalues . this method and its theoretical foundations are presented in detail in zheligovsky ( 2001 ) , available only in russian . for the reader s convenience , the essentials are given in . in we turn to cases with moderate scale separation ( i.e. when @xmath15 is finite ) . first , we consider instances of flows for which we already know that there is a negative magnetic eddy diffusivity , and we determine windows in @xmath15 over which dynamo action is present . second , we examine the case @xmath18 , that is , we assume that the magnetic field has a spatial period twice that of the flow , and study how the critical magnetic reynolds number @xmath16 is affected , compared to the case with no scale separation ( @xmath19 ) . in both types of computations we exploit the fact that , for space - periodic boundary conditions , eigenmodes have the form @xmath20 where @xmath21 is a prescribed wavevector . this enables one to solve a problem involving only small scales , but parameterized by @xmath15 and @xmath21 ( cf . roberts , 1972 ; frisch , 1986 ) . solution of the eigenvalue problem is more cpu - intensive than the solution of auxiliary problems involved in the asymptotic case ( because efficient preconditioning techniques are available for the latter ) . numerical methods and codes used for the eigenvalue computations are adapted from zheligovsky ( 1993 ) . in we summarize our main results . here we briefly review the asymptotic expansion arising in the kinematic dynamo problem , when the scale ratio @xmath15 is small . the conducting fluid flow @xmath22 is assumed to be time - independent , and the problem therefore reduces to the eigenvalue problem for the magnetic induction operator @xmath23 in the space of solenoidal vector fields : @xmath24 here @xmath25 is magnetic molecular diffusivity , @xmath26 is a magnetic mode and @xmath27 is the associated eigenvalue ( the real part of which is here referred to as the growth rate of the mode ; negative growth rates correspond to decaying modes ) . the flow is assumed incompressible : @xmath28 @xmath29-periodic in space and parity - invariant : @xmath30 the magnetic mode @xmath31 is supposed to depend on the fast variable @xmath32 and on the slow variable @xmath33 , and is @xmath29-periodic in both fast and slow variables ( the velocity @xmath34 is independent of slow variables ) . introduction of the two spatial scales results in a `` decomposition of the gradient '' @xmath35 in and . here and in what follows the subscripts @xmath32 and @xmath36 refer to differential operators in fast and slow variables , respectively . let @xmath37 and @xmath38 denote the mean ( over the periodicity cube ) and the fluctuating part of a vector field , respectively : @xmath39 parity invariance implies @xmath40 . solutions to the eigenproblem - are sought in the form of asymptotic series @xmath41 @xmath42 as shown in , substitution of these expansions in leads to a hierarchy of equations . by application of standard techniques for singular perturbation expansions , terms in the expansion eigenvalues are successively determined from solvability conditions , obtained here by just averaging the equations . in particular , it can be shown that @xmath43 , and the leading term @xmath44 in the series can be determined from the following eigenvalue problem @xmath45 subject to the solenoidality condition @xmath46 thus , the leading terms of the expansions in powers of @xmath15 of the mean magnetic field and of the eigenvalue turn out to be , respectively , an eigenfunction and the associated eigenvalue of a partial differential operator comprised only of second order derivatives , which may be regarded as representing an anisotropic diffusion . however , this operator is not necessarily dissipative : it may have eigenvalues with positive real parts . in this context one then may speak of negative eddy diffusivity . as shown in eqn . , the coefficients @xmath47 of the homogenised equation are given by @xmath48 where the tensor field @xmath49 is obtained by solving a set of two auxiliary problems : @xmath50 and @xmath51 which will be referred to as the first and the second auxiliary problem , respectively . here , @xmath52 is the unit vector in the @xmath53-th cartesian coordinate direction and superscripts denote cartesian components of vectors . for solutions to and to exist , we require that zero should not be an eigenvalue of the magnetic induction operator @xmath54 restricted to small - scale solenoidal vector fields with a vanishing mean ; by `` small - scale '' we mean independent of the slow variable . generically , this condition is satisfied . problems and are formulated with a @xmath29-periodicity boundary condition for the vector fields @xmath55 and @xmath56 , which should also have a vanishing mean . it follows that the @xmath57 s are parity - antiinvariant and solenoidal ( @xmath58 ) , and that the @xmath49 s are parity - invariant ( @xmath59 ) . since the homogenized operator defined by - has constant coefficients , its eigenvectors are fourier harmonics : @xmath60 , where @xmath61 and @xmath21 are constant vectors . the wavevector @xmath21 is taken real because we are only interested in modes uniformly bounded in the entire space . substitution in - yields @xmath62 @xmath63 following gama ( 1994 ) , we now ask how common is the phenomenon of negative diffusivity . time - independent periodic flows , satisfying - , are represented here by their fourier series @xmath64 with random amplitude harmonics . for the velocity to be real , @xmath65 , and parity invariance requires re@xmath66=0 . we study two ensembles of flows with different types of fall - offs of the energy spectrum @xmath67 . the latter is defined as the total energy contained in harmonics , whose wavevectors @xmath68 belong to the @xmath69-th spherical shell , i.e. @xmath70 , where @xmath71 is integer . when @xmath72 , the flow is said to have exponential spectrum ; when @xmath73 , it is said to have hyperbolic spectrum . these have respectively rapid and slow fall - offs . the spectra are band - limited and set to zero beyond wavenumber @xmath74 for the exponential case and beyond @xmath75 for the hyperbolic case . the value @xmath76 is used throughout in the exponential case . the flows are produced by the following procedure ( applied only to one half of the set of wavevectors , the second one being related by complex conjugation ) . for @xmath68 in the ball @xmath77 , random imaginary vectors @xmath78 are generated with the imaginary part of each cartesian component uniformly distributed in the interval @xmath79 $ ] ( the ran2 random number generator by press , 1992 , is used ) . then the vectors @xmath78 are projected onto the plane perpendicular to @xmath68 to gain incompressibility . finally , the obtained vectors are normalized , so that the flow has the desired energy spectrum and the root - mean - square velocity is one , when the normalized vectors are employed as the fourier coefficients @xmath66 in . each ensemble consists of 100 flows . for each flow , computations are carried out for three values of molecular diffusivity , @xmath80 , 0.2 and 0.1 . note that when @xmath9 is too large , no dynamo action can take place at large or small scales because the eddy diffusivity will basically be equal to the molecular diffusitivity . ( in contrast , an alpha - effect dynamo is still possible if the system is large enough ) . if @xmath9 is too small , small - scale dynamo instabilities will have much larger growth rates than large - scale instabilities . in both extreme cases it is meaningless to study negative eddy diffusivity large - scale instabilities . the range of values of @xmath9 chosen , between 0.1 and 0.3 , is an empirically determined tradeoff between these extremes . all computations are performed with the resolution of @xmath81 fourier harmonics , employing standard spectral methods based on fast fourier transforms and dealiasing . ( each case requires from 10 to 30 minutes of cpu on one dec alpha ev6 processor , depending on the value of @xmath9 . ) the power spectra of the solutions to and turn out to decrease monotonically very quickly for @xmath82 . for @xmath83 in the hyperbolic case ( resp . exponential case ) , the decrease is typically by 7 - 8 orders of magnitude ( 14 - 15 orders of magnitude ) from maximum to minimum . note that our code reproduces the negative eddy diffusivity values obtained by an alternative numerical method by lanotte ( 1999 ) for the modified taylor green flow . 100 10 eddy diffusivity histograms table 1 . statistics of the eddy diffusivities for the considered ensembles of flows . [ cols="^,^,^,^,^,^,^ " , ] after the homogenized coefficients are determined , the minimum magnetic eddy diffusivity @xmath84 is calculated as the minimum eigenvalue of the problem - . ( note , that for a unit wavevector the eigenvalue is just the opposite of the eddy diffusivity . ) for a given @xmath21 this eigenproblem yields two eigenvalues satisfying a quadratic equation ; the eigenvalues can thus be complex conjugate ( a similar phenomenon was observed for three - dimensional eddy viscosities ; see wirth , 1995 ) . then the minimum of the real part of the complex eddy diffusivity , which determines the maximum rate of instability of a large - scale magnetic perturbation , is sought . however , in our computations complex eddy diffusivities appear quite rarely . histograms of the obtained magnetic eddy diffusivities are shown in . the statistics of eddy diffusivities shown in table 1 indicate that the lower the molecular diffusivity , the higher the chance that a flow will have a negative `` eddy correction '' ( correction to the molecular value coming from the flow ) and also the higher the chance that the resulting eddy viscosity will be negative . note however that the eddy correction ( given , at least implicitly , by - ) has a complicated , generally non - monotonic , nonlinear dependence on the molecular diffusivity . table 1 also shows that flows with the exponential energy spectrum are more efficient in lowering the minimum of the eddy diffusivity than those with the hyperbolic spectrum . this suggests that low wavenumber harmonics of the underlying flow are more important for lowering than high wavenumber harmonics . examples of magnetic eigenmodes for @xmath83 are shown in figures [ f : sample1 ] and [ f : sample2 ] . they correspond to two samples having exponential velocity spectra . the upper of each figure shows regions where @xmath85 is less than 0.2% of its maximum , hence revealing the immediate neighborhood of stagnation points . note that in consequence of parity invariance about the origin and of @xmath29-periodicity , within a basic @xmath29-periodicity cube there are in total eight points about which the flow is parity - invariant . these stagnation points are located at the vertices of a cube with side @xmath86 and they form a uniform @xmath86-periodic grid in space . the magnetic structures seen in the lower left and right of each figure are clearly located near some of those stagnation points . these structures are magnetic flux ropes of the kind considered by galloway and zheligovsky ( 1994 ) . note that only the fluctuating part of the magnetic eigenmodes are shown ( both the periodicity - box averaged magnetic field @xmath87 and the overall slow spatial factor @xmath88 are omitted ) . in @xmath87 is very small ( about two orders of magnitude smaller than the fluctuating field ) , whereas for the case of it is of the same order of magnitude as the fluctuating field . 100 10 eigenmode sample 1 ( exponential spectrum ) 100 10 eigenmode sample 2 ( exponential spectrum ) 100 10 eddy diffusivity vs growth rate ( exponential spectrum ) 100 10 eddy diffusivity vs growth rate ( hyperbolic spectrum ) a physically important question is : which instability arises first , when magnetic molecular diffusivity is decreased , a large - scale or a small - scale one ? similarly to what was done by gama(1994 ) , for each considered pair of a flow and a molecular diffusivity , it is checked that small - scale magnetic fields with a vanishing mean are stable , i.e. have negative growth rate ( see figures [ f : dotsntrianglese ] and [ f : dotsntrianglesh ] ) . 100 10 infinitely low eddy diffusivity several `` outliers '' with relatively large negative eddy diffusivities can be seen on the histograms shown on for @xmath83 . this can be explained as follows . we see on figures [ f : dotsntrianglese ] and [ f : dotsntrianglesh ] that in the case of outliers the growth rate of _ small - scale _ magnetic modes ( plotted horizontally ) is negative , but close to zero . furthermore , we have checked that the corresponding eigenvalue is real . in other words , we are close to a non - oscillatory small - scale instability threshold . now , we observe that the expression of the eddy diffusivity , derived in and , involves the inverse of the small - scale magnetic induction operator . it is thus not surprising that large eddy diffusivities can be obtained . depending on whether a certain inequality is satisfied at the threshold , this eddy diffusivity can be either large negative or large positive ( see details in ) . we have checked numerically that the outlier eddy diffusivities are approximately inversely proportional to the small dominant eigenvalue ( see ) . note , that if we are exactly at the threshold , our two - scale expansion breaks down . a modified expansion can be carried out which takes into account the enlargement of the kernel of the magnetic induction operator @xmath89 . in the presence of a large - scale perturbation this can lead to an eigenvalue @xmath90 rather than the @xmath91-dependence obtained in the general parity - invariant case . in other words , the large - scale dynamics can be governed by a first - order differential operator . for intermediate values of @xmath15 , corresponding to moderate scale separation , direct evaluation of the most unstable magnetic modes is desirable . computations are simplified by the fact that eigenmodes have the form , all the dependence on the slow variable @xmath33 being absorbed into an exponential factor @xmath92 . since the flow is assumed to depend only on the fast variable @xmath32 , the eigenvalue problem is reduced to a small - scale one for a _ modified operator _ by changing @xmath93 in the magnetic induction operator and in the solenoidality condition . to find the most unstable magnetic mode one must , in principle , allow for arbitrary orientations of the vector @xmath21 . throughout this section we assume only `` binary '' @xmath21 s such that each component takes only the values 0 or 1 , so that @xmath94 may be regarded as the maximum scale - separation factor over cartesian directions . since our simulations are computationally demanding , we choose for each flow only one @xmath21-direction as follows . for a given flow and for the molecular diffusivity @xmath83 , we obtain the most unstable @xmath21-directions _ in the asymptotic regime _ @xmath95 from the results of . it turns out that for the cases reported in this section , these directions are approximately parallel to binary vectors with one or two non - zero components . these binary vectors are then employed when @xmath15 is finite for all values of @xmath9 studied . 100 10 growth rate vs scale ratio ( a ) 100 10 growth rate vs scale ratio ( b ) 100 10 growth rate vs scale ratio iii 100 10 growth rate vs scale ratio iv figures [ f : growthvsepsilona]-[f : growthvsepsilond ] show the dominant growth rates for four different flows selected out of the ensemble of flows with hyperbolic spectrum , which sustain negative eddy diffusivity . actually , when @xmath21 has integer components , it is only necessary to make computations for @xmath96 . indeed , first note , that the modified magnetic induction operators for two values of @xmath15 differing by 1 have the same spectrum and their eigenfunctions differ by a factor @xmath97 . second , by parity invariance of the velocity , for two opposite values of @xmath15 the spectra are also the same and eigenfunctions are related by parity @xmath98 . ( from this one easily deduces that the spectrum of the modified operator is complex conjugate , eigenvalues @xmath27 and @xmath99 being associated to eigenfunctions @xmath26 and @xmath100 , respectively . ) hence , outside the interval @xmath96 the plots can be continued by symmetry about the vertical line @xmath18 and by 1-periodicity . the results reported here are obtained with the resolution of @xmath81 fourier harmonics . for the `` worst '' case , @xmath101 , they differ only in the third significant digit from those obtained with the resolution @xmath102 . note , that energy spectra of the computed magnetic modes decrease by at least 4 orders of magnitude for @xmath103 and by at least 5 for @xmath80 , indicating that the employed resolution is adequate . ( for @xmath101 the decrease is only by 3 orders of magnitude , which may seem marginal . however , it is well - known that eigenvalues are less sensitive to insufficient resolution , than the fine structure of eigenmodes . ) it is clear from figures [ f : growthvsepsilona]-[f : growthvsepsilond ] that for all instances where the eddy diffusivity is negative ( i.e. the graph go to zero by negative values as @xmath104 ) there is also an instability for any finite @xmath15 . the converse is , however , not always true ; for example , for the last flow ( figure [ f : growthvsepsilond ] ) and @xmath105 there is a dynamo for @xmath106 , which disappears below @xmath107 ( the eddy diffusivity is positive ) . furthermore , inspection of figures [ f : growthvsepsilonb]-[f : growthvsepsilond ] reveals that there can be one or more windows in scale separation , where the dynamo ( positive growth rate ) is present , whereas in figures [ f : growthvsepsilona ] the dynamo appears to be present for all scale separations , provided the eddy diffusivity is negative . in the two windows belong to two different analytical branches of eigenvalues and eigenmodes , with the transition occurring near @xmath108 . note , that all the growth rates vanish ( quadratically ) when @xmath95 . the reason is that the small-@xmath15 eigenvalues , obtained by the two - scale expansion of , are analytic continuations of the zero eigenvalue in the small - scale problem . the presence of the eigenvalue zero is due to the structure of the magnetic induction operator , which is the curl of a first - order operator . ( hence , its adjoint , acting also in the space of divergenceless functions @xmath109 has constant vectors in its kernel . ) conversely , it can be demonstrated by spatial averaging of @xmath110 , that for a mode with a non - vanishing mean the associated eigenvalue equals 0 . ( note , that in we restricted the operator to functions of vanishing spatial mean and thus the growth rate zero was absent from figures [ f : dotsntrianglese ] and [ f : dotsntrianglesh ] . ) 100 10 critical rm s : small - scale vs double period we now turn to the case , where there is a finite scale separation of a factor two ( @xmath18 ) . the question we address is : how does a mere doubling of the spatial period of magnetic modes ( relative to that of the basic flow ) affect generation ? more precisely : to lower the critical magnetic reynolds number , does it help to introduce such a scale separation ? we systematically compare critical numbers when @xmath19 ( no scale separation ) and @xmath18 for an ensemble of flows with hyperbolic energy spectrum . we define our magnetic reynolds numbers @xmath6 in terms of the _ largest _ scale available @xmath111 ( the inverse of the minimum wavenumber of the magnetic field ) and of the r.m.s . velocity @xmath8 . in computations @xmath112 and hence @xmath113 . when @xmath18 we face again the problem , mentioned earlier , of choosing the vectors @xmath21 . for each flow we now search for the critical @xmath16 , allowing all 7 possible binary @xmath114 . more precisely , we find an approximate ( minimum ) critical @xmath16 and the corresponding @xmath21 by running our code with @xmath102 fourier harmonics , and then refine the critical @xmath16 with @xmath81 fourier harmonics for the obtained @xmath21 . the secant method is used to find stationary modes . iterations arere terminated when the real part of the dominant eigenvalue was below @xmath115 in magnitude . this usually requires 4 - 5 iterations . as far as the accuracy of the calculation is concerned , note that in all @xmath18 runs with @xmath116 the energy spectrum decreases by at least 4 orders of magnitude , an indication that @xmath81 harmonics provide sufficient resolution . in the most demanding runs , with @xmath19 and @xmath117 , the energy spectra decrease only by 23 orders of magnitude but , as pointed out before , eigenvalues are computed nevertheless quite accurately ( critical @xmath16 s , computed with the @xmath102 and @xmath81 resolutions , differ in the third significant digit ) . shows all the critical @xmath16 s calculated for 28 cases . two cases not shown have critical @xmath118 and @xmath119 when @xmath18 , but do not display dynamo action for any @xmath120 when @xmath19 . the results demonstrate that scale separation lowers the critical @xmath16 in all but 9 cases out of the 30 studied . note that nore ( 1997 ) have studied a flow ( a variant of taylor green ) for which a scale separation @xmath18 seems to lower significantly the critical magnetic reynolds number , as is the case for the helical abc flows studied by galanti ( 1992 ) , archontis ( 2000 ) and dorch ( 2000 ) . however the nore ( 1997 ) definition of the magnetic reynolds number uses the scale of the flow and not that of the magnetic field . if instead our definition is used , their flow appears to be just a borderline case . three main results are presented in this work . first , in , we demonstrated that parity - invariant flows possessing a negative magnetic eddy diffusivity are quite common . specifically , we generated two random ensembles of flows ; both had time - independent randomly chosen fourier components , one with an exponentially decreasing energy spectrum , the other with a power - law decrease ( referred to as `` hyperbolic '' because the exponent is -1 ) . one hundred independent selections were made in each case and three different molecular magnetic diffusivities were used , all corresponding to small - scale magnetic reynolds numbers not exceeding ten . none of the 600 cases investigated was a small - scale dynamo . for the lowest diffusivity considered , 86% of the `` exponential '' spectrum flows and 53% of the `` hyperbolic '' spectrum flows produced a negative eddy diffusivity and , hence , can act as a large - scale dynamo ( for more details , see table 1 ) . second , several instances with particularly negative eddy diffusivities were encountered . the theory of these `` outliers '' ( given in details in ) relates them to a near - critical state , where an eigenvalue of the small - scale dynamo problem is about to pass through zero ( there is also an additional inequality ensuring that a large _ negative _ value is obtained ) . finally , we presented numerical evidence that , for a parity - invariant flow , even a moderate scale separation between the flow and the magnetic field is generally beneficial for magnetic field generation . specifically , we showed that the critical magnetic reynolds number based on the largest scale present ( that of the magnetic field ) is often lowered when the scale of twice the spatial period of the velocity is introduced ( 70% of the instances studied had this property ; see ) . for a few flows we also studied the dependence on the scale - separation factor , when it is varied from a 1:2 ratio to a 1:20 ratio . one instance ( shown in ) has a window lacking dynamo action between low and large values , where it is present . for helical flow , such as abc , the lowering of the critical magnetic reynolds number , when a moderate separation of scale is assumed , is understood in terms of symmetry breaking , stagnation points and other geometrical features ( archontis 2000 ) . it would be of interest to find which geometrical features are important for the lowering mechanism when the flow is parity - invariant . * acknowledgments*. we are grateful to the authors of lanotte ( 1999 ) for making available to us the source of the code used to obtain the results reported in their paper , to s. fauve , s. gama , a. lanotte , a. noullez , y. ponty and m. vergassola for useful discussions and to an anonymous referee for several suggestions . computational facilities were provided by the program `` simulations interactives et visualisation en astronomie et mcanique ( sivam ) '' at the observatoire de la cte dazur . visits of o.m . podvigina and v.a . zheligovsky were supported by the french ministry of education . this work was also supported by the european union under contract hprn - ct-2000 - 00162 . [ a : expansion ] we construct here complete asymptotic expansions of magnetic eigenmodes and of their eigenvalues for small values of the scale ratio @xmath15 . assumptions concerning the flow are listed in . the kinematic dynamo problem reduces to the eigenvalue problem @xmath121 here @xmath26 is a magnetic mode , @xmath27 is its eigenvalue , and @xmath122 acting in the space of solenoidal vector fields , is the magnetic induction operator ( it acts like @xmath89 , but in a larger domain ) . the magnetic mode @xmath31 is supposed to be solenoidal and @xmath29-periodic in both fast and slow variables . solutions to the eigenproblem are sought in the form of asymptotic series @xmath123 and @xmath124 . denote @xmath125 where @xmath37 and @xmath38 refer to the mean and fluctuating part ( see ) , respectively . the expansion is constructed under the assumption that zero is not an eigenvalue of the operator @xmath89 acting in the space of small - scale periodic solenoidal vector fields with vanishing mean . ( by `` small - scale '' we understand `` depending only on the fast variables '' . ) the decomposition of the gradient @xmath126 , applied to the solenoidality condition , expanded in powers of @xmath15 and separated into mean and fluctuating part , gives @xmath127 @xmath128 for all @xmath129 . ( @xmath130 is assumed for all @xmath131 . ) substituting the expansions and the decomposition of the gradient into , we obtain : @xmath132\epsilon^n=0 . \label{serieseqn}\end{aligned}\ ] ] we proceed by successively equating to zero the mean and the fluctuating part of each term of this series . in our context , averaging is equivalent to obtaining the solvability condition for a partial differential equation in fast variables . transform the leading @xmath134 term of using vector algebra identities and : @xmath135 averaging of this equation yields @xmath136 . then the solution of can be expressed as @xmath137 where the vector fields @xmath57 are the solutions of the first auxiliary problem ; the solutions exist since it is assumed that the kernel of @xmath89 is empty . ( the first and second auxiliary equations were stated in . ) by taking the divergence of the first auxiliary equation , one obtains @xmath138 . since the velocity is parity - invariant , the domain of @xmath89 is a direct sum of two invariant subspaces , one of which is comprised of parity - invariant vector fields ( i.e. @xmath139 ) , and the second of the parity - antiinvariant ones ( @xmath140 ) . thus , @xmath141 is a parity - antiinvariant matrix : @xmath142 @xmath143 . the @xmath144 term of , after the use of , takes the form @xmath145 & & -({\bf v}\cdot\nabla_{\bf y}){\bf h}_0=\lambda_1({\bf h}_0+{\bf g}_0 ) . \label{eigeneq1}\end{aligned}\ ] ] substituting in and averaging , we obtain @xmath146 so far , developments strictly followed those of vishik ( 1987 ) , but here the point of divergence is reached . since @xmath34 is parity - invariant and @xmath141 is parity - antiinvariant , we have @xmath147 , and therefore implies @xmath148 . upon inserting into the fluctuating part of , we obtain @xmath149 from which one finds @xmath150 here , the nine vector fields @xmath49 are solutions to the second auxiliary equation . note @xmath59 by the symmetry properties of @xmath34 , @xmath57 and of the operator @xmath89 . by taking divergence of the second auxiliary equation one finds @xmath151 . hence , for @xmath152 is automatically satisfied . the @xmath154 term of takes the form @xmath155 & & + \,\,\nabla_{\bf y}\times({\bf v}\times{\bf g}_1 ) -({\bf v}\cdot\nabla_{\bf y}){\bf h}_0+{\bf g}_0 ) . \label{eigeneq2}\end{aligned}\ ] ] upon substitution of , averaging produces the second - order equation @xmath156 which defines the eddy diffusivity operator . so , the leading terms of the expansions of the mean magnetic field , @xmath157 , and of the associated eigenvalue , @xmath44 , satisfy the eigenvalue equation for this operator . its eigenvectors are fourier harmonics : @xmath158 , where @xmath61 and @xmath21 are constant vectors satisfying and . note that the fluctuating part of the leading term of the eigenmode , @xmath159 , is now completely determined by . at this point our construction goes beyond that of lanotte ( 1999 ) and , in the spirit of vishik ( 1987 ) , we show how to extend the analysis to higher orders in @xmath15 . the fluctuating part of reads @xmath160 @xmath161 @xmath162 here , only the terms involving @xmath163 and @xmath164 are not yet determined . however , as one can immediately see , @xmath32-dependent prefactors in front of these terms are identical to those , which are in in front of @xmath165 and @xmath166 , respectively . this implies a representation @xmath167 where the vector field @xmath168 can be uniquely determined from @xmath169 @xmath170 @xmath171 . as already stated , @xmath44 is one of the two eigenvalues of - . in what follows we assume that the other eigenvalue , @xmath172 , is distinct from @xmath44 and we denote by @xmath173 the associated eigenvector . equations arising at higher orders in @xmath15 are now solved recursively . assume that all the equations up to order @xmath174 have been solved , so that the following information has been obtained : @xmath175 vector fields @xmath176 and @xmath177 for all @xmath178 ; @xmath175 representations of @xmath177 of the form @xmath179 for @xmath180 and @xmath181 with known vector fields @xmath182 , @xmath183 ; @xmath175 quantities @xmath184 for @xmath185 . upon substitution of for @xmath180 the average of the equation at order @xmath186 becomes @xmath187 where the right - hand side at this stage is a known vector field . projecting this equation out in the direction of @xmath157 one can uniquely determine @xmath188 . in the complementary invariant subspace the operator @xmath189 is invertible , and hence @xmath190 can be determined up to an arbitrary multiple of @xmath157 . this additive term can be neglected : one can demand that @xmath190 belong to the complementary subspace . this is a normalization condition related to the fact that eigenvectors can be multiplied by arbitrary ( analytic ) functions of @xmath15 . as a result , @xmath191 is now determined by with @xmath181 . upon use of for @xmath180 the fluctuating part of the equation at order @xmath186 becomes @xmath192 @xmath193 @xmath194 @xmath195 only the terms involving @xmath196 and derivatives of @xmath197 are not yet determined . like in , @xmath32-dependent prefactors in front of these terms are identical to those in front of @xmath165 and derivatives of @xmath198 in , respectively . hence , yields a representation of the form for @xmath199 . an equation for @xmath200 can be obtained from by replacing @xmath199 by @xmath200 and omitting all terms involving @xmath196 and derivatives of @xmath201 . thus , a complete formal decomposition of magnetic modes and their eigenvalues has been constructed . a step - by - step analysis of the solution reveals that @xmath202 ( with suitable scalars @xmath203 ) and @xmath204 thus the expanded eigenmode admits a representation in the form of a plane wave in the slow variable . ( for this reason it was sufficient to demand , for constructions presented under heading @xmath171 , that the two eigenvalues @xmath44 and @xmath172 of @xmath205 be distinct in the subspace of solenoidal vector fields @xmath206 with @xmath207const , instead of requiring that @xmath189 be invertible in the whole domain . ) the plane - wave representation does not come as a surprise , since the domain of the magnetic induction operator @xmath208 splits into invariant subspaces , each comprised of vector fields of the form and categorized by wavevectors @xmath21 . if the homogenized operator @xmath205 turns out to be elliptic , one can prove , following vishik ( 1986 , 1987 ) , that any point @xmath44 of the spectrum of @xmath205 is associated with an analytic branch of eigenvalues of the original magnetic induction operator @xmath208 . the proof relies on the general perturbation theory for linear operators ( kato , 1980 ) ; it will not be presented here . [ a : loweddydiff ] in this appendix we discuss a mechanism for appearance of negative eddy diffusivities with arbitrarily large magnitudes . let @xmath209 denote the basis of small - scale @xmath29-periodic magnetic modes with vanishing mean and @xmath210 the associated eigenvalues : @xmath211 ( assuming in this section , for the sake of simplicity , that @xmath89 is diagonalizable ) . suppose that when molecular diffusivity @xmath9 is decreased , small - scale magnetic fields with vanishing mean lose stability in such a way that the dominant eigenvalue of the magnetic induction operator @xmath89 passes through zero . let @xmath212 be the critical diffusivity : @xmath213 generically , the loss of stability occurs either in the proper subspace comprised of parity - invariant fields , or in that of parity - antiinvariant fields , implying two possible variants of the mechanism . suppose small - scale parity - invariant magnetic fields are the first to lose stability . solutions @xmath214 to the first auxiliary problem continuously depend on @xmath9 for @xmath215 . the second auxiliary problem can be reexpressed as @xmath216 where @xmath217 and @xmath218 are decompositions of the r.h.s . of and of @xmath219 in the basis of magnetic modes ; summation over all parity - invariant eigenvectors is assumed in the two sums . from , @xmath220 . since generically the coefficients @xmath221 do not vanish , we have in the limit @xmath222 @xmath223 then the equation for the eigenvalue @xmath44 asymptotically reduces to @xmath224 where @xmath225 . equations and yield two eigenvalues . it is easily checked that one of them , denoted @xmath226 , vanishes and thus is of no interest for us ; the associated eigenvector @xmath173 is orthogonal to the wavevector @xmath21 and satisfies @xmath227 the second eigenvalue is @xmath228 the associated eigenvector being @xmath229 ( we consider a generic case , where @xmath230 does not vanish ) . therefore , if @xmath231 , then , in the limit @xmath222 , the minimum magnetic eddy diffusivity becomes arbitrarily large negative : @xmath232 alternatively , suppose now that the loss of stability occurs in the subspace of small - scale parity - antiinvariant magnetic fields . then the mechanism is similar . in this case already the solutions @xmath214 to the first auxiliary problem tend to infinity . indeed , can be represented as @xmath233 where @xmath234 and @xmath235 are decompositions of the r.h.s . of the first auxiliary problem and of @xmath236 in the basis of magnetic modes ; summation over parity - antiinvariant eigenvectors is assumed in both sums . thus , @xmath237 . since generically @xmath238 , we have in the limit @xmath222 @xmath239 let @xmath240 denote the solution to @xmath241 then @xmath242 and asymptotically reduces to @xmath243 where @xmath244 equations and yield two eigenvalues . again , one can easily check that one of them , @xmath226 , vanishes ; the associated eigenvector @xmath173 is orthogonal to @xmath21 and satisfies @xmath245 the second is @xmath246 with the associated vector @xmath247 ( in the generic case @xmath248 ) . therefore , if @xmath249 , when @xmath222 the minimum magnetic eddy diffusivity becomes again arbitrarily large negative : @xmath250 [ a : chebyshev ] we discuss here three chebyshev iterative methods suitable for large systems of linear or non - linear equations . they employ extremal properties of chebyshev polynomials and their roots . the first method ( ) , discussed in , is applicable to problems with negative - definite self - adjoint jacobians . an extension ( ) , discussed in , is an upgrade of , which eliminates these restrictions . the second method ( ) , discussed in , is designed for discretizations of elliptic differential equations , such as the auxiliary problems and . and were introduced by podvigina and zheligovsky ( 1997 ) and is presented in detail in zheligovsky ( 2001 ) . for general background on chebyshev iterative methods , see axelsson ( 1996 ) . let @xmath254 be a solution to . we first consider the case , where the jacobian matrix @xmath255 is diagonalizable and its spectrum is ( real and ) negative . then the solution can be computed by iterations @xmath256 @xmath257 provided @xmath258 is chosen in the basin of attraction of @xmath259 . here @xmath260 where @xmath261 for any @xmath262 ( the optimal value @xmath263 is usually unknown ) , and where the @xmath264 s are chebyshev polynomials of the first kind . the @xmath265 s can be evaluated recursively : @xmath266 consider how the discrepancy vectors @xmath267 evolve in the course of iterations - . introduce the eigenvalues @xmath268 of @xmath255 ( here,@xmath269 ) and the associated eigenvectors @xmath270 , and decompose the initial discrepancy in this basis : @xmath271 it is then easily shown that @xmath272 ( the @xmath273 correction does not appear if the mapping @xmath274 is linear ) . at this point , the standard strategy is to use the least deviation principle for chebyshev polynomials : the iterative process - with @xmath275 ensures maximum uniform decay of the discrepancy vector component in the proper subspace of @xmath255 associated with the part of the spectrum in the interval @xmath276 . the alternative procedure used by podvigina and zheligovsky ( 1997 ) relies , in addition , on the extremal property of the largest root of a chebyshev polynomial : for @xmath277 , the iterative process gives optimal reduction of the discrepancy vector component in the proper subspace of @xmath255 associated with eigenvalues @xmath268 closest to zero . therefore , a suitable algorithm , called here , consists of a succession of iterative sequences - with @xmath275 and @xmath277 . as usually , lowering of the condition number of the system allows to improve efficiency of the algorithm . if , in addition , @xmath255 is self - adjoint , such lowering can be achieved by replacing with an equivalent system @xmath278 where @xmath279 @xmath280 being a self - adjoint and positive - definite linear preconditioning operator . it may be checked that for such a @xmath280 the mapping @xmath281 has the required properties stated at the beginning of this subsection . in the case of an arbitrary smooth mapping @xmath274 , to which we turn now , a modified system @xmath282 can be considered , where @xmath283 @xmath284 being arbitrary positive definite self - adjoint operators and@xmath285 being the operator adjoint to @xmath286 . the method discussed above can be applied for the solution of the modified system , since @xmath287 has all the required properties . \2 . in any of the following iterative sequences @xmath277 , unless @xmath27 was zero in the previous sequence and that sequence was short ( i.e. it consisted of less than @xmath289 $ ] iterations , where @xmath290 $ ] denotes the integer part of a number ) , in which case in the next sequence @xmath288 . \4 . in an iterative sequence with @xmath277 the errors @xmath291 can initially grow monotonically , if the respective errors @xmath292 decrease monotonically . ( an increase of @xmath293 at the first iteration indicates that @xmath294 is underestimated . ) \5 . after the initial phase allowed by rules 3 or 4 is completed , an iterative sequence is terminated if both discrepancy reduction rate values , @xmath295 and @xmath296 , are smaller than at the previous iteration . we present here an alternative to , better suited to problems arising from discretization of elliptic partial differential equations , when they involve not just leading - order elliptic operators ( such as the laplacian ) , but also lower - order non - self - adjoint perturbations . we do not attempt to present this method in full generality ( see zheligovsky , 2001 ) . in what follows the resolution of the discretization , assumed to be large , is the control parameter , which we denote by @xmath303 . construction of the modified mapping can be avoided under the conditions , that the basic mapping @xmath304 can be split : @xmath305 and if a linear invertible preconditioning operator @xmath306 can be selected , such that : \1 . all eigenvalues @xmath307 of the linear operator @xmath308 are negative . ( if @xmath306 is a self - adjoint positive - definite operator and @xmath309 is a self - adjoint negative - definite operator , this condition is automatically verified . ) the ratio @xmath310 is small for large @xmath303 . here @xmath311 , and @xmath294 is a `` tight '' upper bound for the spectral radius of @xmath312 , i.e. a bound , which stays of the order of the spectral radius for large resolutions . the operators @xmath313 and @xmath314 nearly commute : @xmath315 . ( note , that this condition is not that restrictive : it holds automatically in the absence of preconditioning , if @xmath313 is for example the laplacian . ) a decomposition arises naturally if the problem is a discretization of a dissipative partial differential equation : then the mapping @xmath316 can be identified with the higher - order elliptic part of the equation , and @xmath230 with the remaining terms . auxiliary problems and arising from the two - scale expansion fall into this class . under the conditions stated above a modification of the method discussed in , named , can be applied . solution proceeds by a succession of chebyshev iterative sequences - , but the first iteration in all sequences is made differently . suppose a termination condition ( for example , the one considered in the previous subsection ) for a chebyshev sequence number @xmath317 is verified for the first time in this sequence for the iterate @xmath318 ( we then call @xmath69 the termination signal number ) . then the sequence is continued by @xmath319 further iterations , and we start the next sequence , by selecting @xmath320 subsequent iterations are defined as in : @xmath321 ( the quantities @xmath322 and @xmath265 were defined by - and @xmath281 by . ) in this iterative process , errors behave in the following way . discrepancy vectors @xmath323 can be represented in terms of chebyshev polynomials of the first ( @xmath264 ) and of the second ( @xmath324 ) kind : @xmath325 where @xmath326 , @xmath327 is the identity operator , and vectors @xmath328 and @xmath329 , which we call the basic error vectors , can be determined from @xmath330 and @xmath331 . ( the terms @xmath332 do not emerge if the mapping @xmath274 is linear ; in what follows they are neglected . ) now , suppose that for the ( @xmath333)-nd sequence the termination signal number @xmath69 was the same as for the ( @xmath317)-st iteration ( in computations , successive chebyshev sequences often do have equal lengths ) . then it can be shown that @xmath175 within the succession of iterative sequences , the basic error vectors evolve as follows : @xmath335 where @xmath336 @xmath337 , @xmath307 being eigenvalues of @xmath312 , and the coefficients @xmath338 can be found from the decomposition of @xmath339 in the basis of eigenvectors @xmath340 of @xmath312 : @xmath341 thus , reduction of the basic error vectors @xmath328 is predominantly controlled by @xmath312 ( note that @xmath342 ) . the standard first chebyshev iteration is recovered when one sets @xmath343 and hence @xmath344 . it may be shown that our procedure leads to factors @xmath345 , and thus to enhanced convergence , for eigenvalues sufficiently small in absolute value ( provided @xmath277 or @xmath27 is small ) . for the other eigenvalues good convergence is ensured as usually . ( actually , because of the presence of the non - self - adjoint perturbation @xmath346 , fast convergence is guaranteed only for eigenvalues large in absolute value ; for those of intermediate magnitude slowdown might occur , but has not been observed in any of the cases we studied . ) of course , these convergence arguments hold only in so far as the resolution @xmath303 is high enough and thus @xmath347 is small enough to justify keeping only the leading terms in . in the application of the generalized chebyshev method to the auxiliary problems and , we take @xmath348 ( more precisely , @xmath349 is the fourier galerkin discretization of this operator ) . we select the preconditioning operator @xmath350 for some @xmath351 by analogy with podvigina and zheligovsky ( 1997 ) and podvigina ( 1999a , b ) . then @xmath352 is a pseudodifferential operator of order @xmath353 , and @xmath346 of order @xmath354 . hence , condition 2 is inherited from the non - discretized versions of the operators for @xmath355 . condition 3 is satisfied for @xmath356 . indeed , in the commutator @xmath357 the leading symbols cancel out and therefore its order is by 1 less than the order of each of the operators @xmath358 and @xmath359 ; thus @xmath360 , @xmath361 , where @xmath303 is the maximum wavenumber of the fourier harmonics , retained in the abridged series , this completing the argument . since condition 3 is also trivially satisfied for @xmath362 , it is natural to consider the larger interval @xmath363 , and we find the convergence to be the best for @xmath364 . when applied to and , proved superior to . gailitis a. , lielausis o.a . , dementev s. , platacis e. , cifersons a. , gerbeth g. , gundrum t. , stefani f. , christen m. , hanel h. and will g. `` detection of a flow induced magnetic field eigenmode in the riga dynamo facility '' , _ phys lett . _ * 84 * , 43654368 ( 2000 ) . podvigina o.m . and zheligovsky v.a . `` an optimized iterative method for numerical solution of large systems of equations based on the extremal property of zeroes of chebyshev polynomials '' , _ j. sci . computing _ * 12 * , 433464 ( 1997 ) . steenbeck m. , krause f. , rdler k .- h . `` a calculation of the mean electromotive force in an electrically conducting fluid in turbulent motion , under the influence of coriolis forces '' , _ z. naturforsch _ * 21a * , 369376 ( 1966 ) . stieglitz r. and mller u. `` an experimental demonstration of a homogeneous two scale dynamo '' , _ proceed . nato arw `` dynamo and dynamics , a mathematical challenge '' _ , august 21 - 26 , 2000 , institut dtudes scientifiques de cargse , to appear ( 2001 ) . zheligovsky v.a . `` a chebyshev iterative method with operator splitting for evaluation of roots of large systems of equations '' ( in russian ) , accepted _ proc . international workshop on nonlinear problems of hydrodynamic stability , moscow , 2000_. institute of mechanics , lomonosov moscow state university , moscow univ . press ( 2001 ) . also available at + http://www.obs-nice.fr/etc7/vlad.ps.gz
our technology is based on various electromagnetic phenomena . for designing electronics devices , the maxwell s equation is therefore of essential importance . the mathematical structure of the electromagnetic field is governed by a u(1 ) gauge symmetry , i.e. , an invariance of physical laws under phase transformations . the gauge symmetry is equivalent to the conservation of the electric charge , and was established when a symmetry breaking of unified force occured immediately after the big bang . the beautiful mathematical structure of charge electromagnetism was therefore determined when our universe started , and there is no way to modify its laws . interestingly , charge electromagnetism is not the only electromagnetism allowed in the nature . in fact , electromagnetism arises whenever there is a u(1 ) gauge symmetry associated with conservation of some effective charge . in solids , there are several systems which have the u(1 ) gauge symmetry as a good approximation . solids could thus display several types of effective electromagnetic fields . a typical example is a ferromagnetic metal . in ferromagnetic metals , conduction electron spin ( mostly @xmath0 electron ) is coupled to the magnetization ( or localized spins of @xmath1 electrons ) by an interaction called the @xmath2 interaction , which tends to align the electron spin parallel ( or anti - parallel ) to the localized spin . this interaction is strong in most 3@xmath1 ferromagnetic metals , and as a result , conduction electron s spin originally consisting of three components , reduces to a single component along the localized spin direction . the remaining component is invariant under a phase transformation , i.e. , has a u(1 ) gauge symmetry just like the electric charge does . a spin electromagnetic field thus emerges that couples to conduction electron s spin . interaction with magnetization as it moves in the presence of a magnetization texture , resulting in a spin gauge field . magnetization texture is therefore equivalent to an effective electromagnetic field for conduction electron spin . ] the subject of the present paper is this spin electromagnetic field . spin electromagnetic field drives electron s spin , and thus plays essential roles in spintronics . there is a gauge field for the spin electromagnetic field , a spin gauge field , which couples to spin current of the conduction electron . the gauge coupling describes the effects of spin current on the localized spin dynamics . as we shall see , when a spin - polarized electric current is applied , the adiabatic spin gauge field leads to spin - transfer torque and moves the magnetization structure ( sec . [ secmagdynamics ] ) . the world of spin electromagnetic field is richer than that of electric charge , since the electron s spin in solids is under influence of various interactions such as spin - orbit interaction . we shall show that even magnetic monopoles can emerge ( sec . [ sec : adiabatic ] ) a spin electromagnetic field was first discussed in the context of a voltage generated by a canting of a driven domain wall by l. berger @xcite , and mathematically rigorous formulation was given by g. volovik @xcite . the idea of effective gauge field was shown to be extended to the cases with spin relaxation @xcite , and rashba interaction @xcite . some of the phenomena discussed in this paper overlaps those in the paper by r. raimondi in this lecture series , studied base on the boltzmann equation approach @xcite . ] let us start with a brief introduction of ferromagnetic metals ( fig . [ figfm ] ) . ferromagnets have magnetization , namely , an ensemble of localized spins . denoting the localized spin as @xmath3 , the magnetization is @xmath4 , where @xmath5 and @xmath6 are gyromagnetic ratio and lattice constant , respectively . as the electron has negative charge , the localized spin and magnetization points opposite direction . in @xmath7 transition metals , localized spins are aligned spins of @xmath7 electrons . ferromagnetic metals have finite conductivity , indicating that there are conduction electrons , mainly @xmath8 electrons . the conduction electrons and @xmath1 electrons are coupled via @xmath2 mixing . as a result , there arises an exchange interaction between conduction electron spin , @xmath9 , and localized spin , which reads @xmath10 where @xmath11 represents the strength . in this article , the localized spin is treated as classical variable , neglecting the conduction of @xmath1 electrons . the dynamics of localized spin is described by the landau - lifshiz - gilbert ( llg ) equation , @xmath12 where @xmath13 is a unit vector representing the direction of localized spin , @xmath14 is the total magnetic field acting on the spin . the last term of the right hand side represents the relaxation ( damping ) of localized spin , called the gilbert damping effect and @xmath15 is the gilbert damping constant . the gilbert damping constant in most metallic ferromagnets are of the order of @xmath16 . we shall now start studying phenomena arising from the exchange interaction , eq . ( [ hsd ] ) , between localized spin and conduction electron . , title="fig : " ] , title="fig : " ] we consider a ferromagnetic domain wall , which is a structure where localized spins ( or magnetization ) rotate spatially ( fig . [ figdwelec ] ) . its thickness , @xmath17 , in typical ferromagnets is @xmath18 nm . let us consider here what happens when a conduction electron goes through a domain wall . the wall is a macroscopic object for electrons , since thickness is much larger than the typical length scale of electron , the fermi wavelength , @xmath19 , which is atomic scale in metals . the electron is interacting with localized spin via the @xmath2 exchange coupling , eq . ( [ hsd ] ) . we consider the case of positive @xmath11 , but the sign does not change the scenario . the @xmath2 interaction tends to align parallel the localized spin and conduction electron spin . if localized spin is spatially uniform , therefore , the conduction electron is also uniformly polarized , and electron transport and magnetism are somewhat decoupled . interesting effects arise if the localized spins are spatially varying like the case of a domain wall . we choose the @xmath20 axis along the direction localized spins change . the lowest energy direction ( magnetic easy axis ) for localized spins is chosen as along @xmath20 axis . ( the mutual direction between the localized spin and direction of spin change is irrelevant in the case without spin - orbit interaction . ) the wall in this case is with localized spins inside the wall changing within the plane of localized spin , and such wall is called the nel wall . at @xmath21 the localized spin is @xmath22 , and is @xmath23 at @xmath24 , and those states are represented a @xmath25 and @xmath26 , respectively . for @xmath27 electron , the potential in the left regime is low because of @xmath2 exchange interaction , while that in the right region is high ( dotted lines in fig . [ figdwpotentialenergy ] ) . for conduction electron with spin @xmath25 and @xmath26 as a result of @xmath2 exchange interaction . dotted lines are the cases neglecting spin flip inside the wall , while solid lines are with spin flip . [ figdwpotentialenergy ] ] that is , the localized spin structure due to a domain wall acts as a spatially varying magnetic field , resulting in potential barriers , @xmath28 and @xmath29 . considering the domain wall centered at @xmath30 having profile of @xmath31 conduction electron s schrdinger equation with energy @xmath32 reads @xmath33\psi = e\psi , \label{elecseqindw}\end{aligned}\ ] ] @xmath34 begin the two - component wave function . if the spin direction of the conduction electron is fixed along the @xmath20 axis , the potential barrier represetned by the term proportional to @xmath35 leads to reflection of electron , but in reality , the electron spin can rotate inside the wall as a result of the term proportional to @xmath36 in eq . ( [ elecseqindw ] ) . the mixing of @xmath27 and @xmath37 electron leads to the smooth potential barrier plotted as solid lines in fig . [ figdwpotentialenergy ] . let us consider an incident @xmath27 electron from the left . if the electron is slow , the electron spin can keep the lowest energy state by gradually rotating its direction inside the wall . this is the adiabatic limit . as there is no potential barrier for the electron in this limit , no reflection arises from the domain wall , resulting in a vanishing resistance ( fig . [ figdwelec](a ) ) in contrast , if the electron is fast , the electron spin can not follow the rotation of the localized spin , resulting in a reflection and finite resistance ( fig . [ figdwelec](b ) ) . the condition for slow and fast is determined by the relation between the time for the electron to pass the wall and the time for electron spin rotation . the former is @xmath38 for electron with fermi velocity @xmath39 ( spin - dependence of the fermi wave vector is neglected and @xmath40 is the electron mass ) . the latter time is @xmath41 , as the electron spin is rotated by the @xmath2 exchange interaction in the wall . therefore , if @xmath42 is satisfied , the electron is in the adiabatic limit @xcite . the condition of adiabatic limit here is the case of clean metal ( long mean free path ) ; in dirty metals , it is modified @xcite . the transmission of electron through a domain wall was calculated by g. g. cabrera and l. m. falicov @xcite , and its physical aspects were discussed by l. berger @xcite . linear response formulation and scattering approach were presented in refs . the adiabaticity condition was discussed by x. waintal and m. viret@xcite . as we discussed above , in the adiabatic limit , the electron spin gets rotated after passing through the wall ( fig . [ figdwelec](a ) ) . the change of spin angular momentum , @xmath43 , must be absorbed by the localized spins . ( angular momentum dissipation as a result of spin relaxation is slow compared to the exchange of the angular momentum via the @xmath2 exchange interaction . ) to absorb the spin change of @xmath44 , the domain wall must shift to the right , resulting in an increase of the spins @xmath27 . we consider for simplicity the case of cubic lattice with lattice constant @xmath6 . the distance of the wall shift @xmath45 necessary to absorb the electron s spin angular momentum of @xmath44 is then @xmath46a$ ] ( fig . [ figdwdisplacement ] ) ) . if we apply a spin - polarized current through the wall with the density @xmath47 ( spin current density is defined to have the same unit of a / m@xmath48 as the electric current density . ) the rate of the angular momentum change of the conduction electron per unit time and area is @xmath49 . as the number of the localized spins in the unit area is @xmath50 , the wall must keep moving a distance of @xmath51 per unit time . namely , when a spin current density is applied , the wall moves with the speed of @xmath52 this effect was pointed out by l. berger @xcite in 1986 , and is now called the spin - transfer effect after the papers by j. slonczewski @xcite . results in a change of the spin of the localized spins @xmath53 . the angular momentum change is therefore @xmath44 if @xmath54 . [ figdwdisplacement ] ] from the above considerations in the adiabatic limit , we have found that a domain wall is driven by spin - polarized current , while the electrons do not get reflected and no resistance arises from the wall . these two facts naively seem inconsistent , but are direct consequence of the fact that a domain wall is a composite structure having both linear momentum and angular momentum . the adiabatic limit is the limit where angular momentum is transfered between the electron and the wall , while no linear momentum is transfered . relates the two spin configurations @xmath55 and @xmath56 as @xmath57 . right : the overlap of the wave functions at sites @xmath58 and @xmath59 is @xmath60 . [ figu],title="fig:",scaledwidth=27.0% ] relates the two spin configurations @xmath55 and @xmath56 as @xmath57 . right : the overlap of the wave functions at sites @xmath58 and @xmath59 is @xmath60 . [ figu],title="fig:",scaledwidth=27.0% ] transport of conduction electrons in the adiabatic ( strong @xmath2 ) limit is theoretically studied by calculating the quantum mechanical phase attached to the wave function of electron spin . we here consider a conduction electron hopping from a site @xmath58 to a neighboring site at @xmath61 ( @xmath62 is a vector connecting neighboring sites)(fig . [ figu ] ) . the localized spin direction at those sites are @xmath63 and @xmath64 , respectively , and the electron s wave function at the two sites are @xmath65 , @xmath66 and @xmath67 , @xmath68 are the polar angle of @xmath69 and @xmath70 , respectively ( fig . [ figu ] ) . the wave functions are concisely written by use of matrices , @xmath71 and @xmath72 , which rotates the spin state @xmath55 to @xmath56 ( fig . [ figu ] ) , as @xmath73 and @xmath74 . the rotation matrix is given by @xcite ( neglecting irrelevant phase factors ) @xmath75 the overlap of the electron wave functions at the two sites is thus @xmath76 . when localized spin texture is slowly varying , we can expand the matrix product with respect to @xmath62 as @xmath77 to obtain @xmath78 where @xmath79 since @xmath80 , @xmath81 is real . a vector @xmath82 here plays a role of a gauge field , similarly to that of the electromagnetism , and it is called ( adiabatic ) spin gauge field . by use of eq . ( [ udef ] ) , this gauge field reads ( the factor of @xmath83 represents the magnitude of electron spin ) @xmath84 for a general path @xmath85 , the phase is written as an integral along @xmath85 as @xmath86 existence of path - dependent phase means that there is an effective magnetic field , @xmath87 , as seen by rewriting the integral over a closed path by use of the stokes theorem as @xmath88 where @xmath89 represents the curvature or effective magnetic field . this phase @xmath81 , arising from strong @xmath2 interaction , couples to electron spin , and is called the spin berry s phase . time - derivative of phase is equivalent to a voltage , and thus we have effective electric field defined by @xmath90 where @xmath91 ( for a gauge invariant expression of @xmath92 , we need to include the time component of the gauge field , @xmath93 @xcite . ) in terms of vector @xmath94 the effective fields read @xmath95 , of a hedgehog monopole having a monopole charge of @xmath96 and the one with @xmath97 . at the center , @xmath69 has a singularity and this gives rise to a finite monopole charge . , title="fig:",scaledwidth=30.0% ] , of a hedgehog monopole having a monopole charge of @xmath96 and the one with @xmath97 . at the center , @xmath69 has a singularity and this gives rise to a finite monopole charge . , title="fig:",scaledwidth=30.0% ] these two fields couple to the electron spin and are called spin electromagnetic fields ( @xmath82 is spin gauge field ) . they satisfy the faraday s law , @xmath98 as a trivial result of their definitions . defining the spin magnetic charge as @xmath99 we see that @xmath100 as a local identity , since spin vector with fixed length has only two independent variables , and therefore @xmath101 . however , there is a possibility that the volume integral , @xmath102 , is finite ; in fact , using the gauss s law we can write ( @xmath103 represents a surface integral ) @xmath104 and it follows that @xmath105integer since @xmath106 is a winding number , an integer , of a mapping from a sphere in the coordinate space to a sphere in spin space . if the mapping is topologically non - trivial as a result of a singularity , the monopole charge is finite . typical nontrivial structures of @xmath94 are shown in fig . [ fighh ] . the singular structure with a single monopole charge is called the hedgehog monopole . the faraday s law similarly reads @xmath107 , which vanishes locally but is finite when integrated , indicating that topological monopole current @xmath108 exists . the other two maxwell s equations describing @xmath109 and @xmath110 are derived by evaluating the induced spin density and spin current based on linear response theory @xcite . and spin magnetic field @xmath111 act oppositely for electrons with opposite spin , and thus are useful for generation of spin current . [ figesbs],title="fig : " ] and spin magnetic field @xmath111 act oppositely for electrons with opposite spin , and thus are useful for generation of spin current . [ figesbs],title="fig : " ] the spin electromagnetic fields are real fields detectable in transport measurements . they couples to the spin polarization of the electrons ( fig . [ figesbs ] ) , and because spin density and spin current in ferromagnetic metals is always accompanied with electric charge and current , respectively , the effects of the spin magnetic fields are observable in electric measurements . the electric component @xmath92 is directly observable as a voltage generation from magnetization dynamics , and the voltage signals of @xmath112v order have been observed for the motion of domain walls and vortices @xcite . the spin magnetic field causes an anomalous hall effect of spin , i.e. , the spin hall effect called the topological hall effect . the spin electric field arises if magnetization structure carrying spin magnetic field becomes dynamical due to the lorentz force from @xmath87 according to @xmath113 , where @xmath114 denotes the electron spin s velocity . the topological hall effect due to skyrmion lattice turned out to induce hall resistivity of 4n@xmath115 cm @xcite . although those signals are not large , existence of spin electromagnetic fields is thus confirmed experimentally . it was recently shown theoretically that spin magnetic field couples to helicity of circularly polarized light ( topological inverse faraday effect ) @xcite , and an optical detection is thus possible . as discussed in the previous section , the spin gauge field are measured by transport experiments . here we study the opposite effect , the effects of spin gauge field on magnetization dynamics when spin current is applied . the spin gauge field is expected to couple to the spin current of the electron , @xmath116 , via the minimal coupling , @xmath117 , \label{has}\ ] ] where @xmath118 is the electron density , and @xmath119 is the electron spin density , @xmath120 ( @xmath121 ) representing the density of electron with spin @xmath122 . the field @xmath123 is the time component of spin gauge field ( eq . ( [ asdef ] ) with spatial derivative replaced by time derivative ) . ( for rigorous derivation of the coupling , see eqs . ( [ lezexpression2])([gaugeh ] ) . ) as the spin gauge field is written in terms of localized spin variables , @xmath124 and @xmath66 , as a result of eq . ( [ asdef ] ) , this interaction describes how the spin current and electron density affects the magnetization dynamics . here we study the adiabatic limit , where the contribution second order in @xmath82 ( the second term of the right hand side of eq . ( [ has ] ) is neglected . including the gauge interaction , the lagrangian for the localized spin reads @xmath125 -h_s , \label{ls1 } \end{aligned}\ ] ] where @xmath126 is the hamiltonian . we see that the magnitude of localized spin is modified to be the effective one @xmath127 including the spin polarization of the conduction electron . writing the gauge field terms explicitly , we have @xmath128 -h_s , % -h_{\rm na } \label{lsjstot}\end{aligned}\ ] ] where @xmath129 . the velocity @xmath130 here agrees with the phenomenological one , eq . ( [ vst ] ) , if electron spin polarization is neglected ( i.e. , if @xmath131 ) . in the adiabatic limit , therefore , the time - derivative of the localized spin in the equation of motion is replaced by the galilean invariant form with a moving velocity of @xmath130 when a spin current is present . the equation of motion derived from the lagrangian ( [ lsjstot ] ) reads @xmath132 where @xmath133 is the effective magnetic field due to @xmath126 . from eq . ( [ llmoving ] ) , it is obvious that the magnetization structure flows with velocity @xmath130 , and this effect is in fact the spin - transfer effect discussed phenomenologically in sec . [ secstt ] . it should be noted that the effect is mathematically represented by a simple gauge interaction of eq . ( [ ls1 ] ) . the equation of motion ( [ llmoving ] ) is the landau - lifshiz - gilbert ( llg ) equation including adiabatic spin - transfer effect . it was theoretically demonstrated that the spin - transfer torque induces a red shift of spin wave , resulting in instability of uniform ferromagnetic state under spin - polarized current @xcite . in reality , there is nonadiabatic contribution described by spin - flip interactions . such contribution leads to a mixing of the electron spins resulting in a scattering of the conduction electron and a finite resistance due to the magnetization structure @xcite . this scattering gives rise to a force on the magnetization structure as a counter action @xcite . as we have seen , the concept of adiabatic spin gauge field is useful to give a unified description of both electron transport properties in the presence of magnetization structure and the magnetization dynamics in the presence of spin - polarized current . so far we discussed that an effective spin gauge field emerges by looking into the quantum mechanical phase factor attached to conduction electron in the presence of magnetization structures . existence of effective gauge field is straightforwardly seen in field - theoretic description . a field - theoretical description is based on the lagrangian of the system , @xmath134 where @xmath135 is the field hamiltonian . here @xmath136 describes the free electron part in terms of field operators for conduction electron , @xmath137 and @xmath138 , where @xmath121 denotes spin . the @xmath2 exchange interaction is represented by @xmath139 we are interested in the case where @xmath140 changes in space and time slowly compared to the electron s momentum and energy scales . how the electron feels when flowing through such slowly varying structure is described by introducing a rotating frame where the @xmath2 exchange interaction is locally diagonalized . in sec . [ sec : adiabatic ] , we introduced a unitary matrix @xmath141 , and this matrix is used here to introduce a new electron operator as @xmath142 the new operator @xmath143 describes the low energy dynamics for the case of strong @xmath2 exchange interaction . in fact , the @xmath2 exchange interaction for this electron is diagonalized to be @xmath144 where @xmath145 . instead , the kinetic term for the new electron is modified , because derivative of the electron field is modified as @xmath146 where @xmath147 here @xmath148 is a @xmath149 matrix , whose componets are represented by using pauli matrices as @xmath150 equation ( [ covariantderivative ] ) indicates that the new electron field @xmath143 is interacting with an effective gauge field , @xmath148 . this gauge field has three components , is non - commutative and is called the su(2 ) gauge field . the three components explicitly read @xmath151 due to eq . ( [ covariantderivative ] ) , the kinetic term @xmath152 is written in terms of @xmath143 electron as @xmath153 [ ( \nabla+ i\acals ) \ahat].\ ] ] similarly , time - component of the gauge field @xmath154 arises from the time - derivative term ( @xmath155 ) of the lagrangian ( [ lhat ] ) . the lagrangian in terms of @xmath143 electron therefore reads @xmath156 . \label{lezexpression1}\end{aligned}\ ] ] if we introduce electron density operator , @xmath157 , and operators for spin dnesity and spin current density as @xmath158 , \end{aligned}\ ] ] it reads @xmath159 . \label{lezexpression2}\end{aligned}\ ] ] in the case of @xmath160 ( large @xmath11 ) , the electron with spin @xmath161 has high energy because of strong spin splitting , eq . ( [ hsddiag ] ) , and is neglected . in this case , only the @xmath20 component of the gauge field , @xmath162 , survives . this component is thus essentially a u(1 ) gauge field , which coincides with the u(1 ) gauge field we have obtained from the argument of electron s phase factor , namely , @xmath163 . the total hamiltonian in the limit of large @xmath164 therefore reduces to the one for a charged particle in the presence of a u(1 ) gauge field @xmath82 ; @xmath165 [ ( \nabla+ i\asv ) \ahat_\uparrow ] -\frac{j_{sd}s}{2 } \ahat_\uparrow^\dagger \ahat_\uparrow \rt ] . \label{gaugeh}\ ] ] the field - theoretic method present here is highly useful , as it leads to a conclusion of the existence of an effective gauge field for spin simply by carrying out a unitary transformation to diagonalize strong @xmath2 exchange interaction . in reality , there is a deviation from the adiabatic limit we have considered so far . one origin is the fact that the magnetization structure is not in the slowly varying limit , but has a finite length scale of spatial modulation . this effect , we call the non adiabaticity , leads to reflection of conduction electron by magnetization structures as in fig . [ figdwelec](b ) , resulting in a force on the magnetization structure when an electric current is applied @xcite . in terms of torque , the effect of the force due to reflection is represented by a non - local torque , as it arises from finite momentum transfer @xcite . another effect we need to take into account is the relaxation ( damping ) of spin schematically shown in the fig . [ fignonadiabatic ] . in metallic ferromagnets , the damping mostly arises from the spin - orbit interaction , as seen from the fact that the gilbert damping parameter @xmath15 and the @xmath166 value has a correlation of @xmath167 as shown in ref . spin relaxation generates a torque perpendicular to the motion of the spin , resulting in a canting of the precession axis . similarly , when a spin current @xmath168 is applied , the spin relaxation thus was argued to induce a torque perpendicular to the spin - transfer torque , i.e. , @xmath169 where @xmath170 is a coefficient representing the effect of spin relaxation @xcite . ] those effects of non adiabaticity and spin relaxation can be calculated from a microscopic viewpoint @xcite . let us go back to the llg equation for localized spin interacting with conduction electron spin via the @xmath2 exchange interaction . the total hamiltonian is @xmath171 , where @xmath126 and @xmath172 are the hamiltonian for localized spin and conduction electron , respectively . the equation of motion for localized spin is given by @xmath173 where @xmath174 and @xmath175 are the effective magnetic field arising from the localized spin and conduction electron , respectively . the field @xmath176 is represented by the expectation value of electron spin density , @xmath177 , and all the effects from the conduction electron is included in this field ; equation ( [ llgfromh ] ) is exact if @xmath177 is evaluated exactly . field theoretic approach is suitable for a systematic evaluation of the electron spin density . we move to a rotated frame where the electron spin is described choosing the local @xmath20 axis along the localized spin . in the case we are interested , namely , when the effect of non adiabaticity and damping are weak , these effects are treated perturbatively . the spin density in the laboratory frame is written in terms of the spin in the rotated frame @xmath178 as @xmath179 , where @xmath180 is a rotation matrix , @xmath181 being the vector which define the unitary rotation . the perpendicular components ( denoted by @xmath182 ) of electron spin density in the rotated frame are calculated as @xcite @xmath183 the effect of spin relaxation is included in @xmath184 and @xmath185 , both proportional to the spin relaxation time , @xmath186 @xcite . the first term of eq . ( [ serotresult ] ) represents the renormalization of the localized spin as a result of electron spin polarization and the second term , induced in the presence of applied spin current , describes the adiabatic spin - transfer torque . using the identity @xmath187 we see that eq . ( [ serotresult ] ) leads to @xmath188 + \gamma \bv_{s}\times \nv , \label{llgfromh2}\end{aligned}\ ] ] which is the llg equation taking into account the torque due to electrons . here @xmath189 and @xmath190 , neglecting other origins for gilbert damping and nonadiabatic torque . let us briefly discuss dynamics of a domain wall based on the llg equation ( [ llgfromh2 ] ) including the current - induced torques . we consider an one - dimensional and rigid wall , neglecting deformation . for a domain wall to be created , the system must have an easy axis magnetic anisotropy energy . we also include the hard - axis anisotropy energy , which turns out to govern the domain wall motion . choosing the easy and the hard axises along the @xmath20 and the @xmath191 directions , respectively , the anisotropy energy is represented by the hamiltonian @xmath192,\end{aligned}\ ] ] where @xmath193 and @xmath194 are the easy- and hard - axis anisotropy energies ( both are positive ) . we need to take into account of course the exchange coupling , which is essential for ferromagnetism , which in the continuum expression reads @xmath195 the domain wall solution obtained by minimizing @xmath196 and @xmath197 is eq . ( [ dwsolrest ] ) with @xmath198 . considering a rigid wall , we assume that @xmath199 . the low energy dynamics of the wall is then described by two variables ( called the collective coordinates ) , the center coordinate of the wall , @xmath200 , and the angle @xmath201 out - of the easy plane @xcite . the wall profile including the collective coordinates is @xmath202 the equation of motion for domain wall is obtained by putting the wall profile ( [ dwsol ] ) in eq . ( [ llgfromh2 ] ) and integrating over spatial coordinate as @xmath203 where @xmath204 is spin polarization of the current , and both @xmath205 and @xmath206 have dimension of velocity . when @xmath207 , the wall velocity when a constant @xmath208 is applied is easily obtained as @xcite @xmath209 and @xmath210 is the intrinsic threshold current density @xcite . namely , the wall can not move if the applied current is lower than the threshold value . this is because the torque supplied by the current is totally absorbed by the wall by tilting the out of plane angle to be @xmath211 when the current is weak ( @xmath212 ) and thus the wall can not move . this effect is called the intrinsic pinning effect @xcite . for larger current density , the torque carried by the current induces an oscillation of the angle similar to the walker s breakdown in an applied magnetic field , and the wall speed also becomes an oscillating function of time . when nonadiabaticity parameter @xmath170 is finite , the behavior changes greatly and intrinsic pinning effect is removed and the wall can move with infinitesimal applied current as long as there is no extrinsic pinning . in fact , when the applied current density is @xmath213 , where @xmath214 the solution of eq . ( [ eqs ] ) is an oscillating function given by @xcite @xmath215 where @xmath216 the time - average of the wall speed is @xmath217 for current density satisfying @xmath218 , the oscillation in eq . ( [ dwvelocity1 ] ) is replaced by an exponential decay in time , and the wall velocity reaches a terminal value of @xmath219 the angle of the wall also reaches a terminal value determined by @xmath220 the averaged wall speed ( eq . ( [ averagewallvelocity ] ) ) is plotted in fig . [ figvj ] . as function of applied spin - polarized current @xmath221 for @xmath222 . intrinsic pinning threshold @xmath223 exists only for @xmath224 . the current density where derivative of @xmath225 is discontinuous corresponds to @xmath226 . [ figvj ] ] the intrinsic pinning is a unique feature of current - driven domain wall , as the wall can not move even in the absence of pinning center . in the unit of a / m@xmath48 , the intrinsic pinning threshold is @xmath227 for device applications , this threshold needs to be lowered by reducing the hard - axis anisotropy and wall width @xcite . at the same time , the intrinsic pinning is promising for stable device operations . in fact , in the intrinsic pinning regime , the threshold current and dynamics is insensitive to extrinsic pinning and external magnetic field@xcite , as was confirmed experimentally @xcite . this is due to the fact that the wall dynamics in the intrinsic pinning regime is governed by a torque ( right hand side of the second equation of eq . ( [ eqs ] ) ) , which governs the wall velocity @xmath228 , while pinning and magnetic field induce force , which governs @xmath229 ; the forces due to sample irregularity therefore does not modify the motion induced by a torque in the intrinsic pinning regime . experimentally , intrinsic pinning is observed in perpendicularly magnetized materials @xcite , perhaps due to relatively low intrinsic pinning threshold , while materials with in - plane magnetization mostly are in the extrinsic pinning regime governed by the nonadiabatic parameter @xmath170 and extrinsic pinning . in this regime , the threshold current of the wall motion is given by @xcite @xmath230 where @xmath231 represents strength of extrinsic pinning potential like those generated by geometrical notches and defects . control of nonadiabaticity parameter is therefore expected to be useful for driving domain walls at low current density . of recent interest from the viewpoint of low current operation is to use multilayer structures . for instance , heavy metal layers turned out to lower the threshold current by exerting a torque as a result of spin hall effect @xcite , and synthetic antiferromagnets turned out to be suitable for fast domain wall motion at low current @xcite . physics tends to focus on infinite systems or bulk system approximated as infinite , as one of the most important objective of physics is to search for beautiful general law supported by symmetries . in the condensed matter physics today , studying such beautiful systems seems to be insufficient anymore . this is because demands to understand physics of interfaces and surfaces has been increasing rapidly as present devices are in nanoscales to meet the needs for fast processing of huge data . systems with lower symmetry are therefore important subjects of material science today . surfaces and interfaces have no inversion symmetry , and this leads to emergence of an antisymmetric exchange interaction ( dzyaloshinskii - moriya interaction ) @xcite in magnetism . as for electrons , broken inversion symmetry leads to a peculiar spin - orbit interaction , called the rashba interaction @xcite , whose hamiltonian is @xmath232 where @xmath233 is the vector of pauli matrices and @xmath234 is a vector representing the strength and direction of the interaction . the form of the interaction is the one derived directly from the dirac equation as a relativistic interaction , but the magnitude can be strongly enhanced in solids having heavy elements compared to the vacuum case . as is obvious from the form of the hamiltonian , the rashba interaction induces electromagnetic cross correlation effects where a magnetization and an electric current are induced by external electric and magnetic field , @xmath235 and @xmath14 , respectively , like represented as @xmath236 where @xmath237 and @xmath238 are coefficients , which generally depend on frequency . the emergence of spin accumulation from the applied electric field , mentioned in ref . @xcite , was studied by edelstein @xcite in detail , and the effect is sometimes called edelstein effect . the generation of electric current by magnetic field or magnetization , called the inverse edelstein effect @xcite , was recently observed in multilayer of ag , bi and a ferromagnet @xcite . equation ( [ rashbah ] ) indicates that when a current density @xmath239 is applied , the conduction electron has average momentum of @xmath240 ( @xmath118 is electron density ) , and thus an effective magnetic field of @xmath241 acts on the conduction electron spin ( @xmath242 is the gyromagnetic ratio ) . when the @xmath2 exchange interaction between the conduction electron and localized spin is strong , this field multiplied by the the spin polarization , @xmath243 , is the field acting on the localized spin . namely , the localized spin feels a current - induced effective magnetic field of @xmath244 one may argue more rigorously using field theoretic description . considering the case of @xmath2 exchange interaction stronger than the rashba interaction , we use a unitary transformation to diagonalize the @xmath2 exchange interaction ( eq . ( [ ucdef ] ) ) . the rashba interaction in the field representation then becomes @xmath245 where @xmath246 is the spin current in the rotated frame , @xmath247 is given in eq . ( [ rij ] ) . terms containing spatial derivatives of magnetization structure is neglected , considering the slowly - varying structure . in this adiabatic limit , spin current is polarized along the @xmath20 direction , i.e. , @xmath248 . we therefore obtain using @xmath249 , @xmath250 which results in the same expression as eq . ( [ br ] ) . the strength of the rashba - induced magnetic field is estimated ( choosing @xmath251 ) as @xmath252(jm)@xmath116(a / m@xmath48 ) ; for a strong rashba interaction @xmath253 ev like at surfaces @xcite , @xmath254 t at @xmath255 a / m@xmath48 . this field appears not very strong , but is sufficient at modify the magnetization dynamics . in fact , for the domain wall motion , when the rashba - induced magnetic field is along the magnetic easy axis , the field is equivalent to that of an effective @xmath170 parameter of @xmath256 where @xmath17 is the wall thickness . if @xmath253 ev , @xmath257 becomes extremely large like @xmath258 for @xmath259 nm . note that @xmath170 arising from spin relaxation is the same order as gilbert damping constant , namely of the order of @xmath16 . such a large effective @xmath170 is expected to leads to an extremely fast domain wall motion under current @xcite . experimentally , it was argued that fast domain wall motion observed in pt / co / alo was due to the rashba interaction @xcite , but the result is later associated with the torque generated by spin hall effect in pt layer @xcite . it was recently shown theoretically that strong rashba - induced magnetic field works as a strong pinning center when introduced locally , and that this rashba pinning effect is useful for highly reliable control of domain walls in racetrack memories @xcite . since the interaction ( [ hrrot ] ) is the one coupling to the spin current , the rashba interaction is regarded as a gauge field acting on electron spin as far as the linear order concerns . the gauge field defined by eq . ( [ hrrot ] ) is @xmath260 existence of a gauge field naturally leads to an effective electric and magnetic field @xcite @xmath261 in the presence of electron spin relaxation , the electric field has a perpendicular component @xcite @xmath262 , \label{erp}\ ] ] where @xmath257 is a coefficient representing the strength of spin relaxation . for the case of strong rashba interaction of @xmath263 ev , as realized in bi / ag , the magnitude of the electric field is @xmath264kv / m if the angular frequency @xmath265 of magnetization dynamics is 10 ghz . the magnitude of relaxation contribution is @xmath266v / m if @xmath267 . the effective magnetic field in the case of spatial length scale of 10 nm is high as well ; @xmath268 t . generated by magnetization precession . electric current @xmath239 is induced as a result of motive force @xmath269 in the direction perpendicular to both @xmath270 and rashba field @xmath234 . [ figes_alpha ] ] the rashba - induced electric fields , @xmath271 and @xmath269 , are important from the viewpoint of spin - charge conversion . in fact , results ( [ erbr])([erp ] ) indicates that a voltage is generated by a dynamics magnetization if the rashba interaction is present , even in the case of spatially uniform magnetization , in sharp contrast to the conventional adiabatic effective electric field from the spin berry s phase of eq . ( [ esbsdef ] ) . in the case of a think film with rashba interaction perpendicular to the plane and with a precessing magnetization , the component @xmath272 has no dc component , while the relaxation contribution @xmath269 has a dc component perpendicular to @xmath273 . the geometry of this ( spin - polarized ) current pumping effect , @xmath274 , is therefore the same as the one expected in the case of inverse edelstein effect ( fig . [ figes_alpha ] ) . in the present form , there is a difference between the rashba - induced electric field effect and the system in ref . @xcite , that is , the former assumes a direct contact between the rashba interaction and magnetization while they are separated by a ag spacer in ref . it is expected , however , that the rashba - induced electric field becomes long - ranged and survives in the presence of a spacer if we include the electron diffusion processes . the spin - charge conversion observed in junctions will then be interpreted as due to the rashba - induced electromagnetic field . for this scenario to be justified , it is crucial to confirm the existence of magnetic component , @xmath275 , which can be of the order of 100 t . in the setup of fig . [ figes_alpha ] , @xmath275 is along @xmath276 . the field can therefore be detected by measuring `` giant '' in - plane spin hall effect when a current is injected perpendicular to the plane . the idea of effective gauge field is useful for extending the discussion to include other degrees of freedom , like optical properties . in fact , the fact that the rashba interaction coupled with magnetization leads to an effective vector potential @xmath277 ( eq . ( [ avrdef ] ) ) for electron spin indicates that the existence of intrinsic spin flow . such intrinsic flow affects the optical properties , as incident electromagnetic waves get doppler shift when interacting with flowing electrons , resulting in a transmission depending on the direction ( directional dichroism ) , as was theoretically demonstrated in refs . the magnitude of the directional dichroism for the case of wave vector @xmath278 is given by @xmath279 . the vector @xmath280 is called in the context of multiferroics the toroidal moment , and it was argued to acts as an effective vector potential for light @xcite . . edelstein effect ( e ) generates spin density , @xmath281 , from the applied electric field , and inverse edelstein effect ( ie ) generates current @xmath282 from magnetization @xmath283 . [ figeie ] ] it was shown also that rashba conductor , even without magnetization , shows peculiar optical properties such as negative refraction as a result of spin - charge mixing effects @xcite . in fact , spin - charge mixing effects of eq . ( [ spinchargemixing ] ) leads to a current generated by applied electric field , @xmath235 , given by @xmath284 , \end{aligned}\ ] ] where @xmath285 is a coefficient ( fig . [ figeie ] ) . as it is opposite to the applied field , the mixing effect results in a softening of the plasma frequency as for the @xmath235 having components perpendicular to @xmath234 . the electric permitivity of the system is therefore anisotropic ; choosing @xmath234 along the @xmath20 axis , we have @xmath286 where @xmath287 is the bare plasma frequency ( @xmath288 is the electron density ) , and @xmath289 is the plasma frequency reduced by the spin mixing effect . @xcite ( @xmath290 represents the correlation function representing the rashba - edelstein effect , and its real part is negative . ) the frequency region @xmath291 is of interest , as the system is insulating ( @xmath292 ) in the direction of the rashba field but metallic in the perpendicular direction ( @xmath293 ) . the dispersion in this case becomes hyperbolic , and the group velocity and phase velocity along @xmath278 can have opposite direction , resulting in negative refraction . rashba system is , therefore a natural hyperbolic metamaterial @xcite . a great advantage of rashba conductors are that the metamaterial behavior arises in the infrared or visible light region , which is not easily accessible in fabricated systems . for instance , in the case of bitei with rashba splitting of @xmath294 ev@xcite , the plasma frequency is @xmath295 hz ( corresponding to a wavelength of @xmath296 m ) for @xmath297 m@xmath298 and @xmath299 ev @xcite . we then have @xmath300 ( @xmath301 hz , corresponding to the wavelength of @xmath302 m ) , and hyperbolic behavior arises in the infrared regime . the directional dichroism arises in the infrared - red light regime @xcite . another interesting effect of spin gauge field pointed out recently is to induce the dzyaloshinskii - moriya interaction . dzyaloshinskii - moriya ( dm ) interaction is an antisymmetric exchange interaction between magnetic atoms that can arise when inversion symmetry is broken . in the continuum limit , it is represented as @xmath303 where @xmath304 is the strength , @xmath15 and @xmath305 denotes the spin and spatial direction , respectively . it was recently discussed theoretically that the interaction is a result of doppler shift due to an intrinsic spin current generated by broken inversion symmetry @xcite . in fact , spin current density , @xmath116 , which is odd and even under spatial inversion and time - reversal , respectively , is induced by spin - orbit interaction in systems with broken inversion symmetry . spatial variation of localized spins observed by the flowing electron spin is then described by a covariant derivative , @xmath306 where @xmath307 is a coefficient . this covariant derivative leads to the magnetic energy generated by the electron of @xmath308 . we see that the second term proportional to @xmath116 is the dm interaction , and thus the coefficient is @xmath309 . more rigorous derivation is performed by deriving an effective hamiltonian . the electrons interacting strongly with localized spin is described by a lagrangian ( [ lezexpression2 ] ) , where @xmath310 is an su(2 ) gauge field describing the spatial and temporal variation of localized spin . to discuss dm interaction , we include a spin - orbit interaction with broken inversion symmetry , @xmath311 c , \label{hso}\end{aligned}\ ] ] where @xmath312 is a vector representing the broken inversion symmetry . ( multiorbital cases are treated similarly @xcite . ) as is obvious from this form linear in spatial derivative and pauli matrix , the spin - orbit interaction generates a spin current proportional to @xmath312 . from eq . ( [ lezexpression2 ] ) , the effective lagrangian for localized spin to the linear order in derivative is @xmath313 where @xmath314 is the expectation value of the spin current density in the rotated frame . in terms of the spin current in the laboratory frame , @xmath315 , the effective hamiltonian reads @xmath316 where @xmath317 and @xmath318 is a component of @xmath315 perpendicular to the local magnetization direction , @xmath94 . we therefore see that the dm coefficient is indeed given by the expectation value of the spin current density of the conduction electrons . the first principles calculation based on this spin current expression turns out to have advantage of shorter calculation time than previous methods@xcite by evaluating twist energy of magnetization @xcite . it has been noted that spin wave dispersion is modified in the presence of dm interaction , resulting in doppler shift of spin waves @xcite . the spin wave doppler shift is natural from our physical interpretation of dm interaction , as dm interaction itself is a consequence of flowing electron spin current . we have discussed various magnetic and electron transport properties in metallic ferromagnets from the view points of effective gauge field . the concept of gauge field turned out to be highly useful to describe novel electromagnetic cross correlation effects and optical properties .
there is an increasing interest in the study of dynamic systems of fractional ( where `` fractional '' actually means `` non - integer '' ) order . extending derivatives and integrals from integer to non - integer order has a firm and longstanding theoretical foundation . leibniz mentioned this concept in a letter to lhopital over three hundred years ago . following lhopital s and leibniz s first inquisition , fractional calculus was primarily a study reserved to the best minds in mathematics . fourier , euler , and laplace are among the many that contributed to the development of fractional calculus . along the history , many found , using their own notation and methodology , definitions that fit the concept of a non - integer order integral or derivative . the most famous of these definitions among mathematicians , that have been popularized in the literature of fractional calculus , are the ones of riemann liouville and grunwald letnikov . on the other hand , the most intriguing and useful applications of fractional derivatives and integrals in engineering and science have been found in the past one hundred years . in some cases , the mathematical notions evolved in order to better meet the requirements of physical reality . the best example of this is the caputo fractional derivative , nowadays the most popular fractional operator among engineers and applied scientists , obtained by reformulating the `` classical '' definition of riemann liouville fractional derivative in order to be possible to solve fractional initial value problems with standard initial conditions @xcite . particularly in the last decade of the xx century , numerous applications and physical manifestations of fractional calculus have been found . fractional differentiation is nowadays recognized as a good tool in various different fields : physics , signal processing , fluid mechanics , viscoelasticity , mathematical biology , electrochemistry , chemistry , economics , engineering , and control theory ( see , , @xcite ) . the fractional calculus of variations was born in 1996 with the work of riewe @xcite , and is nowadays a subject under strong current research ( see @xcite and references therein ) . the fractional calculus of variations extends the classical variational calculus by considering fractional derivatives into the variational integrals to be extremized . this occurs naturally in many problems of physics and mechanics , in order to provide more accurate models of physical phenomena ( see , , @xcite ) . the aims of this paper are twofold . firstly , we extend the notion of caputo fractional derivative to the fractional derivative @xmath0 , which is a convex combination of the left caputo fractional derivative of order @xmath1 and the right caputo fractional derivative of order @xmath2 . this idea goes back at least as far as @xcite , where based on the riemann liouville fractional derivatives , the symmetric fractional derivative was introduced . approach @xcite is obtained in our framework as a particular case , by choosing parameter @xmath3 to be @xmath4 . although the symmetric fractional derivative of riemann liouville introduced by klimek is a useful tool in the description of some nonconservative models , this type of differentiation does not seems suitable for all kinds of variational problems . indeed , the hypothesis that admissible trajectories @xmath5 have continuous symmetric fractional derivatives implies that @xmath6 ( @xcite ) . therefore , the advantage of the fractional caputo - type derivative @xmath0 here introduced lies in the fact that using this derivative we can describe a more general class of variational problems . it is also worth pointing out that the fractional derivative @xmath0 allows to generalize the results presented in @xcite . our second aim is to introduce the subject of multiobjective fractional variational problems . this seems to be a completely open area of research , never considered before in the literature . knowing the importance and relevance of multiobjective problems of the calculus of variations in physics , engineering , and economics ( see @xcite and the references given there ) , and the usefulness of fractional variational problems , we trust that the results now obtained will open interesting possibilities for future research . main results of the the paper provide methods for identifying pareto optimal solutions . necessary and sufficient pareto optimality conditions are obtained by converting a multiobjective fractional variational problem into a single or a family of single fractional variational problems with an auxiliary scalar functional , possibly depending on some parameters . the paper is organized as follows . section [ sec2 ] presents some preliminaries on fractional calculus , essentially to fix notations . in section [ main:1 ] we introduce the fractional derivative @xmath0 and provide the necessary concepts and results needed in the sequel . our main results are stated and proved in section [ ssec : pro ] and section [ main:2 ] . the fractional variational problems under our consideration are formulated in terms of the fractional derivative @xmath0 . we discuss the fundamental concepts of a variational calculus such as the the euler lagrange equations for the elementary ( subsection [ ssec : el ] ) , isoperimetric ( subsection [ sec : iso ] ) , and lagrange ( subsection [ sec : lagr ] ) problems , as well as sufficient optimality ( subsection [ ssec : suf ] ) and transversality ( subsection [ sec : tran ] ) conditions . section [ main:2 ] deals with the multiobjective fractional variational calculus . we present pareto optimality conditions ( subsection [ sec : par : op ] ) and examples illustrating our results ( subsection [ sec : ex ] ) . in this section we review the necessary definitions and facts from fractional calculus . for more on the subject we refer the reader to the books @xcite . let @xmath7)$ ] and @xmath8 . we begin by defining the left and the right riemann liouville fractional integrals ( rlfi ) of order @xmath1 of a function @xmath9 . the left rlfi is given by @xmath10,\ ] ] and the right rlfi by @xmath11,\ ] ] where @xmath12 represents the gamma function , , @xmath13 moreover , @xmath14 if @xmath9 is a continuous function . the left and the right riemann liouville derivatives are defined with the help of the respective fractional integrals . the left riemann liouville fractional derivative ( rlfd ) is given by @xmath15,\ ] ] and the right rlfd by @xmath16.\ ] ] let @xmath17)$ ] , where @xmath18)$ ] represents the space of absolutely continuous functions on @xmath19 $ ] . then the caputo fractional derivatives are defined as follows : the left caputo fractional derivative ( cfd ) by @xmath20,\ ] ] and the right cfd by @xmath21,\ ] ] where @xmath1 is the order of the derivative . the operators are obviously linear . we now present the rule of fractional integration by parts for rlfi ( see , , @xcite ) . let @xmath8 , @xmath22 , @xmath23 , and @xmath24 . if @xmath25)$ ] and @xmath26)$ ] , then @xmath27 in the discussion to follow , we will also need the following formulae for fractional integrations by parts : @xmath28 they can be derived using equations , the identity , and performing integration by parts . let @xmath30 and @xmath31 $ ] . we define the fractional derivative operator @xmath0 by @xmath32 which acts on @xmath17)$ ] in the expected way : @xmath33 note that @xmath34 and @xmath35 . the operator is obviously linear . using equations it is easy to derive the following rule of fractional integration by parts for @xmath0 : @xmath36^{x = b}_{x = a}\\ + ( 1-\gamma)\left[-f(x){{_ai_x^{1-\beta}}}g(x)\right]^{x = b}_{x = a } + \int_a^b f(x){d^{\beta,\alpha}_{1-\gamma}}g(x)dx,\end{gathered}\ ] ] where @xmath37 . let @xmath38 and @xmath39:[a , b]\rightarrow { \mathbb{r}}^{n}$ ] with @xmath40)$ ] , @xmath41 ; @xmath42 with @xmath43 and @xmath44 $ ] , @xmath41 . then , @xmath45.\ ] ] let @xmath46 denote the set of all functions @xmath47\rightarrow { \mathbb{r}}^{n}$ ] such that @xmath48 exists and is continuous on the interval @xmath19 $ ] . we endow @xmath46 with the following norm : @xmath49 where @xmath50 is a norm in @xmath51 . along the work we denote by @xmath52 , @xmath53 ( @xmath54 ) , the partial derivative of function @xmath55 with respect to its @xmath56th argument . let @xmath57 . for simplicity of notation we introduce the operators @xmath58^{\alpha,\beta}_{\gamma}$ ] and @xmath59 by @xmath60^{\alpha,\beta}_{\gamma}(x ) & : = \left(x,\mathbf{y}(x),{{^cd}^{\alpha,\beta}_{\gamma}}\mathbf{y}(x)\right ) \ , , \\ _ { \lambda}\{\mathbf{y}\}^{\alpha,\beta}_\gamma(x ) & : = \left(x,\mathbf{y}(x),{{^cd}^{\alpha,\beta}_{\gamma}}\mathbf{y}(x),\lambda_1,\ldots,\lambda_r\right ) \ , . \end{split}\ ] ] we are concerned with the problem of finding the minimum of a functional @xmath61 , where @xmath62 is a subset of @xmath46 . the formulation of a problem of the calculus of variations requires two steps : the specification of a performance criterion , and the statement of physical constraints that should be satisfied . the performance criterion @xmath63 , also called cost functional ( or objective ) , must be specified for evaluating quantitatively the performance of the system under study . we consider the following cost : @xmath64^{\alpha,\beta}_{\gamma}(x ) \ , dx,\ ] ] where @xmath65 $ ] is the independent variable ; @xmath66 is a real vector variable , the functions @xmath67 are generally called trajectories or curves ; @xmath68 stands for the fractional derivative of @xmath69 ; and @xmath70\times\mathbb{r}^{2n } ; \mathbb{r})$ ] is the lagrangian . enforcing constraints in the optimization problem reduces the set of candidate functions and leads to the following definition . a trajectory @xmath71 is said to be an admissible trajectory , provided it satisfies all the constraints of the problem along the interval @xmath19 $ ] . the set of admissible trajectories is defined as @xmath72 . we now define what is meant by a minimum of @xmath63 on @xmath73 . a trajectory @xmath74 is said to be a local minimizer for @xmath63 on @xmath73 , if there exists @xmath75 such that @xmath76 for all @xmath77 with @xmath78 . the concept of variation of a functional is central to the solution of problems of the calculus of variations . the first variation of @xmath63 at @xmath71 in the direction @xmath79 is defined as @xmath80 provided the limit exists . a direction @xmath79 , @xmath81 , is said to be an admissible variation for @xmath63 at @xmath77 if * @xmath82 exists ; and * @xmath83 for all sufficiently small @xmath84 . the following well known result offers a necessary optimality condition for the problems of the calculus of variations , based on the concept of variations . [ nesse_con ] let @xmath63 be a functional defined on @xmath73 . suppose that @xmath67 is a local minimizer for @xmath63 on @xmath73 . then , @xmath85 for each admissible variation @xmath86 at @xmath67 . let us begin with the following fundamental problem : @xmath87^{\alpha,\beta}_{\gamma}(x ) \ , dx \longrightarrow \min\\\ ] ] over all @xmath71 satisfying the boundary conditions @xmath88 where @xmath89 are given . the next theorem gives the fractional euler lagrange equation for the problem . [ theo e - l1 ] let @xmath90 be a local minimizer to problem . then , @xmath67 satisfies the following system of @xmath91 fractional euler lagrange equations : @xmath92^{\alpha,\beta}_{\gamma}(x ) + { d^{\beta_{i-1},\alpha_{i-1}}_{1-\gamma_{i-1}}}\partial_{n+i } l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)=0 , \quad i=2,\ldots n+1,\ ] ] for all @xmath93 $ ] . suppose that @xmath67 is a local minimizer for @xmath63 . let @xmath86 be an arbitrary admissible variation for problem , , @xmath94 , @xmath41 . based on the differentiability properties of @xmath95 and theorem [ nesse_con ] , a necessary condition for @xmath67 to be a local minimizer is given by @xmath96 that is , @xmath97^{\alpha,\beta}_{\gamma}(x ) h_{i-1}(x ) + \sum_{i=2}^{n+1}\partial_{n+i}l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x){^cd^{\alpha_{i-1},\beta_{i-1}}_{\gamma_{i-1}}}h_{i-1}(x)\biggr]dx=0.\ ] ] using formulae of integration by parts in the second term of the integrand function , we get @xmath98^{\alpha,\beta}_{\gamma}(x)+{d^{\beta_{i-1},\alpha_{i-1}}_{1-\gamma_{i-1}}}\partial_{n+i } l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)\right]h_{i-1}(x)dx\\ + \gamma\left.\left[\sum_{i=2}^{n+1}h_{i-1}(x)\left({{_xi_b^{1-\alpha_{i-1}}}}\partial_{n+i}l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)\right)\right]\right|^{x = b}_{x = a}\\ -(1-\gamma)\left.\left[\sum_{i=2}^{n+1}h_{i-1}(x)\left({{_ai_x^{1-\beta_{i-1}}}}\partial_{n+i}l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)\right)\right]\right|^{x = b}_{x = a}=0.\end{gathered}\ ] ] since @xmath94 , @xmath41 , by the fundamental lemma of the calculus of variations we deduce that @xmath99^{\alpha,\beta}_{\gamma}(x ) + { d^{\beta_{i-1},\alpha_{i-1}}_{1-\gamma_{i-1}}}\partial_{n+i}l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)=0 , \quad i=2,\ldots , n+1,\ ] ] for all @xmath93 $ ] . observe that if @xmath1 and @xmath2 go to @xmath100 , then @xmath101 can be replaced with @xmath102 and @xmath103 with @xmath104 ( see @xcite ) . thus , if @xmath105 or @xmath106 , then for @xmath107 we obtain a corresponding result in the classical context of the calculus of variations ( see , , ( * ? ? ? * proposition 6.1 ) ) . let @xmath108 . assume that @xmath109 , @xmath110 , @xmath41 , @xmath111 , but @xmath112 is free . then , @xmath113 is free and by equations and we obtain @xmath114^{\alpha,\beta}_{\gamma}(x ) \left.-(1-\gamma ) { _ ai_x^{1-\beta_l}}\partial_{n+1+l } l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)\bigr]\right|_{x = b}=0,\ ] ] where @xmath115 . let us consider now the case when @xmath109 , @xmath110 , @xmath41 , @xmath111 , and @xmath112 is free but restricted by a terminal condition @xmath116 . then , in the optimal solution @xmath67 , we have two possible types of outcome : @xmath117 or @xmath118 . if @xmath117 , then there are admissible neighboring paths with terminal value both above and below @xmath112 , so that @xmath113 can take either sign . therefore , the transversality conditions is @xmath119^{\alpha,\beta}_{\gamma}(x ) \left.-(1-\gamma ) { _ ai_x^{1-\beta_l}}\partial_{n+1+l } l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)\bigr]\right|_{x = b}=0\ ] ] for @xmath120 . the other outcome @xmath118 only admits the neighboring paths with terminal value @xmath121 . assuming , without loss of generality , that @xmath122 , this means that @xmath123 . hence , the transversality condition , which has it root in the first order condition , must be changed to an inequality . for a minimization problem , the @xmath124 type of inequality is called for , and we obtain @xmath125^{\alpha,\beta}_{\gamma}(x ) \left.-(1-\gamma ) { _ ai_x^{1-\beta_l}}\partial_{n+1+l } l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)\bigr]\right|_{x = b } \leq 0\ ] ] for @xmath118 . combining and , we may write the following transversality condition for a minimization problem : @xmath126^{\alpha,\beta}_{\gamma}(x ) \left.-(1-\gamma ) { _ ai_x^{1-\beta_l}}\partial_{n+1+l } l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)\bigr]\right|_{x = b } \leq 0 , \quad y_l(b)\leq y^{b}_l,\\ ( y_l(b)-y^{b}_l ) \bigl[\gamma { _ xi_b^{1-\alpha_l } } \partial_{n+l+1}l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x ) \left.-(1-\gamma ) { _ ai_x^{1-\beta_l}}\partial_{n+1+l } l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)\bigr]\right|_{x = b}=0.\end{gathered}\ ] ] let us consider now the isoperimetric problem that consists of minimizing over all @xmath71 satisfying @xmath127 isoperimetric constraints @xmath128^{\alpha,\beta}_{\gamma}(x)dx = l_j , \quad j=1,\ldots , r,\ ] ] where @xmath129\times\mathbb{r}^{2n } ; \mathbb{r})$ ] , @xmath130 , and boundary conditions . necessary optimality conditions for isoperimetric problems can be obtained by the following general theorem . [ nes : iso ] let @xmath131 be functionals defined in a neighborhood of @xmath67 and having continuous first variations in this neighborhood . suppose that @xmath67 is a local minimizer of subject to the boundary conditions and the isoperimetric constrains . assume that there are functions @xmath132 such that the matrix @xmath133 , @xmath134 , has maximal rank @xmath127 . then there exist constants @xmath135 such that the functional @xmath136 satisfies @xmath137 for all @xmath79 suppose now that assumptions of theorem [ nes : iso ] hold . then , equation is fulfilled for every @xmath138 . let us consider function @xmath86 such that @xmath139 . then , we have @xmath140dx,\end{gathered}\ ] ] where the function @xmath141\times { \mathbb{r}}^{2n}\times { \mathbb{r}}^r \rightarrow { \mathbb{r}}$ ] is defined by @xmath142^{\alpha,\beta}_{\gamma}(x ) -\sum_{j=1}^{r}\lambda_{j } g^{j}[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x).\ ] ] on account of the above , and similarly in spirit to the proof of theorem [ theo e - l1 ] , we obtain @xmath143 therefore , we have the following necessary optimality condition for the fractional isoperimetric problems : [ th : b : el - cv ] let assumptions of theorem [ nes : iso ] hold . if @xmath67 is a local minimizer to the isoperimetric problem given by , and , then @xmath67 satisfies the system of @xmath91 fractional euler lagrange equations for all @xmath93 $ ] . suppose now that constraints are characterized by inequalities @xmath144^{\alpha,\beta}_{\gamma}(x)dx \leq l_j , \quad j=1,\ldots , r.\ ] ] in this case we can set @xmath145^{\alpha,\beta}_{\gamma}(x ) -\frac{l_j}{b - a}\right)dx + \int_{a}^{b}(\phi_j(x))^2dx=0,\ ] ] @xmath130 , where @xmath146 have the some continuity properties as @xmath147 . therefore , we obtain the following problem : @xmath148 subject to @xmath127 isoperimetric constraints @xmath149^{\alpha,\beta}_{\gamma}(x ) -\frac{l_j}{b - a}+(\phi_j(x))^2\right]dx=0 , \quad j=1,\ldots , r,\ ] ] and boundary conditions . assuming that assumptions of theorem [ th : b : el - cv ] are satisfied , we conclude that there exist constants @xmath150 , @xmath130 , for which the system of equations @xmath151 @xmath152 , @xmath153 and @xmath154 hold for all @xmath93 $ ] . note that it is enough to assume that the regularity condition holds for the constraints which are active at the local minimizer @xmath67 ( constraint @xmath155 is active at @xmath67 if @xmath156 ) . indeed , suppose that @xmath157 constrains , say @xmath158 for simplicity , are active at the local minimizer @xmath67 , and there are functions @xmath159 such that the matrix @xmath160 has maximal rank @xmath161 . since the inequality constraints @xmath162 are inactive , the condition is trivially satisfied by taking @xmath163 . on the other hand , since the inequality constraints @xmath158 are active and satisfy a regularity condition at @xmath67 , the conclusion that there exist constants @xmath150 , @xmath130 , such that holds follow from theorem [ th : b : el - cv ] . moreover , is trivially satisfied for @xmath164 . let us consider the following lagrange problem , which consists of minimizing over all @xmath71 satisfying @xmath127 independent constraints ( @xmath165 ) @xmath166^{\alpha,\beta}_{\gamma}(x)=0 , \quad j=1,\ldots , r,\ ] ] and boundary conditions . in mechanics , constraints of type are called nonholonomic . by the independence of the @xmath127 constraints @xmath129\times\mathbb{r}^{2n } ; \mathbb{r})$ ] it is meant that it should exist a nonvanishing jacobian determinant of order @xmath127 , such as @xmath167 . of course , any @xmath127 of @xmath168 , @xmath169 , can be used , not necessarily the first @xmath127 . [ lagrange ] a function @xmath67 which is a solution to problem , subject to @xmath127 independent constraints ( @xmath165 ) satisfies , for suitably chosen functions @xmath170 , @xmath130 , the system of @xmath91 fractional euler lagrange equations @xmath171^{\alpha,\beta}_{\gamma}(x ) + { d^{\beta_{i-1},\alpha_{i-1}}_{1-\gamma_{i-1}}}\partial_{n+i}f[\mathbf{y},\mathbf{\lambda}]^{\alpha,\beta}_{\gamma}(x)=0 , \quad x\in[a , b ] , \quad i=2,\ldots , n+1,\ ] ] where @xmath172^{\alpha,\beta}_{\gamma}(x ) = l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x ) + \sum_{j=1}^r\lambda_j(x)g^j[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)$ ] . suppose that @xmath173 is the solution to problem defined by , , and . let @xmath174 be an arbitrary admissible variation , , @xmath94 , @xmath41 , and @xmath175^{\alpha,\beta}_{\gamma}(x)=0 $ ] , @xmath176 where @xmath177 is a small parameter . because @xmath173 is a solution to problem defined by , , and , it follows that @xmath96 that is , @xmath178^{\alpha,\beta}_{\gamma}(x ) h_{i-1}(x ) + \sum_{i=2}^{n+1}\partial_{n+i}l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x ) { ^cd^{\alpha_{i-1},\beta_{i-1}}_{\gamma_{i-1}}}h_{i-1}(x)\biggr]dx=0,\ ] ] and , for @xmath130 , @xmath179^{\alpha,\beta}_{\gamma}(x)h_{i-1}(x ) + \sum_{i=2}^{n+1}\partial_{n+i}g^j[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x ) { ^cd^{\alpha_{i-1},\beta_{i-1}}_{\gamma_{i-1}}}h_{i-1}(x)=0.\ ] ] multiplying the @xmath180th equation of the system by the unspecified function @xmath181 , for all @xmath130 , integrating with respect to @xmath182 , and adding the left - hand sides ( all equal to zero for any choice of the @xmath170 ) to the integrand of , we obtain @xmath183^{\alpha,\beta}_{\gamma}(x ) + \sum_{j=1}^r\lambda_j(x)\partial_ig^j[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)\right)h_{i-1}(x)\right.\\ & \left . \quad + \sum_{i=2}^{n+1}\left(\partial_{n+i}l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x ) + \sum_{j=1}^r\lambda_j(x)\partial_{n+i}g^j[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)\right)({^cd^{\alpha_{i-1},\beta_{i-1}}_{\gamma_{i-1}}}h_{i-1}(x))\right]dx\\ & = \int_a^b\left[\sum_{i=2}^{n+1}\partial_i f[\mathbf{y},\mathbf{\lambda}]^{\alpha,\beta}_{\gamma}(x ) h_{i-1}(x ) + \sum_{i=2}^{n+1}\partial_{n+i}f[\mathbf{y},\mathbf{\lambda}]^{\alpha,\beta}_{\gamma}(x ) ( { ^cd^{\alpha_{i-1},\beta_{i-1}}_{\gamma_{i-1}}}h_{i-1}(x))\right]dx\\ & = 0 , \end{split}\ ] ] where @xmath172^{\alpha,\beta}_{\gamma}(x ) = l[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x ) + \sum_{j=1}^r\lambda(x ) g^j[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)$ ] . integrating by parts , @xmath184^{\alpha,\beta}_{\gamma}(x ) + { d^{\beta_{i-1},\alpha_{i-1}}_{1-\gamma_{i-1}}}\partial_{n+i } f[\mathbf{y},\mathbf{\lambda}]^{\alpha,\beta}_{\gamma}(x)\right ] h_{i-1}(x)dx=0.\ ] ] because of , we can not regard the @xmath91 functions @xmath185 as being free for arbitrary choice . there is a subset of @xmath127 of these functions whose assignment is restricted by the assignment of the remaining @xmath186 . we can assume , without loss of generality , that @xmath187 are the functions of the set whose dependence upon the choice of the arbitrary @xmath188 is governed by . we now assign the functions @xmath189 to be the set of @xmath127 functions that make vanish ( for all @xmath182 between @xmath190 and @xmath191 ) the coefficients of @xmath187 in the integrand of . that is , @xmath189 are chosen so as to satisfy @xmath192^{\alpha,\beta}_{\gamma}(x ) + { d^{\beta_{i-1},\alpha_{i-1}}_{1-\gamma_{i-1}}}\partial_{n+i } f[\mathbf{y},\mathbf{\lambda}]^{\alpha,\beta}_{\gamma}(x)=0 , \quad i=2,\ldots , r+1 , \quad x\in[a , b].\ ] ] with this choice gives @xmath193^{\alpha,\beta}_{\gamma}(x ) + { d^{\beta_{i-1},\alpha_{i-1}}_{1-\gamma_{i-1}}}\partial_{n+i } f[\mathbf{y},\mathbf{\lambda}]^{\alpha,\beta}_{\gamma}(x)\right ] h_{i-1}(x)dx=0.\ ] ] since the functions @xmath188 are arbitrary , we may employ the fundamental lemma of the calculus of variations to conclude that @xmath194^{\alpha,\beta}_{\gamma}(x ) + { d^{\beta_{i-1},\alpha_{i-1}}_{1-\gamma_{i-1}}}\partial_{n+i}f[\mathbf{y},\mathbf{\lambda}]^{\alpha,\beta}_{\gamma}(x)=0,\ ] ] @xmath195 , for all @xmath93 $ ] . in order to determine the @xmath196 unknown functions @xmath197 , @xmath189 , we must consider the system of @xmath196 equations , consisting of , , and , together with the @xmath198 boundary conditions . assume now that the constraints , instead of , are characterized by inequalities : @xmath199^{\alpha,\beta}_{\gamma}(x)\leq 0 , \quad j=1,\ldots , r.\ ] ] in this case we can set @xmath200^{\alpha,\beta}_{\gamma}(x)+(\phi_j(x))^2=0 , \quad j=1,\ldots , r,\ ] ] where @xmath146 have the some continuity properties as @xmath147 . therefore , we obtain the following problem : @xmath201 subject to @xmath127 independent constraints ( @xmath165 ) @xmath202^{\alpha,\beta}_{\gamma}(x)+(\phi_j(x))^2=0 , \quad j=1,\ldots , r,\ ] ] and boundary conditions . applying theorem [ lagrange ] we get the following result . [ lagrange : ineq ] a set of functions @xmath203 , @xmath204 , which is a solution to problem , satisfies , for suitably chosen @xmath170 , @xmath130 , the following system of equations : @xmath205 @xmath152 , where @xmath206 , and @xmath207 , @xmath130 , hold for all @xmath93 $ ] . in this section we provide sufficient optimality conditions for the elementary and the isoperimetric problem of the @xmath0 fractional calculus of variations . similarly to what happens in the classical calculus of variations , some conditions of convexity are in order . given a function @xmath208\times\mathbb{r}^{2n } ; \mathbb{r})$ ] , we say that @xmath209 is jointly convex in @xmath210 , if @xmath211 for all @xmath212,@xmath213\times\mathbb{r}^{2n}$ ] . [ suff ] let @xmath214 be jointly convex in @xmath210 . if @xmath67 satisfies the system of @xmath91 fractional euler lagrange equations , then @xmath67 is a global minimizer to problem . the proof is similar to the proof of theorem 3.3 in @xcite . [ suff : iso ] let @xmath215 be jointly convex in @xmath210 , for some constants @xmath216 , @xmath130 . if @xmath217 satisfies the system of @xmath91 fractional euler lagrange equations , then @xmath217 is a minimizer to the isoperimetric problem defined by , and . by theorem [ suff ] , @xmath217 minimizes @xmath218 . that is , for all functions satisfying condition we have @xmath219^{\alpha,\beta}_{\gamma}(x)\ , dx-\sum_{j=1}^{r}\bar{\lambda}_{j}\int_a^bg^{j}[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x)\ , dx\\ \geq \int_a^b l[\mathbf{y}^0]^{\alpha,\beta}_{\gamma}(x)\ , dx-\sum_{j=1}^{r}\bar{\lambda}_{j}\int_a^bg^{j}[\mathbf{y}^0]^{\alpha,\beta}_{\gamma}(x)\ , dx.\end{gathered}\ ] ] restricting to the isoperimetric constraints , we obtain that @xmath220^{\alpha,\beta}_{\gamma}(x)\ , dx -\sum_{j=1}^{r}\bar{\lambda}_{j}l_{j}\ , dx\geq \int_a^b l[\mathbf{y}^0]^{\alpha,\beta}_{\gamma}(x)\ , dx -\sum_{j=1}^{r}\bar{\lambda}_{j}l_{j}\ , dx.\ ] ] therefore , @xmath220^{\alpha,\beta}_{\gamma}(x)\ , dx \geq \int_a^b l[\mathbf{y}^0]^{\alpha,\beta}_{\gamma}(x)\ , dx\ ] ] as desired . choosing @xmath221 in theorem [ suff : iso ] one can easily obtain ( * ? ? ? * theorem 3.10 ) . multiobjective optimization is a natural extension of the traditional optimization of a single - objective function . if the objective functions are commensurate , minimizing one - objective function minimizes all criteria and the problem can be solved using tradicional optimization techniques . however , if the objective functions are incommensurate , or competing , then the minimization of one objective function requires a compromise in another objective . here we consider multiobjective fractional variational problems with a finite number @xmath222 of objective ( cost ) functionals @xmath223^{\alpha,\beta}_{\gamma}(x ) \ , dx , \ldots , \int_a^b l^d[\mathbf{y}]^{\alpha,\beta}_{\gamma}(x ) \ , dx \right)\longrightarrow \min\ ] ] subject to the boundary conditions @xmath224 @xmath89 , and @xmath127 ( @xmath165 ) independent constraints @xmath225^{\alpha,\beta}_{\gamma}(x)\leq 0 , \quad j=1,\ldots , r,\ ] ] where @xmath226\times\mathbb{r}^{2n } ; \mathbb{r})$ ] , @xmath227 , @xmath130 . we would like to find a function @xmath71 , satisfying constraints and , that renders the minimum value to each functional @xmath228 , @xmath229 , simultaneously . the competition between objectives gives rise to the necessity of distinguish between the difference of multiobjective optimization and traditional single - objective optimization . competition causes the lack of complete order for multiobjective optimization problems . the concept of pareto optimality is therefore used to characterize a solution to the multiobjective optimization problem . for the usefulness of variational analysis and pareto optimal allocations in welfare economics , we refer the reader to @xcite . we define @xmath230 [ def : paret ] a function @xmath231 is called a pareto optimal solution to problem if does not exist @xmath232 with @xmath233 definition [ def : paret ] introduces the notion of _ global pareto optimality_. another important concept is the one of _ local pareto optimality_. [ def : locparet ] a function @xmath231 is called a local pareto optimal solution to problem if there exists @xmath234 for which does not exist @xmath232 with @xmath78 and @xmath235 naturally , any global pareto optimal solution is locally pareto optimal . for enhanced notions of pareto optimality of constrained multiobjective problems , the reader is referred to @xcite . we obtain a sufficient condition for pareto optimality by modifying the original multiobjective fractional problem into the following weighting problem : @xmath236^{\alpha,\beta}_{\gamma}(x ) \ , dx \longrightarrow \min\ ] ] subject to @xmath237 , where @xmath238 for all @xmath227 , and @xmath239 . [ pareto : suf ] the solution of the weighting problem is pareto optimal if the weighting coefficients are positive , that is , @xmath240 for all @xmath227 . moreover , the unique solution of the weighting problem is pareto optimal . let @xmath231 be a solution to problem with @xmath240 for all @xmath227 . suppose that @xmath241 is not pareto optimal . then , there exists @xmath67 such that @xmath242 for all @xmath227 and @xmath243 for at least one @xmath180 . since @xmath240 for all @xmath227 , we have @xmath244 . this contradicts the minimality of @xmath241 . now , let @xmath241 be the unique solution to . if @xmath241 is not pareto optimal , then @xmath245 . this contradicts the uniqueness of @xmath241 . therefore , by varying the weights over the unit simplex @xmath246 ones obtains , in principle , different pareto optimal solutions . the next theorem provides a necessary and sufficient condition for pareto optimality . the result is analogous to that valid for the finite dimensional case ( see , , chapter 3.1 and chapter 3.3 of @xcite ) . [ pareto : nes : suf ] a function @xmath231 is pareto optimal to problem if and only if it is a solution to the scalar fractional variational problem @xmath247^{\alpha,\beta}_{\gamma}(x ) \ , dx \longrightarrow \min\ ] ] subject to @xmath232 and @xmath248^{\alpha,\beta}_{\gamma}(x ) \ , dx \leq \int_a^b l^j[\bar{\mathbf{y}}]^{\alpha,\beta}_{\gamma}(x ) \ , dx , \quad j=1,\ldots , d,\quad j\neq i,\ ] ] for each @xmath227 . suppose that @xmath241 is pareto optimal . then @xmath249 for all @xmath250 , so @xmath251 . if @xmath241 does not minimize @xmath252 on the constrained set @xmath253 for some @xmath250 , then there exists @xmath232 such that @xmath254 and @xmath255 for all @xmath256 . this contradicts the pareto optimality of @xmath241 . now , suppose that @xmath241 minimize each @xmath252 on the constrained set @xmath253 . if @xmath241 is not pareto optimal , then there exists @xmath67 such that @xmath242 for all @xmath227 and @xmath243 for at least one @xmath180 . this contradicts the minimality of @xmath67 for @xmath257 on @xmath258 . [ pareto : nes ] for a function @xmath232 to be pareto optimal to problem , it is necessary to be a solution to the fractional isoperimetric problems @xmath247^{\alpha,\beta}_{\gamma}(x ) \ , dx \longrightarrow \min\ ] ] subject to @xmath232 and @xmath248^{\alpha,\beta}_{\gamma}(x ) \ , dx = \int_a^b l^j[\bar{\mathbf{y}}]^{\alpha,\beta}_{\gamma}(x ) \ , dx , \quad j=1,\ldots , d,\quad j\neq i,\ ] ] for all @xmath227 . therefore , necessary optimality conditions for the fractional isoperimetric problems ( see theorem [ th : b : el - cv ] ) are also necessary for fractional pareto optimality . we illustrate our results with two multiobjective fractional variational problems . let @xmath259 , @xmath260 $ ] , where @xmath261 is the mittag leffler function : @xmath262 when @xmath263 , the mittag leffler function is simply the exponential function : @xmath264 . we note that the left caputo fractional derivative of @xmath265 is @xmath265 ( @xcite , p. 98 ) : @xmath266 consider the following multiobjective fractional variational problem ( @xmath267 , @xmath105 , and @xmath268 ) : @xmath269 subject to @xmath270 observe that @xmath265 satisfies the necessary pareto optimality conditions ( see remark [ pareto : nes ] ) . indeed , as shown in ( * ? ? ? * example 1 ) , @xmath265 is a solution to the isoperimetric problem @xmath271 subject to @xmath272 consider now the following fractional isoperimetric problem : @xmath273 subject to @xmath274 let us apply theorem [ nes : iso ] . the equality @xmath275 holds if and only if @xmath276 with @xmath277 ( see ( * ? ? ? * corollary 2.1 ) ) . hence , @xmath265 does not satisfy equation @xmath278 . the augmented function is @xmath279 and the corresponding fractional euler lagrange equation gives @xmath280 a solution to this equation is @xmath281 and @xmath282 . therefore , by remark [ pareto : nes ] , @xmath282 is a candidate pareto optimal solution to problem . [ ex:2 ] consider the following multiobjective fractional variational problem : @xmath283 subject to @xmath284 where @xmath9 is a fixed function . in this case we have @xmath267 , @xmath105 , and @xmath268 . by theorem [ pareto : suf ] , pareto optimal solutions to problem can be found by considering the family of problems @xmath285 subject to @xmath286 where @xmath287 $ ] . let us now fix @xmath288 . by theorem [ theo e - l1 ] , a solution to problem satisfies the fractional euler lagrange equation @xmath289 moreover , by theorem [ suff ] , a solution to is a global minimizer to problem . therefore , solving equation for @xmath287 $ ] , we are able to obtain pareto optimal solutions to problem . in order to solve equation , firstly we use corollary 2.1 of @xcite to get the following equation : @xmath290 the minimizer of example [ ex:2 ] for @xmath291 and @xmath292 . ] equation needs to be solved numerically . we did numerical simulations using the matlab solver ` fode ` for linear fractional - order differential equations ( fode ) with constant coefficients , developed by farshad merrikh bayat @xcite . the results for @xmath293 , @xmath291 , and different values of the parameter @xmath288 can be seen in figure [ fig ] . numerical results for different values of @xmath1 show that when @xmath294 the fractional solution converges to the solution of the classical problem of the calculus of variations . the authors would like to express their gratitude to ivo petras , for having called their attention to @xcite as well as for helpful discussions on numerical aspects and available software packages for fractional differential equations . , on the solution of nonlinear fractional - order differential equations used in the modeling of viscoplasticity . in : f. keil , w. mackens , h. vo , j. werther ( eds . ) : _ scientific computing in chemical engineering ii . computational fluid dynamics , reaction engineering , and molecular properties_. springer - verlag , heidelberg , 1999 , 217224 . , necessary optimality conditions for fractional action - like integrals of variational calculus with riemann liouville derivatives of order @xmath295 , math . methods appl . sci . * 30 * ( 2007 ) , no . 15 , 19311939 . arxiv : math - ph/0702099
particle physics is at a crossroads . behind us is the standard model ( sm ) , the remarkably successful theory of all known elementary particles and their interactions . ahead of us is an equally remarkable array of possibilities for new phenomena at the weak scale . never before has an energy scale been so widely anticipated to yield profound insights , and never before have there been so many ideas about exactly what these insights could be . in this article , we review the current state of experimental searches for supersymmetry , the most widely studied extension of the sm . there are at present many reasons to expect new physics at the weak scale @xmath0 . chief among these is the higgs boson , an essential component of the sm that has yet to be discovered . at the same time , there are also strong motivations for new phenomena beyond the higgs boson . these motivations include naturalness , dark matter , and unification . the physical mass of the sm higgs boson is given by @xmath1 where @xmath2 is the bare mass parameter present in the lagrangian , and the quantum corrections are @xmath3 where @xmath4 is a dimensionless gauge or yukawa coupling , and @xmath5 is the energy scale at which the sm is no longer a valid description of nature . because @xmath6 is proportional to @xmath7 ( `` quadratically divergent '' ) , it is natural to expect the higgs mass to be pulled up to within an order of magnitude of @xmath5 by quantum corrections @xcite . given that unitarity and precision constraints require @xmath8 to be at the weak scale @xcite , this implies @xmath9 , and new physics should appear at the current energy frontier . of course , the higgs boson may not be a fundamental scalar , but in this case , too , its structure requires new physics at the weak scale @xcite . for these reasons , naturalness is among the most robust motivations for new physics at an energy scale accessible to accelerator - based experiments . in the last decade , a wealth of cosmological observations have constrained the energy densities of baryons , non - baryonic dark matter , and dark energy , in units of the critical density , to be @xcite @xmath10 the non - baryonic dark matter must be stable or very long - lived and dominantly cold or warm . none of the particles of the sm satisfies these conditions , and so cosmology requires new particles . perhaps the simplest production mechanism for dark matter is thermal freeze out @xcite . in this scenario , a new particle is initially in thermal contact with the sm , but as the universe cools and expands , this particle loses thermal contact and its energy density approaches a constant . under very general assumptions , this relic energy density satisfies @xmath11 where @xmath12 is the dark matter s thermally - averaged annihilation cross section . it is a tantalizing fact that , when this cross section is typical of weak - scale particles , that is , @xmath13 , where @xmath14 , @xmath15 is near the observed value of @xmath16 given in eq . ( [ omegas ] ) . if thermal freeze out is the mechanism by which dark matter is produced in the early universe , then , cosmological data therefore also point to the weak scale as the natural scale for new physics . the sm is consistent with the observed properties of all known elementary particles . it also elegantly explains why some phenomena , such as proton decay and large flavor - changing neutral currents , are not observed . the latter fact is highly non - trivial , as evidenced by the intellectual contortions required of model builders who try to extend the sm . at the same time , the sm contains many free parameters with values constrained by experiment , but not explained . the number of free parameters may be reduced in unified theories , in which the symmetries of the sm are extended to larger symmetries . in particular , grand unified theories , in which the su(3 ) @xmath17 su(2 ) @xmath17 u(1 ) gauge structure is extended to larger groups , are significantly motivated by the fact that the sm particle content fits perfectly into multiplets of su(5 ) and larger groups @xcite , potentially explaining the seemingly random assignment of quantum numbers , such as hypercharge . one straightforward implication of the simplest ideas of grand unification is that the gauge couplings of the sm must unify when extrapolated to higher scales through renormalization group evolution . the gauge couplings do not unify at any scale given the particle content of the sm , but they do unify at the value @xmath18 at @xmath19 if the sm is minimally extended by supersymmetry ( susy ) _ and _ the supersymmetric particles are at the weak scale @xcite . this unification is highly non - trivial , not only because the couplings are now so precisely measured , but also because @xmath20 is in the perturbative regime and @xmath21 is in the narrow range that is both high enough to suppress proton decay and low enough to avoid quantum gravitational effects . this unification is only logarithmically sensitive to the superpartner mass scale , and the degree of its success is somewhat model - dependent ; see , _ e.g. _ , the review by raby in ref . @xcite . in conjunction with the previous two motivations , however , it provides still more evidence for new physics at the weak scale , and selects supersymmetry as a particularly motivated possibility . there are two main areas where new phenomena could appear in particle physics . deviations from sm predictions could show up in measurements performed with increasing precision . examples are the anomalies observed in the forward - backward asymmetry in the production of @xmath22 pairs in @xmath23 collisions at the @xmath24 peak ( see ref . @xcite , particularly section 7.3.5 ) , or in the anomalous magnetic moment of the muon ( see , _ e.g. _ , the review by hcker and marciano in ref . even if such anomalies receive experimental confirmation at a sufficient significance level , their interpretation will however remain ambiguous , because it will involve virtual contributions to the relevant amplitudes of yet undiscovered , therefore most likely very massive , particles . the alternative approach is to try to observe directly the production of these new particles , which is among the goals of the experiments at colliders operating at the highest possible energies . the large hadron collider ( lhc ) at cern will soon occupy the energy frontier . when it comes into operation , @xmath25 collisions will take place at a center - of - mass energy of 10 tev , and of 14 tev later on . the instantaneous luminosity will be raised first to @xmath26 and progressively to @xmath27 . with the enormous data samples accumulated , the two general purpose experiments at the lhc , atlas @xcite and cms @xcite , will be in a position to explore in great detail the physics at the tev scale . since this is an entirely new domain , and since there are strong reasons to expect new phenomena at that scale , as advocated in the preceding section of this review , it may well be that ground breaking discoveries are made at the lhc , even after a short period of operation , once the detectors are properly aligned , calibrated , and well understood . until then , the most constraining results on searches for new phenomena at high energy have been or are still being obtained at lep , hera , and the tevatron . providing a comprehensive account of such searches for supersymmetry is the purpose of this review . the large @xmath23 collider ( lep ) at cern operated from 1989 to 2000 . in a first phase ( lep1 ) , the center - of - mass energy was set at or close to 91 gev , the peak of the @xmath24 boson resonance . four experiments , aleph @xcite , delphi @xcite , l3 @xcite , and opal @xcite studied millions of @xmath24 decays that allowed them to perform stringent precision tests of the sm . from the end of 1995 on , the energy was progressively increased ( lep2 ) to reach 209 gev in the center of mass during the last year of operation . altogether , each of the experiments collected a total of @xmath28 fb@xmath29 of data , of which @xmath30 pb@xmath29 in 2000 at and above 204 gev , the data set most relevant for new particle searches . at desy , the hera collider operation was terminated in june 2007 . there , @xmath31 collisions were collected by two experiments , h1 @xcite and zeus @xcite , at a center - of - mass energy of @xmath32 gev . this was an asymmetric collider , with @xmath33 and proton beam energies of 30 and 820 gev , respectively . an upgrade took place in 2001 ( hera2 ) , leading to higher luminosities than in the previous phase ( hera1 ) , and allowing operation with polarized @xmath33 beams . the data sets collected at hera1 and hera2 with electron or positron beams altogether correspond to an integrated luminosity of @xmath34 fb@xmath29 per experiment . until the lhc comes into operation , the highest energy collisions are provided by the tevatron @xmath35 collider at fermilab . during its first phase of operation ( run i ) , the center - of - mass energy was set to 1.8 tev , and a data sample of @xmath36 pb@xmath29 was collected by each of the two experiments , cdf @xcite and d @xcite . the highlight of that period was the discovery of the top quark in 1995 . major upgrades of the accelerator complex and of the two detectors took place for the second phase ( run ii ) , which began in 2001 . the center - of - mass energy was raised to 1.96 tev , and the instantaneous luminosity was progressively increased to regularly approach or exceed @xmath37 in 2008 . more than 5 fb@xmath29 of integrated luminosity had been delivered by the tevatron by the end of fiscal year ( fy ) 2008 , and it is expected that another @xmath38 fb@xmath29 of luminosity will be provided per additional year of operation . at the time of writing , running in fy 2009 is underway , running in fy 2010 is increasingly likely , and running in fy 2011 is kept as an option . all general purpose detectors at colliders share similar features . a cylindrical `` barrel '' structure parallel to the beam axis surrounds the interaction region , and is closed by `` end caps '' perpendicular to the beam . the first elements encountered beyond the beam pipe are charged - particle detectors , with those closest to the interaction point benefiting from the highest spatial precision . this tracking system is immersed in an axial magnetic field provided by a solenoidal magnet . beyond the tracking system , electromagnetic calorimeters provide electron and photon identification and energy measurement . these are followed by hadron calorimeters for the measurement of jet energies . finally , track detectors are used to identify and measure the muons which have penetrated through the calorimeters and possibly additional absorber material . non - interacting particles , such as neutrinos , are detected by an apparent non - conservation of energy and momentum . in @xmath23 annihilation , the missing energy and momentum can be directly inferred from a measurement of the final state particles , by comparison with the center - of - mass energy of the collision . in hadronic or @xmath39 collisions , the partons participating in the hard process carry only a fraction of the beam energy , and the beam remnants associated with the spectator partons largely escape undetected in the beam pipe . as a consequence , only conservation of the momentum in the direction transverse to the beams can be used , and the relevant quantity is the missing transverse energy , rather than the total missing energy . the mass reach in @xmath40 collisions at the tevatron is expected to be substantially larger than at lep because of the higher center - of - mass energy . however , since the initial partons participating in the hard process carry fractions @xmath41 and @xmath42 of the beam energy , the effective center - of - mass energy is only @xmath43 . because of the rapidly falling parton distribution functions ( pdfs ) as a function of those energy fractions , increasingly large integrated luminosities are needed to probe larger and larger @xmath44 values . at hera , furthermore , the center - of - mass energy in the @xmath45 collision can not be fully used for new particle production , except in some very specific instances . this is in contrast to @xmath23 or @xmath46 annihilation , and to @xmath47 fusion . as a consequence , the most constraining results on new particle searches typically come from lep and from the tevatron . in the following , all limits quoted are given at a confidence level of 95% . supersymmetry ( susy ) @xcite is an extension of poincar symmetry , which encompasses the known spacetime symmetries of translations , rotations , and boosts . as with the poincar and internal symmetries , susy transforms particle states to other particle states . in contrast to these other symmetries , however , susy relates states of different spin , transforming fermions into bosons and vice versa . none of the known particles can be supersymmetric partners of other known particles . as a result , susy predicts many new particle states . if susy were exact , these particles would be degenerate with known particles . since this is experimentally excluded , if susy is a symmetry of nature , it must be broken . susy is the most studied extension of the sm because it directly addresses several of the motivations for new physics discussed in sec . [ sec : introduction ] . in supersymmetric theories , the quadratically divergent loop contributions to the higgs boson mass from sm particles are canceled by similar contributions from superpartners , ameliorating the gauge hierarchy problem @xcite . supersymmetric theories also include excellent dark matter candidates , in the form of neutralinos @xcite and gravitinos @xcite , that may naturally have the desired relic density . finally , susy is strongly motivated by the hope for unifying forces , as it makes gauge coupling unification possible in simple grand unified theories ( guts ) @xcite . it is important to note that _ all _ of these virtues are preserved only if the superpartner mass scale is around the weak scale . the existence of susy in nature , although not necessarily at the weak scale , is also motivated by string theory and the beautiful mathematical properties of susy that are beyond the scope of this review . for these reasons , this review is devoted to searches for susy at colliders . in this section , we present brief summaries of the supersymmetric spectrum , parameters , and unifying frameworks to establish our conventions and notation . more extensive phenomenological reviews of susy may be found in refs . @xcite . in this review , we focus our attention on the minimal supersymmetric extension of the standard model ( mssm ) , the supersymmetric model with minimal field content . bosonic superpartners are given names with the prefix `` s , '' and fermionic superpartners are denoted by the suffix `` ino . '' squarks and sleptons are collectively known as `` sfermions , '' and the entire group of superpartner particles are often called `` sparticles . '' the particle content of the mssm is in fact slightly more than a doubling of the sm particle content . this is because , in addition to introducing superpartners for all known particles , the mssm requires two electroweak higgs doublets . there are two reasons for this . first , in the sm , mass terms are generated for up- and down - type particles by yukawa couplings to @xmath48 and @xmath49 , respectively , where @xmath49 is the sm higgs field . in susy , yukawa couplings are generalized to terms in a superpotential , a function of superfields that contain both sm particles and their superpartners , which generates the sm yukawa couplings as well as all other terms related to these by susy . complex - conjugated fields are not allowed in the superpotential , however . as a result , two separate higgs fields , denoted @xmath50 and @xmath51 , are required to generate masses through the superpotential terms @xmath52 where @xmath53 , @xmath54 , @xmath55 , @xmath56 , and @xmath57 are the su(2 ) quark doublet , up - type quark singlet , down - type quark singlet , lepton doublet , and lepton singlet superfields , respectively , and the @xmath4 couplings are yukawa couplings . second , susy requires that the sm higgs field have fermion partners , the higgsinos . the introduction of these additional fermions charged under sm gauge groups ruins anomaly cancellation , making this theory mathematically untenable . the introduction of an additional higgs doublet , with its extra higgsinos , restores anomaly cancellation . the mssm higgs boson sector therefore consists of eight degrees of freedom . as in the sm , three of these are eaten to make massive @xmath58 and @xmath24 bosons , but five remain , which form four physical particles : @xmath59 where @xmath60 and @xmath61 are the cp - even neutral higgs bosons , with @xmath60 lighter than @xmath61 , @xmath62 is the cp - odd neutral higgs boson , and @xmath63 is the charged higgs boson . the remaining supersymmetric particle content of the mssm is straightforward to determine and consists of the following states : @xmath64 each sm chiral fermion has a ( complex ) scalar partner , denoted by the appropriate chirality subscript . the dimensionless couplings of all of these particles are fixed by susy to be identical to those of their sm partners . note , however , that , as described in the appropriate sections below , the states in each line of eq . ( [ sparticles ] ) ( except for the last one ) may mix , and mass eigenstates are in general linear combinations of these gauge eigenstates . finally , most analyses of susy include the supersymmetric partner of the graviton : @xmath65 although not technically required as a part of the mssm , when susy is promoted to a local symmetry , it necessarily includes gravity , and the resulting supergravity theories include both gravitons and gravitinos . the gravitino is therefore present if susy plays a role in unifying the sm with gravity , as in string theory . if susy were exact , the gravitino s properties would be determined precisely by the graviton s , and it would be massless and have gravitational couplings suppressed by the reduced planck mass @xmath66 . however , just as goldstone bosons appear when conventional symmetries are spontaneously broken , a fermion , the goldstino @xmath67 , appears when susy is broken . the gravitino then becomes massive by eating the goldstino . in terms of @xmath68 , the mass dimension-2 order parameter of susy breaking , the gravitino mass becomes @xmath69 and , very roughly , its interactions in processes probing energy scale @xmath57 may be characterized by a dimensionless coupling @xmath70 light gravitinos couple more strongly . as we will see below , in well - motivated supersymmetric theories , these properties may take values in the range @xmath71 where we have assumed colliders probing @xmath72 . as noted above , if susy exists in nature , it must be broken . although many different lagrangian terms could be added to break susy , only some of these are allowed if susy is to stabilize the gauge hierarchy . these terms , known as `` soft '' susy - breaking terms , include most , but not all , lagrangian terms with mass dimension 3 and below @xcite . for the mssm , they are @xmath73 + \text{h.c.}\right\ } \nonumber \\ & + & \ ! \ ! \lambda_u a_u h_u \tilde{q } \tilde{u } + \lambda_d a_d h_d \tilde{q } \tilde{d } + \lambda_e a_e h_d \tilde{l } \tilde{e } \nonumber \\ & + & \ ! \ ! m_{h_u}^2 |h_u|^2 + m_{h_d}^2 |h_d|^2 + \left ( b h_u h_d + \text{h.c . } \right ) . \label{soft}\end{aligned}\ ] ] these lines are sfermion masses , gaugino masses , trilinear scalar couplings ( `` @xmath62-terms '' ) , and higgs boson couplings . in addition to the parameters above , there are two other key parameters : the @xmath74 parameter , which enters in the higgsino mass terms @xmath75 , and @xmath76 which parameterizes how the sm higgs vacuum expectation value is divided between the two neutral higgs scalars . the interactions of eq . ( [ soft ] ) conserve @xmath77-parity @xcite . with @xmath78 , where @xmath79 and @xmath56 are the baryon and lepton numbers , respectively , and @xmath80 the spin , all superpartners are odd and all sm particles are even under @xmath77-parity . this implies that all interactions involve an even number of superpartners , and so the lightest superpartner is stable , and a potential dark matter candidate . @xmath77-parity violation generically violates both baryon and lepton number , leading to too - rapid proton decay , which is why , for most of this review , we limit ourselves to the @xmath77-parity conserving case . even restricting ourselves to the @xmath77-parity - preserving terms of eq . ( [ soft ] ) , however , we see that susy introduces many new parameters . note that the terms involving sfermions need not be flavor - diagonal , and so the sfermion masses and @xmath62-terms are in fact matrices of parameters in the most general case . at the same time , fully general flavor mixing terms violate low energy constraints on flavor - changing neutral currents . in addition , arbitrary complex parameters also violate bounds on cp - violation from , for example , @xmath81 and the electric dipole moments of the electron and neutron . these considerations motivate unifying frameworks , to which we now turn . in collider searches , it is desirable to consider theories that are both viable and simple enough to be explored fully . for this reason , it is common to work in simple model frameworks that reduce the number of independent susy parameters . in some cases , these model frameworks also motivate particular collider signatures that might otherwise appear highly unlikely or fine - tuned . in the most common unifying frameworks , susy is assumed to be broken in some other sector . susy breaking is then mediated to the mssm through a mechanism that defines the framework . this sets susy - breaking parameters at some high energy scale . renormalization group evolution to the weak scale then determines the physical soft susy - breaking parameters and the physical spectrum of the mssm . a representative example of renormalization group evolution is shown in fig . [ fig : mssmrun ] . in this evolution from the high scale to the weak scale , gauge couplings increase masses and yukawa couplings decrease masses . this is central to understanding the sparticle spectrum of many models . in addition , it explains why @xmath82 becomes negative at the weak scale it is the only particle to receive large negative contributions from yukawa couplings without compensating large positive contributions from the strong coupling . when @xmath50 becomes tachyonic , it breaks electroweak symmetry , and this feature , known as `` radiative electroweak symmetry breaking , '' is a virtue of many supersymmetric frameworks . note , however , that radiative electroweak symmetry breaking makes essential use of the large top quark mass , and so shifts the burden of understanding why electroweak symmetry is broken to the question of why the top quark is heavy . to the weak scale in a representative msugra model . from ref . [ fig : mssmrun ] , width=321 ] in this section , we discuss several common unifying frameworks that have been used in collider searches , namely , models with gravity- , gauge- , and anomaly - mediated susy breaking . each of these has its distinctive characteristics . as a rough guide , in fig . [ fig : modelspectra ] we show representative spectra resulting from each of these frameworks . these spectra may be generated using publicly available computer programs , including isajet @xcite , softsusy @xcite , spheno @xcite , and suspect @xcite . , the wino mass parameter at the weak scale . in the gravity - mediated case , two example spectra are presented : one for `` no - scale '' models with @xmath83 , and another for @xmath84 . in the anomaly - mediated case , the sleptons are tachyonic in the minimal case additional effects are required to raise these to a viable range . from ref . [ fig : modelspectra ] , width=321 ] in gravity - mediated susy - breaking models @xcite , sometimes referred to as supergravity ( sugra ) models , susy breaking in a hidden sector is mediated to the mssm through terms suppressed by the reduced planck mass @xmath85 . for example , sfermion masses are @xmath86 . for these to be at the weak scale , @xmath87 must be around @xmath88 . given eqs . ( [ gravitinomass ] ) and ( [ gravitinocoupling ] ) , the gravitino also has a weak scale mass and couples with gravitational strength in sugra models . without a quantum theory of gravity , the structure of gravity - mediated susy parameters is unconstrained and generically violates low - energy constraints . to make these theories viable , _ ad hoc _ unifying assumptions must be made . by far the most common assumptions are those of minimal supergravity ( msugra ) , which is specified by 4 continuous and 1 discrete parameter choice : @xmath89 where the first three parameters are the universal scalar mass ( including the two higgs scalars ) , unified gaugino mass , and universal @xmath62-parameter , which are all specified at the grand unified theory ( gut ) scale @xmath19 . the remaining susy parameters @xmath90 and the dimension-2 higgs boson mass parameter @xmath79 are determined by requiring that the higgs potential at the weak scale give correct electroweak symmetry breaking . at tree - level , this requires @xmath91 gaugino mass unification is motivated by the unification of gauge couplings at @xmath21 in the mssm . it leads to the prediction that the bino , wino , and gluino masses are in the ratio @xmath92 at the weak scale , as evident in fig . [ fig : modelspectra ] . scalar mass universality is on much less solid ground . even in guts , for example , the higgs scalars are not necessarily in the same multiplet as the squarks and sleptons . this motivates a slightly less restrictive framework , the non - universal higgs model ( nuhm ) in which @xmath93 is the universal sfermion mass , but @xmath94 and @xmath95 are treated as independent parameters . one may exchange these new degrees of freedom for the more phenomenological parameters @xmath74 and @xmath96 at the weak scale : @xmath97 the nuhm framework is employed in some mssm higgs boson studies discussed in sec . [ sec : higgs ] . in gauge - mediated susy - breaking ( gmsb ) models @xcite , in addition to the gravity - mediated contributions to soft parameters discussed above , each sparticle receives contributions to its mass determined by its gauge quantum numbers . these new contributions to sfermion masses are @xmath98 , where @xmath99 is the mass scale of the messenger particles that transmit the susy breaking . the gmsb contributions are flavor - blind , and do not violate low energy bounds . for these to be dominant , one requires @xmath100 , and so we find that @xmath101 in gmsb scenarios , and the lightest supersymmetric particle ( lsp ) is always the gravitino . in gmsb models , the collider signatures are determined by the next - to - lightest supersymmetric particle ( nlsp ) and its lifetime , or equivalently , the gravitino s mass . if the nlsp is the lightest neutralino , the collider signature is either missing energy or prompt photons , @xmath24 or higgs bosons from @xmath102 @xcite ; if the nlsp is a slepton , the signature is typically either long - lived heavy charged particles or multi - lepton events @xcite . a third class of susy models are those with anomaly - mediated susy - breaking ( amsb ) @xcite . these are extra dimensional scenarios in which susy is broken on another 3-dimensional subspace , and transmitted to our world through the conformal anomaly . as with all anomalies , this effect is one - loop suppressed . the fundamental scale of susy breaking as characterized by the gravitino mass is therefore @xmath103 , with mssm sparticle masses one - loop suppressed and at the weak scale . the amsb contributions to sparticle masses are completely determined by the sparticle s gauge and yukawa couplings . this leads to a highly predictive spectrum . unfortunately , one of these predictions is @xmath104 , but various mechanisms have been proposed to solve this tachyonic slepton problem ; see , e.g. , refs . @xcite . the gaugino masses are determined by the corresponding gauge group beta functions . in particular , amsb predicts @xmath105 ; because the su(2 ) coupling is nearly scale - invariant in the mssm , the wino mass is the smallest . amsb scenarios therefore motivate supersymmetric models with @xmath106 lsp and @xmath107 nlsp . this triplet may be extremely degenerate , with the chargino traveling macroscopic distances before decaying to soft and invisible decay products , which provides a distinctive and challenging signature for collider searches @xcite . the mssm higgs potential is @xmath108 where the parameters @xmath74 , @xmath82 , and @xmath109 are as defined in sec . [ sec : parameters ] , and susy implies that all the quartic couplings are determined by the su(2 ) and u(1 ) hypercharge gauge couplings , denoted @xmath110 and @xmath111 , respectively . @xmath112 automatically conserves cp , since any phase in the @xmath79 parameter can be eliminated by a redefinition of the higgs fields . assuming these parameters are such that the potential admits a stable , symmetry - breaking minimum , the three parameter combinations @xmath113 , @xmath114 , and @xmath79 may be exchanged for the two vacuum expectation values @xmath115 , @xmath116 , and one physical higgs boson mass , conveniently taken to be @xmath96 . the @xmath58 boson mass fixes @xmath117 , leaving one additional degree of freedom , usually taken to be @xmath118 . thus , at tree - level , the entire mssm higgs boson sector is determined by two parameters , @xmath96 and @xmath119 . in terms of these parameters , the physical higgs boson masses are @xmath120 where @xmath121 . the cp - even mass eigenstates are related to the gauge eigenstates through @xmath122 where the rotation angle @xmath123 satisfies @xmath124 with @xmath125 . equation ( [ cpevenmasses ] ) implies that @xmath126 , a rather disastrous relation , given that experimental bounds exclude @xmath127 . the results presented so far , however , are valid only at tree - level . large radiative corrections from top squark / quark loops ( see , _ e.g. _ , ref . @xcite ) , @xmath128 can lift @xmath8 to values above the experimental bounds . note , however , that for @xmath129 , @xmath130 at tree - level , and so large values of @xmath8 are not possible for @xmath131 . from considerations of the higgs mass alone , @xmath132 is possible . however , such values imply very large top yukawa couplings , which become infinite well below the gut or planck scales . in addition , in simple frameworks , @xmath132 is incompatible with radiative electroweak symmetry breaking ; for a review of bounds on @xmath119 , see ref . @xcite . the neutralinos and charginos of the mssm are the mass eigenstates that result from the mixing of the electroweak gauginos @xmath133 and @xmath134 with the higgsinos . the neutral mass terms are @xmath135 where @xmath136 and @xmath137 the neutralino mass eigenstates are @xmath138 , where * n * diagonalizes @xmath139 . in order of increasing mass , the four neutralinos are labeled @xmath140 , @xmath141 , @xmath142 , and @xmath143 . the charged mass terms are @xmath144 where @xmath145 and @xmath146 the chargino mass eigenstates are @xmath147 and @xmath148 , where the unitary matrices * u * and * v * are chosen to diagonalize @xmath149 , and @xmath150 is lighter than @xmath151 . sleptons are promising targets for colliders , as they are among the lightest sparticles in many models . as noted in sec . [ sec : superpartners ] , sleptons include both left- and right - handed charged sleptons and sneutrinos . the mass matrix for the charged sleptons is @xmath152 in the basis @xmath153 , where @xmath154 . the sneutrino has mass @xmath155 these masses are given in third - generation notation ; in the presence of flavor mixing , these generalize to full six - by - six and three - by - three matrices . the left - right mixing is proportional to lepton mass , and is therefore expected to be insignificant for selectrons and smuons , but may be important for staus , especially if @xmath119 is large . through level repulsion , this mixing lowers the lighter stau s mass . as noted in sec . [ sec : unifying ] , yukawa couplings also lower scalar masses through renormalization group evolution . both of these effects imply that in many scenarios , the lighter stau is the lightest slepton , and often the lightest sfermion . the mass matrix for up - type squarks is @xmath156 in the basis @xmath157 , and for down - type squarks is @xmath158 in the basis @xmath159 . large mixing is expected in the stop sector , and possibly also in the sbottom sector if @xmath119 is large . because of these mixings and the impact of large yukawa couplings in renormalization group evolution , the 3rd generation squarks are the lightest squarks in many models @xcite as already explained in sec . [ sec : supersymmetry ] , two higgs doublets are needed in the mssm to give mass to both up- and down - type quarks . under the assumption that the higgs sector is cp conserving , the physical states are two neutral cp - even higgs bosons ( @xmath60 and @xmath61 , ordered by increasing mass ) , a neutral cp - odd higgs boson ( @xmath62 ) , and a doublet of charged higgs bosons ( @xmath160 ) . further details on the higgs sector of the mssm have been given in sec . [ sec : supersymmetry ] . here , we focus on searches for the neutral higgs bosons of the mssm , while searches for charged higgs bosons will be discussed in sec . [ sec : charged_higgs ] . it has been seen that two parameters are sufficient to fully determine the mssm higgs sector at tree level . these are commonly taken to be the @xmath62 boson mass @xmath96 and @xmath119 , the ratio of the vacuum expectation values of the higgs fields giving mass to the up- and down - type quarks . this picture is modified significantly , however , by large radiative corrections , arising essentially from an incomplete cancellation of the top and stop loops . in particular , the important prediction @xmath161 is invalidated . among the many parameters of the mssm , a few have been identified as being most relevant for the determination of higgs boson properties . in addition to @xmath96 and @xmath119 , an effective susy breaking scalar mass , @xmath162 , which sets the scale of all squark masses , and a term controlling the amount of mixing in the stop sector , @xmath163 , play the leading role . ( in eq . ( [ radcorr ] ) , the stop mass is directly related to @xmath162 , and stop mixing is neglected . ) the model is further specified by a weak gaugino mass , @xmath164 , the gluino mass , @xmath165 , and the susy higgs mass term @xmath74 . the relation @xmath166 then allows the trilinear higgs - squark coupling @xmath62 ( assumed to be universal ) to be calculated . for large values of @xmath119 , mixing in the sbottom sector becomes relevant too ; it is controlled by @xmath167 . finally , the top quark mass @xmath168 needs to be specified . a few benchmark scenarios @xcite were agreed upon to interpret the searches for mssm higgs bosons . the most widely considered are the so - called `` @xmath8-max '' and `` no - mixing '' ones , where @xmath169 tev , @xmath170 gev , @xmath171 gev and @xmath172 gev . in @xmath8-max , @xmath163 is set equal to @xmath173 ( in the on - shell renormalization scheme ) , while it is set to 0 in the no - mixing scenario . the largest value of @xmath8 is obtained for large @xmath96 and @xmath119 , and is maximized ( minimized ) in the @xmath8-max ( no - mixing ) scenario . in the @xmath8-max scenario , the maximum value of @xmath8 is @xmath174 gev . at lep , the neutral higgs bosons of the mssm have been searched for in two production processes , the higgsstrahlung process @xmath175 @xcite and the associated production @xmath176 @xcite . both processes are mediated by @xmath177-channel @xmath24 boson exchange . with the notations of sec . [ sec : supersymmetry ] , the cross sections are @xmath178 where @xmath179 and @xmath123 are defined in eqs . ( [ beta ] ) and ( [ alpha ] ) , @xmath180 is the sm higgs boson production cross section , @xmath177 is the square of the center - of - mass energy , and @xmath181[1-(m_i - m_j)^2/s ] \ , \\ \bar\lambda & = & \lambda^{3/2}_{ha}[\lambda^{1/2}_{hz}(\lambda_{hz } + 12m^2_z / s ) ] \ .\end{aligned}\ ] ] it is apparent from the above formulae that the two processes are complementary . in practice , the higgsstrahlung process dominates for values of @xmath119 close to unity , while associated production dominates for large values of @xmath119 , if kinematically allowed . in large regions of the mssm parameter space , the @xmath60 decay branching fractions are similar to those of the sm higgs boson . for a mass of 115 gev , these are 74% into @xmath182 , 7% into both @xmath183 and @xmath47 , 8% into @xmath184 , and 4% into @xmath185 @xcite . the @xmath62 boson couples only to fermions , so that its decay branching fraction into @xmath182 is always close to 90% , with most of the rest going into @xmath183 . these same branching fractions also hold for the @xmath60 boson for large values of @xmath119 @xcite . searches for higgs bosons were performed at lep first in @xmath24 boson decays during the lep1 era , and subsequently at increasing center - of - mass energies at lep2 , up to 209 gev in 2000 . in the following , only the searches performed at the highest energies are described . the four lep experiments carried out searches for the sm higgs boson produced via higgsstrahlung , @xmath186 , and the results were combined to maximize the sensitivity . or @xmath187 @xcite , was also considered , but its contribution was found to be negligible in practice . ] four final state topologies were analyzed to cope with the various decay modes of the higgs and @xmath24 bosons : a four - jet topology with two @xmath188-tagged jets , for @xmath189 ; a two @xmath188-tagged jets and two - lepton topology , for @xmath190 , with @xmath191 or @xmath74 ; a two @xmath188-tagged jets and missing energy topology , for @xmath192 ; and a two - jet and two-@xmath193 topology for @xmath194 and @xmath195 . a few candidate events were observed at the edge of the sensitivity domain , but the overall significance was only at the level of 1.7 @xmath196 . a lower mass limit was therefore derived , excluding a sm higgs boson with mass smaller than 114.4 gev @xcite . the higgs boson mass lower limit depends on the strength of the @xmath197 coupling , and the lep collaborations also provided , as a function of the mass of a sm - like higgs boson , an upper limit on @xmath198 , where @xmath199 is a multiplicative factor by which the sm @xmath197 coupling is reduced @xcite . by sm - like , it is meant that the decay branching fractions are similar to those expected from a sm higgs boson . this result is shown in fig . constraints on the mssm parameter space can be deduced from this , since in that case @xmath200 . the upper bound on the factor @xmath198 by which the square of the sm @xmath197 coupling is multiplied , as provided by the lep experiments @xcite . the full curve is the observed limit , the dashed curve the median expected limit in the absence of signal , and the green and yellow bands are the 68% and 95% probability regions around the expected limit.,width=321 ] for higgs boson masses accessible at lep , the structure of the mssm higgs sector is such that the @xmath60 and @xmath62 masses are similar whenever associated production is relevant , i.e. , for large values of @xmath119 . searches for @xmath201 associated production were performed in the four @xmath188-jet final state for @xmath202 and in the two @xmath188-jet and two-@xmath193 topology for @xmath203 . the constraint that the @xmath60 and @xmath62 boson candidate masses should be similar was imposed . the backgrounds from multijet and @xmath204 production were largely reduced by the @xmath188-jet identification requirements , leaving @xmath205 as an irreducible background . no significant excess over the sm background expectation was observed , and production cross section upper limits were derived as a function of @xmath206 . for each benchmark scenario , a scan was performed as a function of @xmath96 and @xmath119 , and in each point of the scan the cross section upper limit was compared to the corresponding prediction , taking into account the slight modifications expected for the values of the @xmath60 and @xmath62 branching fractions into @xmath182 and @xmath183 , as well as the non - negligible difference between @xmath8 and @xmath96 which develops at lower values of @xmath119 . if the cross section upper limit was found to be smaller than the prediction , the @xmath207 ) set was declared excluded . the result of the combination of the searches in the @xmath208 and @xmath201 channels by the four lep experiments @xcite is shown in fig . [ mhtb ] , projected onto the @xmath209 plane in the @xmath8-max and no - mixing scenarios . in the derivation of those results , contributions of the @xmath186 and @xmath210 processes were also taken into account whenever relevant , where @xmath61 is the heavier cp - even higgs boson . domains excluded at 95% cl ( light green ) and 99.7% cl ( dark green ) by the four lep experiments @xcite in the @xmath209 plane in the @xmath8-max ( top ) and no - mixing ( bottom ) benchmark scenarios , with @xmath211 gev . the yellow regions are not accessible theoretically . the dashed lines represent the boundaries of the domains expected to be excluded at 95% cl in the absence of signal . the upper boundaries of the physical regions are indicated for four top quark masses : 169.3 , 174.3 , 179.3 , and 183 gev , from left to right.,title="fig:",width=321 ] + domains excluded at 95% cl ( light green ) and 99.7% cl ( dark green ) by the four lep experiments @xcite in the @xmath209 plane in the @xmath8-max ( top ) and no - mixing ( bottom ) benchmark scenarios , with @xmath211 gev . the yellow regions are not accessible theoretically . the dashed lines represent the boundaries of the domains expected to be excluded at 95% cl in the absence of signal . the upper boundaries of the physical regions are indicated for four top quark masses : 169.3 , 174.3 , 179.3 , and 183 gev , from left to right.,title="fig:",width=321 ] in the most conservative scenario , i.e. , @xmath8-max , it can be seen in fig . [ mhtb ] that the lower limit on the mass of the sm higgs boson holds also for @xmath8 as long as @xmath119 is smaller than about 5 , and that values of @xmath119 between @xmath212 and 2 are excluded for the current average value of the top quark mass , @xmath213 gev @xcite . a lower mass limit of 93 gev is obtained for @xmath206 for large values of @xmath119 . the benchmark scenarios were chosen such that the higgs bosons do not decay into susy particles . an interesting possibility is that the @xmath214 decay mode is kinematically allowed , where @xmath215 is the lsp . if @xmath77-parity is conserved , the lsp is stable and , since it is weakly interacting , the higgs boson decay final state is invisible . searches for such an `` invisible '' higgs boson were performed by the lep experiments , and the combination @xcite yields a mass lower limit identical to that set on the sm higgs boson if the production cross section is the sm one , as is the case for low values of @xmath119 . to cope with fine - tuned choices of mssm parameters , the lep collaborations considered yet other possibilities , e.g. , that the @xmath216 decay mode is kinematically allowed , or that the @xmath217 decay is suppressed . for example , dedicated searches for @xmath218 and for @xmath208 , with @xmath219 in a flavor - independent way , have been performed @xcite . in the end , the sensitivity of the standard searches is only slightly reduced , except for rather extreme parameter choices leading , for instance , to @xmath220 gev , while at the same time @xmath221 . this last possibility is however less unnatural in extensions of the mssm , such as the nmssm where an additional higgs singlet field is introduced @xcite . finally , the possibility that cp is violated in the higgs sector has also been considered . while cp is conserved at tree level , radiative corrections may introduce such a cp violation if the relative phase of @xmath74 and @xmath62 is not vanishing . in such a case , the three mass eigenstates all share properties of @xmath60 , @xmath61 and @xmath62 , so that the signatures of higgs boson production are less distinct . the constraints are accordingly weaker . a dedicated `` cpx '' scenario @xcite was set up to perform quantitative studies . as an example , a region around @xmath222 gev and @xmath223 is not excluded for @xmath224 gev , @xmath170 gev , @xmath225 tev , and @xmath226 tev , when @xmath227 tev and arg@xmath228 . further details can be found in ref . @xcite . at the tevatron , i.e. , in @xmath40 collisions at 1.96 tev , the dominant production mechanism for the sm higgs boson is via gluon fusion , @xmath229 @xcite . in the mass range that is of interest for a sm - like higgs boson of the mssm , namely @xmath230 gev , the dominant decay mode is @xmath231 . such a two - jet final state is totally overwhelmed by standard jet production via the strong interaction , even after @xmath188-jet identification . this is why the sm higgs boson searches at the tevatron have been performed in the associated production processes @xmath232 @xcite , which proceed via @xmath177-channel @xmath58 or @xmath24 exchanges in a similar way to the higgsstrahlung in @xmath23 collisions . in spite of cross sections an order of magnitude smaller than that of gluon fusion , these processes offer better discrimination against the multijet background , by making use of the leptonic decays of the @xmath58 and @xmath24 ( @xmath233 , @xmath234 and @xmath235 ) . these searches for the sm higgs boson apply equally well for the @xmath60 boson of the mssm in the low @xmath119 regime . their sensitivity is , however , still not sufficient to provide any significant constraint . the situation is much more favorable for large values of @xmath119 . in this regime , the @xmath62 boson is almost mass degenerate with either the @xmath60 or @xmath61 boson , depending on whether @xmath96 is less than or greater than @xmath236 , where @xmath236 is the maximum value that @xmath8 can take , e.g. , 135 gev in the @xmath8-max scenario . in the following , the two nearly degenerate higgs bosons are collectively denoted @xmath237 . their couplings to @xmath188 quarks and @xmath193 leptons are enhanced by a factor @xmath119 with respect to the sm couplings . as a result , the contribution of the @xmath188 quark loop to their production via gluon fusion is enhanced by a factor @xmath238 . although this is not sufficient to render feasible a detection in the @xmath239 decay mode , this is not the case for the @xmath240 decay mode , which has a branching fraction of @xmath241 . both cdf and d required one of the two @xmath193 leptons to decay leptonically ( @xmath242 ) to ensure proper triggering . three final state topologies were considered : @xmath243 , @xmath244 , and @xmath245 , all with missing transverse energy from the @xmath193 decay neutrinos . here @xmath246 denotes a @xmath193 lepton decaying into hadrons and a neutrino . the dominant , irreducible background comes from @xmath24 production with @xmath247 , but there also remains a substantial component from @xmath248+jet , where the jet is misidentified as a @xmath193 lepton . this background was reduced , for instance , by requiring a low transverse mass of the lepton and the . the final discriminating variable was chosen to be the visible mass @xmath249 , constructed from the @xmath193 visible products and from the . the distribution of @xmath250 obtained by cdf @xcite in a 1.8 fb@xmath29 data sample is shown in fig . [ cdfttmvis ] . from this distribution , as well as from a similar one in the @xmath245 channel , a cross section upper limit on @xmath237 production was derived , which in turn was translated into exclusion domains in the @xmath251 plane within benchmark scenarios . the result obtained in the @xmath8-max and no - mixing scenarios is shown in fig . [ cdfttmatb ] . similar results have been obtained by d @xcite . the calculations of ref . @xcite were used to derive these results as well as those reported in the rest of this section . visible mass distribution in the @xmath252 channels from the cdf search for @xmath253 @xcite . the signal contribution indicated corresponds to the cross section upper limit set with this data.,width=321 ] domains in the @xmath251 plane excluded by the cdf search for @xmath253 @xcite . the domains excluded at lep are also indicated.,width=321 ] because of the enhanced coupling of @xmath237 to @xmath188 quarks at high @xmath119 , the production of higgs bosons radiated off a @xmath188 quark may be detectable in the @xmath239 decay mode in spite of the large background from multijet events produced via the strong interaction ( `` qcd background '' ) . this process can be described in the so - called four - flavor or five - flavor schemes , and it has been shown that the two approaches yield very similar results @xcite . in the four - flavor scheme , the main contribution comes from gluon fusion , @xmath254 , while the main one in the five - flavor scheme comes from @xmath255 . because one of the final state @xmath188 quarks ( a spectator @xmath188 quark in the five - flavor scheme ) tends to be emitted with a low transverse momentum , the searches required only three @xmath188 jets to be identified . the signal was searched for by inspecting the mass distribution of the two jets with highest transverse momenta in the sample of events with three @xmath188-tagged jets . further discrimination against the qcd background was provided by the mass of the charged particles in the tagged jets ( at cdf @xcite ) or by the inclusion of additional kinematic variables in a likelihood discriminant ( at d @xcite ) . the qcd background was modeled using a combination of information from control samples in the data , where one of the jets is not @xmath188-tagged , and from monte carlo simulations of the various processes contributing to the background ( @xmath256 , @xmath257 , @xmath258 , @xmath259 , @xmath260 , etc . , where @xmath261 represents a light quark , @xmath262 , @xmath263 , @xmath177 , or a gluon ) . the mass distribution obtained by cdf in a 1.9 fb@xmath29 data sample is shown in fig . [ cdfbbbmbb ] , with the individual background contributions displayed . no signal was observed , and production cross section upper limits were derived , from which exclusion domains in the @xmath251 plane were determined in various benchmark scenarios . the d result obtained with 2.6 fb@xmath29of data in the @xmath8-max scenario is shown in fig.[dzbbbmatb ] . in the derivation of the cross section upper limits and exclusion domains , special attention was given to a proper handling of the higgs boson width , which is enhanced by a factor @xmath264 at tree level and therefore becomes large with respect to the mass resolution . ( this is not the case for @xmath240 because of the degradation of the mass resolution due to the missing neutrinos . ) . it should also be noted that the exclusion domain is quite sensitive to the model parameters . it is smaller in the no - mixing scenario , and also if @xmath74 is positive . these effects due to potentially large susy loop corrections to the production cross sections and decay widths tend to cancel in the search for @xmath240 described above @xcite . fit to the mass of the two jets with highest transverse momenta in the cdf sample of events with three @xmath188-tagged jets @xcite . the contributions of the various multijet backgrounds and of a signal with a mass of 150 gev are indicated.,width=321 ] domain in the @xmath251 plane excluded by the d search for @xmath239 in events with three @xmath188-tagged jets in the @xmath8-max scenario @xcite.,width=321 ] finally , higgs bosons produced in association with @xmath188 quarks can also be searched for in the @xmath240 decay mode . although the branching fraction is an order of magnitude smaller than the one of @xmath239 , the signal is much easier to disentangle from the background . a d analysis @xcite was performed where one of the @xmath193 leptons decays into a muon and neutrinos , while the other decays into hadrons and a neutrino . furthermore , a @xmath188-tagged jet was required , at which point the main background comes from top quark pair production , @xmath265 . a neural network was used to discriminate signal and @xmath266 background , taking advantage of the large differences in their kinematic properties . the result , based on 1.2 fb@xmath29 of data , is shown in fig . [ dzbttmatb ] . given the limited amount of integrated luminosity used up to now , this channel appears to be quite promising . domain in the @xmath251 plane excluded by the d search for @xmath240 produced in association with a @xmath188 quark in the @xmath8-max scenario @xcite . the light green shaded region is the extension of the lep exclusion region to @xmath267.,width=321 ] many extensions of the sm involve more than one complex doublet of higgs fields . two - higgs doublet models ( 2hdms ) fall into three main categories . in type i models , all quarks and leptons couple to the same higgs doublet . in type ii models , down - type fermions couple to the first higgs doublet , and up - type fermions couple to the second higgs doublet . flavor - changing neutral currents are naturally avoided in type i and type ii 2hdms . in type iii models , fermions couple to both doublets , and flavor - changing neutral currents must be avoided using other strategies . in addition to the three neutral higgs bosons discussed in the previous section , 2hdms involve a pair of charged higgs bosons , @xmath160 . most of the experimental results on charged higgs bosons have been obtained within the context of type ii 2hdms , of which the mssm is a specific instance . further details on extended higgs boson sectors may be found in ref . @xcite . in type ii 2hdms , the charged higgs boson decay width into a fermion pair @xmath268 is @xmath269 where @xmath270 is the number of colors , and we have approximated @xmath271 in the phase space factor . charged higgs bosons therefore decay into the heaviest kinematically - allowed fermions : @xmath272 at large @xmath119 and @xmath273 at low @xmath119 for charged higgs boson masses to which current accelerators are sensitive . at lep , charged higgs bosons are produced in pairs through @xmath274 @xcite . the production cross section depends only on sm parameters and on the mass of the charged higgs boson . the process @xmath275 has a significantly lower cross section . the charged higgs boson can decay into @xmath276 or @xmath277 . in searches for type i 2hdm higgs bosons , the decay @xmath278 @xcite was also considered , as in refs . the interpretation of the search results generally assumed that br(@xmath279 ) + br(@xmath280 ) = 1 , where the dominant @xmath281 flavors are @xmath276 , due to the cabbibo suppression of @xmath282 . this assumption leads to the consideration of three topologies for pair - produced charged higgs bosons : four jets from @xmath283 , two jets , a @xmath193 lepton and missing energy from @xmath284 and two charge conjugate , acoplanar momenta on a plane transverse to the beam axis . if this angle is less than @xmath285 , the @xmath193 leptons are said to be acoplanar . ] @xmath193 leptons from @xmath286 . direct searches for pair production of charged higgs bosons have been published by all four lep experiments @xcite . each topological analysis began with a general selection for the expected number of jets and @xmath193 leptons , followed by more sophisticated techniques . the main difficulty in these analyses was separating the signal from the nearly identical signature of @xmath287 production ; selection criteria usually included a mass - dependent optimization . techniques such as linear discriminants , likelihood estimators , and jet - flavor tagging were used in these analyses . the @xmath288 channel had additional complexity due to the missing neutrinos , which removed the possibility of reconstructing the @xmath160 candidate masses and of improved discrimination from the equal - mass constraint . however , final states with @xmath193 leptons can benefit from extracting information about their polarization ; the @xmath289 lepton from a @xmath290 boson ( a scalar ) is produced in a helicity state opposite to that of a @xmath289 lepton from @xmath291 decay . the lep experiments have combined the results of their searches for charged higgs bosons into one result @xcite based on common assumptions . the total dataset has an integrated luminosity of 2.5 fb@xmath29 , collected at center - of - mass energies between 189 and 209 gev . the possible decays were restricted to @xmath292 and @xmath293 in a general 2hdm framework . the combined mass limit is shown in fig . [ fig : lep - hpm - combo ] as a function of br(@xmath294 ) . a lower bound of 78.6 gev holds for any value of the branching ratio . limit on the charged higgs boson mass as a function of br(@xmath294 ) , from the combined data of the four lep experiments at center - of - mass energies from 189 to 209 gev . the expected exclusion limit is shown as a thin solid line and the observed limit as a thick solid line ; the shaded region is excluded @xcite.,width=321 ] at the tevatron , pair production of charged higgs bosons is expected to occur at a very low rate . however , in contrast to searches at lep , advantage can be taken of the large mass of the top quark , which opens new ways to search for evidence of charged higgs bosons . two approaches have been considered , depending on whether the charged higgs boson is lighter or heavier than the top quark . in the first case , the top quark can decay into a @xmath290 boson and a @xmath188 quark @xcite . for heavier charged higgs bosons , resonant production of a single @xmath290 boson followed by the decay @xmath295 is the most promising process @xcite . for a charged higgs boson mass of 100 gev , the branching ratios for the top quark decay into @xmath296 , under the assumption that br(@xmath297)+br(@xmath298)=1 , and for the various @xmath290 decay channels , as a function of tan@xmath179 . from ref . @xcite.,width=321 ] in the sm , the top quark decays almost exclusively into a @xmath58 boson and a @xmath188 quark , and the possible signatures of @xmath299 pair production are associated with the various combinations of @xmath58-boson decay channels . if the charged higgs boson is lighter than the top quark , the decay @xmath298 will compete with the standard @xmath297 mode . the decay of the charged higgs boson , with branching ratios different from those of the @xmath58 boson , will modify the fractions of events observed in the various topologies , compared to the sm expectations . pair production are roughly 50% in six jets ; 14% in each of @xmath300 , @xmath74 , and @xmath301 ; 1% in each of @xmath302 , @xmath303 , and @xmath304 ; and 2% in each of @xmath245 , @xmath305 , and @xmath306 . in each of these channels , there are two @xmath188 jets . ] the qualitative aspects and magnitude of these modifications depend on the model parameters . the dependence on @xmath119 of the top quark decay ratio to @xmath296 and of the various charged - higgs boson decay channels is shown in fig . [ fig : tev - br ] for @xmath307 gev and a typical set of mssm parameters , with qcd , susy - qcd and electroweak radiative corrections to the top and bottom quark yukawa couplings calculated with the cpsuperh code @xcite . the dominant @xmath290 boson decay channels are @xmath276 at low values of @xmath119 and @xmath293 at high values ; with this set of parameters , @xmath290 boson decays to @xmath308 are also allowed , although always at a small rate . the @xmath309 decay mode becomes relevant for charged higgs boson masses closer to the top quark mass . it can be seen that charged higgs bosons will be most prominent at high and low values of @xmath119 . two simplified models address each of these regions : the tauonic model , with br(@xmath294)=1 , and the leptophobic model , with br(@xmath292)=1 . the tauonic model is a very good approximation to the mssm with @xmath310 , while purely leptophobic charged higgs bosons can be found in some multi - higgs - doublet models @xcite . analyses based on measurements of @xmath299 final states include an earlier cdf search in @xmath311 of data @xcite and a recent d analysis of @xmath312 of data @xcite . the yields observed in the various topologies were compared to what would be expected in models with charged higgs bosons , taking into account the @xmath298 and @xmath160 decay branching ratios predicted as a function of the higgs boson mass and @xmath119 . in particular , no excess of final states involving @xmath193 leptons was observed , nor was any disappearance of final states with one or two leptons , jets and , as would be expected at large and small @xmath119 , respectively . figure [ fig : d0-h - from - t ] displays the exclusion domain in the plane of the charged higgs boson mass and @xmath119 from the d analysis @xcite , for leptophobic and tauonic models . the cdf analysis excludes br(@xmath298 ) @xmath313 0.4 for a tauonic @xmath160 boson @xcite . limit on the mass of the charged higgs boson as a function of @xmath119 from the d search in top quark decays @xcite.,width=321 ] in a recent analysis based on a data sample of 2.2 fb@xmath29 , the cdf collaboration used a different approach to search for a leptophobic charged higgs boson in top quark decays @xcite . the search was performed in the @xmath314 final states with two @xmath188-tagged jets , where the lepton ( electron or muon ) , the neutrino ( responsible for the missing ) , and a @xmath188 jet were the signature of a @xmath315 decay , while the other top quark of the @xmath299 pair was assumed to decay to either @xmath316 or @xmath317 . the @xmath299 events were fully reconstructed , taking the masses of the @xmath58 boson and of the top quark into account as constraints to assign correctly each of the @xmath188 jets to its parent @xmath318 or @xmath319 . templates of the mass of the dijet system reconstructed from the non-@xmath188 jets were used to extract limits on the branching ratio of @xmath320 , as shown in fig . [ fig : cdf - h - from - t-2 ] . for a leptophobic charged higgs boson , upper limit on the branching ratio br(@xmath320 ) as a function of the higgs boson mass from a cdf search in top quark decays @xcite.,width=321 ] if the charged higgs boson is heavier than the top quark , it will decay dominantly into @xmath321 . the resonant production of such a charged higgs boson leads to a final state similar to the one resulting from single top @xmath177-channel production , @xmath322 . therefore the analyses developed for the search for single top production can be applied to the search for a charged higgs boson . such an analysis was performed by the dcollaboration @xcite , in the topology arising from a subsequent @xmath323 decay . the large @xmath160 mass , reconstructed from the decay products imposing the @xmath58 boson and top quark mass constraints , was used as discriminating variable . no excess was observed over sm background predictions , and upper limits were set on the production of a charged higgs boson . the results are , however , not sensitive to type ii 2hdms , but provide some exclusion in type i 2hdms . as explained in sec . [ sec : supersymmetry ] , the main features of susy models for phenomenology are related to the type of mediation mechanism for susy breaking , to the choice of soft breaking terms , and to whether or not @xmath77-parity is assumed to be conserved . the most widely studied models involve gravity - mediation of susy breaking . in the minimal form of such models , msugra , @xmath77-parity is conserved , and only five parameters are needed beyond those already present in the standard model : a universal gaugino mass @xmath324 , a universal scalar mass @xmath93 , and a universal trilinear coupling @xmath325 , all defined at the scale of grand unification , and @xmath119 and the sign of @xmath74 . the low energy parameters , including @xmath90 , are determined by the renormalization group equations and by the condition of electroweak symmetry breaking . in addition , it is commonly assumed that the lsp is the lightest neutralino @xmath215 . a somewhat less constrained model keeps @xmath74 and @xmath96 as independent low energy parameters , which is in effect equivalent to decoupling the higgs scalar masses from the masses of the other scalars . such a model was largely used at lep . many studies have been performed where the assumption of @xmath77-parity conservation is dropped , while keeping unchanged the other features of those msugra inspired models . if @xmath77-parity is violated , the superpotential is allowed to contain lepton or baryon number - violating terms @xcite @xmath326 where @xmath56 and @xmath53 are lepton and quark doublet superfields , @xmath57 and @xmath55 are lepton and down - type quark singlet superfields , and @xmath327 , @xmath328 and @xmath329 are generation indices . these terms are responsible for new couplings through which the lsp decays to sm particles . the simultaneous occurrence of different coupling types is however strongly constrained , e.g. , by the bounds on the proton lifetime , which is why it is commonly assumed that only one of the @xmath77-parity violating terms is present in the superpotential . in models with gauge - mediated susy breaking ( gmsb ) , the lsp is a very light gravitino @xmath330 , and the phenomenology is governed by the nature of the nlsp . in the minimal such model , mgmsb , all susy particle masses derive from a universal scale @xmath5 , and in most of the parameter space the nlsp is either the lightest neutralino @xmath215 or the lighter stau @xmath331 , the latter occurring preferentially at large @xmath119 . the couplings of the gravitino depend on yet another parameter , the susy - breaking scale @xmath87 , which can be traded for the lifetime of the nlsp . anomaly - mediation of susy breaking ( amsb ) generically leads to a neutralino lsp which is almost a pure wino @xmath106 , and has a small mass splitting with the lighter chargino . as a consequence , this chargino may acquire a phenomenologically relevant lifetime , possibly such that it behaves like a stable particle . most of the searches for susy particles were performed within a `` canonical scenario , '' the main features of which are borrowed from msugra : @xmath77-parity conservation , universal gaugino mass terms , a universal sfermion mass term , and a neutralino lsp . because of @xmath77-parity conservation , susy particles are produced in pairs , and each of the produced susy particles decays into sm particles accompanied by an lsp . since the lsp is neutral and weakly interacting , it appears as missing energy , which is the celebrated signature of susy particle production . alternatively , if @xmath77-parity is not conserved , the lsp decays to sm particles , so that no missing energy is expected beyond that possibly arising from neutrinos . the signature of susy particle production is therefore to be sought in an anomalously large multiplicity of jets or leptons . the @xmath77-parity violating couplings can also make it possible that susy particles are produced singly , rather than in pairs . in @xmath77-parity conserving scenarios other than the canonical one , additional or different features are expected . in gmsb , each of the pair - produced susy particles decays into sm particles and an nlsp . the nlsp further decays into its sm partner and a gravitino . with a neutralino nlsp in the mass range explored up to now , the dominant decay is @xmath332 , so that the final state contains photons , with missing energy due to the escaping gravitinos . with a stau nlsp , the decay is @xmath333 . if the stau lifetime is so long that it escapes the detector before decaying , the final state from stau pair production does not exhibit any missing energy , but rather appears as a pair of massive stable particles . a similar final state may also arise from chargino pair production in amsb . long - lived gluinos can lead to spectacular signatures if they are brought to rest by energy loss in the detector material . except for the gluino , all susy particles are produced in a democratic way in @xmath23 collisions via electroweak interactions . it is therefore natural that the searches at lep were targeted toward the lightest ones . the results of these searches could further be combined within a given model , thus providing constraints on the model parameters . in contrast , it is expected that the most copiously produced susy particles in hadron collisions , such as @xmath40 at the tevatron , will be colored particles , namely squarks and gluinos . their detailed signature however depends on the mass pattern of the other susy particles , which may be present in their decay chains . this is why a specific model , usually msugra , is needed to express the search results in terms of mass constraints . thanks to lower masses and more manageable backgrounds , the search for gauginos produced via electroweak interactions can be competitive at hadron colliders for model parameter configurations where their leptonic decays are enhanced . in @xmath23 collisions , the production cross sections of susy particles are similar to those of their sm partners , except for the phase space reduction due to their larger masses . the data collected at the highest lep energies , up to 209 gev , are therefore the most relevant for susy particle searches . mixing effects may however reduce these cross sections , as is the case for instance for neutralinos with a small higgsino component , in which case the large integrated luminosity accumulated by the lep experiments at lower energies also contributes to the search sensitivity . although the center - of - mass energy of 1.96 tev in @xmath40 collisions at the tevatron allows higher new particle masses to be probed , large integrated luminosities are needed because of the rapid pdf fall off at high @xmath334 , as explained in sec . [ subsec : context ] . the search for susy particles at the tevatron is also rendered more challenging than at lep because of the large cross sections of the background processes . in the searches for squarks and gluinos , signal production cross sections of the order of 0.1 pb at the edge of the sensitivity domain are to be compared to the total inelastic cross section of 80 mb . in the searches for gauginos , with similar signal production cross sections in the mass range probed , the main backgrounds are @xmath233 and @xmath335 , with cross sections at the 2.7 nb and 250 pb level per lepton flavor . in @xmath39 collisions at hera , the most promising susy particle production process is single squark resonant production via an @xmath77-parity violating @xmath336 or @xmath337 coupling , with a cross section depending not only on the squark mass , but also on the value of the coupling involved . the decay of the squark produced could be either direct , via the same @xmath338 coupling as for its production , or indirect through a cascade leading to the lsp , which in turn decays to two quarks and a neutrino or an electron . the mass reach at hera is the full center - of - mass energy of 320 gev , but the production of squarks with masses close to this bound involves quarks at large @xmath334 values , so that the effective reach is substantially smaller , even for large values of the @xmath338 coupling . as mentioned above , the characteristic signature of susy particle production in the canonical scenario is missing energy carried away from the detector by the lsps at the end of the decay chains . the main channels for susy particle searches in @xmath23 collisions are slepton @xcite , chargino @xcite and neutralino @xcite production . squark pair production @xcite can also be relevant in some specific cases @xcite . * sleptons : * in @xmath23 annihilation , the search for susy particles that involves the least set of hypotheses for its interpretation is the search for smuons . pair production proceeds via @xmath339 exchange in the @xmath177-channel . because of the small mass of the muon , the smuon mass eigenstates can be identified with the interaction eigenstates , of which @xmath340 is the lighter one in models with slepton and gaugino mass unification . the search results were interpreted under this assumption , which is furthermore conservative , as the coupling of the @xmath340 to the @xmath24 boson is smaller than that of the @xmath341 . only one parameter is needed to calculate the smuon pair production cross section , the smuon mass @xmath342 . the sole decay mode of a @xmath340 nlsp is @xmath343 , so that smuon pair production leads to a final state consisting of two acoplanar muons with missing energy and momentum . the topology of this final state also depends on the mass of the lsp . if @xmath344 is small , the final state is very similar to that arising from @xmath58 pair production , with both @xmath58 bosons decaying to a muon and a neutrino . if the @xmath345 mass difference is small , the final state muons carry little momentum , so that the selection efficiency is reduced . in that configuration , the main background comes from `` @xmath346 interactions , '' @xmath347 , where the spectator electrons @xmath348 escape undetected in the beam pipe . the lsp mass @xmath344 is therefore needed , in addition to the smuon mass , to interpret the search results . the constraints obtained in the @xmath349 plane by the four lep experiments @xcite are shown in fig . [ smuonlep ] . if the assumption that the smuon is the nlsp is dropped , further specification of the model is needed to turn the search results into mass constraints . an example is shown in fig . [ smuonlep ] in the case of gaugino mass unification , for the specified values of @xmath74 and @xmath119 . a slight reduction of the excluded domain is observed for low values of @xmath344 , due to the competition of the @xmath350 decay mode , with @xmath351 . depending on @xmath344 , smuon masses smaller than 95 to 99 gev are excluded , except for @xmath345 mass differences below 5 gev . region in the ( @xmath352 ) plane excluded by the searches for smuons at lep @xcite . the dotted contour is drawn under the assumption that the smuon decay branching ratio into @xmath353 is 100% . , width=321 ] because of the larger @xmath193 mass , compared to the muon mass , the hypothesis that the stau mass eigenstates can be identified with the interaction eigenstates may not hold , especially for large values of @xmath119 that enhance the off - diagonal elements of the mass matrix in eq . ( [ staumass ] ) . the coupling to the @xmath24 boson of the lighter stau mass eigenstate @xmath354 may therefore be reduced with respect to the smuon coupling , and even vanish . moreover , because there is at least one neutrino in each @xmath193 decay , the visible energy of the final state arising from stau pair production is smaller than in the case of smuons , so that the selection efficiency is reduced . the mass lower limits obtained at lep are therefore lower for staus than for smuons , from 86 to 95 gev , depending on @xmath344 , provided the @xmath355 mass difference is larger than 7 gev @xcite . as for smuons , the selectron mass eigenstates can be identified with the interaction eigenstates . but because of the contribution of @xmath318-channel neutralino exchange to selectron pair production , the gaugino sector of the model , mass spectrum and field contents , has to be specified to interpret the results of the searches for acoplanar electrons . with gaugino mass unification and for @xmath356 and @xmath171 gev , a selectron mass lower limit of 100 gev was obtained for @xmath357 gev @xcite . neutralino @xmath318-channel exchange can furthermore mediate associated @xmath358 production . this process is useful if the @xmath359 mass difference is small , because the electron from the @xmath360 decay can be energetic enough to lead to an apparent single electron final state . both gaugino and slepton mass unifications have to be assumed for the masses of the two selectron species to be related . under these assumptions , a lower limit of 73 gev was set on @xmath361 , independent of the @xmath362 mass difference @xcite . from the measurement of the invisible width of the @xmath24 boson @xcite , a general mass lower limit of 45 gev can be deduced for a sneutrino lsp or nlsp . * charginos and neutralinos : * as evident from eq . ( [ chargino ] ) , three parameters are sufficient to fully specify the masses and field contents in the chargino sector . these may be taken to be @xmath164 , @xmath74 , and @xmath119 . the lighter of the two charginos will simply be denoted `` chargino '' in the following . to specify the neutralino mass matrix of eq . ( [ neutralino ] ) , one more parameter , @xmath363 , is needed . if gaugino mass unification is assumed , the two gaugino masses are related by @xmath364 . unless otherwise specified , this relation is assumed to hold in the following . charginos are pair produced via @xmath177-channel @xmath339 and @xmath318-channel @xmath365 exchanges , the two processes interfering destructively . the three - body final states @xmath366 are reached in chargino decays via virtual @xmath58 or sfermion exchange . if kinematically allowed , two - body decays such as @xmath367 are dominant . similarly , neutralino pair or associated production proceed via @xmath177-channel @xmath24 and @xmath318-channel selectron exchanges , and @xmath368 three - body decays to @xmath369 via virtual @xmath24 or sfermion exchange ; whenever kinematically allowed , two - body decays such as @xmath370 are dominant . if sfermions are heavy , chargino decays are mediated by virtual @xmath58 exchange , so that the final states arising from chargino pair production are the same as for @xmath58 pairs , with additional missing energy from the two neutralino lsps : all hadronic ( @xmath371 ) , mixed ( @xmath372 ) , and fully leptonic ( @xmath373 ) . selections were designed for these three topologies and for various @xmath374 regimes , with no excess observed over sm backgrounds . from a scan over @xmath164 , @xmath74 , and @xmath119 , a chargino mass lower limit of 103 gev was derived for @xmath375 gev @xcite . for smaller sneutrino masses , the limit is reduced by the destructive interference in the production . this limit holds for @xmath376 tev . for larger @xmath164 values , the selection efficiency decreases rapidly as the @xmath377 mass difference becomes smaller . if this mass difference becomes so small that even the @xmath378 decay mode is closed , the chargino becomes long lived . searches for charged massive stable particles , in which advantage is taken of their larger ionization power , were designed to cope with this configuration . for slightly larger mass differences , the visible final state is so soft that even triggering becomes problematic . chargino pair production can however be `` tagged '' by an energetic photon from initial state radiation , @xmath379 , providing access to those almost invisible charginos , although at a reduced effective center of mass energy . the combination of these analysis techniques allowed chargino masses smaller than 92 gev to be excluded , irrespective of the @xmath377 mass difference @xcite . for lower sfermion masses , the sensitivity of the former analyses is reduced first because of the destructive interference between the @xmath177-channel @xmath339 and @xmath318-channel sneutrino exchanges , and second because of the opening of two - body decays . the latter effect is specifically detrimental in the `` corridor '' of small @xmath380 mass differences , where the final state from the @xmath381 decays becomes invisible in practice . gaugino mass unification allows indirect limits on charginos to be obtained , based on constraints on the parameter space resulting from searches for pair or associated neutralino production , e.g. , @xmath382 or @xmath383 . in order to relate all production cross sections and decay branching fractions , it is however necessary to fully specify the sfermion spectrum , which is done with the assumption of sfermion mass unification . the results of the chargino and neutralino searches are then expressed as exclusion domains in the @xmath384 plane for selected values of @xmath119 and @xmath93 . the invisible two - body decay @xmath370 can however cause a large sensitivity reduction . since this configuration occurs for low @xmath93 values , constraints arising from the slepton searches can be used to mitigate this effect . with gaugino and sfermion mass unification , the slepton masses are related to the model parameters by @xmath385 , so that a limit on @xmath386 can be turned into a limit on @xmath164 for given values of @xmath119 and @xmath93 . after a proper combination of the searches for charginos , neutralinos and sleptons , an example of which is shown in fig . [ m2mucombi ] , it turns out that the chargino mass limit obtained in the case of heavy sfermions is only moderately degraded . regions in the @xmath384 plane excluded by the lep i constraints ( 1 ) , and by the searches for charginos ( 2 ) , neutralinos ( 3 ) and sleptons ( 4 ) at lep ii , for @xmath387 and @xmath388 gev . the region ( 5 ) is excluded by the higgs boson searches at lep ii . this figure is from ref . @xcite.,width=321 ] direct searches for the lightest neutralino had been performed at lower energy @xmath23 colliders , pep and petra , in the reaction @xmath389 , where the photon from initial state radiation is used to tag the production of an invisible final state . at lep , at or above the @xmath24 resonance , the irreducible background from @xmath390 is too large to obtain competitive results . furthermore , production via @xmath177-channel @xmath24 exchange may simply vanish , e.g. , if the lsp is a pure photino , while production via @xmath318-channel selectron exchange can be made negligible if selectrons are sufficiently heavy . indirect limits on the mass of the lsp can however be obtained within constrained models . with gaugino mass unification , @xmath344 is typically half the chargino mass . as a result , the chargino mass limit translates into a @xmath391 mass lower limit of 52 gev for heavy sfermions and large @xmath119 . if sfermion mass unification is used in addition , a limit of 47 gev is obtained at large @xmath119 , independent of @xmath93 . this limit is set by searches for sleptons in the corridor . for low values of @xmath119 , constraints from the higgs boson searches can be used , as was shown in sec . [ subsec : lephiggs ] for benchmark scenarios . a complete scan over @xmath93 , @xmath324 , @xmath74 and @xmath119 was performed and , for each parameter set , the maximum @xmath60 mass predicted was compared to the experimental limit , and the constraints from chargino and slepton searches were included . the translation of the scan result in terms of excluded domain in the @xmath392 plane is shown in fig . [ lsp ] , from which a @xmath391 mass lower limit of 47 gev is derived @xcite . within the more constrained msugra scenario , wherein @xmath74 is calculated from the other parameters , this limit becomes 50 gev @xcite . lower mass limit for the lightest neutralino as a function of @xmath119 , inferred in the conventional scenario from searches at lep for charginos , sleptons , and neutral higgs bosons @xcite . the dashed contour is the limit obtained for large @xmath93.,width=321 ] * squarks : * on general grounds , the mass reach for strongly interacting particles is expected to be substantially higher at the tevatron than at lep . for some specific configurations , however , the searches at the tevatron become inefficient , in which cases the results obtained at lep remain of interest . this is particularly relevant for third generation squarks which may be substantially lighter than the other squarks , as motivated in sec . [ sec : supersymmetry ] . the lighter third generation mass eigenstates are simply denoted stop and sbottom , @xmath393 and @xmath394 , in the following . in the mass range accessible at lep , and given the chargino mass limit which effectively forbids @xmath395 , the stop is expected to decay into a charm quark and a neutralino , @xmath396 , as long as @xmath397 @xcite . because this decay is a flavor - changing loop process , the stop lifetime can be large enough to compete with the hadronization time , and the simulation programs were adjusted to take this feature into account . the final state from stop pair production exhibits an acoplanar jet topology , for which no signal was observed above standard model backgrounds . as already explained for staus , the amount of mixing between the weak eigenstates can be such that the stop does not couple to the @xmath24 boson . in this worst case scenario , stop mass lower limits ranging from 96 to 99 gev were obtained , depending on the @xmath391 mass , as long as @xmath398 gev @xcite . for smaller @xmath399 mass differences , long - lived @xmath77-hadrons may be produced in the stop hadronization process . the production of such @xmath77-hadrons and their interaction in the detector material were taken into account in a dedicated search , from which a stop mass lower limit of 63 gev was derived , valid for any @xmath400 @xcite . for specific parameter choices , and in spite of the slepton mass limits , it can be that the @xmath401 decay is kinematically allowed , in which case it is dominant . from a search for events exhibiting jets , leptons and missing energy , a stop mass lower limit of 96 gev was obtained , valid for sneutrino masses smaller than 86 gev @xcite . the case of a light sbottom is much simpler , as the tree - level @xmath402 decay mode is dominant . from searches for acoplanar @xmath188-flavored jets , a mass lower limit of about 95 gev was obtained in the worst case scenario where the sbottom does not couple to the @xmath24 @xcite . the program most widely used for the calculation of susy particle production cross sections at the tevatron is prospino @xcite , which provides next - to - leading order accuracy . the results reported below were generally obtained with the cteq6.1 m pdf set @xcite . various codes were used to calculate the low energy susy spectrum from initial parameters at the grand unification scale : suspect @xcite , softsusy @xcite , and isajet @xcite . this may introduce slight inconsistencies when comparing results in different channels or from different experiments in terms of parameters at the high scale . the production of susy particles was in general simulated with pythia @xcite , with decays modeled with sdecay @xcite or with isasugra as implemented in pythia . typically , sm backgrounds were simulated with alpgen @xcite for the production of @xmath58 and @xmath24 bosons in association with jets , or with pythia otherwise . as already mentioned in sec . [ subsec : strategies ] , the channels most relevant for susy particle searches at hadron colliders are the production of squarks and gluinos on the one hand , of electroweak gauginos on the other . for squarks and gluino , the search is conducted in events exhibiting a jets+topology @xcite , while for electroweak gauginos , it is conducted in the trilepton final state @xcite . * generic squarks and gluinos : * depending on the squark and gluino mass hierarchy , different pair production processes via the strong interaction are expected to contribute in @xmath40 collisions at the tevatron : @xmath403 and , to a lesser extent , @xmath404 , if @xmath405 ; @xmath406 if @xmath407 ; and all of these processes , as well as @xmath408 , if the squark and gluino masses are similar . if @xmath409 , squarks are expected to decay directly into a quark and a gaugino , @xmath410 , where @xmath411 is dominantly @xmath215 for @xmath412 , and @xmath413 or @xmath414 for @xmath415 . if @xmath416 , gluinos are expected to decay via virtual squark exchange into a quark , an antiquark , and a gaugino , @xmath417 , where @xmath411 is typically @xmath413 or @xmath414 @xcite . the heavier gauginos further decay into a fermion - antifermion pair and an lsp , @xmath215 , so that there is always some missing in the final state . more detailed predictions can be made only within a specific model such as msugra . the aforementioned production processes have been searched for by cdf and d in topologies involving at least two jets , four jets and three jets , all with large @xmath418 . initial and final state radiation of soft jets can increase further those jet multiplicities . a first class of background to squark and gluino production arises from processes with intrinsic @xmath418 , such as ( @xmath233)+jets , where the lepton escapes detection , or ( @xmath419)+jets , which is irreducible . monte carlo simulations were used to estimate those backgrounds , after calibration on events where leptons from @xmath233 or @xmath335 are detected . another class of background is due to multijet production by strong interaction . although there is no intrinsic in such events , fake can arise from jet energy mismeasurements ( and also real from semileptonic decays of heavy flavor hadrons ) . in such events , the distribution decreases quasi - exponentially , and the direction of the tends to be close to that of a mismeasured jet . requiring sufficiently large and applying topological selection criteria allows this background to remain under control . while d applied criteria tight enough to reduce this background to a negligible level , cdf estimated its remaining contribution based on simulations calibrated on control samples . no excesses of events were observed over sm backgrounds , which was translated into exclusion domains in the plane of squark and gluino masses . to this end , a specific susy model had to be chosen , so that the masses and decay modes of all the gauginos involved in the decay chains could be determined . the model used by both cdf and d was msugra , with @xmath420 , @xmath421 , and @xmath223 ( cdf ) or 3 ( d ) . the production of all squark species was considered , except for the third generation ( cdf ) or for stops ( d ) , and the squark mass quoted was the average of the masses of the squarks considered . finally , the large theoretical uncertainties associated to the choices of pdfs and of the factorization and renormalization scales had to be taken into account when turning cross section upper limits into exclusion domains in terms of masses . based on an integrated luminosity of 2.1 fb@xmath29 , d excluded the domain shown in fig . [ sqgl ] , from which lower limits of 379 and 308 gev were derived for the squark and gluino masses , respectively , as well as a lower limit of 390 gev if @xmath422 @xcite . similar results were obtained by the cdf collaboration @xcite . region in the ( @xmath423 ) plane excluded by d @xcite and by earlier experiments . the red curve corresponds to the nominal scale and pdf choices . the yellow band represents the uncertainty associated with these choices . the blue curves represent the indirect limits inferred from the lep chargino and slepton searches.,width=321 ] * third generation squarks : * as already mentioned , a stop nlsp decays into a charm quark and a neutralino as long as @xmath397 . the final state from stop pair production therefore consists in acoplanar charm jets and . because only one of the squark species is now produced , the cross section is smaller than for generic squarks , and the mass reach is therefore lower . as a consequence , the jets are softer , and there is also less . the corresponding loss of sensitivity was attenuated by making use of heavy - flavor tagging , which resulted in the exclusion domain shown in fig . [ stop ] , obtained by d @xcite from an analysis of 1 fb@xmath29 of data . it can be seen that a stop mass of 150 gev is excluded for @xmath424 gev . in spite of the larger mass reach at the tevatron , the lep results remain the most constraining for @xmath425 mass differences smaller than @xmath426 gev . similar searches were performed for a sbottom nlsp decaying into @xmath427 @xcite , with better sensitivity due to a more efficient heavy - flavor tagging for @xmath188 than for @xmath428 quarks . a mass lower limit of 222 gev was obtained by d for @xmath429 gev , based on 310 pb@xmath29 of data . region in the ( @xmath430 ) plane excluded by d @xcite and by earlier experiments . the solid curve corresponds to the nominal scale and pdf choices . the yellow band represents the uncertainty associated with these choices.,width=321 ] other mass hierarchies were considered , where the stop or sbottom is not the nlsp . three - body stop decays , @xmath401 , are dominant if kinematically allowed and when @xmath431 is not , which is possible for some model parameter choices in spite of the mass limits on charged sleptons available from lep . the final states investigated by d comprised two muons or a muon and an electron , with @xmath188 jets and . based on an analysis of 400 pb@xmath29 of data , the largest stop mass excluded is 186 gev , for @xmath432 gev @xcite . if the chargino is lighter than the stop , the @xmath433 decay is dominant . a search was performed by cdf in the two lepton , two @xmath188 jets and final state , with a sensitivity depending on the branching fraction of the chargino leptonic decay , @xmath434 , which is enhanced for light sleptons . an example of an excluded domain in the @xmath435 plane is shown in fig . [ stoptocha ] @xcite , based on 2.7 fb@xmath29 of data . in both of those searches , the background from top quark pair production was a major challenge . yet another mass hierarchy was considered by cdf , namely that where the sbottom is the only squark lighter than the gluino . in such a configuration , the @xmath436 decay is dominant , and gluino pair production then leads to a final state of four @xmath188 jets and . this search was performed in a data sample of 2.5 fb@xmath29 , and lead to excluded sbottom masses as large as 325 gev for gluino and lsp masses of 340 and 60 gev , respectively @xcite . regions in the ( @xmath430 ) plane excluded by cdf @xcite for @xmath437 gev and for various values of the branching fraction for the @xmath434 decay.,width=321 ] * charginos and neutralinos : * the associated production of charginos and neutralinos , @xmath438 , is an electroweak process mediated by @xmath177-channel @xmath58 and @xmath318-channel squark exchanges . leptonic decays , @xmath439 and @xmath440 , are mediated by @xmath58 and @xmath24 exchange , respectively , and by slepton exchange . if sleptons are light , leptonic decays can be sufficiently enhanced for searches in final states consisting of three leptons and to become sensitive in spite of production cross sections of a fraction of a picobarn . an additional challenge is the rather small energy carried by the final state leptons in the chargino and neutralino mass domain to which the searches at the tevatron are currently sensitive . in both the cdf @xcite and d @xcite analyses , only two leptons were required to be positively identified as electrons or muons . leptons identified . ] allowing the third lepton to be detected as an isolated charged particle track provided sensitivity to final states including a @xmath193 lepton that decays into hadrons . in the cdf analysis , the trilepton final state was split into topologies with different signal to background ratios , the purest being when the three leptons are positively identified as electrons or muons with tight criteria . in the d approach , different selections were optimized according to the amount of energy available to the lepton candidates . the ultimate background for these trilepton searches is associated @xmath441 production . the d search , based on an integrated luminosity of 2.3 fb@xmath29 , excludes regions in the msugra parameter space as shown in fig . [ tril ] for @xmath420 , @xmath442 and @xmath443 . it can be seen that the domain excluded at lep is substantially extended by these trilepton searches . the interruption in the exclusion domain is due to configurations where the small @xmath444 mass difference results in one of the final state leptons carrying too little energy , thus preventing efficient detection . requiring only two leptons to be identified , but with same charge sign in order to reduce the otherwise overwhelming sm backgrounds , should provide sensitivity in that region , as was shown by d in an analysis based on a smaller data sample @xcite . the same - sign dilepton signature was also considered in an earlier cdf analysis @xcite . regions in the ( @xmath445 ) plane excluded by the d search for trileptons @xcite.,width=321 ] searches for susy with @xmath77-parity violation were performed at lep , the tevatron and hera . both @xmath77-parity conserving pair production of susy particles and @xmath77-parity violating resonant single susy particle production were considered . the produced particles were subsequently subject to either direct or indirect ( via a cascade to the lsp ) @xmath77-parity violating decays . unless otherwise specified , a single @xmath77-parity violating coupling is assumed to be non - vanishing in the following , large enough for the lifetime of the lsp to be safely assumed to be negligible . * searches at lep : * extensive searches for pair production were performed at lep , involving all possible @xmath77-parity violating couplings . the possible final states are numerous , ranging from four leptons and missing energy for @xmath391 pair production , with decays mediated by a @xmath4-type coupling , e.g. , @xmath446 , to ten hadronic jets and no missing energy for chargino pair production , with @xmath447 followed by a @xmath391 decay into three quarks via a @xmath448-type coupling , e.g. , @xmath449 . the results of these searches are at least as constraining as in the canonical scenario @xcite . the production of a sneutrino resonance via a @xmath450 coupling was also investigated . no signal was observed , and mass lower limits almost up to the center - of - mass energy were set for sufficiently large values of the @xmath77-parity violating coupling involved @xcite . * searches at hera : * as explained in sec . [ subsec : strategies ] , the hera @xmath39 collider is most effective in the searches for @xmath77-parity violating resonant single squark production via a @xmath338-type coupling . direct and indirect squark decays were investigated , and the search results were combined to lead to squark mass lower limits up to 275 gev @xcite , within mild model assumptions , for a @xmath338 coupling of 0.3 , i.e. , with electromagnetic strength . * searches at the tevatron : * a fully general search for all @xmath77-parity violating couplings is not possible at the tevatron , as it was at lep . for instance , @xmath448 couplings lead to multijet final states with no or little missing energy , which can not be distinguished from standard multijet production . searches have therefore been designed for specific choices of couplings leading to distinct signatures . gaugino pair production followed by indirect decays has been extensively studied by both cdf @xcite and d @xcite in the case of a @xmath4-type coupling . the final state is expected to contain four charged leptons , with flavors depending on the indices in the @xmath451 coupling , and due to two neutrinos . for @xmath452 tev , @xmath223 , and @xmath443 , the chargino mass lower limits obtained by d from an analysis of 360 pb@xmath29 of data are 231 , 229 , and 166 gev for the @xmath453 , @xmath454 , and @xmath455 couplings , respectively , with reduced sensitivity in the last case due to the occurrence of @xmath193 leptons in the final state . stop pair production , with @xmath456 via a @xmath457 coupling has been searched by cdf @xcite in the topology where one @xmath193 lepton decays into an electron or a muon , and the other into hadrons . from an analysis of 322 pb@xmath29 of data , a stop mass lower limit of 151 gev was derived . resonant smuon or sneutrino production could be mediated by a @xmath458 coupling . with indirect decays , the final state would exhibit at least one muon and two jets . this topology was investigated by d @xcite , and an excluded domain was set in the @xmath459 plane , leading to a smuon mass lower limit of 363 gev for @xmath460 , and for @xmath420 , @xmath223 , and @xmath421 . resonant sneutrino production mediated by a @xmath461 coupling was also investigated by cdf and d @xcite , now assuming that the sneutrino decays directly via a @xmath4-type coupling . the final states considered were @xmath302 , @xmath245 , @xmath303 , and @xmath462 . the sneutrino mass limits obtained depend on the product of the two couplings involved . as already explained , the lsp in gmsb is a very light gravitino , and the phenomenology depends essentially on the nature of the nlsp , a neutralino or a stau , possibly almost mass degenerate with @xmath463 and @xmath340 , and on its lifetime . * neutralino nlsp : * in the mass range of current interest , a neutralino nlsp decays into a photon and a gravitino , @xmath464 . pair production of such a neutralino at lep would therefore lead , assuming prompt decays , to a final state of two acoplanar photons and missing energy . as can be seen in fig . [ gaga ] , no excess was observed above the sm background from @xmath465 . in gmsb , @xmath391 has a large bino component , so that pair production in @xmath23 interactions proceeds via selectron @xmath318-channel exchange . an excluded domain in the @xmath466 plane was therefore derived @xcite , ruling out the gmsb interpretation ( in terms of selectron pair production ) of an anomalous @xmath467 event that had been observed by cdf @xcite during run i of the tevatron . mass of the invisible system recoiling against pairs of photons at lep @xcite.,width=321 ] searches were also performed at lep for photons not pointing toward the interaction point , which could arise from non - prompt decays of a neutralino nlsp . for even longer lifetimes , the phenomenology becomes identical to that of the canonical scenario . the results of these various searches for a neutralino nlsp were combined with those in various topologies expected to arise from heavier susy particle production to lead to a robust neutralino mass lower limit of 54 gev within the minimal gmsb framework @xcite . searches for acoplanar photons with large were performed at the tevatron by both cdf @xcite and d @xcite . this topology is expected to arise whenever susy particles are pair produced , which subsequently decay to a neutralino nlsp with negligible lifetime . no excess of events was observed over the backgrounds due to photon misidentification or from fake , all determined from data . these results were interpreted within the `` snowmass slope sps 8 '' benchmark gmsb model @xcite where the only free parameter is the effective susy breaking scale @xmath5 . the other parameters were fixed as follows : @xmath468 messenger , a messenger mass of @xmath469 , @xmath470 , and @xmath443 . neutralino nlsp masses smaller than 138 gev are excluded by the cdf analysis , based on 2 fb@xmath29 of data . the possibility of non - prompt neutralino nlsp decays was also investigated by cdf @xcite , making use of the timing information of their calorimeter . no signal of delayed photons was observed in a data sample of 570 pb@xmath29 , from which an excluded domain in the plane of the mass and lifetime of the nlsp was inferred , as shown in fig . [ cdfgaga ] together with the result of ref . @xcite . domain excluded by cdf in the plane of neutralino - nlsp mass and lifetime @xcite.,width=321 ] * stau nlsp : * for prompt @xmath471 decays , the final state arising from stau pair production at lep is the same as in the canonical scenario with a very light @xmath391 . for very long lifetimes , the searches for long lived charginos already reported apply . searches for in - flight decays along charged particle tracks were designed to address intermediate lifetimes . the combination of all these searches allowed a stau nlsp mass lower limit to be set from 87 to 97 gev , depending on the stau lifetime , as shown in fig . [ lepstau ] @xcite . domain excluded at lep @xcite in the @xmath472 plane , where @xmath193 is the lifetime in seconds and @xmath473 is the mass of a stau nlsp in gmsb . the green shaded region is the excluded domain , and the dashed blue curve is the expected boundary of the exclusion region.,width=283 ] a number of searches were performed at lep and at the tevatron in other non - canonical scenarios . * stable charged particles : * in anomaly mediated susy breaking , the lsp is wino - like , and the @xmath377 mass difference is therefore small . as a result , stable charginos are not unlikely . the searches designed at lep in the canonical scenario for large @xmath164 values apply here equally well . at the tevatron , a search was performed by d for pairs of charged massive stable particles that could result from chargino pair production . such particles would behave like slow moving muons that could be detected as delayed signals in the muon system . no significant excess of delayed muons was observed in 1.1 fb@xmath29 of data , and a mass lower limit of 206 gev was set on long - lived wino - like charginos @xcite . a search for stable stops was performed by cdf in 1 fb@xmath29 of data , using a high muon trigger and their time - of - flight detector . stable stops hadronize to form @xmath77-hadrons which behave like slow muons . a model for the interactions of those @xmath77-hadrons with the detector material was constructed , within which a stop mass lower limit of 249 gev was derived @xcite . * stable or long - lived gluinos : * models have been built where the gluino could be the lsp and therefore stable , if @xmath77-parity is conserved @xcite . alternatively , gluinos may decay , but with long lifetimes . this occurs , for example , in models with `` split susy , '' unnatural models in which all squarks and sleptons are very heavy , but the gauginos remain at the electroweak scale @xcite . since gluino decays are mediated by squark exchange , the gluino becomes long - lived . although gluinos can not be produced directly in @xmath23 interactions , they could be produced via gluon splitting , e.g. , @xmath474 , and hadronize into metastable `` @xmath77-hadrons . '' the qcd predictions for four - jet events would therefore be modified . gluinos could also be produced in the decay of heavier squarks . dedicated analyses were performed at lep @xcite , leading to a mass lower limit of 27 gev for a stable gluino . a search for long - lived gluinos was also performed by d with 410 pb@xmath29 of data @xcite . after hadronization into an @xmath77-hadron , a long - lived gluino could come to rest in the calorimeter and decay later on , during a bunch crossing different from the one during which it was created @xcite . the main decay mode expected is @xmath475 , leading to an hadronic shower originating from within the calorimeter and not pointing toward the @xmath40 interaction region , in an otherwise empty event . no excess of this anomalous topology was observed over the background due to cosmic muons or to the beam halo . the gluino mass lower limits derived depend on the lifetime @xmath476 , the branching fraction @xmath477 for the decay mode considered , the @xmath215 mass , and the cross section @xmath478 for the conversion of a neutral @xmath77-hadron into a charged one in the calorimeter . as an example , a mass limit of 270 gev was obtained for @xmath479 hours , @xmath480 , @xmath481 gev , and @xmath482 mb . supersymmetry is one of the most promising ideas for extending the standard model . when realized at the weak scale , many susy models provide natural , and even elegant , solutions to the most pressing problems in particle physics today by stabilizing the gauge hierarchy , providing dark matter candidates , and accommodating force unification , both with and without gravity . in addition , the general framework of weak - scale susy is flexible enough to encompass a wide variety of new phenomena , including extended higgs sectors , missing energy , long - lived and metastable particles , and a host of other signatures of new physics . searches for susy are therefore also searches for other forms of new physics which , even if less profoundly motivated , are , of course , also important to pursue . in this review , we have comprehensively summarized the state of the art in searches for susy at the high energy frontier . although this is a continuously evolving subject , this review provides a snapshot of the field at a particularly important time , when final results from lep and hera are in hand , the tevatron experiments have reported deep probes of many supersymmetric models with several @xmath483 of data , and the lhc will soon begin operation . this review has summarized searches for both supersymmetric higgs bosons and standard model superpartners . in the higgs sector , susy requires a light neutral higgs boson . this higgs boson could be standard - model like , but it could also have non - standard couplings . in addition , it is accompanied by other higgs bosons , both neutral or charged . the most stringent constraints on a sm - like higgs boson currently come from lep , with a mass lower limit of 114.4 gev that applies in the mssm at low @xmath119 . furthermore , the lep experiments set a lower limit of 93 gev on the lightest neutral higgs boson of the mssm , independent of @xmath119 . the mssm parameter space has now been further restricted by the tevatron experiments . for example , @xmath119 values larger than 40 are excluded for @xmath484 gev . for charged higgs bosons , lep excludes masses below 78.6 gev , and the tevatron experiments have extended this mass limit to @xmath485 gev for very large values of @xmath119 . for superpartners , the bounds are , of course , model - dependent , but the main results may be summarized as follows . the searches at lep have constrained the masses of all susy particles , except for the gluino and the lsp , to be larger than approximately 100 gev in most susy scenarios . furthermore , an indirect lower limit on the mass of a neutralino lsp has been set at 47 gev in the mssm with gaugino and sfermion mass unification . the higher center - of - mass energy at the tevatron has allowed tighter mass limits to be obtained for strongly interacting susy particles : 379 and 308 gev for squarks and gluinos , respectively , within the msugra framework at low @xmath119 . in that same model , domains beyond the lep reach were also probed by searches for associated chargino - neutralino production . in the near future , the first indication for susy at high energy colliders could be the observation of a light neutral higgs boson at the tevatron . of course , such a discovery is not proof of susy only the discovery of superpartners would unambiguously establish susy as being realized in nature . once collisions begin at the lhc and the detectors are sufficiently understood , it will not take more than @xmath28 fb@xmath29 to discover squarks and gluinos with masses less than @xmath38 tev @xcite . a new era will then begin during which the whole susy spectrum will have to be deciphered , and the properties of the susy model established . many more fb@xmath29 will be needed for that purpose , and to unravel the spectrum of susy higgs bosons . the work of jlf was supported in part by nsf grants phy0239817 and phy0653656 , nasa grant nng05gg44 g , and the alfred p. sloan foundation . jfg is supported by the cnrs / in2p3 ( france ) . the work of jn was supported by doe grant de fg0291er40664 . s. weinberg , `` implications of dynamical symmetry breaking , '' http://dx.doi.org/10.1103/physrevd.13.974[_phys . * d13 * ( 1976 ) 974996 ] . s. weinberg , `` implications of dynamical symmetry breaking : an addendum , '' http://dx.doi.org/10.1103/physrevd.19.1277[_phys . rev . _ * d19 * ( 1979 ) 12771280 ] . l. susskind , `` dynamics of spontaneous symmetry breaking in the weinberg - salam theory , '' http://dx.doi.org/10.1103/physrevd.20.2619[_phys . rev . _ * d20 * ( 1979 ) 26192625 ] . g. t hooft , in recent developments in gauge theories . proceedings , nato advanced study institute , cargese , france , august 26 - september 8 , 1979 , new york , plenum ( 1980 ) 438 p. ( nato advanced study institutes series : series b , physics , 59 ) . l. reina , `` tasi 2004 lecture notes on higgs boson physics , '' http://arxiv.org/abs/hep-ph/0512377[arxiv:hep-ph/0512377 ] . c. t. hill and e. h. simmons , `` strong dynamics and electroweak symmetry breaking , '' http://dx.doi.org/10.1016/s0370-1573(03)00140-6 [ _ phys . rept . _ * 381 * ( 2003 ) 235402 ] , http://arxiv.org/abs/hep-ph/0203079[arxiv:hep-ph/0203079 ] . collaboration , e. komatsu _ et al . _ , `` five - year wilkinson microwave anisotropy probe ( wmap ) observations : cosmological interpretation , '' http://dx.doi.org/10.1088/0067-0049/180/2/330 [ _ astrophys . j. suppl . _ * 180 * ( 2009 ) 330376 ] , http://arxiv.org/abs/0803.0547[arxiv:0803.0547 [ astro - ph ] ] . y. b. zeldovich _ adv . astrophys . _ * 3 * ( 1965 ) 241 . chiu , `` symmetry between particle and anti - particle populations in the universe , '' http://dx.doi.org/10.1103/physrevlett.17.712[_phys . * 17 * ( 1966 ) 712 ] . g. steigman , `` cosmology confronts particle physics , '' http://dx.doi.org/10.1146/annurev.ns.29.120179.001525[_ann . nucl . part . * 29 * ( 1979 ) 313338 ] . r. j. scherrer and m. s. turner , `` on the relic , cosmic abundance of stable weakly interacting massive particles , '' http://dx.doi.org/10.1103/physrevd.33.1585[_phys . rev . _ * d33 * ( 1986 ) 1585 ] . h. georgi and s. l. glashow , `` unity of all elementary particle forces , '' http://dx.doi.org/10.1103/physrevlett.32.438[_phys . * 32 * ( 1974 ) 438441 ] . s. dimopoulos and h. georgi , `` softly broken supersymmetry and su(5 ) , '' http://dx.doi.org/10.1016/0550-3213(81)90522-8[_nucl . phys . _ * b193 * ( 1981 ) 150 ] . s. dimopoulos , s. raby , and f. wilczek , `` supersymmetry and the scale of unification , '' http://dx.doi.org/10.1103/physrevd.24.1681[_phys . _ * d24 * ( 1981 ) 16811683 ] . n. sakai , `` naturalness in supersymmetric guts , '' http://dx.doi.org/10.1007/bf01573998[_zeit . phys . _ * c11 * ( 1981 ) 153 ] . l. e. ibanez and g. g. ross , `` low - energy predictions in supersymmetric grand unified theories , '' http://dx.doi.org/10.1016/0370-2693(81)91200-4[_phys . lett . _ * b105 * ( 1981 ) 439 ] . m. b. einhorn and d. r. t. jones , `` the weak mixing angle and unification mass in supersymmetric su(5 ) , '' http://dx.doi.org/10.1016/0550-3213(82)90502-8[_nucl . _ * b196 * ( 1982 ) 475 ] . collaboration , c. amsler _ et al . _ , `` review of particle physics , '' http://dx.doi.org/10.1016/j.physletb.2008.07.018[_phys . _ * b667 * ( 2008 ) 1 ] . collaborations , lep electroweak working group and sld electroweak group and sld heavy flavour group , `` precision electroweak measurements on the @xmath24 resonance , '' http://dx.doi.org/10.1016/j.physrep.2005.12.006[_phys . * 427 * ( 2006 ) 257 ] , http://arxiv.org/abs/hep-ex/0509008[arxiv:hep-ex/0509008 ] . collaboration , g. aad _ et al . _ , `` the atlas experiment at the cern large hadron collider , '' http://dx.doi.org/10.1088/1748-0221/3/08/s08003[_j . inst . _ * 3 * ( 2008 ) s08003 ] . collaboration , g. aad _ et al . _ , `` expected performance of the atlas experiment - detector , trigger and physics , '' http://arxiv.org/abs/0901.0512[arxiv:0901.0512 [ hep - ex ] ] . , `` cms , the compact muon solenoid : technical proposal , '' cern - lhcc-94 - 38 . collaboration , g. l. bayatian _ _ , `` cms technical design report , volume ii : physics performance , '' http://dx.doi.org/10.1088/0954-3899/34/6/s01[_j . * g34 * ( 2007 ) 9951579 ] . collaboration , d. decamp _ et al . _ , `` aleph : a detector for electron - positron annihilations at lep , '' http://dx.doi.org/10.1016/0168-9002(90)91831-u[_nucl . instrum . _ * a294 * ( 1990 ) 121178 ] . collaboration , d. buskulic _ et al . _ , `` performance of the aleph detector at lep , '' http://dx.doi.org/10.1016/0168-9002(95)00138-7[_nucl . _ * a360 * ( 1995 ) 481506 ] . collaboration , p. a. aarnio _ et al . _ , `` the delphi detector at lep , '' http://dx.doi.org/10.1016/0168-9002(91)90793-p[_nucl . _ * a303 * ( 1991 ) 233276 ] . collaboration , p. abreu _ et al . _ , `` performance of the delphi detector , '' http://dx.doi.org/10.1016/0168-9002(96)00463-9[_nucl . instrum . meth . _ * a378 * ( 1996 ) 57100 ] . collaboration , b. adeva _ et al . _ , `` the construction of the l3 experiment , '' _ nucl . _ * a289 * ( 1990 ) 35102 . collaboration , k. ahmet _ et al . _ , `` the opal detector at lep , '' http://dx.doi.org/10.1016/0168-9002(91)90547-4[_nucl . _ * a305 * ( 1991 ) 275319 ] . collaboration , i. abt _ et al . _ , `` the h1 detector at hera , '' http://dx.doi.org/10.1016/s0168-9002(96)00893-5[_nucl . * a386 * ( 1997 ) 310347 ] . , `` the zeus detector : status report 1993 , '' zeus - status - rept-1993 . http://www-zeus.desy.de/bluebook/bluebook.html . collaboration , d. e. acosta _ et al . _ , `` measurement of the @xmath486 meson and @xmath487hadron production cross sections in @xmath35 collisions at @xmath488 gev , '' http://dx.doi.org/10.1103/physrevd.71.032001[_phys . rev . _ * d71 * ( 2005 ) 032001 ] , http://arxiv.org/abs/hep-ex/0412071[arxiv:hep-ex/0412071 ] . collaboration , v. m. abazov _ , `` the upgraded d0 detector , '' http://dx.doi.org/10.1016/j.nima.2006.05.248[_nucl . instrum . meth . _ * a565 * ( 2006 ) 463537 ] , http://arxiv.org/abs/physics/0507191[arxiv:physics/0507191 ] . y. a. golfand and e. p. likhtman , `` extension of the algebra of poincare group generators and violation of p invariance , '' _ jetp lett . _ * 13 * ( 1971 ) 323326 . d. v. volkov and v. p. akulov , `` is the neutrino a goldstone particle ? , '' http://dx.doi.org/10.1016/0370-2693(73)90490-5[_phys . * b46 * ( 1973 ) 109110 ] . j. wess and b. zumino , `` supergauge transformations in four - dimensions , '' http://dx.doi.org/10.1016/0550-3213(74)90355-1[_nucl . _ * b70 * ( 1974 ) 3950 ] . l. maiani , in vector bosons and higgs bosons in the weinberg - salam theory of weak and electromagnetic interactions . proceedings : summer school on particle physics , gif - sur- yvette , 3 - 7 sep 1979 , paris , france : natl . phys . part . ( 1979 ) 201 p. m. j. g. veltman , `` the infrared - ultraviolet connection , '' _ acta phys . _ * b12 * ( 1981 ) 437 . e. witten , `` dynamical breaking of supersymmetry , '' http://dx.doi.org/10.1016/0550-3213(81)90006-7[_nucl . * b188 * ( 1981 ) 513 ] . h. goldberg , `` constraint on the photino mass from cosmology , '' http://dx.doi.org/10.1103/physrevlett.50.1419[_phys . * 50 * ( 1983 ) 1419 ] . j. r. ellis , j. s. hagelin , d. v. nanopoulos , k. a. olive , and m. srednicki , `` supersymmetric relics from the big bang , '' http://dx.doi.org/10.1016/0550-3213(84)90461-9[_nucl . _ * b238 * ( 1984 ) 453476 ] . h. pagels and j. r. primack , `` supersymmetry , cosmology and new tev physics , '' http://dx.doi.org/10.1103/physrevlett.48.223[_phys . * 48 * ( 1982 ) 223 ] . j. l. feng , a. rajaraman , and f. takayama , `` superweakly - interacting massive particles , '' http://dx.doi.org/10.1103/physrevlett.91.011302 [ _ phys . lett . _ * 91 * ( 2003 ) 011302 ] , http://arxiv.org/abs/hep-ph/0302215[arxiv:hep-ph/0302215 ] . s. p. martin , `` a supersymmetry primer , '' http://arxiv.org/abs/hep-ph/9709356[arxiv:hep-ph/9709356 ] . n. polonsky , `` supersymmetry : structure and phenomena . extensions of the standard model , '' _ lect . notes phys . _ * m68 * ( 2001 ) 1169 , http://arxiv.org/abs/hep-ph/0108236[arxiv:hep-ph/0108236 ] . m. drees , r. godbole , and p. roy , `` theory and phenomenology of sparticles : an account of four - dimensional n=1 supersymmetry in high energy physics , '' hackensack , usa : world scientific ( 2004 ) 555 p. l. girardello and m. t. grisaru , `` soft breaking of supersymmetry , '' http://dx.doi.org/10.1016/0550-3213(82)90512-0[_nucl . * b194 * ( 1982 ) 65 ] . p. fayet , `` spontaneously broken supersymmetric theories of weak , electromagnetic and strong interactions , '' http://dx.doi.org/10.1016/0370-2693(77)90852-8[_phys . _ * b69 * ( 1977 ) 489 ] . g. r. farrar and p. fayet , `` phenomenology of the production , decay , and detection of new hadronic states associated with supersymmetry , '' http://dx.doi.org/10.1016/0370-2693(78)90858-4[_phys . _ * b76 * ( 1978 ) 575579 ] . f. e. paige , s. d. protopopescu , h. baer , and x. tata , `` isajet 7.69 : a monte carlo event generator for @xmath489 , @xmath490 , and @xmath491 reactions , '' http://arxiv.org/abs/hep-ph/0312045[arxiv:hep-ph/0312045 ] . b. c. allanach , `` softsusy : a c++ program for calculating supersymmetric spectra , '' http://dx.doi.org/10.1016/s0010-4655(01)00460-x [ _ comput . * 143 * ( 2002 ) 305331 ] , http://arxiv.org/abs/hep-ph/0104145[arxiv:hep-ph/0104145 ] . w. porod , `` spheno , a program for calculating supersymmetric spectra , susy particle decays and susy particle production at @xmath491 colliders , '' http://dx.doi.org/10.1016/s0010-4655(03)00222-4[_comput . _ * 153 * ( 2003 ) 275315 ] , http://arxiv.org/abs/hep-ph/0301101[arxiv:hep-ph/0301101 ] . a. djouadi , j .- l . kneur , and g. moultaka , `` suspect : a fortran code for the supersymmetric and higgs particle spectrum in the mssm , '' http://dx.doi.org/10.1016/j.cpc.2006.11.009[_comput . * 176 * ( 2007 ) 426455 ] , http://arxiv.org/abs/hep-ph/0211331[arxiv:hep-ph/0211331 ] . m. e. peskin , `` theoretical summary lecture for eps hep99 , '' http://arxiv.org/abs/hep-ph/0002041[arxiv:hep-ph/0002041 ] . a. h. chamseddine , r. l. arnowitt , and p. nath , `` locally supersymmetric grand unification , '' http://dx.doi.org/10.1103/physrevlett.49.970[_phys . * 49 * ( 1982 ) 970 ] . r. barbieri , s. ferrara , and c. a. savoy , `` gauge models with spontaneously broken local supersymmetry , '' http://dx.doi.org/10.1016/0370-2693(82)90685-2[_phys . lett . _ * b119 * ( 1982 ) 343 ] . n. ohta , `` grand unified theories based on local supersymmetry , '' http://dx.doi.org/10.1143/ptp.70.542[_prog * 70 * ( 1983 ) 542 ] . l. j. hall , j. d. lykken , and s. weinberg , `` supergravity as the messenger of supersymmetry breaking , '' http://dx.doi.org/10.1103/physrevd.27.2359[_phys . rev . _ * d27 * ( 1983 ) 23592378 ] . l. alvarez - gaume , j. polchinski , and m. b. wise , `` minimal low - energy supergravity , '' http://dx.doi.org/10.1016/0550-3213(83)90591-6[_nucl . * b221 * ( 1983 ) 495 ] . h. p. nilles , `` supersymmetry , supergravity and particle physics , '' http://dx.doi.org/10.1016/0370-1573(84)90008-5[_phys . rept . _ * 110 * ( 1984 ) 1162 ] . m. dine , w. fischler , and m. srednicki , `` supersymmetric technicolor , '' http://dx.doi.org/10.1016/0550-3213(81)90582-4[_nucl . phys . _ * b189 * ( 1981 ) 575593 ] . s. dimopoulos and s. raby , `` supercolor , '' http://dx.doi.org/10.1016/0550-3213(81)90430-2[_nucl . phys . _ * b192 * ( 1981 ) 353 ] . c. r. nappi and b. a. ovrut , `` supersymmetric extension of the su(3 ) x su(2 ) x u(1 ) model , '' http://dx.doi.org/10.1016/0370-2693(82)90418-x[_phys . lett . _ * b113 * ( 1982 ) 175 ] . l. alvarez - gaume , m. claudson , and m. b. wise , `` low - energy supersymmetry , '' http://dx.doi.org/10.1016/0550-3213(82)90138-9[_nucl . phys . _ * b207 * ( 1982 ) 96 ] . m. dine , a. e. nelson , and y. shirman , `` low - energy dynamical supersymmetry breaking simplified , '' http://dx.doi.org/10.1103/physrevd.51.1362[_phys . rev . _ * d51 * ( 1995 ) 13621370 ] , http://arxiv.org/abs/hep-ph/9408384[arxiv:hep-ph/9408384 ] . m. dine , a. e. nelson , y. nir , and y. shirman , `` new tools for low - energy dynamical supersymmetry breaking , '' http://dx.doi.org/10.1103/physrevd.53.2658[_phys . rev . _ * d53 * ( 1996 ) 26582669 ] , http://arxiv.org/abs/hep-ph/9507378[arxiv:hep-ph/9507378 ] . d. r. stump , m. wiest , and c. p. yuan , `` detecting a light gravitino at linear collider to probe the susy breaking scale , '' http://dx.doi.org/10.1103/physrevd.54.1936[_phys . rev . _ * d54 * ( 1996 ) 19361943 ] , http://arxiv.org/abs/hep-ph/9601362[arxiv:hep-ph/9601362 ] . s. dimopoulos , m. dine , s. raby , and s. d. thomas , `` experimental signatures of low energy gauge mediated supersymmetry breaking , '' http://dx.doi.org/10.1103/physrevlett.76.3494[_phys . * 76 * ( 1996 ) 34943497 ] , http://arxiv.org/abs/hep-ph/9601367[arxiv:hep-ph/9601367 ] . j. l. feng and t. moroi , `` tevatron signatures of longlived charged sleptons in gauge mediated supersymmetry breaking models , '' http://dx.doi.org/10.1103/physrevd.58.035001[_phys . _ * d58 * ( 1998 ) 035001 ] , http://arxiv.org/abs/hep-ph/9712499[arxiv:hep-ph/9712499 ] . m. drees and x. tata , `` signals for heavy exotics at hadron colliders and supercolliders , '' http://dx.doi.org/10.1016/0370-2693(90)90508-4[_phys . _ * b252 * ( 1990 ) 695702 ] . j. l. goity , w. j. kossler , and m. sher , `` production , collection and utilization of very longlived heavy charged leptons , '' http://dx.doi.org/10.1103/physrevd.48.5437[_phys . * d48 * ( 1993 ) 54375439 ] , http://arxiv.org/abs/hep-ph/9305244[arxiv:hep-ph/9305244 ] . l. randall and r. sundrum , `` out of this world supersymmetry breaking , '' http://dx.doi.org/10.1016/s0550-3213(99)00359-4[_nucl . * b557 * ( 1999 ) 79118 ] , http://arxiv.org/abs/hep-th/9810155[arxiv:hep-th/9810155 ] . g. f. giudice , m. a. luty , h. murayama , and r. rattazzi , `` gaugino mass without singlets , '' _ jhep _ * 12 * ( 1998 ) 027 , http://arxiv.org/abs/hep-ph/9810442[arxiv:hep-ph/9810442 ] . a. pomarol and r. rattazzi , `` sparticle masses from the superconformal anomaly , '' _ jhep _ * 05 * ( 1999 ) 013 , http://arxiv.org/abs/hep-ph/9903448[arxiv:hep-ph/9903448 ] . z. chacko , m. a. luty , i. maksymyk , and e. ponton , `` realistic anomaly - mediated supersymmetry breaking , '' _ jhep _ * 04 * ( 2000 ) 001 , http://arxiv.org/abs/hep-ph/9905390[arxiv:hep-ph/9905390 ] . e. katz , y. shadmi , and y. shirman , `` heavy thresholds , slepton masses and the mu term in anomaly mediated supersymmetry breaking , '' _ jhep _ * 08 * ( 1999 ) 015 , http://arxiv.org/abs/hep-ph/9906296[arxiv:hep-ph/9906296 ] . j. l. feng , t. moroi , l. randall , m. strassler , and s .- f . su , `` discovering supersymmetry at the tevatron in wino lsp scenarios , '' http://dx.doi.org/10.1103/physrevlett.83.1731[_phys . * 83 * ( 1999 ) 17311734 ] , http://arxiv.org/abs/hep-ph/9904250[arxiv:hep-ph/9904250 ] . r. barbieri and m. frigeni , `` the supersymmetric higgs searches at lep after radiative corrections , '' http://dx.doi.org/10.1016/0370-2693(91)91106-6[_phys . _ * b258 * ( 1991 ) 395398 ] . h. e. haber , `` introductory low - energy supersymmetry , '' http://arxiv.org/abs/hep-ph/9306207[arxiv:hep-ph/9306207 ] . j. r. ellis and s. rudaz , `` search for supersymmetry in toponium decays , '' http://dx.doi.org/10.1016/0370-2693(83)90402-1[_phys . lett . _ * b128 * ( 1983 ) 248 ] . m. s. carena , s. heinemeyer , c. e. m. wagner , and g. weiglein , `` suggestions for improved benchmark scenarios for higgs boson searches at lep2 , '' http://arxiv.org/abs/hep-ph/9912223[arxiv:hep-ph/9912223 ] . m. s. carena , s. heinemeyer , c. e. m. wagner , and g. weiglein , `` suggestions for benchmark scenarios for mssm higgs boson searches at hadron colliders , '' http://dx.doi.org/10.1140/epjc/s2002-01084-3[_eur . phys . j. _ * c26 * ( 2003 ) 601607 ] , http://arxiv.org/abs/hep-ph/0202167[arxiv:hep-ph/0202167 ] . j. r. ellis , m. k. gaillard , and d. v. nanopoulos , `` a phenomenological profile of the higgs boson , '' _ nucl . phys . _ * b106 * ( 1976 ) 292 . b. w. lee , c. quigg , and h. b. thacker , `` weak interactions at very high - energies : the role of the higgs boson mass , '' http://dx.doi.org/10.1103/physrevd.16.1519[_phys . _ * d16 * ( 1977 ) 1519 ] . b. l. ioffe and v. a. khoze , `` what can be expected from experiments on colliding @xmath491 beams with @xmath57 approximately equal to 100 gev ? , '' _ sov . j. part . nucl . _ * 9 * ( 1978 ) 50 . g. pocsik and g. zsigmond , `` on the production of neutral higgs bosons in the weinberg - salam model with two higgs doublets , '' http://dx.doi.org/10.1007/bf01546210[_zeit . phys . _ * c10 * ( 1981 ) 367 ] . a. djouadi , j. kalinowski , and m. spira , `` hdecay : a program for higgs boson decays in the standard model and its supersymmetric extension , '' http://dx.doi.org/10.1016/s0010-4655(97)00123-9[_comput . * 108 * ( 1998 ) 5674 ] , http://arxiv.org/abs/hep-ph/9704448[arxiv:hep-ph/9704448 ] . d. r. t. jones and s. t. petcov , `` heavy higgs bosons at lep , '' http://dx.doi.org/10.1016/0370-2693(79)91234-6[_phys . lett . _ * b84 * ( 1979 ) 440 ] . collaborations , r. barate _ et al . _ , `` search for the standard model higgs boson at lep , '' http://dx.doi.org/10.1016/s0370-2693(03)00614-2[_phys . lett . _ * b565 * ( 2003 ) 6175 ] , http://arxiv.org/abs/hep-ex/0306033[arxiv:hep-ex/0306033 ] . collaborations , s. schael _ et al . _ , `` search for neutral mssm higgs bosons at lep , '' http://dx.doi.org/10.1140/epjc/s2006-02569-7[_eur . phys . j. _ * c47 * ( 2006 ) 547587 ] , http://arxiv.org/abs/hep-ex/0602042[arxiv:hep-ex/0602042 ] . collaborations , tevatron electroweak working group , `` combination of cdf and d0 results on the mass of the top quark , '' http://arxiv.org/abs/0903.2503[arxiv:0903.2503 [ hep - ex ] ] . collaborations , lep higgs working group , `` searches for invisible higgs bosons : preliminary combined results using lep data collected at energies up to 209 gev , '' http://arxiv.org/abs/hep-ex/0107032[arxiv:hep-ex/0107032 ] . collaborations , lep higgs working group , `` flavor independent search for hadronically decaying neutral higgs bosons at lep , '' http://arxiv.org/abs/hep-ex/0107034[arxiv:hep-ex/0107034 ] . r. dermisek and j. f. gunion , `` consistency of lep event excesses with an @xmath492 decay scenario and low - fine - tuning nmssm models , '' http://dx.doi.org/10.1103/physrevd.73.111701[_phys . _ * d73 * ( 2006 ) 111701 ] , http://arxiv.org/abs/hep-ph/0510322[arxiv:hep-ph/0510322 ] . m. s. carena , j. r. ellis , a. pilaftsis , and c. e. m. wagner , `` renormalization group - improved effective potential for the mssm higgs sector with explicit cp violation , '' http://dx.doi.org/10.1016/s0550-3213(00)00358-8[_nucl . * b586 * ( 2000 ) 92140 ] , http://arxiv.org/abs/hep-ph/0003180[arxiv:hep-ph/0003180 ] . m. s. carena , j. r. ellis , a. pilaftsis , and c. e. m. wagner , `` cp - violating mssm higgs bosons in the light of lep 2 , '' http://dx.doi.org/10.1016/s0370-2693(00)01215-6[_phys . _ * b495 * ( 2000 ) 155163 ] , http://arxiv.org/abs/hep-ph/0009212[arxiv:hep-ph/0009212 ] . f. wilczek , `` decays of heavy vector mesons into higgs particles , '' http://dx.doi.org/10.1103/physrevlett.39.1304[_phys . * 39 * ( 1977 ) 1304 ] . h. m. georgi , s. l. glashow , m. e. machacek , and d. v. nanopoulos , `` higgs bosons from two gluon annihilation in proton proton collisions , '' http://dx.doi.org/10.1103/physrevlett.40.692[_phys . _ * 40 * ( 1978 ) 692 ] . s. l. glashow , d. v. nanopoulos , and a. yildiz , `` associated production of higgs bosons and @xmath24 particles , '' http://dx.doi.org/10.1103/physrevd.18.1724[_phys . * d18 * ( 1978 ) 17241727 ] . , `` search for neutral mssm higgs bosons decaying to tau pairs with 1.8 fb@xmath29 of data , '' cdf note 9071 . t. hahn , s. heinemeyer , f. maltoni , g. weiglein , and s. willenbrock , `` sm and mssm higgs boson production cross sections at the tevatron and the lhc , '' http://arxiv.org/abs/hep-ph/0607308[arxiv:hep-ph/0607308 ] . j. m. campbell _ et al . _ , `` higgs boson production in association with bottom quarks , '' http://arxiv.org/abs/hep-ph/0405302[arxiv:hep-ph/0405302 ] . , `` search for higgs bosons produced in association with @xmath188 quarks , '' cdf note 9284 . m. s. carena , s. mrenna , and c. e. m. wagner , `` mssm higgs boson phenomenology at the tevatron collider , '' http://dx.doi.org/10.1103/physrevd.60.075010[_phys . rev . _ * d60 * ( 1999 ) 075010 ] , http://arxiv.org/abs/hep-ph/9808312[arxiv:hep-ph/9808312 ] . m. s. carena , s. mrenna , and c. e. m. wagner , `` the complementarity of lep , the tevatron and the lhc in the search for a light mssm higgs boson , '' http://dx.doi.org/10.1103/physrevd.62.055008[_phys . * d62 * ( 2000 ) 055008 ] , http://arxiv.org/abs/hep-ph/9907422[arxiv:hep-ph/9907422 ] . , `` a search for neutral higgs bosons at high @xmath119 in the mode @xmath495 in run iib data , '' d0 note 5727-conf . j. f. gunion , h. e. haber , g. l. kane , and s. dawson , `` errata for the higgs hunter s guide , '' http://arxiv.org/abs/hep-ph/9302272[arxiv:hep-ph/9302272 ] . l. n. chang and j. e. kim , `` possible signature of charged higgs particles in high - energy @xmath491 annihilation , '' http://dx.doi.org/10.1016/0370-2693(79)90531-8[_phys . lett . _ * b81 * ( 1979 ) 233 ] . a. g. akeroyd , `` three - body decays of higgs bosons at lep2 and application to a hidden fermiophobic higgs , '' http://dx.doi.org/10.1016/s0550-3213(98)00845-1[_nucl . * b544 * ( 1999 ) 557575 ] , http://arxiv.org/abs/hep-ph/9806337[arxiv:hep-ph/9806337 ] . collaboration , j. abdallah _ et al . _ , `` search for charged higgs bosons at lep in general two higgs doublet models , '' http://dx.doi.org/10.1140/epjc/s2004-01732-6[_eur . phys . j. _ * c34 * ( 2004 ) 399418 ] , http://arxiv.org/abs/hep-ex/0404012[arxiv:hep-ex/0404012 ] . collaboration , g. abbiendi _ et al . _ , `` search for charged higgs bosons in @xmath491 collisions at @xmath496 = 189 - 209 gev , '' http://arxiv.org/abs/0812.0267[arxiv:0812.0267 [ hep - ex ] ] . collaboration , a. heister _ et al . _ , `` search for charged higgs bosons in @xmath497 collisions at energies up to @xmath496 = 209 gev , '' http://dx.doi.org/10.1016/s0370-2693(02)02380-8[_phys . lett . _ * b543 * ( 2002 ) 113 ] , http://arxiv.org/abs/hep-ex/0207054[arxiv:hep-ex/0207054 ] . collaboration , p. achard _ et al . _ , `` search for charged higgs bosons at lep , '' http://dx.doi.org/10.1016/j.physletb.2003.09.057[_phys . lett . _ * b575 * ( 2003 ) 208220 ] , http://arxiv.org/abs/hep-ex/0309056[arxiv:hep-ex/0309056 ] . collaborations , lep higgs working group , `` search for charged higgs bosons : preliminary combined results using lep data collected at energies up to 209 gev , '' http://arxiv.org/abs/hep-ex/0107031[arxiv:hep-ex/0107031 ] . j. f. gunion , h. e. haber , f. e. paige , w .- k . tung , and s. s. d. willenbrock , `` neutral and charged higgs detection : heavy quark fusion , top quark mass dependence and rare decays , '' http://dx.doi.org/10.1016/0550-3213(87)90600-6[_nucl . phys . _ * b294 * ( 1987 ) 621 ] . , `` a search for charged higgs bosons in @xmath299 events , '' d0 note 5715-conf . j. s. lee _ et al . _ , `` cpsuperh : a computational tool for higgs phenomenology in the minimal supersymmetric standard model with explicit cp violation , '' http://dx.doi.org/10.1016/s0010-4655(03)00463-6 [ _ comput . * 156 * ( 2004 ) 283317 ] , http://arxiv.org/abs/hep-ph/0307377[arxiv:hep-ph/0307377 ] . y. grossman , `` phenomenology of models with more than two higgs doublets , '' http://dx.doi.org/10.1016/0550-3213(94)90316-6[_nucl . * b426 * ( 1994 ) 355384 ] , http://arxiv.org/abs/hep-ph/9401311[arxiv:hep-ph/9401311 ] . collaboration , a. abulencia _ et al . _ , `` search for charged higgs bosons from top quark decays in @xmath35 collisions at @xmath498 1.96 tev , '' http://dx.doi.org/10.1103/physrevlett.96.042003[_phys . * 96 * ( 2006 ) 042003 ] , http://arxiv.org/abs/hep-ex/0510065[arxiv:hep-ex/0510065 ] . , `` a search for charged higgs in lepton + jets @xmath299 events using @xmath499 of cdf data , '' cdf note 9322 . collaboration , v. m. abazov _ et al . _ , `` search for charged higgs bosons decaying to top and bottom quarks in @xmath490 collisions , '' http://dx.doi.org/10.1103/physrevlett.102.191802[_phys . _ * 102 * ( 2009 ) 191802 ] , http://arxiv.org/abs/0807.0859[arxiv:0807.0859 [ hep - ex ] ] . v. d. barger , g. f. giudice , and t. han , `` some new aspects of supersymmetry r - parity violating interactions , '' http://dx.doi.org/10.1103/physrevd.40.2987[_phys . * d40 * ( 1989 ) 2987 ] . g. r. farrar and p. fayet , `` searching for the spin 0 leptons of supersymmetry , '' http://dx.doi.org/10.1016/0370-2693(80)90007-6[_phys . lett . _ * b89 * ( 1980 ) 191 ] . a. bartl , h. fraas , and w. majerotto , `` gaugino - higgsino mixing in selectron and sneutrino pair production , '' http://dx.doi.org/10.1007/bf01548825[_z . phys . _ * c34 * ( 1987 ) 411 ] . v. d. barger , r. w. robinett , w. y. keung , and r. j. n. phillips , `` production of gauge fermions at colliders , '' http://dx.doi.org/10.1016/0370-2693(83)90519-1[_phys . * b131 * ( 1983 ) 372 ] . a. bartl , h. fraas , and w. majerotto , `` signatures of chargino production in @xmath491 collisions , '' http://dx.doi.org/10.1007/bf01557609[_z . phys . _ * c30 * ( 1986 ) 441 ] . j. r. ellis , j. m. frere , j. s. hagelin , g. l. kane , and s. t. petcov , `` search for neutral gauge fermions in @xmath491 annihilation , '' http://dx.doi.org/10.1016/0370-2693(83)90343-x[_phys . * b132 * ( 1983 ) 436 ] . d. a. dicus , s. nandi , w. w. repko , and x. tata , `` electron - positron annihilation cross - section in su(2)@xmath500 x u(1 ) supergravity , '' http://dx.doi.org/10.1103/physrevlett.51.1030[_phys . * 51 * ( 1983 ) 1030 ] . a. bartl , h. fraas , and w. majerotto , `` production and decay of neutralinos in @xmath491 annihilation , '' http://dx.doi.org/10.1016/0550-3213(86)90104-5[_nucl . phys . _ * b278 * ( 1986 ) 1 ] . f. almeida and d. h. schiller , `` two jet distributions and energy - energy correlation for spin 0 quarks in @xmath491 annihilation , '' http://dx.doi.org/10.1007/bf01648784[_z . phys . _ * c21 * ( 1983 ) 103 ] . a. bartl , h. eberl , s. kraml , w. majerotto , and w. porod , `` production of stop , sbottom , and stau at lep2 , '' http://dx.doi.org/10.1007/s002880050337[_z . phys . _ * c73 * ( 1997 ) 469476 ] , http://arxiv.org/abs/hep-ph/9603410[arxiv:hep-ph/9603410 ] . collaborations , lep susy working group , `` sleptons , '' note lepsusywg/04 - 01.1 . http://lepsusy.web.cern.ch/lepsusy/welcome.html . collaboration , a. heister _ et al . _ , `` absolute lower limits on the masses of selectrons and sneutrinos in the mssm , '' http://dx.doi.org/10.1016/s0370-2693(02)02471-1[_phys . lett . _ * b544 * ( 2002 ) 7388 ] , http://arxiv.org/abs/hep-ex/0207056[arxiv:hep-ex/0207056 ] . collaboration , p. achard _ et al . _ , `` search for scalar leptons and scalar quarks at lep , '' http://dx.doi.org/10.1016/j.physletb.2003.10.010[_phys . lett . _ * b580 * ( 2004 ) 3749 ] , http://arxiv.org/abs/hep-ex/0310007[arxiv:hep-ex/0310007 ] . collaborations , lep susy working group , `` charginos , large @xmath93 , '' note lepsusywg/01 - 03.1 . http://lepsusy.web.cern.ch/lepsusy/welcome.html . collaboration , r. barate _ et al . _ , `` search for charginos and neutralinos in @xmath497 collisions at center - of - mass energies near 183 gev and constraints on the mssm parameter space , '' http://dx.doi.org/10.1007/s100520050627[_eur . phys . j. _ * c11 * ( 1999 ) 193216 ] . collaborations , lep susy working group , `` interpretation in constrained mssm and limit on the lsp mass , '' note lepsusywg/04 - 07.1 . http://lepsusy.web.cern.ch/lepsusy/welcome.html . hikasa and m. kobayashi , `` light scalar top at @xmath491 colliders , '' http://dx.doi.org/10.1103/physrevd.36.724[_phys . * d36 * ( 1987 ) 724 ] . collaborations , lep susy working group , `` stops and sbottoms , '' note lepsusywg/04 - 02.1 . http://lepsusy.web.cern.ch/lepsusy/welcome.html . collaboration , a. heister _ et al . _ , `` search for scalar quarks in @xmath497 collisions at @xmath496 up to 209 gev , '' _ phys . * b537 * ( 2002 ) 520 , http://arxiv.org/abs/hep-ex/0204036[arxiv:hep-ex/0204036 ] . w. beenakker , r. hopker , m. spira , and p. m. zerwas , `` squark and gluino production at hadron colliders , '' http://dx.doi.org/10.1016/s0550-3213(97)00084-9[_nucl . * b492 * ( 1997 ) 51103 ] , http://arxiv.org/abs/hep-ph/9610490[arxiv:hep-ph/9610490 ] . j. pumplin _ et al . _ , `` new generation of parton distributions with uncertainties from global qcd analysis , '' _ jhep _ * 07 * ( 2002 ) 012 , http://arxiv.org/abs/hep-ph/0201195[arxiv:hep-ph/0201195 ] . d. stump _ et al . _ , `` inclusive jet production , parton distributions , and the search for new physics , '' _ jhep _ * 10 * ( 2003 ) 046 , http://arxiv.org/abs/hep-ph/0303013[arxiv:hep-ph/0303013 ] . t. sjostrand , s. mrenna , and p. skands , `` pythia 6.4 physics and manual , '' _ jhep _ * 05 * ( 2006 ) 026 , http://arxiv.org/abs/hep-ph/0603175[arxiv:hep-ph/0603175 ] . m. muhlleitner , a. djouadi , and y. mambrini , `` sdecay : a fortran code for the decays of the supersymmetric particles in the mssm , '' http://dx.doi.org/10.1016/j.cpc.2005.01.012[_comput . * 168 * ( 2005 ) 4670 ] , http://arxiv.org/abs/hep-ph/0311167[arxiv:hep-ph/0311167 ] . m. l. mangano , m. moretti , f. piccinini , r. pittau , and a. d. polosa , `` alpgen , a generator for hard multiparton processes in hadronic collisions , '' _ jhep _ * 07 * ( 2003 ) 001 , http://arxiv.org/abs/hep-ph/0206293[arxiv:hep-ph/0206293 ] . i. hinchliffe and l. littenberg , `` phenomenological consequences of supersymmetry , '' in proc . of dpf summer study on elementary particle physics and future facilities , aspen , colo . , jun 28 - jul 16 , 1982 . g. l. kane and j. p. leveille , `` experimental constraints on gluino masses and supersymmetric theories , '' http://dx.doi.org/10.1016/0370-2693(82)90968-6[_phys . * b112 * ( 1982 ) 227 ] . j. r. ellis and h. kowalski , `` supersymmetric particles at the cern @xmath502 collider , '' http://dx.doi.org/10.1016/0550-3213(84)90291-8[_nucl . phys . _ * b246 * ( 1984 ) 189 ] . h. baer , k. hagiwara , and x. tata , `` gauginos as a signal for supersymmetry at @xmath490 colliders , '' http://dx.doi.org/10.1103/physrevd.35.1598[_phys . rev . _ * d35 * ( 1987 ) 1598 ] . nath and r. l. arnowitt , `` supersymmetric signals at the tevatron , '' http://dx.doi.org/10.1142/s0217732387000446[_mod . lett . _ * a2 * ( 1987 ) 331341 ] . r. barbieri , f. caravaglios , m. frigeni , and m. l. mangano , `` production and leptonic decays of charginos and neutralinos in hadronic collisions , '' http://dx.doi.org/10.1016/0550-3213(91)90040-5[_nucl . phys . _ * b367 * ( 1991 ) 2859 ] . r. m. barnett , j. f. gunion , and h. e. haber , `` gluino decay patterns and signatures , '' http://dx.doi.org/10.1103/physrevd.37.1892[_phys . * d37 * ( 1988 ) 1892 ] . h. baer , x. tata , and j. woodside , `` gluino cascade decay signatures at the tevatron collider , '' http://dx.doi.org/10.1103/physrevd.41.906[_phys . rev . _ * d41 * ( 1990 ) 906 ] . collaboration , v. m. abazov _ et al . _ , `` search for squarks and gluinos in events with jets and missing transverse energy using 2.1 fb@xmath29 of @xmath490 collision data at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1016/j.physletb.2008.01.042[_phys . * b660 * ( 2008 ) 449457 ] , http://arxiv.org/abs/0712.3805[arxiv:0712.3805 [ hep - ex ] ] . collaboration , t. aaltonen _ et al . _ , `` inclusive search for squark and gluino production in @xmath35 collisions at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1103/physrevlett.102.121801[_phys . * 102 * ( 2009 ) 121801 ] , http://arxiv.org/abs/0811.2512[arxiv:0811.2512 [ hep - ex ] ] . collaboration , v. m. abazov _ et al . _ , `` search for scalar top quarks in the acoplanar charm jets and missing transverse energy final state in @xmath490 collisions at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1016/j.physletb.2008.05.037[_phys . lett . _ * b665 * ( 2008 ) 18 ] , http://arxiv.org/abs/0803.2263[arxiv:0803.2263 [ hep - ex ] ] . collaboration , v. m. abazov _ et al . _ , `` search for pair production of scalar bottom quarks in @xmath490 collisions at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1103/physrevlett.97.171806 [ _ phys . * 97 * ( 2006 ) 171806 ] , http://arxiv.org/abs/hep-ex/0608013[arxiv:hep-ex/0608013 ] . collaboration , t. aaltonen _ et al . _ , `` search for direct pair production of supersymmetric top and supersymmetric bottom quarks in @xmath502 collisions at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1103/physrevd.76.072010[_phys . rev . _ * d76 * ( 2007 ) 072010 ] , http://arxiv.org/abs/0707.2567[arxiv:0707.2567 [ hep - ex ] ] . collaboration , v. m. abazov _ et al . _ , `` search for the lightest scalar top quark in events with two leptons in @xmath490 collisions at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1016/j.physletb.2007.11.086[_phys . lett . _ * b659 * ( 2008 ) 500508 ] , http://arxiv.org/abs/0707.2864[arxiv:0707.2864 [ hep - ex ] ] . , `` search for pair production of stop quarks mimicking top events signatures , '' cdf note 9439 . collaboration , t. aaltonen _ et al . _ , `` search for gluino - mediated sbottom production in @xmath35 collisions at @xmath494 tev , '' http://dx.doi.org/10.1103/physrevlett.102.221801[_phys . * 102 * ( 2009 ) 221801 ] , http://arxiv.org/abs/0903.2618[arxiv:0903.2618 [ hep - ex ] ] . collaboration , t. aaltonen _ et al . _ , `` search for supersymmetry in @xmath490 collisions at @xmath496 = 1.96 tev using the trilepton signature of chargino - neutralino production , '' http://dx.doi.org/10.1103/physrevlett.101.251801[_phys . lett . _ * 101 * ( 2008 ) 251801 ] , http://arxiv.org/abs/0808.2446[arxiv:0808.2446 [ hep - ex ] ] . collaboration , v. m. abazov _ et al . _ , `` search for associated production of charginos and neutralinos in the trilepton final state using 2.3 fb@xmath29 of data , '' http://arxiv.org/abs/0901.0646[arxiv:0901.0646 [ hep - ex ] ] . collaboration , v. m. abazov _ et al . _ , `` search for supersymmetry via associated production of charginos and neutralinos in final states with three leptons , '' http://dx.doi.org/10.1103/physrevlett.95.151805 [ _ phys . * 95 * ( 2005 ) 151805 ] , http://arxiv.org/abs/hep-ex/0504032[arxiv:hep-ex/0504032 ] . collaboration , t. aaltonen _ et al . _ , `` search for chargino - neutralino production in @xmath490 collisions at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1103/physrevlett.99.191806[_phys . * 99 * ( 2007 ) 191806 ] , http://arxiv.org/abs/0707.2362[arxiv:0707.2362 [ hep - ex ] ] . collaborations , lep susy working group , `` rp violation with lle couplings , '' note lepsusywg/02 - 10.1 . http://lepsusy.web.cern.ch/lepsusy/welcome.html . collaboration , a. heister _ et al . _ , `` search for supersymmetric particles with r parity violating decays in @xmath497 collisions at @xmath496 up to 209 gev , '' http://dx.doi.org/10.1140/epjc/s2003-01311-5[_eur . phys . j. _ * c31 * ( 2003 ) 116 ] , http://arxiv.org/abs/hep-ex/0210014[arxiv:hep-ex/0210014 ] . collaboration , j. abdallah _ et al . _ , `` search for supersymmetric particles assuming r - parity non- conservation in @xmath491 collisions at @xmath496 = 192 gev to 208 gev , '' http://dx.doi.org/10.1140/epjc/s2004-01871-8[_eur . phys . j. _ * c36 * ( 2004 ) 123 ] , http://arxiv.org/abs/hep-ex/0406009[arxiv:hep-ex/0406009 ] . collaboration , p. achard _ et al . _ , `` search for r parity violating decays of supersymmetric particles in @xmath497 collisions at lep , '' http://dx.doi.org/10.1016/s0370-2693(01)01367-3[_phys . * b524 * ( 2002 ) 6580 ] , http://arxiv.org/abs/hep-ex/0110057[arxiv:hep-ex/0110057 ] . collaboration , g. abbiendi _ et al . _ , `` search for r parity violating decays of scalar fermions at lep , '' http://dx.doi.org/10.1140/epjc/s2004-01596-8[_eur . phys . j. _ * c33 * ( 2004 ) 149172 ] , http://arxiv.org/abs/hep-ex/0310054[arxiv:hep-ex/0310054 ] . collaboration , r. barate _ et al . _ , `` search for r - parity violating decays of supersymmetric particles in @xmath497 collisions at center - of - mass energies from 189 gev to 202 gev , '' http://dx.doi.org/10.1007/s100520100633[_eur . phys . j. _ * c19 * ( 2001 ) 415428 ] , http://arxiv.org/abs/hep-ex/0011008[arxiv:hep-ex/0011008 ] . collaboration , a. heister _ et al . _ , `` search for r - parity violating production of single sneutrinos in @xmath491 collisions at @xmath496 = 189 gev to 209 gev , '' http://dx.doi.org/10.1007/s10052-002-1010-2[_eur . phys . j. _ * c25 * ( 2002 ) 112 ] , http://arxiv.org/abs/hep-ex/0201013[arxiv:hep-ex/0201013 ] . collaboration , j. abdallah _ et al . _ , `` search for resonant sneutrino production at @xmath496 = 183 gev to 208 gev , '' http://dx.doi.org/10.1140/epjc/s2003-01158-8[_eur . phys . j. _ * c28 * ( 2003 ) 1526 ] , http://arxiv.org/abs/hep-ex/0303033[arxiv:hep-ex/0303033 ] . collaboration , m. acciarri _ et al . _ , `` search for r - parity breaking sneutrino exchange at lep , '' http://dx.doi.org/10.1016/s0370-2693(97)01200-8[_phys . * b414 * ( 1997 ) 373381 ] . collaboration , g. abbiendi _ et al . _ , `` tests of the standard model and constraints on new physics from measurements of fermion pair production at 189 gev at lep , '' http://dx.doi.org/10.1007/s100520050718[_eur . phys . j. _ * c13 * ( 2000 ) 553572 ] , http://arxiv.org/abs/hep-ex/9908008[arxiv:hep-ex/9908008 ] . collaboration , a. aktas _ et al . _ , `` search for bosonic stop decays in r - parity violating supersymmetry in e+ p collisions at hera , '' http://dx.doi.org/10.1016/j.physletb.2004.08.032[_phys . * b599 * ( 2004 ) 159172 ] , http://arxiv.org/abs/hep-ex/0405070[arxiv:hep-ex/0405070 ] . collaboration , a. aktas _ et al . _ , `` search for squark production in r parity violating supersymmetry at hera , '' http://dx.doi.org/10.1140/epjc/s2004-01972-4[_eur . phys . j. _ * c36 * ( 2004 ) 425440 ] , http://arxiv.org/abs/hep-ex/0403027[arxiv:hep-ex/0403027 ] . collaboration , s. chekanov _ et al . _ , `` search for stop production in r - parity - violating supersymmetry at hera , '' http://dx.doi.org/10.1140/epjc/s10052-007-0240-8[_eur . phys . j. _ * c50 * ( 2007 ) 269281 ] , http://arxiv.org/abs/hep-ex/0611018[arxiv:hep-ex/0611018 ] . collaboration , a. abulencia _ et al . _ , `` search for anomalous production of multi - lepton events in @xmath490 collisions at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1103/physrevlett.98.131804 [ _ phys . * 98 * ( 2007 ) 131804 ] , http://arxiv.org/abs/0706.4448[arxiv:0706.4448 [ hep - ex ] ] . collaboration , v. m. abazov _ et al . _ , `` search for r - parity violating supersymmetry via the ll anti - e couplings @xmath453 , @xmath454 or @xmath455 in @xmath490 collisions at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1016/j.physletb.2006.05.077 [ _ phys . * b638 * ( 2006 ) 441449 ] , http://arxiv.org/abs/hep-ex/0605005[arxiv:hep-ex/0605005 ] . collaboration , t. aaltonen _ et al . _ , `` search for pair production of scalar top quarks decaying to a @xmath193 lepton and a @xmath188 quark in @xmath35 collisions at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1103/physrevlett.101.071802[_phys . lett . _ * 101 * ( 2008 ) 071802 ] , http://arxiv.org/abs/0802.3887[arxiv:0802.3887 [ hep - ex ] ] . collaboration , v. m. abazov _ et al . _ , `` search for resonant second generation slepton production at the tevatron , '' http://dx.doi.org/10.1103/physrevlett.97.111801[_phys . * 97 * ( 2006 ) 111801 ] , http://arxiv.org/abs/hep-ex/0605010[arxiv:hep-ex/0605010 ] . collaboration , a. abulencia _ et al . _ , `` search for new high mass particles decaying to lepton pairs in @xmath35 collisions at @xmath503 tev , '' http://dx.doi.org/10.1103/physrevlett.95.252001 [ _ phys . * 95 * ( 2005 ) 252001 ] , http://arxiv.org/abs/hep-ex/0507104[arxiv:hep-ex/0507104 ] . collaboration , d. e. acosta _ et al . _ , `` search for new physics using high mass tau pairs from 1.96 tev @xmath35 collisions , '' http://dx.doi.org/10.1103/physrevlett.95.131801[_phys . * 95 * ( 2005 ) 131801 ] , http://arxiv.org/abs/hep-ex/0506034[arxiv:hep-ex/0506034 ] . collaboration , a. abulencia _ et al . _ , `` search for high - mass resonances decaying to e @xmath74 in @xmath35 collisions at @xmath498 1.96 tev , '' http://dx.doi.org/10.1103/physrevlett.96.211802[_phys . * 96 * ( 2006 ) 211802 ] , http://arxiv.org/abs/hep-ex/0603006[arxiv:hep-ex/0603006 ] . collaboration , v. m. abazov _ et al . _ , `` search for scalar neutrino superpartners in @xmath504 final states in @xmath35 collisions at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1103/physrevlett.100.241803[_phys . * 100 * ( 2008 ) 241803 ] , http://arxiv.org/abs/0711.3207[arxiv:0711.3207 [ hep - ex ] ] . collaborations , lep susy working group , `` acoplanar two - photon events , '' note lepsusywg/04 - 09.1 . http://lepsusy.web.cern.ch/lepsusy/welcome.html . collaboration , f. abe _ et al . _ , `` searches for new physics in diphoton events in @xmath35 collisions at @xmath505 tev , '' http://dx.doi.org/10.1103/physrevd.59.092002[_phys . rev . _ * d59 * ( 1999 ) 092002 ] , http://arxiv.org/abs/hep-ex/9806034[arxiv:hep-ex/9806034 ] . collaboration , a. heister _ et al . _ , `` search for gauge mediated susy breaking topologies in @xmath497 collisions at center - of - mass energies up to 209 gev , '' http://dx.doi.org/10.1007/s10052-002-1005-z[_eur . phys . j. _ * c25 * ( 2002 ) 339351 ] , http://arxiv.org/abs/hep-ex/0203024[arxiv:hep-ex/0203024 ] . collaboration , g. abbiendi _ et al . _ , `` searches for gauge - mediated supersymmetry breaking topologies in @xmath491 collisions at lep2 , '' http://dx.doi.org/10.1140/epjc/s2006-02524-8[_eur . phys . j. _ * c46 * ( 2006 ) 307341 ] , http://arxiv.org/abs/hep-ex/0507048[arxiv:hep-ex/0507048 ] . , `` limits on gauge - mediated supersymmetry - breaking models using diphoton events with missing transverse energy at cdf ii , '' cdf note 9625 . collaboration , v. m. abazov _ et al . _ , `` search for supersymmetry in di - photon final states at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1016/j.physletb.2007.12.006[_phys . lett . _ * b659 * ( 2008 ) 856863 ] , http://arxiv.org/abs/0710.3946[arxiv:0710.3946 [ hep - ex ] ] . b. c. allanach _ et al . _ , `` the snowmass points and slopes : benchmarks for susy searches , '' http://dx.doi.org/10.1007/s10052-002-0949-3 [ _ eur . phys . j. _ * c25 * ( 2002 ) 113123 ] , http://arxiv.org/abs/hep-ph/0202233[arxiv:hep-ph/0202233 ] . collaboration , a. abulencia _ et al . _ , `` search for heavy , long - lived particles that decay to photons at cdf ii , '' http://dx.doi.org/10.1103/physrevlett.99.121801[_phys . * 99 * ( 2007 ) 121801 ] , http://arxiv.org/abs/0704.0760[arxiv:0704.0760 [ hep - ex ] ] . collaborations , lep susy working group , `` gmsb with sleptons as nlsp , '' note lepsusywg/02 - 09.2 . http://lepsusy.web.cern.ch/lepsusy/welcome.html . collaboration , v. m. abazov _ et al . _ , `` search for long - lived charged massive particles with the d0 detector , '' http://dx.doi.org/10.1103/physrevlett.102.161802[_phys . * 102 * ( 2009 ) 161802 ] , http://arxiv.org/abs/0809.4472[arxiv:0809.4472 [ hep - ex ] ] . collaboration , t. aaltonen _ et al . _ , `` search for long - lived massive charged particles in 1.96 tev @xmath506 collisions , '' http://arxiv.org/abs/0902.1266[arxiv:0902.1266 [ hep - ex ] ] . s. raby , `` gauge - mediated susy breaking with a gluino lsp , '' http://dx.doi.org/10.1016/s0370-2693(98)00050-1[_phys . lett . _ * b422 * ( 1998 ) 158162 ] , http://arxiv.org/abs/hep-ph/9712254[arxiv:hep-ph/9712254 ] . h. baer , k .- cheung , and j. f. gunion , `` a heavy gluino as the lightest supersymmetric particle , '' http://dx.doi.org/10.1103/physrevd.59.075002[_phys . rev . _ * d59 * ( 1999 ) 075002 ] , http://arxiv.org/abs/hep-ph/9806361[arxiv:hep-ph/9806361 ] . s. raby and k. tobe , `` the phenomenology of susy models with a gluino lsp , '' http://dx.doi.org/10.1016/s0550-3213(98)00703-2[_nucl . * b539 * ( 1999 ) 322 ] , http://arxiv.org/abs/hep-ph/9807281[arxiv:hep-ph/9807281 ] . n. arkani - hamed and s. dimopoulos , `` supersymmetric unification without low energy supersymmetry and signatures for fine - tuning at the lhc , '' _ jhep _ * 06 * ( 2005 ) 073 , http://arxiv.org/abs/hep-th/0405159[arxiv:hep-th/0405159 ] . g. f. giudice and a. romanino , `` split supersymmetry , '' http://dx.doi.org/10.1016/j.nuclphysb.2004.11.048[_nucl . phys . _ * b699 * ( 2004 ) 6589 ] , http://arxiv.org/abs/hep-ph/0406088[arxiv:hep-ph/0406088 ] . g. f. giudice and a. romanino , `` split supersymmetry : erratum , '' _ nucl . phys . _ * b706 * ( 2005 ) 65 . collaboration , a. heister _ et al . _ , `` search for stable hadronizing squarks and gluinos in @xmath491 collisions up to @xmath496 = 209 gev , '' http://dx.doi.org/10.1140/epjc/s2003-01376-0[_eur . phys . j. _ * c31 * ( 2003 ) 327342 ] , http://arxiv.org/abs/hep-ex/0305071[arxiv:hep-ex/0305071 ] . collaboration , j. abdallah _ et al . _ , `` search for an lsp gluino at lep with the delphi detector , '' http://dx.doi.org/10.1140/epjc/s2002-01111-5[_eur . phys . j. _ * c26 * ( 2003 ) 505525 ] , http://arxiv.org/abs/hep-ex/0303024[arxiv:hep-ex/0303024 ] . collaboration , v. m. abazov _ et al . _ , `` search for stopped gluinos from @xmath35 collisions at @xmath496 = 1.96 tev , '' http://dx.doi.org/10.1103/physrevlett.99.131801[_phys . * 99 * ( 2007 ) 131801 ] , http://arxiv.org/abs/0705.0306[arxiv:0705.0306 [ hep - ex ] ] . a. arvanitaki , s. dimopoulos , a. pierce , s. rajendran , and j. g. wacker , `` stopping gluinos , '' http://dx.doi.org/10.1103/physrevd.76.055007[_phys . rev . _ * d76 * ( 2007 ) 055007 ] , http://arxiv.org/abs/hep-ph/0506242[arxiv:hep-ph/0506242 ] .
as scientists and as educators we ask ourselves , `` how do we know what we know ? '' if we had to derive everything we used from scratch , we would not have the time to make much progress . but it is worthwhile to derive _ some _ fundamental parameters that we use in our area of expertise . in 1987 mit physicist philip morrison hosted a television series called `` ring of truth : an inquiry into how we know what we know . '' one purpose of the show was to demonstrate simple measurements . in the episode called `` mapping '' he measured the circumference of the earth with the `` van of eratosthenes '' . he and his crew went to the north edge of kansas and measured the elevation angle of the bright southern star antares as it transited the meridian . then they drove 370 miles down highway 183 to the south edge of kansas and measured antares again . they found that it was 5 degrees higher at the second location . the distance corresponded to 1/72 of a circle , so the implied circumference of the earth was 26,600 miles , which is a bit more than 6 percent greater than the correct value ( 24,901 miles ) . the cosmological distance ladder is a sequence of steps used by astronomers to derive distances within the solar system , throughout the galaxy , and beyond to the farthest galaxies detectable @xcite . it hinges on simple geometry and the principles of surveying . for example , using the positions of two observers separated by some baseline on the earth , we can determine the distance to the moon , a nearby planet or an asteroid . in this paper we determine the sizes of the first three rungs of the cosmological distance ladder . these are : 1 ) the radius of the earth ; 2 ) the distance to the moon ; and 3 ) the distance to the sun ( i.e. the size of the astronomical unit , or au ) . the first two can be accomplished without a telescope . the third was attempted by various pre - telescopic astronomers such as tycho brahe in 1582 @xcite , but was not accomplished until 1672 by gian domenico cassini and john flamsteed @xcite , who made observations of mars when it was prominently visible in the middle of the night and therefore about as close to the earth as it gets . ( mars is roughly 0.6 au distant at such a time . when it is on the other side of the sun it is 2.6 au distant . ) measuring the size of the au requires telescopic measurements or distances to nearby planets or asteroids determined with radar . once the scale of the solar system is known , we can determine the distances to nearby stars via trigonometric stellar parallaxes if our positional measurements are good to better than 0.1 arc second . then , classically , we determine the distance to the hyades star cluster @xcite and tie other star clusters to the hyades distance . with the discovery of certain standard candles and standardizable candles in star clusters ( e.g. rr lyrae stars and cepheids ) , we can calibrate the mean intrinsic brightness of these pulsating stars . they are useful for distance determinations throughout our galaxy . with the hubble space telescope we can determine distances to other galaxies using the cepheid period - luminosity relation as far as 25 million parsecs @xcite . ( 1 parsec equals 3.086 @xmath2 10@xmath4 km , or 206265 au . ) if we observe type ia supernovae in some galaxies whose distances are known via cepheids , we can calibrate the intrinsic brightness of these supernovae . since type ia supernovae are typically 4 billion times brighter than the sun at maximum brightness , they can be used to determine distances halfway across the observable universe with a 4-m class telescope . the methodology of this paper is to use only our own data , if possible . we make minimal use of the _ astronomical almanac _ and minimal use of telescopes . many of the results here can be obtained independently by students using very unsophisticated and inexpensive equipment . in fig . [ triangle ] we show the astronomical triangle . the northern sky turns around the north celestial pole ( ncp ) , near the direction of the star polaris . the ncp is @xmath5 degrees above the horizon . @xmath5 is the latitude . the azimuth ( @xmath6 ) is measured clockwise around the horizon and is equal to 0@xmath7 deg at the north point on the horizon , 90@xmath7 deg at the east point on the horizon , etc . the hour angle @xmath8 divided by 15 deg / hr is the number of hours that an object is west of the celestial merdian . @xmath8 is negative for objects in the eastern sky . the declination ( @xmath9 ) of a celestial object is the number of degrees the object is north or south of the celestial equator . if we happen to know the declination of the sun and we determine how high it is above the horizon when it transits the _ celestial meridian _ , we can determine our latitude . at local apparent noontime the elevation angle of the sun will be : @xmath10 = 0 mm we are concerned here with the sun and moon as observed at mid - northern latitudes . these objects transit the celestial meridian between the zenith and the south point on the horizon . = 9 mm to determine the elevation angle of the sun we can set up a vertical pointed stick or _ gnomon_. even better is a vertical stick with a small sphere at the top , like that shown in fig . [ gnomon ] . it is easier to measure the center of the elliptical shadow of the sphere on the ground than it is to measure the end of the darker part of the shadow of a vertical pointed stick . the sun s declination ranges from @xmath11 to @xmath12 over the course of the year . @xmath13 is the _ obliquity of the ecliptic _ , the tilt of the earth s axis of rotation to the plane of its orbit . ( the ecliptic is the apparent path of the sun through the constellations of the zodiac owing to the earth s orbit around the sun . ) on the first day of summer we have : @xmath14 = 0 mm and on december 21st we have : @xmath15 eq . 2 plus eq . 3 gives : @xmath16 / 2 \ ; \ ; .\ ] ] and eq . 2 minus eq . 3 gives us : @xmath17 / 2 \ ; \ ; .\ ] ] in other words , if we determine the maximum elevation angle of the sun on june 21st and december 21st , the average gives us 90@xmath18 , and half the difference gives us @xmath13 . = 9 mm in table [ gnomon_data ] we give the raw data from two solstice experiments carried out in college station , texas . in fig . [ gnomon_xy ] we show the x - y positions of the end of the gnomon s shadow for three key times of the year . note that when the declination of the sun is negative the positions of the end of the shadow on the ground trace out an upward pointing curve . on the first day of spring ( march 20 or 21 ) or the first day of autumn ( about september 23 ) the points delineate a straight line . when the declination of the sun is positive the points trace out a downward curving locus . a graph of this type allows one to determine which points , if any , have been mismeasured . if a student invents data for such an experiment , such a graph allows one to detect such fraud easily . plots of the data from table [ gnomon_data ] are shown in fig . [ shadow ] . we have converted the time values to the number of minutes since local apparent noontime ( when @xmath19 ) . in the top graph we have added , for graphical purposes only , a hyperbolic fit to the data . in fact , there is no simple function that fits such a data set . the best estimate of the minimum value in the summer solstice plot is obtained from a fourth order polynomial fit to the data within 41 minutes of the meridian transit of the sun . for the 2010 june 21 observations we used a gnomon of height @xmath20 mm . the minimum shadow length was l@xmath21 = 85.84 mm with an average scatter of @xmath1 0.64 mm . this average scatter is what is known as a _ random error _ , as the points scatter randomly above and below an appropriate function fit to the data . @xmath22 is equal to the arctangent of @xmath23 , or 82@xmath7 15 9 @xmath1 3 5 . for 2010 december 21 we used a gnomon of height @xmath24 mm . from a fourth order polynomial fit to all 26 points we find l@xmath21 = 753.98 @xmath1 0.74 mm . @xmath22 is found to be 36@xmath7 06 6 @xmath1 3 4 . 4 gives @xmath5 = 30@xmath7 48 7 @xmath1 3 5 , which is 11 5 north of the true value of 30@xmath7 37 2 from google earth . ( one arc minute of latitude equals one nautical mile , or 1852 m. ) eq . 5 gives @xmath13 = 23@xmath7 04 7 @xmath1 3 5 , which is several standard deviations smaller than the true value of 23@xmath7 26 2 . these values show what kind of uncertainty can be obtained with careful observations . so far we have used no information from the _ astronomical almanac_. however , to determine our _ longitude _ requires that we know the amount of time that the apparent sun is ahead or behind the mean sun . this difference is known as the _ equation of time_. it ranges from @xmath2514.2 minutes to + 16.4 minutes over the course of the year . from four gnomon experiments carried out in college station , texas , on 2006 november 12 , 2007 september 24 , and the two solstices mentioned above we found that the mean sun transits the meridian at 12:24:26 pm standard time , with an uncertainty of @xmath172 sec . since each degree of longitude corresponds to 4 minutes of time , our longitude is 96@xmath7 06 5 @xmath1 18 1 . we are within one standard deviation of the true value of 96@xmath7 20 4 . on 2006 september 3 we carried out a gnomon experiment in south bend , indiana , on the campus of the university of notre dame . for this we needed a value of the declination of the sun from the _ astronomical almanac_. our location was found to be latitude 41@xmath7 51 5 , longitude 85@xmath7 50 5 . in late october of 2007 we drove from south bend through bella vista , arkansas , and on to college station , texas , keeping track of our route , mileage , and the length of any side trips . after subtracting the side trips , the elapsed distance on the odometer was 1248 statute miles . for a section of highway through central illinois we noted that 96.8 miles on the odometer corresponded to 98.0 miles according to the highway markers . the implication is that our odometer exhibited a systematic error . on a subsequent occasion we drove the final 92 percent of the exact same route , but with a different car , and the odometer mileage was 40 miles less ( about 4 percent ) . we conclude that an odometer can not be explicitly trusted . we adjust the original mileage to 1248 @xmath26 = 1263 miles . next , we used a `` map tool '' ( also known as an _ opisometer _ ) and traced our route on a map from the 1980 edition of _ the new international atlas _ by rand mcnally , and determined that if we could have traveled along a great circle route from south bend to college station , the direct route had a length that was 0.744 of the length of the route we actually drove . ( this means that we did not rely on the stated scale of the map . ) thus , the great circle distance from one place to the other was 940 statute miles . consider two locations on the earth with ( latitude , longitude ) = ( @xmath27 , @xmath28 ) and ( @xmath29 , @xmath30 ) , respectively . the length of the great circle arc between them ( @xmath31 ) can easily be obtained from the law of cosines of spherical trigonometry : @xmath32 given the latitudes and longitudes determined by gnomon and clock at both locations , the great circle arc was 13.78 deg . thus , our estimate of the circumference of the earth is 940 @xmath33 = 24557 miles , which is 1.4 percent smaller than the true value . our value for the radius of the earth is 6290 km . the true value is 6378 km . given the simplicity of the tools involved , we feel that we have achieved an accuracy far better than our expectations . for comparison we refer the reader to @xcite , which describes projects `` eratosthenes 2009 '' and `` eratosthenes 2010 america '' . more than 15,000 students at more than 200 schools determined the radius of the earth using this method . they obtained 6290 km in 2009 and 6375 km in 2010 . the method used above to determine the circumference of the earth is in principle the same as that used by eratosthenes . it turns out that it was a generation _ before _ eratosthenes that aristarchus of samos ( ca . 310 - 230 bc ) determined the distance to the moon in terms of the radius of the earth . inherent in both methods is the acceptance of the idea that the earth is , for all intents and purposes , spherical . aristarchus cleverly deduced that we can determine the distance to the moon from the geometry of a lunar eclipse . see @xcite and fig . [ lunar_ecl ] . a lunar eclipse can only take place when the moon is opposite the sun . the shadow of the ( spherical ) earth is circular in any plane perpendicular to the earth - sun line . consider the triangle delineated by points a , c , and h. ac is the radius of the earth , r@xmath0 . ch is the distance from the center of the earth to the moon , d@xmath34 . the _ horizontal parallax _ of the moon , p@xmath35 , is given by : @xmath36 = 0 mm thus , the distance to the moon in earth radii is 1/sin(p@xmath35 ) . in fig . [ lunar_ecl ] p@xmath37 is the horizontal parallax of the sun . @xmath38 is the angular radius of the sun . @xmath39 is the angular radius of the earth s shadow at the distance of the moon . for @xmath40xch it is obvious that @xmath41xch + p@xmath37 + p@xmath35 = 180@xmath7 . in other words , 180@xmath42xch = p@xmath37 + p@xmath35 . also , @xmath38 + @xmath41xch + @xmath39 = 180@xmath7 , which means that 180@xmath42xch = @xmath38 + @xmath39 . it follows that : @xmath43 = 9 mm aristarchus believed that the sun was 19 times more distant than the moon . the true ratio is closer to 400 . our value for the radius of the earth and our value of the distance to the sun given in 5 below means that the solar parallax p@xmath44 014 . aristarchus knew that the moon and sun approximately had the same angular diameter because the duration of a total solar eclipse is never more than a few minutes . he took the angular diameter of the moon to be 0.5 deg . in the _ almagest _ , section v 14 , ptolemy states that the earth s shadow at the distance of the moon is 2.6 times the angular diameter of the moon @xcite . we can use two actual measurements of the sun made on 2010 may 1 and 2011 june 6 to provide an angular diameter of 308 @xmath1 0.9 , so @xmath45 154 . in a previous paper @xcite we presented measures of the moon s angular diameter from sightings of the moon through a 6.2 mm hole in a piece of cardboard that slides up and down a yardstick . our mean value for the angular diameter of the moon was 3118 . data in that paper and subsequent data yield a time of lunar perigee of 2011 may 14.70 ut , with a perigee - to - perigee period ( i.e. the _ anomalistic month _ ) of 27.4992 @xmath1 0.0518 d , which is within 1-@xmath38 of the correct value of 27.55455 d. based on our non - telescopic data , the lunar eclipse of 2011 june 15 occurred about 4.6 days after perigee , so the moon s angular size would have been a bit larger than the mean value . using six images obtained from www.slooh.com , we determined that the earth s shadow was 2.56 @xmath1 0.03 times the angular diameter of the full moon during the lunar eclipse of 2011 june 15 . thus , @xmath39 = 0.5 @xmath2 31.18 @xmath2 2.56 @xmath46 3919 . it follows that p@xmath35 = 154 + 3919 @xmath25 014 @xmath46 5517 . 7 gives a lunar distance of 62.3 r@xmath0 with a random error of at least @xmath1 0.7 r@xmath0 . the true mean distance is 60.27 r@xmath0 . we have seen that the geometry of a lunar eclipse allows a determination of the distance to the moon . hipparchus ( ca . 140 bc ) used observations of the _ solar _ eclipse of 14 march 190 bc to derive a distance to the moon between 71 and 83 r@xmath47 @xcite . but can one determine the moon s distance without the use of an eclipse ? simultaneous observations of the moon s position from two widely separated places on the earth would suffice . @xcite shows how this can be done from two observatories widely separated in latitude but on the same meridian of longitude . observations of the moon from the _ same _ location over the course of a night can also be used to determine the distance to the moon . in this case one also uses a large fraction of the earth s diameter as a baseline . the complication is that while we wait for the earth to rotate , the moon is also orbiting the earth . the moon s orbital period with respect to the background stars is 27.32 d. thus , _ on average _ , the moon moves 0.55 deg east per hour against the background of stars as viewed by a hypothetical observer at the earth s center . however , because of the parallax of the moon , the observed motion for someone situated on the surface of the earth is roughly 1/6 deg per hour less @xcite . we need the geocentric vs. topocentric shifts in right ascension and declination , respectively , of the moon owing to our location on the surface of the earth . the _ topocentric _ coordinate system has the observer , standing on the earth s surface , at the origin . let @xmath48 = @xmath49 = @xmath50 . ( right ascension increases to the east , while hour angle increases to the west . ) then eq . 35 of @xcite gives @xmath51 = 0 mm letting @xmath52 and t = tan@xmath53 , eq . 38 of @xcite is @xmath54 } { \rm{cos } \ ; \delta \ ; \rm{cos } \ ; t \ ; - \ ; \left ( \frac{r _ { \earth}}{d _ { moon}}\right ) \ ; \rm{cos } \ ; \phi^{\prime } } \ ; \ ; .\ ] ] here the primed parameters are the topocentric values of declination , hour angle , and latitude , while unprimed values are for a hypothetical observer at the earth s center . eqs . 9 and 10 can not be easily inverted to give d@xmath34/r@xmath0 . but since these equations contain this ratio , we now show how single - site observations can be used to obtain the distance to the moon . what follows is primarily a demonstration of method . = 9 mm in fig . [ moon_rate ] we show the rate at which the moon s apparent position ( for an observer situated in college station , texas ) varies as a function of hour angle for a number of occasions over nearly the full range of distance of the moon from the earth . not surprisingly , the moon moves against the background of stars more slowly on a night when it is near apogee than when it is near perigee . in fig . [ cross_staff ] we show a simple cross staff . the pattern for the device that slides up and down the yardstick was obtained from the university of washington . with such a device we can derive the angular separation of two objects in the sky . for two objects with known right ascensions and declinations @xmath31 is given by an expression similar to eq . 6 : @xmath55 if the moon s angular separation from two stars of known ra and dec is determined , then it is possible to determine the ra and dec of the moon . the easiest way to envision this is to take a star chart and a compass and draw portions of two circles of different radius centered at the locations of the two particular stars . one obtains two numerical solutions of the moon s position , one on each side of the great circle arc joining the two stars . one of those possibilities can easily be eliminated . once an approximate location of the moon s position is found , one can use a computer program to search a box that covers a range of ra and dec to find the celestial location that is the specified number of degrees from each of the two stars of known position . in table [ separations ] we give five sets of observations made on the night of 2011 may 15 ( ut ) , an occasion when the moon was close to the bright star @xmath56 vir and within a day of perigee . on this night the moon was waxing and 93 percent illuminated . thus , it was visible almost the entire night . one of our reference `` stars '' was saturn . given the accuracy of our observations we can assume that the position of saturn was constant on this night . as one can see , the angular separation of saturn and @xmath56 vir was measured to better than 0.2 deg on five occasions . because the moon is not a point source , measuring the angular separation of the moon and saturn or the moon and a star is more difficult than measuring the angular separation of two bright stars or one bright star and a planet . in table [ posns ] we give the derived right ascensions and declinations of the moon from our observations of 2011 may 15 . at the end of the night the reference stars @xmath9 crv and @xmath57 hya were too low in the sky to be seen . the uncertainties of the right ascension and declination of the moon at 08:03 and 08:41 ut were derived with the assumption of an uncertainty of @xmath1 0.3 deg for the angular separation of saturn vs. the moon and @xmath56 vir vs. the moon . because of the large uncertainty in the ra and dec of these final two observations , we exclude data sets four and five from further analysis . the first , second , and third determinations of the position of the moon yield an apparent angular rate of 0.508 @xmath1 0.094 the corresponding range of the moon s distance is 47.3 to 58.5 r@xmath0 with a most likely value of 52.1 r@xmath0 . 2010 october 21 ut was the day after lunar apogee . on this date we made a series of six measurements of the angular separation of the moon vs. jupiter over a 9.1 hour period . ( no stars near the moon were bright enough to be seen with the unaided eye given the quality of the sky . ) thus , we could not use the same method to obtain multiple estimates of the ra and dec of the moon at a given time . the separation of the moon and jupiter increased from 11.31 to 14.77 deg , with a mean rate of increase of 0.403 @xmath1 0.028 deg / hr . under the assumption that the moon was moving directly away from the position of jupiter ( which was only approximately true ) , this angular rate is a lower limit to the rate of change of position of the moon against the background of stars . the implied range of lunar distance was 57.2 to 62.1 r@xmath0 , with a most likely value of 59.5 r@xmath0 . if we take the 1-@xmath38 upper limit of the rate of change of position of the moon on 2011 may 15 and the 1-@xmath38 lower limit of the rate of change of the lunar position on 2010 october 21 , we get a conservative estimate of the range of the moon s distance , namely 47 to 62 r@xmath0 . obviously , the observations and analysis required for this method of determining an estimate of the moon s distance are much more complicated than aristarchus s method using the geometry of a lunar eclipse . in 2008 the authors of this paper were participants in the summer science program ( ssp ) , a residential non - credit enrichment program for incoming high school juniors and seniors . coauthors krisciunas and kim were faculty , while the other coauthors were students . it has been the tradition at ssp to divide the students into teams of three for the observing , but each student then writes computer code to produce a determination of the orbital parameters of a particular asteroid . in table [ orbit_det ] we give the orbital solution of asteroid 8567 (= 1996 hw1 ) by one of us ( j. s. ) along with the parameters obtainable from the horizons website of the jet propulsion laboratory . the code to determine the orbital parameters was debugged using observations of ceres and the orbital parameters of ceres given by jpl horizons . we note that our orbit solution is based on observations extending over a very small arc of the full orbit . ( the orbital period of asteroid 1996 hw1 is 2.93 years . ) it had been an ambition at ssp for some time to observe a near earth object simultaneously from two sites so that its distance could be derived , the length of the astronomical unit could be measured , and the scale of the solar system be determined . we report one such experiment here . on 2008 july 24 ut we took images of asteroid 1996 hw1 at etscorn observatory of the new mexico institute of mining and technology in socorro , new mexico , using a 15 cm takahashi refractor and ccd camera . the geographic position was latitude + 34@xmath7 04@xmath58 217 , longitude w 106@xmath7 54@xmath58 501 . a simultaneous image was obtained in ojai , california , at latitude + 34@xmath7 26@xmath58 040 , longitude w 119@xmath7 11@xmath58 226 using a 25 cm meade reflector and ccd camera . figs . [ pos3 ] and [ pos4 ] show the asteroid at 08:17:28 ut on 2008 july 24 , as viewed from the two sites . the right ascensions and declinations of a number of field stars were determined using * ds9 * and an image of the field obtained from the space telescope science institute digital sky survey . we then determined a transformation from pixel coordinates to right ascension and declination using the iraf programs * ccmap * and * cctrans*. this allowed us to determine the right ascension and declination of the asteroid for each of our images . from the new mexico site we determined the asteroid s topocentric position to be @xmath56 = 21:27:07.42 , @xmath9 = + 15@xmath7 53@xmath58 0277 from the etscorn image and the positions of the field stars , the root - mean - square errors in right ascension and declination were @xmath1 0.31 and @xmath1 0.36 arcsec , respectively . our image obtained in california just barely has the asteroid in the frame , and the guiding was not as good . still , we find a topocentric position of @xmath56 = 21:27:07.77 , @xmath9 = + 15@xmath7 53@xmath58 0225 . the rms errors of the field star positions were @xmath1 0.52 and @xmath1 0.44 arcsec , respectively , for right ascension and declination . the effect on right ascension and declination due to parallax and the finite size of the earth can be calculated much more easily than using eqs . 9 and 10 ( above ) if the planet or asteroid under consideration has a distance @xmath59 considerably greater than the distance to the moon . @xcite gives these relevant equations : @xmath60 @xmath61 = 0 mm the values on the left hand sides of these equations are measured in radians . = 9 mm given that the latitudes of the new mexico and california sites were almost the same , we have almost no leverage to use eq . 13 to determine the distance to the asteroid . we limit ourselves to a consideration of the effect of parallax on the right ascensions . let the new mexico site be `` position 1 '' and the california site be `` position 2 '' . consider the seconds part of the observed right ascension of the asteroid . there exist corrections to the right ascension such that 7.77 @xmath62 = 7.42 @xmath63 . the corrections @xmath64 adjust the observed right ascensions to what would be observed by a hypothetical observer at the center of the earth . so , 0.35 seconds of time equals the difference of the parallactic corrections . since one second of time in right ascension equals 15 cos @xmath9 arc seconds , we have 505 @xmath1 0.61 = @xmath65 . the uncertainty comes from the square root of the sum of squares of the uncertainties of the pixel to ra / dec transformations from * ccmap * in iraf . at the new mexico site at the time fig . [ pos3 ] was taken the hour angle of the asteroid was @xmath251.899 deg , while at the california site the hour angle of the asteroid was @xmath2514.176 deg when fig . [ pos4 ] was taken . 12 we obtain a distance to the asteroid of : @xmath66 = 0 mm the reader will know that the number of arc seconds in a radian is 206265 . if we adopt _ our _ value of the radius of the earth from 3 ( namely , 6290 km ) , we obtain a distance to the asteroid of ( 4.66 @xmath1 0.56 ) @xmath2 10@xmath67 km . = 9 mm using the steeger orbit solution in table [ orbit_det ] , the method of ( * ? ? ? * ff . ) , and the rectangular coordinates of the sun on 2008 july 24 from the _ astronomical ephemeris _ , we determined that asteroid 1996 hw1 was 0.294 au distant when figs . [ pos3 ] and [ pos4 ] were taken . our resulting value for the astronomical unit is ( 1.59 @xmath1 0.19 ) @xmath2 10@xmath3 km , which is roughly 6 percent larger than the true value of 1.496 @xmath2 10@xmath3 km . our value of the solar parallax , p@xmath37 in fig . [ lunar_ecl ] , equals ( 6290/1.59 @xmath2 10@xmath3 ) @xmath2 206265 = 82 , or about 014 , which we used in 4.1 . our distance to the sun in earth radii divided by our distance to the moon in earth radii gives a sun distance that is roughly 406 times the distance to the moon , which is close to the true mean value of 389 . in the interests of readability we have not included all statistical details above . however , given the readership of the _ american journal of physics _ , we thought it prudent to have a discussion of various kinds of errors here . say you want to answer a simple question such as `` how tall am i ? '' by standing up against the wall , placing a ruler level on the top of your head and making a mark on the wall you can find out the answer . but what if you forgot to take your shoes off ? you would have a _ systematic error _ equal to the height of the heels of your shoes . what if you measured your height immediately after getting out of bed after a full night s sleep ? you would find that first thing in the morning you are about 2 cm taller than you will be later that same day . why is this so ? the disks between the vertebrae in your back expand when you are lying down sleeping . thus , if you measured your height first thing in the morning , right after lunch , then after a 6 mile run in the afternoon , you could get three different answers . the `` mean value of your height during the day '' might be 178 @xmath1 1 cm . this standard deviation of @xmath1 1 cm is the `` mean error of the mean '' of the set of measurements . this is referred to as a _ random error _ because values that you measure would scatter about some average value . very often a set of measurements exhibits a gaussian ( i.e. bell - shaped ) distribution . then 68.3 percent of the measurements are within one standard deviation of the mean value , 95.5 percent are within 2-@xmath38 and 99.7 percent are within 3-@xmath38 . this allows us to identify _ outliers _ in the data . if some data value is several ( or many ) standard deviations from the mean ( or expected ) value , we have either underestimated the size of our random errors , or there is some unaccounted source of systematic error . in 2 above we derived a value of the latitude for a particular location in college station , texas , which was 11 5 north of the true value this is the systematic error . usually in science we do not know the `` true '' value , but in this case we do . from a scatter of the measurements of the shadow length of our gnomon we could obtain a numerical value of the root - mean - square scatter of the data about some best fit line . that gives us an estimate of the uncertainty of the minimum shadow length , which translates into a random error for the maximum elevation angle of the sun on some given day , which then leads to a random error for our value of the latitude . our value of the latitude is within 3 of our standard deviations of the true value . our value of the tilt of the earth s axis of rotation to the plane of its orbit is 6 standard deviations from the true value . thus , we have room for improvement in the experiment , should we choose to do it again and again . some sources of uncertainty have not yet been mentioned . for the location of the june 21st and december 21st gnomon experiments we have measured the levelness of the spot where the data were taken , finding that a perfectly straight gnomon which was perfectly squarely set in its base would have been tilted 6 @xmath1 1 arc minute north of the zenith . this would have made the elevation angle too low by that amount at local noontime , but less so at other times . to complicate matters , our gnomon is just a wooden dowel rod which is _ not _ totally straight , and we do not know just how squarely it sits in the block of wood that is its base . regarding the use of the `` map tool '' in 3 , we found that the great circle arc from south bend , indiana , to college station , texas , was 0.7437 @xmath1 0.0009 of the route driven . we have rounded this off to 0.744 and have decided to ignore the numerical uncertainty . in 4.1 we found from six images of the moon in partial lunar eclipse taken from 21:29:55 to 21:52:59 ut that the angular diameter of the earth s shadow was 2.46 , 2.65 , 2.49 , 2.62 , 2.56 , and 2.56 times the angular diameter of the moon . these were obtained using hard copies of the images , a compass , and a ruler . in all instances we could see more than a 180 degree arc of the illuminated part of the moon . the mean ratio was 2.56 @xmath1 0.03 . to find angle @xmath39 ( the angular radius of the earth s shadow ) we used the ( more robust ) mean value of the angular diameter of the moon from krisciunas ( 2010 ) rather than an estimated value of the moon s angular size on 2011 june 15 . in 4.2 and table [ separations ] we give results of an experiment that is at or beyond the capabilities of a ruler plus cardboard cross piece . we found that the systematic error of the angular separation of two bright point sources ( saturn vs. @xmath56 vir ) was + 0.12 deg . the random error was @xmath10.04 deg . these are considerably better than the often quoted uncertainty of 0.20 to 0.25 deg for non - telescopic measurements . but consider our measurements of the angular separation of saturn vs. the moon and @xmath56 vir vs. the moon . our mean systematic errors are as great as 0.39 deg and our random errors are @xmath1 0.23 to @xmath1 0.29 deg . table [ posns ] shows that if we have measured the moon s position with respect to three or more point sources , we can locate the moon s location on the sky to within 0.3 deg on average . but if we have measured the moon s position with respect to only two point sources , the error in position can be too large to be useful for determining the rate of change of the moon s position over the course of a night . when it comes to certain kinds of analysis , there are other errors that are neither systematic or random . in the top half of fig . [ shadow ] we have fitted a hyperbola to the full data set . but this is for illustrative purposes only . careful scrutiny shows that the points do not randomly scatter above and below the curved line . the actual function we need requires the latitude of the site , which is the thing we are trying to _ derive_. in this case , for practical reasons we have fitted a fourth order polynomial to a subset of the data around the time of the minimum shadow length . the bottom line is that we fit some data under the _ assumption _ that we have the appropriate function . what the most appropriate function is also depends on the typical size of the error bars of the data points . if the error bars are large , high order polynomial fits would be unjustified . using a vertical stick , a car , watch , map , and map tool , we have measured the size of the earth from first principles . we used observations of the elevation angle of the sun over the course of a few hours overlapping local noontime on the summer solstice and winter solstice of 2010 to determine the latitude and longitude of college station , texas . we also determined the latitude and longitude of south bend , indiana . the two sites were 13.78 deg apart along a great circle arc . we obtained a value of the radius of the earth of 6290 km , about 1.4 percent too small . aristarchus s method of determining the distance to the moon derives from the geometry of the earth , moon , and sun at the time of a lunar eclipse . one key observable is the angular size of the earth s shadow at the distance of the moon compared to the angular size of the moon . using images available on the web of the lunar eclipse of 2011 june 15 we found that the earth s shadow was 2.56 times the angular diameter of the moon . the corresponding distance of the moon was 62.3 earth radii , about 3.3 percent larger than the true mean value . it is also possible to determine the distance to the moon from observations of its motion against the background of stars . these observations are made from a single site over the course of one night . this method is far more complicated than aristarchus s method , but from such observations we showed that the moon was between 47 and 62 r@xmath0 distant . the true range of the moon s distance is 55.9 to 63.8 r@xmath0 . simultaneous observations from socorro , new mexico , and ojai , california allowed us to determine the distance to the asteroid 1996 hw1 on 2008 july 24 . a determination of the orbital elements of this asteroid by one of us allowed us to calculate that the asteroid was 0.294 au distant on that date . the final result was a calibration of the au equal to ( 1.59 @xmath1 0.19 ) @xmath2 10@xmath3 km , which is 6 percent larger than the true value . given the seeing and tracking constraints associated with our asteroid observations , our asteroid experiment was at the limits of the technology of our small telescopes . to determine the distance to a _ main belt _ asteroid ( which would have been 10 times more distant ) would have required a baseline 10 times bigger than the 1130 km baseline we used if we wanted the parallax to be several arc seconds . such an experiment would be nearly impossible on the surface of the earth . this paper has overflowed with numerical and trigonometrical details . still , the _ qualitative _ results are easy to understand and are worth restating . using very inexpensive equipment we can determine the size of the earth and the distance to the moon . to determine the length of the au we do _ not _ need to organize an international endeavor like what was done for the venus transits of 1761 , 1769 , 1874 , and 1882 @xcite . we can measure the length of the au with carefully timed observations of a near earth asteroid using telescopes comparable to those owned by many amateur astronomers . it is also worth reminding ourselves why astronomers have such an obsession with determining cosmic distances . thanks to kepler s third law the length of the au gets us the distances to all other objects that orbit the sun . once we have the distances to stars like the sun ( converting hydrogen to helium in their cores , but with a range of mass ) via trigonometric stellar parallaxes the next rung of the distance ladder we can use photometric methods to determine distances to star clusters using the fact that `` main sequence stars '' of the same mass have comparable intrinsic brightness . a simple equation relates the apparent brightness of a star with its intrinsic brightness and the distance . we can exploit the cepheid period - luminosity law to calibrate our way across the galaxy and to nearby galaxies . knowing the energy budget of a star helps us determine what it is made of , what is its structure , how it will live , and how it will die . luminosities of type ia supernovae have allowed us to address some the largest questions we can ask , such as , `` what is the ultimate fate of the universe ? '' the interconnectedness of astronomical topics means that the big questions are related to the earth size experiment of eratosthenes carried out more than 22 centuries ago . starting with an instrument as simple as a vertical stick we can connect basic and profound aspects of the universe we inhabit . tom weimar helped with the construction of the gnomon pictured in fig . [ gnomon ] . we thank nick suntzeff for useful discussions . the nmt observations of asteroid 1996 hw1 were obtained with the help of jeff lu , david oh , anna heinz , peter combs , and rebecca mickol . we thank dan klinglesmith and jason speights for observing support at etscorn observatory . the asteroid observations were made as part of the summer science program , which is funded by private donations and sponsored by new mexico institute of mining and technology . elisabeth button kindly made fig . 1 . van helden , a. 1995 , in planetary astronomy from the renaissance to the rise of astrophysics , part b : the eighteenth and nineteenth centuries , r. taton and c. wilson , eds . ( cambridge : cambridge univ . press ) , 153 clcl 11:25:00 & 330 & 11:07:45 & 844 + 11:35:00 & 300.5 & 11:15:00 & 829 + 11:46:30 & 268.5 & 11:25:00 & 807 + 11:55:00 & 245.5 & 11:35:15 & 787 + 12:06:17 & 217 & 11:49:20 & 770 + 12:15:04 & 196 & 11:59:20 & 761.5 + 12:25:02 & 172 & 12:10:00 & 756.5 + 12:35:01 & 151 & 12:15:00 & 754.5 + 12:45:06 & 131 & 12:20:00 & 754 + 12:55:00 & 112.5 & 12:25:15 & 754 + 13:00:10 & 104 & 12:30:00 & 755 + 13:05:00 & 99 & 12:36:00 & 756.5 + 13:10:47 & 92.5 & 12:40:38 & 759.5 + 13:15:00 & 88.5 & 12:45:18 & 761.5 + 13:20:00 & 86.5 & 12:50:00 & 765.5 + 13:25:00 & 86 & 12:55:43 & 771 + 13:30:10 & 86 & 13:00:00 & 775 + 13:37:20 & 92 & 13:05:20 & 782 + 13:45:00 & 101 & 13:10:06 & 788.5 + 13:49:30 & 106 & 13:20:00 & 805 + 13:56:00 & 117 & 13:30:00 & 824.5 + 14:04:30 & 132 & 13:40:00 & 848 + 14:15:00 & 154 & 13:50:00 & 875 + 14:25:00 & 175.5 & 14:01:20 & 910 + 14:38:12 & 206 & 14:10:50 & 946 + 14:45:00 & 224 & 14:20:06 & 987 + 14:55:00 & 247.5 & & + 15:05:00 & 275 & & + 15:15:00 & 303 & & + 15:25:00 & 331.5 & & + clrrl 02:21:00 & @xmath56 vir vs. moon & 3.@xmath741 & 3.@xmath770 & + 0.@xmath729 + 02:24:00 & saturn vs. moon & 12.85 & 13.21 & + 0.36 + 02:28:00 & @xmath9 crv vs. moon & 10.71 & 10.85 & + 0.14 + 02:30:00 & saturn vs. @xmath56 vir & 13.60 & 13.71 & + 0.11 + 04:04:00 & @xmath56 vir vs. moon & 3.28 & 3.40 & + 0.12 + 04:06:40 & saturn vs. moon & 13.34 & 13.86 & + 0.52 + 04:08:00 & @xmath9 crv vs. moon & 11.33 & 11.34 & + 0.01 + 04:10:00 & saturn vs. @xmath56 vir & 13.63 & 13.71 & + 0.08 + 05:54:00 & @xmath56 vir vs. moon & 3.32 & 3.22 & @xmath250.10 + 05:56:00 & saturn vs. moon & 14.52 & 14.53 & + 0.01 + 05:58:30 & @xmath9 crv vs. moon & 11.77 & 11.91 & + 0.14 + 06:01:40 & saturn vs. @xmath56 vir & 13.55 & 13.71 & + 0.16 + 06:04:00 & @xmath57 hya vs. moon & 9.33 & 9.31 & @xmath250.02 + 08:02:00 & @xmath56 vir vs. moon & 3.63 & 3.15 & @xmath250.48 + 08:03:00 & saturn vs. moon & 14.80 & 15.34 & + 0.54 + 08:05:00 & saturn vs. @xmath56 vir & 13.55 & 13.71 & + 0.16 + 08:31:00 & saturn vs. @xmath56 vir & 13.60 & 13.71 & + 0.11 + 08:38:00 & @xmath56 vir vs. moon & 3.35 & 3.17 & @xmath250.18 + 08:41:00 & saturn vs. moon & 15.06 & 15.60 & + 0.54 + cccccc 02:24:00 & 198.@xmath715 @xmath1 0.09 & @xmath2512.@xmath776 @xmath1 0.09 & 198.@xmath727 & @xmath2513.@xmath720 & 0.@xmath746 + 04:04:40 & 198.76 @xmath1 0.04 & @xmath2513.14 @xmath1 0.07 & 198.93 & @xmath2513.56 & 0.45 + 05:56:00 & 199.53 @xmath1 0.02 & @xmath2513.94 @xmath1 0.01 & 199.63 & @xmath2513.90 & 0.10 + 08:03:00 & 199.50 @xmath1 0.57 & @xmath2514.31 @xmath1 0.36 & 200.55 & @xmath2514.25 & 1.02 + 08:41:00 & 200.03 @xmath1 0.62 & @xmath2514.27 @xmath1 0.34 & 200.87 & @xmath2514.34 & 0.81 + clll t@xmath68 & epoch ( ut ) & 2008 june 29.35 & 2007 january 15.00 + t@xmath68 & epoch ( julian date ) & 2454646.85 & 2454115.5 + @xmath69 & mean anomaly at t@xmath68 & 335.@xmath71461 & 155.@xmath733678 + @xmath70 & semi - major axis & 2.0855 au & 2.046041 au + @xmath71 & eccentricity & 0.4575 & 0.449165 + @xmath72 & inclination angle & 8.@xmath75033 & 8.@xmath7437363 + @xmath73 & longitude of ascending node & 177.@xmath76887 & 177.@xmath7216737 + @xmath74 & argument of perihelion & 176.@xmath71618 & 177.@xmath7020070 +
asteroseismology is the only systematic technique to study the insides of a star . pulsators are fortunately found all over the h - r diagram , and allow us the opportunity to look inside different stars in various evolutionary stages . white dwarf stars are the stellar remains of 9899% of stars in the sky @xcite , and contain an archival record of their main sequence lifetime . pulsating white dwarf stars serve as effective instruments to harness this archival record . white dwarf spectra reveal that 80% of them have atmospheres dominated by hydrogen ( das ; @xcite ) . hydrogen atmosphere da white dwarfs are observed to pulsate in an instability strip located in the temperature range 1080012500k for @xmath0 @xcite . recent studies determine that the da instability strip is only about 10001200k in width @xcite . these da variables ( davs ) are known as the zz ceti stars , after the proto type zz ceti ( r548 ) of the class . in this paper we gather the pulsation spectra of all known non - interacting zz ceti stars with the purpose of studying their ensemble pulsation characteristics , and how these properties change across the instability strip . such a systematic study of the zz ceti pulsators was first undertaken by pioneers in the field such as @xcite , @xcite , and @xcite . @xcite was the first to demonstrate the distinct behavior of pulsation periods and amplitudes as a function of temperature systematically for a significant sample of davs . the sample size of known davs is now three times larger than the sample used in the last such characterization of the zz ceti pulsators by @xcite . also , we now have better and internally consistent temperatures for these pulsators , calculated using the ml2/@xmath1 convection description ( see * ? ? ? * ; * ? ? ? this prescription for convection gives the best internal consistency between optical and ultra - violet ( uv ) effective temperatures , trigonometric parallaxes , v magnitudes , and gravitational redshifts @xcite . these two reasons compel us to re - examine the pulsation characteristics of the instability strip . of the 35 new zz ceti variables published in @xcite , hs0951 + 1312 , hs0952 + 1816 , wd1443 + 0134 , wd1524 - 0030 , and wd2350 - 0054 do not have reliable spectroscopic @xmath2 and @xmath3 fits ( see @xcite ) . our sample of new zz ceti variables discovered using spectra from the sloan digital sky survey ( sdss ) then consists of 30 davs from @xcite and 11 davs from @xcite . for these 41 davs , d. eisenstein derived a homogeneous and internally consistent set of @xmath2 and @xmath3 values from the sdss using model atmospheres from d. koester ( see * ? ? ? * ) . we carefully and consistently re - analyzed all of our time - series photometry data on these pulsators , acquired using the prime focus ccd camera argos @xcite on the 2.1 m telescope at mcdonald observatory . we shall henceforth refer to this set of 41 davs as the sdss zz ceti sample with homogeneous spectroscopic fits and homogeneous time - series photometry . @xcite and @xcite have published internally consistent temperatures and @xmath3 fits for 39 davs using their latest model atmospheres ; we will henceforth refer to this second set of 39 davs as the bg04 zz ceti sample . we compiled a corresponding set of 39 pulsation spectra from the literature , and via private communication with our colleagues . the seismic data of the bg04 zz ceti sample was acquired by different observers using different instruments and telescopes . however a substantial amount of time - series data exists on most davs in this sample , and we utilized practically all published pulsation spectra to carefully derive well - averaged values of weighted mean period and pulsation amplitudes which we present in this paper . @xcite show evidence of a relative shift of about 200k between the sdss and bg04 zz ceti instability strips , which also differ in shape and width . the spectra of zz ceti stars in these samples were analyzed using different techniques and with different model atmospheres . as homogeneity is imperative , we can not merge these two samples for analyses that involve the spectroscopic temperature . however we can merge these two samples when formulating plots based solely on seismic data such as figure 3 . during the writing of this manuscript , 25 new zz ceti stars were submitted for publication in separate papers : silvotti et al . 2005a , castanheira et al . 2005 , kepler et al . 2005b , gianninas 2005 , and silvotti et al . this brings the total number of non - interacting zz ceti stars known to 107 . twenty of the new davs in these papers come from the sdss . their spectroscopic @xmath2 and @xmath3 values were derived using the same technique as the 41 sdss davs discovered previously , but with a different algorithm ( version auto23 vs. auto21 used earlier ) . hence we can not include the new davs in our sdss zz ceti sample for plots based on effective temperature , such as figures 1 and 2 . also , we find that the main purpose of the pulsation spectra presented in these papers is to demonstrate discovery of variability . such pulsation spectra may be incomplete and are not well suited for direct inclusion in most plots of this paper where ensemble homogeneity plays a big role . however we will include these variables in figure 3 ( based on seismic data ) for completeness , and show the net effect of the inclusion on the original plot . @xcite , @xcite , @xcite , and @xcite show that the non - linear pulse shapes observed in some variable white dwarfs may arise as a result of relatively thick convection zones . many frequencies are evident in these stars because the non - linearities appear as harmonics and linear combinations in our fourier transforms . we are interested in studying the pulsation characteristics of self - excited real modes in the context of this paper . hence we will disregard all harmonics and linear combination frequencies in the observed pulsation spectra throughout this paper . when our program detects a linear combination or harmonic in an observed pulsation spectrum , we typically hand - pick the lowest amplitude mode as the linear combination mode . however if we find a mode involved in two or more linear combinations , we consider it to be a linear combination mode even if it is not the lowest amplitude mode . our simplistic approach could lead us to incorrect choices ; we could be misinterpreting resonances as linear combination modes in a few cases . however any small discrepancies in values evaluated for a single star should have a minimal impact on the conclusions we draw from the ensemble . short pulsation periods ( 100300s ) in the zz ceti stars typically have pulsation amplitudes of a few percent or smaller . long pulsation periods ( 6001000s ) typically have large amplitudes that can be as high as 10% . this correlation between the pulsation periods and amplitudes of the zz ceti stars has been established for a long time @xcite . however the correlation of these properties with temperature had to wait more than a decade for sufficiently accurate determinations of zz ceti temperatures . subsequently the distinct behavior of pulsation periods and amplitudes as a function of temperature was systematically demonstrated for a significant sample of davs by @xcite and more recently by @xcite . we now demonstrate and discuss these trends for the new as well as previously known davs of our two homogeneous samples . we will begin with showing how the number of observed independent modes changes across the instability strip . we expect a larger ( than intrinsic ) scatter in the number of independent modes we measure for an ensemble of zz ceti stars due to varied detection efficiency . this partially arises because we use different instruments on different telescopes , or observe for different durations of time , and also due to the variety of weather conditions and extinction values . a fraction of the scatter in our detection efficiency is caused because zz ceti stars have different magnitudes and different pulsation amplitudes , and the pulsation amplitudes depend on the effective wavelength of observation @xcite . by using the sdss zz ceti sample alone , we reduce some of this expected scatter ( see subsection 1.1 ) . to reduce the scatter additionally , we choose to limit ourselves to the magnitude range @xmath4 and to a stellar mass @xmath5 to exclude intrinsic low amplitude davs ( see section 2.3 ) . reducing our magnitude range any more will take us further into the realm of small number statistics . for the subset of 23 sdss davs that fall in the chosen range , we find an average of 2.5modes / star for the hotter zz ceti stars ( 1130011800k ) , and a slightly larger value of 3.1modes / star for the cooler zz ceti stars ( 1080011300k ) . if we consider all the sdss davs in our sample , excluding the three massive variables ( wd0923 + 0120 , wd1711 + 6541 , & wd2159 + 1322 ) , then we find an average of 2.9modes / star for the hotter zz ceti stars ( 1130011800k ) and 3.3modes / star for the cooler zz ceti stars ( 1080011300k ) up to an amplitude limit of order a few mma . this may seem contrary to published literature , but we are attempting to restrict ourselves to independent self - excited modes only , rather than plot the total number of observed periodicities in the star . the period spacing for nonradial g - modes with @xmath63 is larger than the asymptotic limit . this increase in the density of available eigenmodes as a function of period is consistent with the increase we find in the number of observed independent modes , comparing the hotter zz ceti stars to the cooler zz ceti stars . pulsations in zz ceti models ( and in stars in general ) are self - driven oscillations . stochastic noise frequencies coincident with the eigenfrequencies are amplified by the driving mechanism to observable amplitudes . the blue edge of the zz ceti instability strip occurs when the star is cool enough to have a hydrogen partial ionization zone sufficiently deep to excite global pulsations . our models suggest that the driving frequency is governed by the thermal timescale at the base of the hydrogen partial ionization zone ( see @xcite ) . as the model star cools , the base of the partial ionization zone moves deeper in the model , and the thermal timescale increases @xcite . numerical calculations by montgomery ( 2005 ) show that the thermal timescale at the base of the ionization zone @xmath7 is proportional to @xmath8 . however recent investigations by @xcite suggest that the thermal timescale is not consistent with the mean pulsation period , although both quantities increase in the models as we traverse from the blue to the red edge of the instability strip . figure 1 shows the mean pulsation period as a function of effective temperature for davs in both the sdss and bg04 zz ceti samples . we determine the weighted mean period ( wmp ) for each dav by weighting each period with the corresponding observed amplitude ( see equation 1 ) . @xmath9 we can now appreciate the significance of figure 1 ; the weighted mean pulsation period provides us with a method of determining the temperature of a zz ceti star independent of spectroscopy . the respective equations of the best fit lines through the sdss and the bg04 zz ceti samples shown in figure 1 are : @xmath10 = -0.835@xmath11 @xmath12 + 10060@xmath13@xmath14 = -0.830@xmath15 @xmath16 + 10240@xmath17 where wmp is in units of seconds and @xmath2 is in units of kelvin . it is reassuring to see an agreement in the values of the slope for both the sdss and bg04 zz ceti samples , and gives us more confidence in using this period - temperature correlation as a means to measure the effective temperature of a dav . we can now determine a temperature for the five variables with unreliable spectroscopic fits we excluded earlier , hs0951 + 1312 , hs0952 + 1816 , wd1443 + 0134 , wd1524 - 0030 , and wd2350 - 0054 ( see table 1 in the appendix ) . we compute a sum of the squares of the * horizontal * offsets of the points in figure 1 respectively from the above best fit lines , and arrive at a seismic estimate of the average uncertainty in our spectroscopic temperature . we determine @xmath18=188.3k for the sdss zz ceti sample , and @xmath18=185.4k for the bg04 zz ceti sample . it is not surprising that the average @xmath2 uncertainties in both these samples turned out to be similar because zz ceti stars in both samples define an instability strip of the same width @xmath1910001200k@xcite . we can expect that if the average @xmath2 uncertainty in the sdss sample had been significantly larger than that of the bg04 sample , the empirical width of the sdss instability strip would also have been correspondingly larger . our current temperature determinations are getting better and may be internally consistent to about 200k at present . our empirical measure of the width of the instability strip appears to have reduced from 1500k @xcite to about 10001200k @xcite . but even with our current atmospheric models we do find exceptions to this period-@xmath2 correlation , where the spectroscopic temperature does not match the pulsation properties of the zz ceti star . for example , @xcite report the zz ceti star wd2350 - 0054 with a dominant period close to 300s as an unusual pulsator because its sdss spectroscopic determination places it below the red edge of the instability strip at @xmath20k . @xcite report a known complex pulsator g29 - 38 @xcite as a hot zz ceti star with @xmath21k . better s / n spectra of these and other such zz ceti stars may reveal temperatures consistent with their observed periods . unless the pulsation characteristics themselves reveal curious features , it seems premature to conclude that the zz ceti star in question ( with a spectroscopic temperature inconsistent with its pulsation spectrum ) is strange or abnormal in any way . measuring the pulsation periods of a zz ceti star is a model - independent and straight forward process as opposed to a spectroscopic determination of its temperature . the instability strip spans 10001200k in temperature and about 1300s in period . the internal uncertainty in measuring temperature using spectra is typically about 200k , which is 1720% of the width of the instability strip . while even low quality photometric data will yield a period precise to at least a few seconds . for the hotter zz ceti stars , with little or no amplitude modulation , the uncertainty in wmp constitutes less than 1% of the width of the instability strip in period . cooler zz ceti stars with a significant amplitude modulation can exhibit pulsation spectra with a wmp different by 3060s and in a few cases even as much as 100200s . for most of the cooler zz ceti stars , the uncertainty in wmp represents less than 5% of the width of the instability strip . we therefore expect that this method of using the wmp to effectively measure the location of a dav within the instability strip is in general more accurate and reliable than spectroscopy . any dispersion in period due to differences in stellar mass , core composition , h / he layer masses , etc . , can increase the uncertainty of determining the temperature of a given dav using this relation . we expect this may be significantly smaller than our present typical spectroscopic @xmath2 uncertainties of a few 100k . note that we do not claim that the relationship between the wmp and the temperature is linear ; a straight line is merely the simplest model fit possible to the observations shown in figure 1 , considering the large amount of scatter . we do not necessarily require a linear relation between wmp and @xmath2 for this method to work as a temperature indicator . our collaborators at the university of texas are presently investigating the interpretation of these data in terms of the theoretical models ( kim et al . 2006 ; in preparation ) . the plot of wmp vs. @xmath2 is similar to comparing two independent temperature scales , each with its own independent source of uncertainties . the uncertainty in spectroscopic @xmath2 depends on the quality of the spectrum , the accuracy and completeness of the model atmosphere , and the details of the algorithm used to fit the observed spectrum with the template spectra . large amplitude variables have a corresponding higher uncertainty in temperature @xcite . the uncertainty in wmp comes from the quality of the photometric observations , and the amplitude modulation of the star . as long as we restrict our relative @xmath2 parameter in units of seconds in the wmp temperature scale , the uncertainty in our measurements is not related to spectroscopic temperature values at all . it is only when we attempt to translate wmp into a temperature in degree kelvin , that we have to use the relation between wmp and spectroscopic temperature . even in this case , we are still better off than the typical 200k uncertainty in spectroscopic temperature because the slopes of the best fit lines in figure 1 do not depend on a few stars , but on 3540 stars . using wmp directly as a temperature scale is non - intuitive at the present time . but this maybe worth thinking about as an alternative scale in the long run , once we improve our understanding of the relation between wmp and @xmath2 , both observationally and theoretically . we presently classify the zz ceti stars into two groups : the hot davs and the cool davs . however a clear dividing point ( in temperature ) between these two classes does not exist to date in published literature . our present classification of these stars is thus vague in this context , although it has certainly proved to be a useful guide . the primary reason such a temperature based classification can not be well defined is because the location and width of the instability strip are model - dependent features . model atmospheres of dav stars treat convection with some parameterization , the choice of which can shift the instability strip in temperature by a few thousand k(@xcite ) . the pulsation characteristics of the zz ceti stars helped us divide them into simple and complex pulsators in the late seventies . a decade later , we recognized that the simple pulsators that exhibit a few modes with short periods ( 200300s ) , small amplitudes ( few mma ) , sinusoidal or saw - toothed pulse shapes , and continued to show the same pulsation spectra over a few decades , were hot zz ceti stars . we also realized that the complex pulsators that exhibit several long periods ( 6001200s ) with large amplitudes ( 40110mma ) , non - sinusoidal pulse shapes with fast rise times and slow decadence , and amplitude modulation over timescales of a few days to a few years , were cool zz ceti stars . although we now use spectroscopic temperature as a means to classify these stars , the classification scheme came about only because we initially used the pulsation characteristics to separate these stars into two groups . we suggest a new zz ceti classification scheme based on the wmp of these variables . we intend to retain the fundamental aspect of the previous scheme in using a temperature based classification . we have shown that wmp is also a temperature scale ( figure 1 ) . we expect that wmp serves as a more accurate and reliable @xmath2 scale than spectroscopic temperature because measuring the pulsation periods of a zz ceti star constitutes a relatively simple , less uncertain , and model - independent exercise compared to measuring its spectroscopic temperature . we redefine the class of hot davs ( hdavs ) as zz ceti stars with wmp@xmath22350s . we redefine the class of cool davs ( cdavs ) as zz ceti stars with wmp@xmath23650s . we suggest introducing a new class of davs , to be called the intermediate davs ( idavs ) , as zz ceti stars with 350@xmath24wmp@xmath24650 . this class merely forms the evolutionary subclass adjoining the hot and cool zz ceti stars and typically encompasses those zz ceti pulsators that show a large range of pulsation periods , e.g. hs0507 + 0435b . for borderline cases such as g29 - 38 , when one season of observations place the dav in one class , and a second season places it in another class , then we suggest choosing the coolest class of the two possibilities . we find several cases where a cdav or an idav exhibit modes typical of the hotter pulsators , but we have not seen an instance of a hot dav exhibiting a mode typical of the cool zz ceti stars . we show our suggested classification for most of the known non - interacting zz ceti stars in tables 1 and 2 . note that the boundaries of 350s and 650s are arbitrary in some sense ; we merely wished to divide the instability strip into three parts and used a period histogram to fine - tune our choice . for simplicity and better readability , tables 1 and 2 in the appendix do not show all of the pulsation spectra we accrued for each star . we show multiple seasons of observations only for those davs that exhibit different frequencies and amplitudes at different times . this also helps us understand the intrinsic changes in weighted mean period and mean pulsation amplitude for such variables . to use such stars in the ensemble , we computed the weighted mean period and mean pulsation amplitude for each season of observations individually , and then included the average value in our analysis . we refer the reader to @xcite , @xcite , @xcite , @xcite , and @xcite for the pulsation spectra of the new davs . the physical quantity of interest in this subsection is the intrinsic pulsation amplitude of modes excited in the zz ceti star . but the measured amplitudes of observed modes are most likely lower than the intrinsic amplitudes due to geometric cancellation . this effect has three independent causes : disk averaging , inclination angle , and limb darkening . we are unable to resolve the disk of the star from earth . hence the observed amplitude of each pulsation mode is lower due to a disk - averaging effect . this explains why the probability of detecting @xmath25=1 modes is higher than the detection probability for @xmath25=2 and @xmath25=3 modes . the inclination angle dictates the distribution of the bright and dark zones in our view for a given mode . eigenmodes with different @xmath26 values exhibit different cancellation patterns ( see @xcite ) . limb darkening effectively reduces the area of the stellar disk in our view , and this reduction in area depends on wavelength . at uv wavelengths , the increased limb darkening decreases the contribution of zones near the limb . as a result , modes of higher @xmath27 are canceled less effectively in the uv compared to the low @xmath27 modes @xcite . of these three independent causes of geometric cancellation , limb darkening is the only one that works in our favor , and provides us with a mode identification technique @xcite . the intrinsic pulsation amplitude depends on the mass of the star . nonradial g - modes have a non - negligible radial component , the amplitude of which scales with stellar mass and plays a role in dictating the amplitude of the nonradial component . the massive pulsators bpm37093 ( @xmath3=8.81 ) , wd0923 + 0120 ( @xmath3=8.74 ) , wd1711 + 6541 ( @xmath3=8.64 ) , and wd2159 + 1322 ( @xmath3=8.61 ) exhibit low amplitudes as a result of their high gravity , and hence we exclude them from section 2.3 . figure 2 shows the square root of total power @xmath28 for 38 stars from the sdss sample ( top ) and 38 stars from the bg04 zz ceti sample ( bottom ) , plotted as a function of their spectroscopic temperature . all of the pulsation amplitudes we report in this paper come from optical whole disk observations . we expect these are mostly low @xmath25 modes . for these reasons , we expect them to be lower than the corresponding intrinsic amplitudes due to geometric cancellation . the points that form the upper envelopes in both panels of figure 2 are then better indicators of the intrinsic amplitude at that temperature . note that in the few cases of under - resolved data , beating of closely spaced periodicities can lead us to determine a relatively smaller or larger amplitude than the observable amplitude . we expect that zz ceti stars with a high pulsation amplitude have a corresponding higher uncertainty in temperature . for large amplitude variables , the surface temperature changes substantially during a pulsation cycle , by as much as a thousand degrees @xcite . depending on which phase of the pulsation cycle ( typical periods of 6001200s for cool davs ) we acquire the spectra and for how long , our measure of their effective temperature can be incorrect by a few hundred degrees . we can attempt to estimate this uncertainty by obtaining time - series photometry on the star simultaneous with the spectroscopic data . at any given temperature , the apparently low amplitude variables could be suffering from extensive geometric cancellation . at the same time , the high amplitude variables with the expectedly least geometric cancellation have a proportionally high uncertainty in temperature due to the large intrinsic temperature fluctuation during a pulsation cycle . this makes the task of interpreting figure 2 difficult . there is fortunately a silver lining to this bleak cloud : zz ceti stars are multi - mode pulsators . if a zz ceti with 3 independent modes still exhibits a small amplitude , then it is unlikely that we are dealing with an unfavorable inclination angle in all three cases and @xmath26 . however we do not fully understand the mode selection mechanism for different @xmath26 values . also , they exhibit different cancellation patterns @xcite . ] . instead of traditional histograms , we adopt a gaussian binning technique where each bin serves as a gaussian function instead of a box ( top - hat ) function . this is a better noise averaging technique because all points contribute to the value of each bin . although the contributions of distant points to the value of an individual bin are small , this method may not be appropriate to study sharp local trends . we are interested in slow trends across the width of the instability strip and do not hesitate to use this technique . we choose gaussian bins with @xmath29 that are an infinitesimal 1k apart from each other . we show the result as a solid line in both panels of figure 2 . both samples show an initial increase in pulsation amplitude , and are also suggestive of a decline prior to the red edge . we established in section 2.2 how the weighted mean pulsation period correlates directly with the effective temperature of the davs for both samples . we now merge both the sdss and the bg04 samples , also including the five davs with unreliable spectroscopic fits ( see section 1.1 ) to plot the pulsation amplitude as a function of the wmp . we no longer have to worry about the internal consistency in their spectroscopic temperatures , and we can use their wmp as a @xmath2 scale . we show these 81 davs as filled circles in figure 3 . we also include the 19 new davs from @xcite , @xcite , and @xcite as filled squares . we have excluded the massive dav wd1337 + 0104 ( @xmath3=8.55 ) from @xcite and the low mass davs he0031 - 5525 ( @xmath3=7.65 ) & wd2135 - 0743 ( @xmath3=7.67 ) from @xcite in figure 3 . figure 3 shows a plot with better statistics than figure 2 due to the larger sample size of 100 davs . we used gaussian bins of width 75s , separated from adjacent bins by an infinitesimal amount of 1s , to produce both the curves shown in figure 3 . the solid line is the histogram determined from the 81 average mass davs of the bg04 and sdss samples , while the dotted line shows the effect of including the new 19 average mass davs . the minor difference between the two lines assures us that our conclusions are robust . we clearly see an initial increase in pulsation amplitude near the blue edge , followed by a gentle rise and then a decline prior to the red edge . we find several reasons why figure 3 looks much more convincing of a decline in amplitude before the red edge compared to figure 2 . changing from a x - axis based on spectroscopic temperature to weighted mean period allows us to include four red edge pulsators with low amplitudes , namely hs0952 + 1816 , wd1443 + 0134 , wd0249 - 0100 , and wd2307 - 0847 . secondly , we get better statistics by combining the two samples that are individually suggestive of a decline in amplitude near the red edge . lastly , we also find that the change in x - axis between figures 2 and 3 moves pulsators left and right , apparently leaving behind cleaner evidence of a decline prior to the red edge . for example , we find that the low amplitude cool dav gd154 moves from 11180k in figure 2 to 1154s in figure 3 , which corresponds to @xmath30k using the slope in the bottom panel of figure 1 as a conversion factor . we have eight pulsators with weighted mean periods of order or greater than 1000s with amplitudes near or less than 25mma . there are no high amplitude pulsators in this range , and the scatter in amplitude near the red edge is much smaller than the center of the zz ceti instability strip . it is unlikely that unfavorable inclination angles can explain the low amplitudes of all the independent modes of all eight pulsators . @xcite established the observed red edge at 11000k , but did not see any decline in pulsation amplitude near the red edge within the instability strip , although they were searching for it . it is for the first time that we now have clear evidence of a decline in pulsation amplitude within the instability strip , just prior to the zz ceti red edge . in other words , the star loses pulsation energy before pulsations shut down at the red edge . this work will have implications and new constraints for our pulsation models . we find that the present large sample of non - interacting davs conforms to the well established trend of increasing pulsation periods with decreasing temperature across the instability strip . investigations of the number of observed independent modes show that hot davs have one more mode on average compared to the cool davs . we also confirm the increase in pulsation amplitude near the blue edge and find strong evidence of a decline in amplitude prior to the red edge . @xcite established the observed zz ceti red edge at 11000k , but did not see any decline in pulsation amplitude near the red edge within the instability strip . we present the first observational evidence that shows the red edge is not an abrupt feature in zz ceti evolution , and that the star loses pulsation energy before pulsations shut down at the red edge . this work poses new constraints on our pulsation models . support for this work was provided by nasa through the hubble fellowship grant hst - hf-01175.01-a awarded by the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , for nasa , under contract nas 5 - 26555 . we thank the referee for helping us make this manuscript a better paper . llllllllllll wd0018 + 0031 & hdav & 11700 & 7.93 & 257.9 5.8 & & & & & & + wd0048 + 1521 & idav & 11290 & 8.23 & 615.3 24.8 & & & & & & + wd0102@xmath310032 & cdav & 11050 & 8.24 & 1043.4 15.3 & 926.3 34.7 & 830.3 35.1 & 752.2 19.4 & & & + wd0111 + 0018 & hdav & 11510 & 8.26 & 292.3 21.9 & 255.3 15.6 & & & & & + wd0214@xmath310823 & hdav & 11570 & 7.92 & 348.1 8.4 & 347.1 8.2 & 297.5 16.0 & 263.5 7.1 & & & + wd0318 + 0030 & cdav & 11040 & 8.07 & 844.9 15.3 & 826.4 27.3 & 695.0 8.9 & 587.1 10.6 & 536.1 11.1 & & + wd0332@xmath310049 & cdav & 11040 & 8.25 & 1143.7 7.4 & 938.4 6.7 & 765.0 15.2 & 402.0 4.1 & & & + wd0332@xmath310049 & cdav & 11040 & 8.25 & 770 23 & 910 10 & & & & & + wd0756 + 2020 & hdav & 11710 & 8.01 & 199.5 6.8 & & & & & & + wd0815 + 4437 & idav & 11620 & 7.93 & 787.5 7.2 & 311.7 18.9 & 309.0 10.2 & 258.3 6.8 & & & + wd0818 + 3131 & hdav & 11800 & 8.07 & 253.3 2.9 & 202.3 3.3 & & & & & + wd0825 + 4119 & idav & 11820 & 8.49 & 653.4 17.1 & 611.0 11.2 & & & & + wd0842 + 3707 & hdav & 11720 & 7.73 & 321.1 4.4 & 309.3 18.0 & 212.3 5.2 & & & & & + wd0847 + 4510 & hdav & 11680 & 8.00 & 200.5 7.0 & 123.4 3.0 & & & & & + wd0906@xmath310024 & idav & 11520 & 8.00 & 769.4 26.1 & 618.8 9.3 & 574.5 23.7 & 457.9 9.8 & 266.6 7.6 & & + wd0913 + 4036 & hdav & 11680 & 7.87 & 320.5 14.7 & 288.7 12.4 & 260.3 16.5 & 203.9 3.8 & & & + wd0923 + 0120 & cdav & 11150 & 8.74 & 655.7 4.4 & & & & & & & + wd0923 + 0120 & cdav & 11150 & 8.74 & 668.9 3.5 & & & & & & & + wd0939 + 5609 & hdav & 11790 & 8.22 & 249.9 7.2 & 48.5 5.9 & & & & & & + wd0942 + 5733 & idav & 11260 & 8.27 & 909.4 7.7 & 694.7 37.7 & 550.5 12.3 & 273.0 9.0 & & & & + wd0949@xmath310000 & idav & 11180 & 8.22 & 711.6 6.0 & 634.2 5.1 & 516.6 16.2 & 365.2 17.7 & 364.1 7.3 & 363.2 12.5 & & + wd0958 + 0130 & hdav & 11680 & 7.99 & 264.4 4.7 & 203.7 2.5 & 121.2 1.6 & & & & & + wd1002 + 5818 & hdav & 11710 & 7.92 & 304.6 5.3 & 268.2 6.8 & & & & & + wd1007 + 5245 & hdav & 11430 & 8.08 & 323.1 10.4 & 290.1 7.7 & 258.8 11.0 & & & & + wd1015 + 0306 & hdav & 11580 & 8.14 & 270.0 8.4 & 255.7 7.3 & 194.7 5.8 & & & & & + wd1015 + 5954 & idav & 11630 & 8.02 & 769.9 5.7 & 453.8 15.5 & 401.7 21.4 & 292.4 8.6 & & & & + wd1015 + 5954 & idav & 11630 & 8.02 & 768.4 7.5 & 455.3 17.7 & 401.4 19.8 & 294.0 9.1 & & & & + wd1015 + 5954 & idav & 11630 & 8.02 & 456.1 19.6 & 399.7 19.2 & 145.5 3.1 & 139.2 4.6 & & & & + wd1054 + 5307 & cdav & 11120 & 8.01 & 869.1 37.4 & & & & & & & + wd1056@xmath310006 & cdav & 11020 & 7.86 & 1024.9 31.6 & 925.4 60.3 & 670.6 12.0 & 603.0 11.5 & & & & + wd1122 + 0358 & cdav & 11070 & 8.06 & 996.1 17.3 & 859.1 34.6 & 740.1 10.0 & & & & & + wd1125 + 0345 & hdav & 11600 & 7.99 & 335.1 2.8 & 265.8 3.3 & 265.5 7.1 & 208.6 2.8 & & & & + wd1157 + 0553 & cdav & 11050 & 8.15 & 1056.2 5.8 & 918.9 15.9 & 826.2 8.1 & 748.5 5.6 & & & + wd1345@xmath310055 & hdav & 11800 & 8.04 & 254.4 2.4 & 195.5 3.9 & 195.2 5.5 & & & & & + wd1354 + 0108 & hdav & 11700 & 8.00 & 322.9 1.9 & 291.6 2.2 & 198.3 6.0 & 173.3 1.1 & 127.8 1.5 & & + wd1355 + 5454 & hdav & 11580 & 7.95 & 324.0 21.8 & & & & & & + wd1417 + 0058 & cdav & 11300 & 8.04 & 980.0 11.3 & 894.6 42.8 & 812.5 32.1 & 749.4 17.9 & 522.0 14.9 & & + wd1502@xmath310001 & idav & 11200 & 8.00 & 687.5 12.0 & 629.5 32.6 & 581.9 11.1 & 418.2 14.9 & & & & + wd1617 + 4324 & cdav & 11190 & 8.03 & 889.7 36.6 & 661.7 21.2 & 626.3 15.4 & & & & & + wd1700 + 3549 & cdav & 11160 & 8.04 & 1164.4 11.4 & 955.3 20.3 & 893.4 54.3 & 552.6 9.3 & & & + wd1711 + 6541 & cdav & 11310 & 8.64 & 1248.2 3.2 & 690.2 3.3 & 606.3 5.2 & 234.0 1.2 & & & & + wd1711 + 6541 & cdav & 11310 & 8.64 & 1186.6 3.3 & 934.8 2.9 & 612.6 5.7 & 561.5 3.0 & 214.3 1.7 & & & + wd1724 + 5835 & hdav & 11540 & 7.89 & 337.9 5.9 & 279.5 8.3 & 189.2 3.2 & & & & & + wd1732 + 5905 & cdav & 10860 & 7.99 & 1248.4 4.5 & 1122.4 8.0 & & & & & & + wd1732 + 5905 & cdav & 10860 & 7.99 & 1336 7.8 & 1090 8.0 & & & & & & + wd2159 + 1322 & cdav & 11710 & 8.61 & 801.0 15.1 & 683.7 11.7 & & & & & & + wd2214@xmath310025 & hdav & 11440 & 8.33 & 255.2 13.1 & 195.2 6.1 & & & & & & + wd1443 + 0134 & cdav & 10830 & & 1085.0 5.2 & 969.0 7.5 & & & & & & + wd1524@xmath310030 & cdav & 11160 & & 873.3 110.8 & 717.5 28.3 & 498.6 21.6 & 255.2 17.9 & & & + wd2350@xmath310054 & hdav & 11690 & & 391.1 3.1 & 304.1 16.3 & 272.8 16.2 & & & & & + wd2350@xmath310054 & hdav & 11710 & & 304.5 13.8 & 272.7 14.8 & 212.6 3.0 & & & & & + wd2350@xmath310054 & hdav & 11680 & & 391.1 7.5 & 304.3 17.0 & 273.3 6.3 & 206.7 3.2 & & & & + hs0951 + 1312 & hdav & 11740 & & 311.7 2.7 & 282.2 9.0 & 258.0 3.6 & 208.0 9.4 & & & & + hs0952 + 1816 & cdav & 10800 & & 1160.9 7.9 & 945.9 10.4 & & & & & & + hs0952 + 1816 & cdav & 10960 & & 1150 4.8 & 883 3.6 & 790 2.9 & 674.7 3.0 & & & & + llllllllllllllll bpm30551 & cdav & 11260 & 8.23 & 920.5 18.5 & 741.4 21.7 & 655.4 17.4 & 442.8 6.5 & & & & & & & & @xcite + bpm30551 & cdav & 11260 & 8.23 & 993.0 6.5 & 936.2 13.0 & 885.6 15.2 & 819.2 19.5 & 738.0 6.5 & 609.6 6.5 & & & & & & @xcite + bpm30551 & cdav & 11260 & 8.23 & 963.8 13.0 & 844.5 19.5 & 799.2 10.9 & 682.7 13.0 & 606.8 13.0 & & & & & & & @xcite + bpm30551 & cdav & 11260 & 8.23 & 920.4 8.7 & 751.6 13.0 & 682.7 10.9 & 606.8 15.2 & 546.1 5.4 & 496.5 4.3 & & & & & & @xcite + bpm30551 & cdav & 11260 & 8.23 & 910.2 8.7 & 862.3 6.5 & 731.4 6.5 & 682.7 13.0 & 607.9 17.4 & & & & & & & @xcite + bpm30551 & cdav & 11260 & 8.23 & 1129.9 10.9 & 1057.0 8.7 & 744.7 18.5 & 606.8 16.3 & & & & & & & & @xcite + bpm30551 & cdav & 11260 & 8.23 & 1137.8 8.7 & 958.1 7.6 & 862.3 16.3 & 744.7 7.6 & 606.8 14.1 & & & & & & & @xcite + bpm30551 & cdav & 11260 & 8.23 & 1092.3 5.4 & 993.0 8.7 & 936.2 9.8 & 744.7 7.6 & 712.3 6.5 & 606.8 11.9 & & & & & & @xcite + bpm30551 & cdav & 11260 & 8.23 & 862.3 6.5 & 799.2 8.7 & 744.7 7.6 & 668.7 7.6 & 612.5 5.4 & & & & & & & @xcite + r548 & hdav & 11990 & 7.97 & 333.6 0.6 & 318.0 0.9 & 274.8 3.8 & 274.3 4.8 & 213.1 7.4 & 212.8 4.7 & 187.3 0.9 & & & & & @xcite + mct0145 - 2211 & idav & 11550 & 8.14 & 823.2 17 & 727.9 19 & 462.2 25 & & & & & & & & & @xcite + bpm31594 & idav & 11540 & 8.11 & 617.9 48.0 & 401.6 16 & 416.1 5 & & & & & & & & & @xcite + hltau76 & idav & 11450 & 7.89 & 933 25.2 & 796 9.7 & 781 9.9 & 541 40.5 & 494 30.2 & 383 21.8 & & & & & & @xcite + g38 - 29 & cdav & 11180 & 7.91 & 1024.0 26.1 & 938.0 26.5 & & & & & & & & & & @xcite + g38 - 29 & cdav & 11180 & 7.91 & 1019.8 12.2 & 910.3 28.3 & & & & & & & & & & @xcite + g191 - 16 & cdav & 11420 & 8.05 & 892.9 100 & & & & & & & & & & & @xcite + hs0507 + 0435b & idav & 11630 & 8.17 & 743.0 13.9 & 557.7 18.7 & 446.2 11.0 & 444.8 13.6 & 355.8 22.7 & 354.9 6.8 & 286.1 3.6 & & & & & @xcite + hs0507 + 0435b & idav & 11630 & 8.17 & 743.4 8.3 & 588.7 3.9 & 583.8 1.6 & 559.6 2.7 & 557.6 17.4 & 557.2 3.1 & 556.5 7.8 & 555.3 18.0 & 446.1 15.1 & 445.3 3.0 & 444.6 12.9 & @xcite + & & & & 355.8 26.1 & 355.4 4.7 & 354.9 11.1 & & & & & & & & & + gd66 & hdav & 11980 & 8.05 & 649.4 2.2 & 441.9 1.6 & 301.7 6.7 & 271.7 8.4 & 196.5 3.6 & & & & & & & @xcite + gd66 & hdav & 11980 & 8.05 & 788.6 5.5 & 461.0 3.6 & 301.7 6.3 & 271.7 30 & 197.2 2.2 & & & & & & & @xcite + gd66 & hdav & 11980 & 8.05 & 304.5 8.8 & 271.1 14.8 & 256.5 9 & 197.8 7 & 123.1 2.3 & & & & & & & @xcite + he0532 - 5605 & idav & 11560 & 8.49 & 688.8 8.3 & 586.4 7.9 & & & & & & & & & & @xcite + kuv08368 + 4026 & idav & 11490 & 8.05 & 618.0 17.4 & 494.5 6.0 & & & & & & & & & & @xcite + gd99 & cdav & 11820 & 8.08 & 1151.0 1.9 & 1088.0 4.3 & 1058.0 8.3 & 1007.0 6.5 & 976.0 2.1 & 924.7 1.7 & 853.2 2.4 & 633.1 2.0 & 228.9 4.5 & 223.6 2.9 & 105.2 2.0 & chynoweth&thompson2005;prv.comm . + g117-b15a & hdav & 11630 & 7.97 & 304.4 8.2 & 271.0 7.3 & 215.2 23.9 & & & & & & & & & @xcite + kuv11370 + 4222 & hdav & 11890 & 8.06 & 462.9 3.5 & 292.2 2.7 & 257.2 5.8 & & & & & & & & & @xcite + g255 - 2 & cdav & 11440 & 8.17 & 898.5 9.4 & 855.4 11.2 & 819.7 11.3 & 775.2 15.2 & 681.2 24.9 & 607.9 13.1 & 568.5 6.6 & & & & & vauclair2005;prv.comm . + g255 - 2 & cdav & 11440 & 8.17 & 985.2 4.8 & 773.4 12.7 & 681.2 27.7 & 568.5 16.5 & & & & & & & & vauclair2005;prv.comm . + bpm37093 & idav & 11730 & 8.81 & 636.7 1.7 & 633.2 1.1 & 613.5 1.1 & 600.7 0.9 & 582.0 1.0 & 565.5 1.2 & 562.6 0.9 & 548.4 1.1 & & & & @xcite + bpm37093 & idav & 11730 & 8.81 & 660.8 0.5 & 637.2 0.7 & 633.5 1.3 & 565.9 0.5 & 549.2 0.8 & 531.1 1.2 & 511.7 0.7 & & & & & @xcite + he1258 + 0123 & cdav & 11410 & 8.04 & 1092.1 14 & 744.6 23 & 528.5 9 & & & & & & & & & @xcite + gd154 & cdav & 11180 & 8.15 & 1190.5 6.3 & 1186.5 16.7 & 1183.5 4.6 & 1092.1 3.0 & 1088.6 5.0 & 1084.0 5.6 & & & & & & @xcite + lp133 - 144 & hdav & 11800 & 7.87 & 327.3 4 & 306.9 4 & 304.5 4 & 209.2 10 & & & & & & & & @xcite + g238 - 53 & hdav & 11890 & 7.91 & 206.2 8 & & & & & & & & & & & @xcite + ec14012 - 1446 & idav & 11900 & 8.16 & 937 11 & 610 57 & 724 21 & 530 15 & 399 13 & & & & & & & @xcite + gd165 & hdav & 11980 & 8.06 & 249.7 0.7 & 192.8 0.9 & 192.7 2.4 & 192.6 1.9 & 166.2 0.4 & 146.4 0.5 & 120.4 1.8 & 120.36 4.8 & 120.3 1.4 & 114.3 0.6 & 107.7 0.4 & @xcite + l19 - 2 & hdav & 12100 & 8.21 & 350.1 1.1 & 348.7 0.5 & 192.6 6.5 & 193.1 0.9 & 192.1 0.8 & 143.4 0.6 & 143.0 0.3 & 118.9 0.3 & 118.7 1.2 & 118.5 2.0 & & @xcite + & & & & 114.2 0.3 & 113.8 2.4 & 113.3 0.6 & & & & & & & & + pg1541 + 651 & cdav & 11600 & 8.10 & 757 14 & 689 45 & 564 15 & 467 3 & & & & & & & & @xcite + r808 & cdav & 11160 & 8.04 & 833 81 & & & & & & & & & & & @xcite + g226 - 29 & hdav & 12270 & 8.28 & 109.5 2.8 & 109.3 1.1 & 109.1 2.5 & & & & & & & & & @xcite + bpm24754 & cdav & 11070 & 8.03 & 1176 22.6 & & & & & & & & & & & @xcite + bpm24754 & cdav & 11070 & 8.03 & 1050 9.1 & & & & & & & & & & & @xcite + bpm24754 & cdav & 11070 & 8.03 & 1086 13.2 & & & & & & & & & & & @xcite + bpm24754 & cdav & 11070 & 8.03 & 978 7.7 & & & & & & & & & & & @xcite + bpm24754 & cdav & 11070 & 8.03 & 1098 6.1 & & & & & & & & & & & @xcite + bpm24754 & cdav & 11070 & 8.03 & 1122 6.7 & & & & & & & & & & & @xcite + g207 - 9 & idav & 11950 & 8.35 & 557.4 6.3 & 318.0 6.4 & 292.0 5.0 & 259.1 1.7 & & & & & & & & @xcite + g185 - 32 & hdav & 12130 & 8.05 & 651.7 0.7 & 537.6 0.6 & 454.6 0.4 & 370.2 1.6 & 301.4 1.1 & 299.8 1.0 & 264.2 0.5 & 215.7 1.9 & 141.9 1.4 & 72.9 0.4 & & @xcite + g185 - 32 & hdav & 12130 & 8.05 & 370.2 1.3 & 301.6 1.5 & 215.7 1.9 & 141.9 1.5 & 72.6 0.7 & & & & & & & @xcite + gd385 & hdav & 11710 & 8.04 & 256.3 10.9 & 256.1 11.4 & & & & & & & & & & @xcite + gd244 & hdav & 11680 & 8.08 & 307.0 14 & 294.6 5.5 & 256.3 5 & 203.3 10.5 & & & & & & & & @xcite + pg2303 + 243 & cdav & 11480 & 8.09 & 900.5 16 & 794.5 56 & 675.4 8 & 623.4 15 & 570.7 8 & & & & & & & @xcite + g29 - 38 & cdav & 11820 & 8.14 & 1015.5 14.5 & 930.9 25.7 & 824.7 20.2 & 677.0 17.6 & 612.9 20.0 & & & & & & & @xcite + g29 - 38 & cdav & 11820 & 8.14 & 934.5 20.5 & 813.8 23.5 & 671.5 23.0 & 623.8 11.8 & & & & & & & & @xcite + g29 - 38 & cdav & 11820 & 8.14 & 859.6 24.6 & 648.7 7.8 & 614.3 31.3 & 498.3 5.8 & 401.3 4.4 & 400.4 7.0 & 399.6 8.8 & 283.9 3.5 & & & & @xcite + g29 - 38 & cdav & 11820 & 8.14 & 915.4 5.9 & 615.1 58.0 & 500.4 8.0 & 474.9 4.8 & 401.3 5.6 & 400.4 5.8 & 399.6 10.0 & 354.9 2.9 & 333.9 2.7 & 283.9 3.5 & 110.1 0.9 & @xcite + g29 - 38 & cdav & 11820 & 8.14 & 770.7 1.5 & 503.5 8.6 & 495.0 11.8 & 401.2 9.7 & 400.5 1.3 & 399.7 4.9 & 283.9 6.4 & 177.1 0.8 & & & & @xcite + g29 - 38 & cdav & 11820 & 8.14 & 809.4 30.1 & 610.3 10.6 & 401.2 11.2 & 399.7 4.5 & 283.9 6.9 & & & & & & & @xcite + g29 - 38 & cdav & 11820 & 8.14 & 894.0 14.0 & 770.8 8.7 & 678.4 9.7 & 612.4 31.6 & 610.7 8.2 & 608.9 8.5 & 551.9 4.4 & 498.3 6.1 & 401.2 6.0 & 399.7 5.7 & 237.0 1.8 & @xcite + & & & & 236.4 1.8 & & & & & & & & + g30 - 20 & cdav & 11070 & 7.95 & 1068 13.8 & & & & & & & & & & & @xcite + ec23487 - 2424 & cdav & 11520 & 8.10 & 993.0 37.7 & 989.3 11.0 & 868.2 12.8 & 804.5 19.3 & & & & & & & & @xcite +
in high - energy scattering , strong interaction plays a significant role both in the high - energy region and in the low - energy region . the scattering of energetic quarks and gluons with radiative corrections occur in the high - energy regime , in which perturbative qcd can describe the effects of the strong interaction . on the other hand , strong interaction is also responsible for pulling the quarks and gluons out of the initial hadrons through the parton distribution functions , and for the hadronization of the energetic scattered partons through the fragmentation functions . these are nonperturbative aspects of the strong interaction in the low - energy regime . furthermore the strong interaction affects all these processes at different energy scales at the same time , therefore the effects of the strong interaction are entangled in a complicated way all through the scattering process from beginning to end . it is important to disentangle the short- and the long - distance effects of the strong interaction to have theoretical predictive power . this is achieved by the factorization theorem in which the scattering cross section is factorized into the hard , collinear and soft parts . the factorization proof in various hard scattering processes has been a long - standing problem in qcd @xcite . recently the advent of the soft - collinear effective theory ( scet ) @xcite facilitates the proof in a straightforward way . scet is formulated such that the collinear and the soft modes are decoupled at the lagrangian level , and the factorization of the scattering cross section into the hard , collinear and soft parts is intuitively transparent . in high - energy processes , final - state hadrons are typically clustered in collimated beams , which are called jets . jets are important tools to exploit information on standard model or beyond . there are many jet definitions depending on the kinematics , or the detector design , etc . @xcite . in this paper we will confine ourselves to the study of the various jet definitions used in @xmath4 annihilation , but it can be extended to @xmath5 scattering or to @xmath6 scattering . the jet definitions are realized by providing the jet algorithms which give rules on how to combine particles in forming a jet . the jet algorithms include the sterman - weinberg algorithm @xcite , the cone - type algorithm @xcite , the jade algorithm @xcite , the @xmath1 algorithm @xcite , the anti-@xmath1 algorithm @xcite , the cambridge / aachen algorithm @xcite to name a few . the basic issue in jet algorithms is how to combine adjacent final - state particles into a jet without ambiguity . the important ingredients of appropriate jet algorithms are such that they should yield infrared safety , and they can be employed conveniently both in theory and experiment . however , the implementation of a convenient jet algorithm in experiment is not always convenient in theory , or vice versa . the factorized form of the dijet cross section in @xmath4 annihilation can be obtained in the framework of scet . the cross section is factorized into the hard , collinear and soft parts . however , the factorization does not hold for any arbitrary jet algorithm if we require that each factorized part be infrared finite . if any factorized part contains infrared divergence for a certain jet algorithm , though the partial sum of the factorized parts is infrared safe , the factorization approach loses its physical meaning . in order to approach the problem about which jet algorithms respect the factorization theorem , we devise a general jet algorithm , motivated by the @xmath1 jet algorithm , in which we introduce a parameter @xmath7 . the original exclusive @xmath1 algorithm is given by @xmath8 where @xmath9 is the center - of - mass energy . for a given pair of partons @xmath10 and @xmath11 , with the energy @xmath12 and @xmath13 respectively , the parameter @xmath14 is constructed with the relative angle @xmath15 . physically this jet algorithm requires that the relative transverse momentum of every pair of final - state partons @xmath10 and @xmath11 be measured . if the smallest value of @xmath14 for a given pair of partons is less than a resolution parameter @xmath16 , they are combined into a jet . this process is repeated until all pairs have @xmath17 . if the relative transverse momentum is replaced by the invariant mass , the jet algorithm becomes the jade algorithm . here we employ the generalized exclusive jet algorithm , which is defined as @xmath18 where the power of the energy in eq . ( [ yij ] ) is replaced by @xmath7 , and the resolution parameter @xmath16 remains dimensionless . as @xmath7 varies , the general jet algorithm includes all the jet algorithms mentioned above for specific values of @xmath7 . for example , the @xmath1 algorithm is obtained for @xmath3 . the jade algorithm is similar to the case @xmath19 . the cambridge / aachen algorithm corresponds to @xmath20 , and the anti-@xmath1 algorithm algorithm in ref . @xcite is proposed for the inclusive jet algorithm . we refer to the exclusive jet algorithm with @xmath21 in eq . ( [ genkt ] ) as the anti-@xmath1 algorithm . ] is given by @xmath21 . we probe all the possible values of @xmath7 and compute the jet and the soft functions to see if they are well defined and free of infrared divergence . the phase space available for the collinear and the soft functions differs for different values of @xmath7 , so does the structure of divergence . the next - to - leading - order contribution of each factorized part will be computed accordingly , in which there are at most two particles in a jet . we compute each contribution using the dimensional regularization with the spacetime dimension @xmath22 regulating both the ultraviolet ( uv ) and the infrared ( ir ) divergences . also the @xmath23 scheme is employed with @xmath24 . the separation of the uv and ir divergences is important to see whether each factorized part is ir finite . so far , it has been proven that the ir divergence cancels in many processes . and , in fact , any physical quantity should be ir safe . in this case , the remaining divergence is of the uv origin . technically , we can put all the scaleless integrals to be zero , and also put all the poles in @xmath25 as the uv poles in @xmath26 . but this is possible only after it is verified that each factorized part is ir finite . we intend to verify which types of the jet algorithms guarantee the ir finiteness of each factorized part , and which do not , using the generalized jet algorithm . as it turns out , the jet algorithm for @xmath2 indeed yields ir - finite jet and soft functions , hence can safely put all the poles as the uv poles . however , for @xmath27 each jet and soft function contains ir divergence , and we can not naively put all the poles as the uv poles , though the jet cross section is ir finite . this verification is possible only when we distinguish the uv and ir divergences , which we explicitly show here at next - to - leading order . the existence of the ir divergence in the factorized part , or the breakdown of the factorization theorem was also considered in ref . @xcite in a different context . in full qcd , only the jet cross section has been computed and the divergence is regarded as the uv divergence . this is justified since the cross section is a physical quantity free of ir divergence . however , the factorization property in high - energy scattering of the jet cross section or any other physical observables such as thrust is important to disentangle the strong interaction and to compute high - order corrections for better accuracy . then it is crucial to see if the factorization works or breaks depending on the jet algorithms . we address this problem by considering the ir finiteness of the jet and the soft functions in various jet algorithms and point out which jet algorithms allow the factorization theorem in the jet cross section . the structure of the paper is as follows : in section [ factor ] , the factorization of the dijet cross section with the jet algorithm is briefly described in the framework of scet . in section [ jetalgo ] , the phase space for the naive collinear , the zero - bin and the soft contributions is analyzed using the generalized jet algorithm . the structure of the phase space shows different behavior for different values of @xmath7 , and we classify them according to the possible divergence structure . in section [ posalp ] , the collinear and soft contributions for @xmath28 are computed . these contributions are considered for various values of @xmath7 , corresponding to the different phase spaces . we verify that the collinear and soft functions are infrared finite for @xmath29 , while they contain infrared divergence for @xmath27 . the case @xmath3 , corresponding to the original @xmath1 algorithm is not well defined in dimensional regularization though it can be inferred that each contribution contains the ir divergence from the consideration of the phase space . in section [ negalp ] , the collinear and soft contributions are computed for negative @xmath7 . in this case , there should be a jet veto needed in calculating the soft function . each contribution is ir finite for any negative @xmath7 , and the generalized jet algorithm with @xmath30 shows similar behavior compared to the cone - type and the sterman - weinberg jet algorithms @xcite . in section [ psdiv ] , we discuss the structure of divergence for various values of @xmath7 . the identification of the uv and ir divergences can be inferred from the structure of the corresponding phase space . it is explained in detail what kind of divergence there will be _ a priori _ by examining the structure of the phase space . in section [ alpha2 ] , we discuss peculiar characteristics of the @xmath1 algorithm with @xmath3 . in dimensional regularization , it is impossible to extract the poles in the jet and the soft functions . furthermore they contain poles in @xmath31 . the behavior of the collinear and soft parts for @xmath3 is explained in detail . in section [ xsec ] , all the contributions are added to compute the dijet cross section . the jet algorithm with @xmath32 yields finite collinear and soft functions . for @xmath33 , the factorization breaks down , that is , each function contains ir divergence . however , if we add all the contributions , the ir divergence and the singularity in @xmath31 cancel and the resultant dijet cross section is ir finite . we also present the resummed results at next - to - leading logarithmic accuracy for the dijet cross sections with their theoretical uncertainty , and compare with the fixed - order results . in section [ conc ] , we summarize the characteristics of the jet and the soft functions in the generalized jet algorithm and give conclusions . in appendix , we collect all the ingredients for the numerical analysis . the factorization theorem of the dijet cross section has been presented in ref . @xcite , and we briefly review the result instead of rederiving it . the factorized dijet cross section is given by @xmath34 here @xmath35 is the invariant - mass squared of the @xmath0 system , and @xmath36 is the born cross section for a given flavor @xmath37 of the quark - antiquark pair with the electric charge @xmath38 , given by @xmath39 @xmath40 is the hard function which is obtained from the matching of the electromagnetic current between the full qcd and scet at leading order as @xmath41 and @xmath42 . to one loop , it is given by @xcite @xmath43 and @xmath44 is a gauge - invariant collinear quark with a collinear wilson line @xmath45 , and @xmath46 is the soft wilson line @xcite @xmath47_,\ ] ] where ` p ' denotes the path ordering . after redefining the collinear fields , the soft interaction is decoupled from the collinear interaction @xcite , and the factorized jet cross section in eq . ( [ facjet ] ) is obtained . the integrated jet function @xmath48 is defined with the jet algorithm as @xmath49 where @xmath50 is the invariant mass squared of the collinear jet . the unintegrated jet function @xmath51 is defined as @xmath52 where @xmath53 specifies the jet algorithm . the soft function @xmath54 with the jet algorithm is given by @xmath55 the jet function @xmath56 and the soft function @xmath54 are normalized to 1 at tree level , and we compute them at next - to - leading order with various values of @xmath7 in the generalized @xmath1 algorithm . we now consider the constraints imposed on the phase space by the jet algorithm in eq . ( [ genkt ] ) . at next - to - leading order ( nlo ) , there are at most two particles in a jet . the dominant contribution for the dijet in @xmath0 annihilation is the quark - antiquark pair forming a back - to - back jet and a gluon is emitted . we choose the jet in the lightlike @xmath57 direction , or in the @xmath58 direction with @xmath59 , @xmath60 and @xmath61 . in applying the jet algorithm , the virtual correction and the real gluon emission are included . these can be obtained by cutting the diagram in fig . [ conf ] , which is part of the matrix elements squared for the jet cross section . if we cut a single line , it corresponds to the virtual correction . when we cut the loop , there are two particles in the final state with momenta @xmath62 ( for a gluon ) and @xmath63 ( for a quark ) in the @xmath57 direction . the following nontrivial consideration of the phase space using the jet algorithm applies to this case with a real gluon emission . , width=188 ] the total momentum @xmath64 of the final state scales as @xmath65 where @xmath66 is the small parameter in scet . we choose the @xmath57 direction as the direction of @xmath67 such that @xmath68 , and @xmath69 . the collinear gluon momentum @xmath70 scales as @xmath71 with this power counting , the collinear momenta of the quark and the gluon can be written as @xmath72 with their energies @xmath73 and the invariant - mass squared @xmath50 is given by @xmath74 the phase space constraint from eq . ( [ genkt ] ) for the jet function changes with the sign of @xmath7 . in terms of the gluon momentum @xmath62 and @xmath9 , the jet algorithm for @xmath75 is given as @xmath76 the nontrivial jet algorithm appears when the resolution parameter @xmath16 is of order @xmath77 in scet . hence we choose @xmath78 from now on . algorithm with @xmath28 . the first column shows the @xmath79 phase space , while the second column shows the @xmath80 space . ( a ) @xmath81 , ( b ) @xmath19 , ( c ) @xmath82 , ( d ) @xmath83 , and ( e ) @xmath27 . the functions @xmath84 and @xmath85 describe the boundaries of the phase space for @xmath86 and @xmath87 respectively , as defined in eq . ( [ f12]).[phplus],width=604 ] by letting @xmath88 , @xmath89 , the phase space is constrained by , or infinity . for the momentum reaching infinity , the dimensionful coefficient is the renormalization scale @xmath90 which is to be combined with the integral in applying the dimensional regularization . see the zero - bin contribution or the soft function below . ] @xmath91 solving for @xmath92 , it becomes @xmath93 where the boundaries are defined in terms of @xmath94 and @xmath95 respectively . the shape of the phase space in the @xmath96 space with @xmath28 changes drastically for the following five cases : ( i ) @xmath97 , ( ii ) @xmath19 , ( iii ) @xmath82 , ( iv ) @xmath3 , ( v ) @xmath27 . the corresponding phase spaces are shown in fig . [ phplus ] . the figures in the first column are the phase spaces in the @xmath79 space , and those in the second column are those in the @xmath80 space . the structure of the divergence can be inferred from the phase space in the @xmath79 space . but when it becomes ambiguous , the phase space in the @xmath80 space plays a complementary role in deducing the origin of the divergence @xcite . this point will be discussed in detail in section [ psdiv ] after the explicit result is presented . for @xmath97 , @xmath98 approaches zero at the endpoints @xmath99 and @xmath9 . at @xmath19 , @xmath98 becomes a nonzero constant at @xmath100 , while it approaches zero at @xmath101 . for @xmath102 , @xmath98 diverges as @xmath103 approaches zero , while it becomes zero at @xmath104 . for @xmath3 , @xmath98 diverges at @xmath100 , while remains finite at @xmath104 . for @xmath105 , @xmath98 diverges at both endpoints @xmath100 and @xmath9 . the different behavior of the phase space near the endpoints gives a clue to whether the naive contribution to the jet function contains uv or ir divergence . in the first column of fig . [ phplus ] , fig . [ phplus ] ( a ) and ( b ) corresponding to @xmath106 , @xmath103 and @xmath98 never reach infinity , and we expect that the naive collinear contribution has ir divergence only . in fig . [ phplus ] ( c ) , ( d ) and ( e ) corresponding to @xmath107 , @xmath98 can reach infinity and we can expect that there may be uv divergence as well as ir divergence when the integration over @xmath103 and @xmath98 is performed . however , this is not exactly true . in most cases the uv and ir divergences can be expected by looking at the @xmath79 phase space alone . but there are some occasions where it is ambiguous . then we turn to the @xmath80 phase space . for example , if we consider the final form of the integral in computing the collinear part with respect to @xmath103 and @xmath108 , the second figure in fig . [ phplus ] ( c ) shows that @xmath108 never reaches infinity , hence there should be only ir divergence for @xmath82 . for @xmath27 , both figures in fig . [ phplus ] ( e ) indicate that there are both ir and uv divergences . in the second figure in fig . [ phplus ] ( d ) , we expect that there is only ir divergence , but the collinear and the soft functions are not well - defined for @xmath3 , and it needs more explanation , as we will see below . in the same format as fig . [ phplus ] . ( a ) @xmath81 , ( b ) @xmath19 , ( c ) @xmath82 , ( d ) @xmath83 , and ( e ) @xmath27 . [ zeroplus],width=415 ] in order to avoid double counting , we subtract the soft limit from the naive collinear contribution , which is referred to as the zero - bin subtraction @xcite . for the zero - bin contribution , we take the soft limit of the gluon @xmath109 and rewrite the first equation in eq . ( [ f12 ] ) . then the phase space for the zero - bin contribution is given as @xmath110 which can be written , by putting @xmath111 , @xmath112 , as @xmath113 the phase space for the zero - bin contribution corresponding to each case in fig . [ phplus ] is shown in fig . [ zeroplus ] . as can be seen clearly in fig . [ zeroplus ] , the zero - bin contribution contains both uv and ir divergences . however , the ir divergence is cancelled in the sum of the naive collinear contribution and the zero - bin contribution for @xmath32 , while the cancellation is incomplete and contains ir divergence for @xmath27 . for the soft function , @xmath14 satisfies the following relations with the appropriate power counting . @xmath114 by replacing @xmath111 , @xmath89 , the boundary from the @xmath58 jet is given by @xmath115 where @xmath116 is a dimensionless parameter given as @xmath117 the boundary from the @xmath57 jet is given by @xmath118 . the phase space for the soft function is shown in fig . [ softplus ] . as can be seen in fig . [ softplus ] , since the region near the origin is totally covered , no jet veto is necessary for @xmath75 . however , the jet veto is needed for @xmath119 , as in the sterman - weinberg jet algorithm @xcite . corresponding to fig . [ phplus ] . the jet veto is not necessary for @xmath28 , and @xmath120 , ( a ) @xmath81 , ( b ) @xmath19 , ( c ) @xmath82 , ( d ) @xmath83 , and ( e ) @xmath27 . [ softplus],width=415 ] for @xmath119 , the minimum in eq . ( [ genkt ] ) is switched compared to the case @xmath28 . it means that the boundary functions are also reversed . that is , the constraint of the phase space for the jet function is given by @xmath121 the phase space for the zero - bin contribution is given by @xmath122 and that for the soft function is given by @xmath123 in contrast to the case with @xmath28 , the jet veto is needed here since the phase space near the origin is not completely covered by the jet algorithm alone . and we take @xmath124 for definite power counting . the phase space for the naive collinear part , the zero - bin contribution , and the soft function for @xmath30 is shown in fig . [ cominus ] . the shapes of the phase spaces are similar to the sterman - weinberg algorithm or the cone - type algorithm @xcite . and we expect that the jet and the soft functions are ir finite for all negative values of @xmath7 . algorithm with @xmath30 . ( a ) naive collinear contribution ( b ) the zero - bin contribution ( c ) the soft function.[cominus],width=604 ] the feynman diagrams for the jet function in the @xmath57 direction , are shown in fig . [ intjet ] . the dashed line is the cut , therefore fig . [ intjet ] ( a ) is the virtual correction , while fig . [ intjet ] ( b ) and ( c ) are the real contributions . the mirror images of ( a ) and ( b ) are omitted in fig . [ intjet ] , but they are included in the computation . each diagram is accompanied by the corresponding zero - bin contribution , which should be subtracted to obtain the collinear jet function . , width=604 ] the dimensional regularization states that @xmath125 where @xmath62 is a momentum variable . the integral vanishes identically if we do not distinguish the uv and ir poles since the integral is a scaleless integral . however , we distinguish the uv and ir poles since we are interested in identifying the sources of the divergence . the virtual correction in fig . [ intjet ] ( a ) is independent of the jet algorithms and the net collinear contribution @xmath126 is given by the naive collinear contribution @xmath127 , subtracted by the zero - bin contribution @xmath128 : @xmath129 the naive collinear contribution for @xmath81 in fig . [ intjet ] ( b ) is given as @xmath131\nonumber \\ & = & \frac{\alpha_s c_f}{2\pi } \bigl [ \frac{1}{\alpha-2}\bigl ( -\frac{1}{{\epsilon_{\mathrm{ir}}}^2 } -\frac{1}{{\epsilon_{\mathrm{ir } } } } \ln \frac{\mu^2}{2^{\alpha-2 } y_c q^2 } - \frac{1}{2}\ln^2 \frac{\mu^2}{2^{\alpha-2 } y_c q^2 } + \frac{\pi^2}{12}\bigr ) \nonumber \\ & & + \frac{1}{{\epsilon_{\mathrm{ir } } } } + \ln \frac{\mu^2}{2^{\alpha-2 } y_c q^2 } + ( 4-\alpha ) \bigl(1-\frac{\pi^2}{12}\bigr ) -\alpha \ln 2 \bigr].\end{aligned}\ ] ] this explicit calculation supports the expectation from the phase space in fig . [ phplus ] ( a ) that the poles are of the ir origin . this can be also seen by the fact that the integral of @xmath132 in eq . ( [ naimb ] ) exists only when @xmath133 . the corresponding zero - bin contribution is given by @xmath134 where @xmath135 is the zero - bin limit of @xmath94 and is given in eq . ( [ f10 ] ) . from fig . [ zeroplus ] , since @xmath103 , @xmath98 can both approach zero and infinity , the zero - bin contribution contains both uv and ir divergences . in order to separate the uv and ir contributions , the integral is decomposed to be written as @xmath136 \nonumber \\ & = & \frac{\alpha_s c_f}{2\pi } \bigl(\frac{1}{{\epsilon_{\mathrm{uv } } } } -\frac{1}{{\epsilon_{\mathrm{ir}}}}\bigr ) \bigl [ \frac{1}{\alpha-2 } \bigl ( \frac{1}{{\epsilon_{\mathrm{uv } } } } + \frac{1}{{\epsilon_{\mathrm{ir } } } } + \ln \frac{\mu^2}{2^{\alpha-2 } y_c q^2 } \bigr ) + \frac{1}{{\epsilon_{\mathrm{uv } } } } + \ln \frac{\mu}{q } \bigr],\end{aligned}\ ] ] where @xmath137 is a small positive number to separate the ir and uv regions . the final result is independent of @xmath137 . this decomposition looks complicated , but it is arranged in such a way that the pole from the integral including the infinity ( zero ) is of the uv ( ir ) origin with @xmath138 ( @xmath133 ) @xcite . and the net collinear contribution from fig . [ intjet ] ( b ) is given by @xmath139 . \nonumber \end{aligned}\ ] ] the naive collinear contribution from fig . [ intjet ] ( c ) is given by @xmath140 \nonumber \\ & = & \frac{\alpha_s c_f}{2\pi } \bigl ( -\frac{1}{2{\epsilon_{\mathrm{ir } } } } -\frac{1}{2 } \ln \frac{\mu^2}{2^{\alpha-2 } y_c q^2 } + \frac{\alpha-3}{2 } + \frac{\alpha}{2}\ln 2 \bigr),\end{aligned}\ ] ] while the zero - bin contribution is subleading and suppressed . adding all the terms with the wavefunction renormalization and the residue at one loop @xmath141 the collinear contribution for @xmath130 at order @xmath142 is ir finite and is given as @xmath143 . \end{aligned}\ ] ] by removing the uv poles , the jet function for @xmath81 at order @xmath142 is given by @xmath144.\end{aligned}\ ] ] note that the phase space for @xmath19 ( naive and zero - bin ) allows the same method of integration with the same specification of the poles as in the case @xmath97 . the only difference is that @xmath98 approaches a nonzero , finite value as @xmath103 approaches zero in the phase space for the naive collinear contribution , hence the case @xmath145 can be included in eq . ( [ mb1 ] ) . the case @xmath19 is very similar to the jade jet algorithm as far as the phase space is concerned . in the jade algorithm , the invariant mass @xmath146 of every pair of the final - state partons @xmath10 and @xmath147 should be less than @xmath148 , where @xmath11 is some fraction ( of order @xmath77 in the context of scet ) to form a jet . at next - to - leading order , the jade algorithm is written as @xmath149 if we compare this phase space with that of fig . [ phplus ] ( b ) , [ zeroplus ] ( b ) and [ softplus ] ( b ) for @xmath19 , the basic structure near the endpoints of the phase space is the same , and the only difference lies in the middle region of fig . [ phplus ] ( b ) . therefore we expect that the divergence structure for the case @xmath19 and the jade algorithm is the same except the finite part in the dijet cross section . in fact , performing the calculation using the jade algorithm , the collinear part is given as @xmath150,\ ] ] which is consistent with the result in ref . @xcite . and the corresponding collinear part for @xmath145 from eq . ( [ mb1 ] ) is given by @xmath151.\ ] ] if we put @xmath152 , all the uv divergence and the logarithmic terms are reproduced except the finite terms , which is why we say that the generalized jet algorithm with @xmath19 is basically identical to the jade algorithm . from eqs . ( [ jadejet ] ) and ( [ jetalpha1 ] ) , the jet functions in the jade algorithm and in the case with @xmath19 are given as @xmath153 by looking at the phase space for the naive collinear contribution in fig . [ phplus ] ( c ) in the @xmath79 space , one might naively expect that there could be a mixed uv and ir divergence since @xmath98 reaches infinity as @xmath103 approaches zero . however it becomes clear when we look at the phase space in the @xmath80 space . it means that , when we perform the integral over @xmath62 , the @xmath98 is first performed and the remaining integral is to be performed with respect to @xmath103 and @xmath108 @xcite . then the divergent behavior is determined by @xmath155 . if the divergence occurs when @xmath156 , it is the uv divergence . on the other hand , if the divergence occurs when @xmath157 , it is the ir divergence . the constraint for @xmath108 from the jet algorithm is given by @xmath158 which approaches zero ( @xmath82 ) as @xmath159 . it can be also seen in fig . [ phplus ] ( c ) with the plot in the @xmath160 space , @xmath108 never approaches infinity . therefore the poles should be of the ir origin . the collinear contributions computed for the case @xmath161 can be exactly performed with the same prescription on the signs of @xmath25 , hence giving the same result for @xmath82 . it also turns out that @xmath162 therefore eq . ( [ mb1 ] ) can be extended to the case @xmath154 , and it is continuous at @xmath19 . the situation is completely different for @xmath27 . in the naive collinear and the zero - bin contributions , the phase space shows that @xmath108 can reach zero or infinity . therefore we expect that the collinear contributions contain both uv and ir poles , which makes the factorization obsolete because the jet function and the soft function are not ir finite . however , as will be seen later , the sum of the jet functions and the soft function is ir finite . in the real gluon emission , the expression for @xmath132 is given by @xmath163 . \end{aligned}\ ] ] but the function @xmath164 describing the boundary of the phase space for @xmath86 diverges as @xmath103 approaches zero by noting that the function @xmath94 is proportional to @xmath165 . it requires the modification of the first integral in eq . ( [ nmb2 ] ) as @xmath166 this manipulation is necessary since the remaining integral converges for @xmath138 , hence the integral over @xmath92 should produce @xmath26 . the boundary of the phase space , @xmath167 for @xmath168 is proportional to @xmath169 , and it also diverges as @xmath103 approaches @xmath9 ( @xmath170 ) . however , the integrand is proportional to @xmath171 , and there is no uv divergence involved . the naive collinear contribution is given as @xmath172.\end{aligned}\ ] ] in the zero - bin contribution , a similar modification of the integral is necessary . it is given as @xmath173 \nonumber \\ & = & \frac{\alpha_s c_f}{2\pi } \bigl [ \frac{-1}{\alpha-2 } \bigl ( \frac{1}{{\epsilon_{\mathrm{uv } } } } -\frac{1}{{\epsilon_{\mathrm{ir}}}}\bigr ) \bigl(\frac{1}{{\epsilon_{\mathrm{uv } } } } + \frac{1}{{\epsilon_{\mathrm{ir } } } } + \ln \frac{\mu^2}{2^{\alpha-2 } y_c q^2 } \bigr ) + \frac{1}{{\epsilon_{\mathrm{ir}}}^2 } -\frac{1}{{\epsilon_{\mathrm{uv}}}{\epsilon_{\mathrm{ir } } } } \nonumber \\ & & - \bigl ( \frac{1}{{\epsilon_{\mathrm{uv } } } } -\frac{1}{{\epsilon_{\mathrm{ir}}}}\bigr ) \ln \frac{\mu}{q } \bigr].\end{aligned}\ ] ] the net collinear contribution is given by @xmath174 . \end{aligned}\ ] ] fig . [ intjet ] ( c ) yields the same results as in the case @xmath175 . summing all the contributions , the net collinear part is given as @xmath176.\end{aligned}\ ] ] in this case , the collinear part is both uv and ir divergent . therefore it does not have any physical meaning for @xmath177 . compared with the collinear part for @xmath175 given in eq . ( [ mb1 ] ) , if we put @xmath178 , the result is the same . however , we stress that this is misleading because the collinear part for @xmath179 is ir finite and there are only uv poles , while that for @xmath27 contains both uv and ir divergences . in this case , simply putting @xmath178 ruins the physical implication on the divergences . but , as we will see later , the sum of the ir divergent collinear part and the soft part , which is also ir divergent , is ir finite . therefore the combined collinear and soft part is physically well - defined , but the factorization is not realized in a strict sense . the jet algorithm with @xmath3 corresponds to the exclusive @xmath1 algorithm . a serious trouble in this case is that the jet function has a pole at @xmath3 and is not well - defined . this is to be contrasted with the case @xmath19 . for @xmath19 , the structure of the phase space remains continuous for @xmath81 and @xmath102 , hence the same divergence structure follows . however , for @xmath3 , the shape of the phase space abruptly changes from @xmath81 to @xmath27 . the structure of the divergence also abruptly changes accordingly . there is only ir divergence for @xmath81 , while there are uv , ir divergences for @xmath27 . therefore the jet function for @xmath3 can not be obtained by taking the limiting case as @xmath180 from either side . this is manifested by the existence of the pole in @xmath31 . however , as we will see later , the dijet cross section in the @xmath1 algorithm is finite , but we emphasize that the factorization breaks down for @xmath3 . since the behavior of the jet function , as well as the soft function , is peculiar , it is worth mentioning what happens at @xmath3 in detail later . , width=604 ] the feynman diagrams for the soft function at one loop is shown in fig . [ feynsoft ] , in which the hermitian conjugate is omitted . the phase space for the soft function is shown in fig . [ softplus ] . the prescription for the uv and ir divergences is the same as in the case of the jet function according to the possible values @xmath108 . the virtual correction in fig . [ feynsoft ] ( a ) is given as @xmath181 which is independent of the jet algorithm . we compute the real gluon emission from fig . [ feynsoft ] ( b ) for various values of @xmath28 . the structure of the phase space for the soft function is shown in fig . [ softplus ] ( a ) , and it is apparent that both the uv and ir divergences may exist . the contribution of the real gluon emission is given by @xmath182,\ ] ] where @xmath183 is the point where two boundaries meet , and @xmath184 describes the boundary for the @xmath58 jet , and is given in eq . ( [ softb ] ) . we divide the phase space symmetrically with respect to the line @xmath185 , and compute only the upper half , hence the factor 2 is multiplied . by carefully separating the uv and ir divergences , the amplitude for the soft function , including the hermitian conjuagates , is given by @xmath186 , \nonumber\end{aligned}\ ] ] where @xmath187 . the soft function at one loop for @xmath188 is given as @xmath189.\ ] ] the ir divergence from the virtual correction is cancelled by that in the real gluon emission , and there remains only uv divergence . this fact can be seen transparently from the argument of the phase space @xcite . the virtual contribution corresponds to the same integral as for the real gluon emission , but the integration region covers all the phase space , with a minus sign . therefore the total soft contribution is given by the integral over the unshaded phase space in fig . [ softplus ] with a minus sign . since the unshaded region never touches the ir region , the soft contribution contains only uv divergence . the calculation procedure is the same for @xmath19 , hence the result can be extended to the case @xmath19 . the soft part at @xmath19 is given as @xmath190 while the soft part in the jade algorithm is given by @xmath191 which is identical to the soft part with @xmath19 by replacing @xmath152 . this should be true because the jet algorithm for @xmath19 in the soft part is identical to the jade algorithm with @xmath152 . the soft functions in these two algorithms at next - to - leading order are given as @xmath192 the contribution from the real gluon emission is given as @xmath193 it turns out that the result is the same as that in the case with @xmath161 . the left - hand - side plot in fig . [ softplus ] ( e ) , @xmath98 approaches infinity as @xmath103 goes to zero . therefore we may expect that there are both uv and ir divergences . however , the plot of @xmath108 with respect to @xmath103 in the right - hand side shows that @xmath108 can never reach infinity , hence the real gluon emission involves only the ir divergence . in order to handle the divergence correctly , we write @xmath194 as @xmath195 \nonumber \\ & = & \frac{\alpha_s c_f}{2\pi } \bigl [ \frac{1}{\alpha-2 } \bigl ( \frac{2}{{\epsilon_{\mathrm{ir}}}^2 } -\frac{2}{{\epsilon_{\mathrm{ir } } } } \ln a + \ln^2 a -\frac{\pi^2}{6}\bigr ) + \frac{1}{{\epsilon_{\mathrm{ir}}}^2 } -\frac{1}{\alpha}\ln^2 a + \frac{\pi^2}{2 } \bigl ( \frac{1}{3\alpha } -\frac{1}{2}\bigr ) \bigr].\end{aligned}\ ] ] the amplitude for the soft function is given by @xmath196 . \end{aligned}\ ] ] note the difference of the soft amplitudes between the cases with @xmath102 and @xmath27 . if we naively put @xmath178 , the result is the same . however , if we distinguish the uv and ir divergences , the meaning of the results is totally different . the case with @xmath102 contains only the uv poles , while the case with @xmath27 is not ir finite . therefore , when @xmath27 , the soft function is not physically meaningful . however , if we add the soft amplitude with the corresponding collinear amplitude , eq . ( [ co2l ] ) , the ir divergence cancels out . it means that the dijet cross section can be computed with ir finiteness , but the factorization breaks down for @xmath27 . one more comment is that the results are also divergent at @xmath3 . the soft function for @xmath3 is not well defined in dimensional regularization . however , if we add the soft amplitude with the corresponding collinear amplitude , the divergent term in @xmath7 , proportional to @xmath31 also cancels out . this will be discussed in detail in section [ alpha2 ] . the generalized @xmath1 algorithm for @xmath197 is similar to the generic cone type jet algorithm or the sterman - weinberg jet algorithm as far as the shape of the phase space is concerned @xcite . the phase space for the naive collinear contribution , the zero - bin contribution and the soft function is shown in fig . [ cominus ] . we can extend the jet algorithm to @xmath20 since the phase space for @xmath20 is obtained by taking the limit @xmath198 from the case with @xmath119 because a jet veto is needed . the jet algorithm with @xmath20 is known as the cambridge / aachen jet algorithm . the naive collinear contribution from fig . [ intjet ] ( b ) is given as @xmath199 \nonumber \\ & = & \frac{\alpha_s c_f}{2\pi } \bigl [ \frac{1}{2{\epsilon_{\mathrm{ir}}}^2 } + \frac{1}{{\epsilon_{\mathrm{ir } } } } \bigl ( 1 + \frac{1}{2 } \ln \frac{\mu^2}{2^{\alpha -2 } y_c q^2 } \bigr ) + \ln \frac{\mu^2}{2^{\alpha -2 } y_c q^2 } + \frac{1}{4 } \ln^2 \frac{\mu^2}{2^{\alpha -2 } y_c q^2 } \nonumber \\ & & + 2 + 2\ln 2 -\frac{5\pi^2}{24 } + ( \alpha-2 ) \bigl(-1+\ln 2 + \frac{\pi^2}{12}\bigr)\bigr].\end{aligned}\ ] ] for the zero - bin contribution , the boundary for the phase space is obtained by taking the zero - bin limit of @xmath167 , which is given by @xmath200 the zero - bin contribution is given as @xmath201 \nonumber \\ & = & \frac{\alpha_s c_f}{2\pi } \bigl [ \frac{1}{2}\bigl ( \frac{1}{{\epsilon_{\mathrm{uv } } } } -\frac{1}{{\epsilon_{\mathrm{ir}}}}\bigr)^2 + \frac{1}{2 } \bigl ( \frac{1}{{\epsilon_{\mathrm{uv } } } } -\frac{1}{{\epsilon_{\mathrm{ir}}}}\bigr ) \ln 2^{\alpha-2 } y_c\bigr].\end{aligned}\ ] ] from the real gluon emission in fig . [ intjet ] ( c ) , the naive collinear contribution is given by @xmath202\nonumber \\ & = & \frac{\alpha_s c_f}{2\pi}\bigl ( -\frac{1}{2{\epsilon_{\mathrm{ir } } } } -\frac{1}{2 } \ln \frac{\mu^2}{2^{\alpha-2 } y_c q^2}+\frac{\alpha -3}{2 } -\frac{\alpha}{2}\ln 2\bigr).\end{aligned}\ ] ] compared to the corresponding calculation for @xmath28 in eq . ( [ mcal+ ] ) , the only difference is in the coefficient of @xmath203 . combining the wave function renormalization and the residue , the collinear amplitude for @xmath197 is given as @xmath204.\end{aligned}\ ] ] and the jet function at one loop for @xmath197 is given as @xmath205.\end{aligned}\ ] ] the contribution from the real gluon emission is given by @xmath206.\end{aligned}\ ] ] with the virtual correction , the amplitude for the soft function is written as @xmath207 and the soft function at one loop for @xmath197 is given by @xmath208 in the limit @xmath209 , which corresponds to the cambridge / aachen jet algorithm , the soft function is given as @xmath210 note that the limit @xmath198 should be taken from @xmath119 . for @xmath28 , there is a pole in @xmath211 , and it may be dangerous to take the limit @xmath209 . however , when the limit @xmath209 is taken , it coincides with the phase space for @xmath119 and the jet veto is needed . therefore the soft function at @xmath20 should be taken from the soft part with @xmath30 . the structure of divergence can be inferred from the phase space argument @xcite . since the phase spaces for different @xmath7 yield different shapes of the phase space , we will discuss and compare the cases for various values of @xmath7 . let us first consider the phase space for the naive collinear contribution in fig . [ phplus ] . in the first column , the phase space is plotted in @xmath79 space , and the second column in @xmath80 space . when we compute the naive collinear contribution , there is an integration with respect to the gluon momentum @xmath62 . if we first integrate over @xmath212 , which takes care of the delta function in the integrand , the remaining integral is to be performed with respect to @xmath103 and @xmath98 , where the phase space in the first column is relevant . if we first integrate over @xmath98 , the remaining integral is to be performed with respect to @xmath103 and @xmath212 , where the phase space in the second column is appropriate . in @xmath79 space , the uv ( ir ) divergence comes from the region where @xmath103 and @xmath98 approaches infinity ( zero ) . if one of them approaches infinity , while the other approaches zero , it is possible that there is a mixed uv - ir divergence . we can infer the structure of divergence in @xmath79 space from the first column of fig . [ phplus ] . for @xmath106 , it is clear that , if there is divergence , it should be of the ir origin from fig . [ phplus ] ( a ) and ( b ) since @xmath103 and @xmath98 can never reach infinity . it is explicitly verified in eqs . ( [ naimb ] ) , ( [ mcal+ ] ) . this is also evident from the second column of the corresponding phase spaces since @xmath103 and @xmath108 never reach infinity . with the same argument , when @xmath27 in fig . [ phplus ] ( e ) , both plots imply that there are uv , ir and mixed divergences . however , the two plots lead to seemingly inconsistent results in fig . [ phplus ] ( c ) , which corresponds to @xmath213 . in @xmath79 space , @xmath98 can reach infinity while @xmath103 can not . therefore we naively expect that there might be ir and mixed divergences . on the other hand , in @xmath80 space , @xmath108 never reaches infinity and there should be only ir divergence . in this case , we have to rely on the phase space in @xmath80 space because there is an additional relation @xmath214 for the real gluon emission . from the phase space constraint in eq . ( [ cocon ] ) , it reads for @xmath215 @xmath216 and indeed @xmath98 can approach infinity as @xmath103 approaches zero for @xmath102 . however , @xmath108 is bounded by @xmath217 which can not reach infinity as @xmath103 approaches zero . since the divergence behavior is determined by the behavior in @xmath108 in @xmath218 dimensions , and @xmath219 , the inference on the divergence behavior is more reliable in @xmath80 space rather than in @xmath79 space . this can also be confirmed in ref . @xcite , in which the jet function has been computed with the transverse momentum fixed . since the transverse momentum is fixed at a finite value , there is no uv divergence , but ir divergence . the computation was performed for the inclusive jet function with the transverse momentum fixed , hence there is no constraint on the phase space as in the jet algorithm . however , the result supports that there is only ir divergence when @xmath108 never reaches infinity . as can be seen in fig . [ phplus ] ( a ) to ( c ) , the phase spaces for @xmath102 in @xmath80 space shows similar shapes and there is only ir divergence . for the phase space of the zero - bin contribution in fig . [ zeroplus ] , all the plots show that the zero - bin contribution should contain both uv and ir divergences . for the soft contribution shown in fig . [ softplus ] , we can conclude that there are both uv and ir divergences for @xmath175 , while there is only ir divergence for @xmath33 , as can be seen in eq . ( [ sbg2 ] ) . however , we can apply the phase space method as in ref . the virtual correction for the soft part has the same integral as in the real gluon emission except the minus sign , and it covers the entire phase space . therefore the soft part is obtained by integrating the matrix element over the unshaded phase space in fig . [ softplus ] with a minus sign . then the soft part for @xmath179 has only uv divergence since the unshaded region never reaches zero . it means that the cancellation of the ir divergence occurs between the virtual and real contributions . on the other hand , for @xmath33 , we expect that there is not only uv divergence , but also ir divergence , which breaks the factorization . the case with @xmath3 ( the @xmath1 algorithm ) is tricky because it contains the pole in @xmath31 , and it can not be computed in dimensional regularization . however , from the phase space diagrams , we expect that the naive collinear contribution contains only ir divergence , while the soft contribution contains both uv and ir divergences . this will be discussed in detail in the next section . we have considered the jet and the soft functions in the generalized jet algorithm by varying @xmath7 . it turns out that the jet algorithm for @xmath2 respects the factorization , that is , the jet and the soft functions are ir finite , and it breaks the factorization for @xmath33 since each part contains ir divergence . the case with @xmath3 shows a peculiar behavior and it is worth mentioning a few characteristics . at @xmath3 , in the @xmath1 algorithm , there is an uncontrollable integral of the form @xmath220 in the naive collinear , the zero - bin and the soft contributions , even though the spacetime dimension is extended to @xmath221 . it is also noted in ref . @xcite . if we use the offshellness to regulate the ir divergence in the soft wilson line by replacing the propagator , the soft part at @xmath3 is given by @xmath222 where @xmath223 represent the offshellness of the quark and the antiquark . and this is consistent with the result in ref . the result clearly shows that the soft part contains the uv and ir divergences . if we look at the plots for the phase space in figs . [ phplus ] ( d ) , [ zeroplus ] ( d ) and [ softplus ] ( d ) , it can be inferred that the naive collinear contribution contains only ir divergence , the zero - bin and the soft contribution have both uv and ir divergences . in spite of the possible argument on the existence of the uv and ir divergence in the jet and the soft functions for @xmath3 , we interpret the result differently in the sense that they are not well defined for @xmath3 . if we consider the jet and soft functions as a function of @xmath7 in the generalized @xmath1 algorithm , they are not continuous at @xmath3 in the sense that @xmath224 when we look at the plots for the phase space , for example , fig . [ phplus ] , the phase space allows finite @xmath108 only for @xmath32 , but @xmath108 can reach infinity for @xmath27 . therefore the divergence structure abruptly changes at @xmath3 . this discontinuity appears as a pole in @xmath225 in our treatment of the generalized jet algorithm as a function of @xmath7 . in contrast , the behavior of the jet function and the soft function near @xmath19 is smooth . the shape of the phase space in the @xmath79 space changes near @xmath19 , that is , @xmath98 approaches a finite value instead of zero near @xmath100 . however , in the @xmath160 space , there is no abrupt change in shape . therefore the jet algorithm is continuous at @xmath19 , hence the factorization works in the jade algorithm , too . interestingly enough , the pole in @xmath225 as well as the ir divergence cancels in the sum of the collinear part and the soft part to yield the jet cross section . therefore the jet cross section is ir finite even at @xmath83 . we may conclude that the @xmath1 algorithm works fine , but it is misleading . the main point is that the factorization , which lies behind the jet algorithm , breaks down for @xmath33 including the @xmath1 algorithm . however , note that the claim that the @xmath1 algorithm breaks factorization may be too strong because the separation of the uv and ir divergence is spoiled by the existence of the pole in @xmath225 . in dimensional regularization , the extraction of the divergence is impossible at @xmath3 , but there may be other ways to obtain the divergence unambiguously . for example , in ref . @xcite , the authors have tried to regulate this integral with the rapidity renormalization technique . it may turn out to be useful , but we will not pursue in this direction here , and our claim stands as it is as far as we use the dimensional regularization to extract uv and ir divergences . let us collect all the results for the resummation of the large logarithms of the form @xmath226 with @xmath227 through the renormalization group behavior . at one loop , the hard function is given by @xmath228 the jet function and the soft function for @xmath179 at one loop are given as @xmath229 , \nonumber \\ \mathcal{s}^{(1 ) } ( q , y_c,\mu ) & = & \frac{\alpha_s c_f}{2\pi } \bigl[\frac{4}{\alpha(\alpha-2 ) } \ln^2 2^{\alpha-2 } y_c \bigl(\frac{q}{\mu}\bigr)^{\alpha } -\pi^2 \bigl ( \frac{2}{3\alpha(\alpha-2 ) } + \frac{1}{2 } \bigr ) \bigr ] . \end{aligned}\ ] ] the jet function @xmath230 in the @xmath58 direction is the same as @xmath231 . the dijet cross section in eq . ( [ facjet ] ) at next - to - leading order ( nlo ) for @xmath29 is written as @xmath232.\ ] ] for example , the dijet cross section for @xmath19 , which is close to the jade algorithm , is given by @xmath233 while the cross section exactly in the jade algorithm is given by @xmath234 this agrees with the full qcd calculation @xcite . for @xmath119 , the jet function and the soft function at one loop are given as @xmath235 , \nonumber \\ \mathcal{s}^{(1 ) } ( q , y_c , \beta , \mu ) & = & \frac{\alpha_s c_f}{2\pi } \bigl ( 4\ln \frac{\mu}{2\beta q } \ln ( 2^{\alpha-2 } y_c ) - \ln^2 ( 2^{\alpha-2 } y_c ) -\frac{\pi^2}{3}\bigr).\end{aligned}\ ] ] the dijet cross section at nlo for @xmath197 is given as @xmath236.\ ] ] an interesting case is the result of the anti-@xmath1 jet algorithm with @xmath21 . the dijet cross section in the anti-@xmath1 algorithm at order @xmath142 is explicitly given by @xmath237.\ ] ] also the cambridge / aachen jet algorithm corresponds to the case @xmath20 , and the dijet cross section in this jet algorithm is given as @xmath238.\ ] ] for @xmath239 , the factorization breaks down since the jet and the soft functions contain ir divergence . however , when the jet function and the soft function are added , these divergent terms cancel and the finite dijet cross section is obtained . the sum of the collinear and the soft parts is given as @xmath240.\end{aligned}\ ] ] note that only this combination is free of ir divergence as well as of the pole in @xmath241 . therefore the dijet cross section has a finite limit as @xmath180 . the dijet cross section for @xmath242 is given as @xmath243.\end{aligned}\ ] ] note that the dijet cross section is the same for @xmath32 and @xmath27 , and it is continuous at @xmath3 . therefore , for the @xmath1 algorithm , the dijet cross section at order @xmath142 is given as @xmath244 the fixed - order dijet cross sections for various values of @xmath7 are independent of the renormalization scale @xmath90 . however , they contain logarithms of @xmath16 . when @xmath245 becomes small , the fixed - order results are not to be trusted and the resummation of the large logarithms should be performed . this is achieved by solving the renormalization group ( rg ) equations for the hard , jet and soft functions . we present the solution of the rg equation and show the numerical results of the resummation effect with the theoretical uncertainties at next - to - leading logarithm ( nll ) order . the rg equation for the hard , jet and soft functions is of the form @xmath246 f_i(\omega_i,\mu),\ ] ] where @xmath247 is the corresponding function and @xmath248 respectively for @xmath249 . the solution of the rg equation can be expressed in terms of the following quantities which are defined as @xcite @xmath250 then @xmath247 is written as @xmath251 \bigl(\frac{\omega_i^2}{\mu_i^2}\bigr)^{-a_i(\mu_i , \mu ) } f_i ( \omega_i,\mu_i),\ ] ] where the scales @xmath252 for @xmath249 are the hard , jet and soft scales respectively the @xmath253 function and the anomalous dimensions can be expanded as a series in @xmath142 as @xmath254 , \nonumber \\ a_i ( \alpha_s ) & = & a_i^0 \frac{\alpha_s}{4\pi } + a_i^1 \bigl(\frac{\alpha_s}{4\pi}\bigr)^2 + \cdots , \ \ b_i ( \alpha_s ) = b_i^0 \frac{\alpha_s}{4\pi } + b_i^1 \bigl(\frac{\alpha_s}{4\pi}\bigr)^2 + \cdots,\end{aligned}\ ] ] the anomalous dimension for the hard coefficient at order @xmath142 is given by @xmath255 the anomalous dimensions of the jet and the soft functions for @xmath256 at order @xmath142 are given as @xmath257 where the jet scale @xmath258 and the soft scale @xmath259 are defined as @xmath260 for @xmath197 , they are given as @xmath261 it can be explicitly verified that @xmath262 for @xmath2 , and the dijet cross section is independent of the renormalization scale . however , the anomalous dimensions for the jet and the soft functions for @xmath33 are not defined since the corresponding amplitudes include ir divergence . in this case the anomalous dimension of the combined collinear and soft parts is given by @xmath263 which exactly cancels that of the hard function @xmath264 . the anomalous dimensions for @xmath175 can be cast into the form @xmath265 and for @xmath266 , @xmath267 where @xmath268 is the cusp anomalous dimension of the hard function @xmath269 . this form is obtained since @xmath262 , and since the cusp anomalous dimensions are independent of @xmath7 , that is , the jet algorithm . to nll order with large logarithms , we need the cusp anomalous dimension to two - loop order , the remaining anomalous dimensions to one loop order and the hard , jet and soft functions at tree level @xcite . the details for the numerical analysis are provided in appendix . the hard , jet and soft scales can be inferred by requiring that the logarithmic terms in the fixed - order results are not significant . the hard scale is set by @xmath270 , and it is varied from @xmath271 to @xmath272 to evaluate the theoretical uncertainty . the jet scale is given by @xmath273 for @xmath2 , and is also varied from @xmath274 to @xmath275 . ( b ) @xmath276 ( c ) @xmath21 . we set @xmath277 gev , and @xmath278 , and for the soft function @xmath279 . [ csecscale],width=604 ] the choice of the soft scale is more delicate . for @xmath179 , from the argument of the logarithm in the soft function in eq . ( [ js+ ] ) , the soft scale is set as @xmath259 . from the power counting , we require that @xmath280 and @xmath281 , where @xmath66 is a small parameter in scet . for @xmath30 , the soft scale is set by @xmath282 . as discussed in the jet algorithm , @xmath253 is also of order @xmath77 . since @xmath77 is set by the jet scale as @xmath283 , we require that @xmath284 in the numerical analysis . we also vary the soft scale from @xmath285 to @xmath286 to evaluate the theoretical uncertainty . [ csecscale ] shows the theoretical uncertainty of the jet cross sections for three values of @xmath7 at nll as we vary the jet and soft scales . the cross sections are normalized by dividing the jet cross section by the total cross section at order @xmath142 , which is given by @xmath287 therefore the ratio @xmath288 represents the fraction for the two jets . the values of @xmath9 and @xmath16 are chosen as 1000 gev , and 0.1 respectively for illustrative purposes . the hard scale is fixed at @xmath9 . and the cases with @xmath19 , 0 and @xmath289 are shown , which correspond to the jade , the cambrige / aachen , and the anti-@xmath1 algorithms respectively . the upper plots show the dependence on the jet scale @xmath290 as it varies from @xmath274 to @xmath275 while the soft scale @xmath291 is fixed at @xmath259 . the lower plots show the dependence on the soft scale @xmath291 as it varies from @xmath285 to @xmath286 while the jet scale @xmath290 is fixed at @xmath258 . as can be seen in the plots , the dependence on the jet scale is mild , but the cross section is rather sensitive to the choice of the soft scale for @xmath20 and @xmath289 . ( b ) @xmath20 ( c ) @xmath21 corresponding to fig . [ csecscale ] . the bands show the theoretical uncertainties . the solid line represents the resummed cross section with @xmath292 , @xmath293 , @xmath294 . the dashed line is the fixed - order cross section with @xmath295 . [ csec],width=604 ] fig . [ csec ] shows the theoretical uncertainty of the resummed and the fixed - order dijet cross sections . in resummed results , the hard , jet and soft scales are varied as @xmath296 , @xmath297 and @xmath298 respectively . in the fixed - order result , we set @xmath299 and @xmath300 is varied from @xmath301 to @xmath302 . the small parameter @xmath66 corresponds to @xmath283 , and the @xmath16 values on the horizontal axis indicate that we set @xmath303 . the fixed - order results diverge as @xmath16 approaches zero , but the resummed results are suppressed for small @xmath16 and converge to zero . the theoretical uncertainty at nll order is not significantly reduced compared to the nlo fixed - order results . we expect that the uncertainty will be reduced at higher orders . in addition to the suppression of the resummed result for small @xmath16 , the fixed - order results tend to be more negative as @xmath7 decreases , while the jet cross sections have meaning only when they are positive . as figure [ csec ] shows , the nlo result and the nll result are consistent except for small values of @xmath16 for @xmath145 . as @xmath7 decreases , the fixed - order result is not reliable . only after the resummation is performed , the jet cross sections become positive , including the behavior for small @xmath16 . it turns out that the cone - type algorithm , which is equivalent to the inclusive @xmath1 algorithm at nlo , and the sterman - weinberg algorithm show similar shapes for the phase spaces @xcite , therefore we expect that the dijet cross section should be resummed in the cone - type algorithm to yield a meaningful answer . since the dijet cross section is ir finite in spite of the fact that both the jet and soft functions are ir divergent , one might expect that the resummed result for the @xmath1 algorithm ( @xmath3 ) can be obtained by taking the limit @xmath180 from the result with @xmath175 . indeed we can obtain the resummed result and verify that the cross section approaches zero for small @xmath16 . however , the theoretical uncertainties blow up in this limit due to the large uncertainties in the jet and soft functions as @xmath180 , and there is no predictive power in the resummed result for @xmath3 . the generalized jet algorithm enables us to see the rich structure of the jet algorithm from the viewpoint of the factorization . it includes familiar jet algorithms with specific values of @xmath7 . the advantage of the generalized jet algorithm is to see the divergence structure of the jet and soft functions in a unified way . we can see systematically how the structure of the divergence changes as @xmath7 varies . the structure of the divergence can be inferred from the observation of the phase spaces in @xmath79 space . when it is ambiguous for @xmath213 , the phase space in @xmath80 space is helpful in identifying the divergence . in explicit computation , we have verified that the expectation from the phase space is correct . to summarize , the jet algorithm for @xmath2 renders the jet and soft functions ir finite , thus guarantees the factorization of the dijet cross section . among the known jet algorithms , this includes the jade algorithm ( corresponding to @xmath19 ) , the cambridge / aachen algorithm ( @xmath20 ) and the sterman - weinberg and the cone - type algorithm ( @xmath30 ) and the anti-@xmath1 algorithm ( @xmath21 ) . on the other hand , the factorization breaks down with the jet algorithms with @xmath33 because the jet and soft functions contain infrared divergence . though the ir divergence is cancelled in the dijet cross section and gives a finite result , it breaks the factorization and the @xmath1 algorithm belongs to this category . the main point of the paper is that the structure of divergence in the collinear and soft parts depends on how the jets are defined . the important criterion is that each factorized part should be ir finite in order to respect the factorization . and according to our analysis of the generalized jet algorithm , only the jet algorithm with @xmath32 factorizes the dijet cross section in a strict sense . note that we have considered the so - called exclusive generalized @xmath1 algorithms @xcite , which is based on the @xmath0 annihilation . the scet version is presented in ref . the inclusive recombination algorithm @xcite is more suitable in hadronic collisions . but it turns out that the inclusive jet algorithm reduces to the cone - type algorithm @xcite at next - to - leading order in which there are three final - state particles . in contrast to @xmath0 annihilation , the partonic center of energy is boosted along the beam direction compared to the hadron - hadron center of energy in hadronic collisions . therefore the physical observables in hadron - hadron scattering , as in lhc , should be described in terms of the boost - invariant quantities such as the transverse momentum , the rapidity and the azimuthal angle . it will be interesting to probe the jet properties arising from the jet algorithms in other types of scattering . and the analysis of the divergence structure in various jet algorithms , which we have probed here , can be extended to deep inelastic scattering and hadron - hadron collisions with the boost invariance along the beam direction to see if the structure of the divergence is sustained or if the kinematics of the scattering changes it using the exclusive or inclusive generalized @xmath1 algorithm . j. chay and i. kim are supported by basic science research program through the national research foundation of korea ( nrf ) funded by the ministry of education(grant no . nrf-2014r1a1a2058142 ) . c. kim is supported by basic science research program through nrf funded by the ministry of science , ict and future planning with grants no . nrf-2012r1a1a1003015 , no . nrf-2014r1a2a1a11052687 . the qcd @xmath253 function and the cusp anomlous dimension in the @xmath23 scheme are given as @xmath254 , \nonumber \\ \gamma_{\mathrm{cusp } } ( \alpha_s ) & = & \gamma_0 \frac{\alpha_s}{4\pi } + \gamma_1 \bigl(\frac{\alpha_s}{4\pi}\bigr)^2+\cdots , \end{aligned}\ ] ] where the expansion coefficients for the qcd @xmath253 function to two - loop order are @xmath304 and the cusp anomalous dimension to two loop order are given as @xcite @xmath305.\ ] ] the coefficients of the anomalous dimensions in the hard function are given by @xmath306 for the jet and the soft functions with @xmath175 , the coefficients are given a @xmath307 and with @xmath197 , they are given as @xmath308 at nll order , the quantities in eq . ( [ rgf ] ) are written as @xmath309 , \nonumber \\ a_i ( \mu_i,\mu ) & = & \frac{a_i^0}{2\beta_0 } \bigl [ \ln \frac{\alpha_s ( \mu)}{\alpha_s ( \mu_i ) } + \bigl ( \frac{a_i^1}{a_i^0 } -\frac{\beta_1}{\beta_0 } \bigr ) \frac{\alpha_s ( \mu ) -\alpha_s ( \mu_i)}{4\pi } \bigr],\end{aligned}\ ] ] and @xmath310 is obtained from @xmath311 when @xmath312 are replaced by @xmath313 . g. p. korchemsky and a. v. radyushkin , renormalization of the wilson loops beyond the leading order , nucl . b * 283 * , 342 ( 1987 ) . i. a. korchemskaya and g. p. korchemsky , on lightlike wilson loops , phys . b * 287 * , 169 ( 1992 ) .
the importance of neutrinos in the explosions of massive stars has long been recognized . this significance is coupled with their ability to carry information to us about the processes and conditions in the core of the supernova so that the neutrino signal from the next galactic supernova will provide us with an opportunity to test the core collapse paradigm . the explosion begins when the runaway process of electron capture in the core a massive star leads to a rapid compression and collapse that is only halted when the degenerate pressure of the neutrons and thermal pressure of non - degenerate particles kicks in at super - nuclear densities . the rapid neutronization during the collapse leads to a large burst of @xmath0 neutrinos that identifies the beginning of the event . as nuclear densities are reached the mean free path of the neutrinos becomes shorter than the size of the proto - neutron star and the neutrinos become trapped . thermal processes within the core create a thermal bath of neutrino - antineutrino pairs of all flavors . the neutrinos slowly diffuse from the core over a period of order @xmath1 seconds , carrying 99% of the gravitational binding energy of the core with them . with such a large neutrino luminosity even a small number of neutrino interactions above the core can create an important impact on the explosion physics . detection of the neutrinos from the next nearby supernova will offer us the opportunity to examine the internal evolution of the explosion since the neutrinos allow us to see down all the way to the proto - neutron star . for example , if the proto - neutron star collapses to a black hole , the resulting neutrino signal will be altered @xcite . in addition , although these neutrinos are only weakly interacting they can have significant effects upon the nucleosynthesis that occurs in the supernova : in the outer layers of the star neutrinos can cause a transformation of the elements synthesized during the preceding thermonuclear burning period in the `` neutrino process '' @xcite , it is thought that a neutrino driven wind occurs at late time in the supernova which may create sufficiently neutron rich material to produce the r - process elements @xcite , and long duration gamma ray bursts ( which are thought to be a rare type of supernova ) produce elements in conditions governed by strong neutrino fluxes @xcite . the results of all these nucleosynthesis processes are quite sensitive to the neutrino spectra @xcite . for reasons pertaining to both hydrodynamics and to element synthesis it is important to understand the details of neutrino propagation in the supernova and the mixing between neutrino flavors . neutrino oscillations also alter what we observe but , simultaneously , also allow the possibly of elucidating yet unknown neutrino oscillation physics , e.g. @xcite . supernova neutrino flavor transformation is a rapidly developing field with the neutrino background terms the subject of intense study @xcite . realistic possibilities include a complete mixing of all flavors @xcite or a partial oscillation between flavors @xcite . an important part of any future observation of supernova neutrinos will be to understand this physics . in parallel , phase effects due to multiple resonances have recently been recognized @xcite distorting the simple picture of neutrino propagation through the outer layers of the supernova that has been used in the past @xcite . the explosion of the star leads to an evolution of the flavor mixing so that to appreciate the effects upon neutrino propagation it is important to understand the density profile . observations of high pulsar velocities and polarized supernova light suggest an inherent asymmetry in the explosion mechanism . though many possible mechanisms for generating the asymmetry have been proposed , recent work by blondin , mezzacappa & demarino @xcite has identified an instability of the standing accretion shock that leads to large dipole and quadrupole moments . more recent work by blondin & mezzacappa @xcite and ohnishi , kotake , and yamada @xcite as well as scheck _ @xcite has both confirmed and furthered the understanding of this instability . an alternative mechanism , acoustic heating , has also been discussed by burrows et al . @xcite . the temporal evolution of the neutrino flavor mixing and the subsequent variation of the neutrino signal means that it may be possible to to detect prominent features of the supernova density profile and to use this information to learn important information about the explosion . this possibility was demonstrated first by schirato and fuller @xcite who used a time - dependent , one - dimensional @xmath2 density profile to demonstrate that the supernova s forward shock wave reaches and disrupts the ` h ' resonance transformation layer a few seconds after the core bounce leading to a detectable change in the charged current neutrino signal . this was followed and elaborated upon by dighe & smirnov @xcite , takahashi _ et al . _ @xcite and fogli _ et al . more recently tom ` as _ @xcite used a more sophisticated numerical model of a progenitor star to investigate the effect of both forward and reverse shocks on the neutrino signal . tom ` as _ et al . _ also present one second of a two - dimensional simulation result to show how the density profile can be greatly complicated by a deformed forward shock followed by strong convection currents . but due to the intensive computational burden of supernova models it is numerically very taxing to to watch these the explosion propagate outward until they affect neutrino transformations . in this paper we present calculations which link hydrodynamical simulations of time dependent density profiles and phase - retaining neutrino oscillations . our presentation begins in section [ section : simulations ] with the results from simulations of supernovae in both one and two dimensions where we artificially heat a density profile constructed so as to mimic the state of the supernova at the point where the accretion shock has stalled . our two - dimensional explosions are heated aspherically so as to create an aspherical supernova . though this 2d model lacks some structural features present in more sophisticated models that generate asphericity via standing accretion shock instabilities , we still obtain results that give a good idea of how density isotropy affects neutrino flavor transformations . in section iii we pass a spectrum of neutrinos through the simulation results and show how the crossing probability is affected by the evolution of the profiles . we finish in section iv by presenting our calculation of two neutrino detector signals - the positron spectra within a water cerenkov detector , and the ratio of charged current to neutral current event rates for a heavy water detector - that demonstrate our ability to extract information about the supernova . our first task in determining the neutrino signal is to simulate the supernova so as to obtain the time evolution of the density profile . a supernova simulation is a complex and computationally intensive problem . at the present time there is not yet a robust , self - consistent model for core - collapse supernovae , and the most sophisticated multi - dimensional simulations take a very long time to compute . but we do not need to simulate the actual core collapse and formation of the proto - neutron star , these events occur deep within the core at a radius of @xmath3 or so , whereas the first neutrino resonance occurs at around @xmath4 for neutrino energies of order @xmath5 , and the second is even further out . what is occurring in the core is irrelevant for our purposes , all we need from the simulation of the supernova is the propagation of the shock through the star . for this reason we employ a simplistic and artificially energetic numerical simulation to create explosions with a range of features that might be present in reality . we do not intend this model to be realistic itself , but rather a tool to help with the difficult problem of reconstructing density features of an exploding star based on changing neutrino signals detected at earth . we use vh-1 @xcite , a hydrodynamic code based on woodward and colella s @xcite piecewise parabolic method . for both the one - dimensional and two - dimensional simulations we employ an exponentially coarsening radial grid that ranges from @xmath6 to @xmath7 where @xmath8 is the radius from the center of the star , and for the two - dimensional simulations an angular grid that ranges from @xmath9 to @xmath10 where @xmath11 is the polar angle . the innermost radial grid increment is set to @xmath12 and increases by 1.3% thereafter . by @xmath13 the radial increment has grown to @xmath14 and by @xmath15 we reach @xmath16 . we map into the code a spherically symmetric progenitor profile intended to represent the state of the supernova about @xmath17 after the collapse . beyond @xmath18 the profile is that of a low metallicity , @xmath19 progenitor model developed by heger et al @xcite that we have allowed to collapse further within vh-1 so that the infall velocities approached free fall . interior to @xmath20 we splice in a slow , outward - moving standing accretion shock profile provided by heywood @xcite that is similar to that found in @xcite . finally , inside @xmath21 we insert a dense , @xmath22 , polytropic core containing approximately @xmath23 . the inner boundary condition inside the dense core at @xmath24 is reflecting . figure ( [ fig:1da ] ) shows the full initial density profile . the initial , @xmath25 , density profile used in our sn simulations . the dense core ( a ) inside @xmath26 contains approximately @xmath23 . the slow , outward - moving standing accretion shock ( b ) is located at @xmath27 . above that we have a collapsing , @xmath19 , progenitor ( c).,width=480 ] to drive the explosion , we mimic neutrino heating by inserting energy into the region above a @xmath28 gain radius . the energy deposition rate per unit volume , @xmath29 , is proportional to the density of material , falls as @xmath30 and decreases exponentially with time @xmath31 over a timescale @xmath32 set to @xmath33 . for the 1d simulations then , @xmath34 the additional factors of @xmath35 and @xmath31 are inserted so as to ramp up the energy deposition over both distance and time thereby avoiding the violent disruption that occurs if we deposit energy either too suddenly or within a small volume . for the 2d models we introduce an angular dependence into the energy deposition prescription so as to match the observation by blondin _ _ @xcite that small perturbations in standing accretion shock models can ultimately lead to aspherical shock modes . we break the spherical symmetry of our 2d simulations by heating the initial profile with a combination of 75% spherical mode heating and 25% @xmath36 mode heating . thus for the 2d cases @xmath29 becomes @xmath37 in either case the total energy , @xmath38 , input during the simulation was recorded . below the gain radius we wish to maintain the spherical density / gravity conditions and to prevent mass from escaping the core so that we do not disturb the shock heating and convective flow in the outer star . in a real supernova the core is stabilized by neutrino emission but in lieu of full implementation of this cooling mechanism our simulations achieve the same result by forcing all radial and angular velocities below @xmath39 to zero . in this way we separate the evolution of the outer layers of the star from the complex behavior near the core . the runtime of the 1d simulations is sufficiently short that we can vary the total energy deposition thereby obtaining a variety of results ranging from weak explosions through to the very powerful . the 2d simulations take much longer to run and so we have just one data set with an explosion energy set to @xmath40 . before presenting our simulation results we mention that in general , numerical schemes tend to spread the shock front over several zones and they therefore become artificially softened . schirato and fuller @xcite account for this artifact by steepening by hand the density profile at the shocks . the shocks in our simulations are sufficiently steep to demonstrate the primary effect on neutrino mixing , therefore we do not implement a similar correction . the density profiles of the sn we obtain from the simulations possess noticeable differences as the energy deposition changes . for weak explosions the profiles are monotonically decreasing functions of the radius , @xmath8 , with a single , forward facing , shock front that moves relatively slowly outwards . as the deposition energy increases a lower - density cavity forms behind the shock . further increases in @xmath38 eventually lead to the formation of a reverse shock behind the lower - density zone . all three features are present in the results of the 2d simulations but the higher dimensionality , coupled with the aspherical heating means , that this sn possesses a much more turbulent / chaotic profile . the density as a function of the radius and angle in a 2d sn model at @xmath41 . the forward shock is located to the left of ` a ' , the reverse shock is the step - up in density found to the right of ` b ' , and one of the many local cavities in the profile between the shocks is to the right of ` c'.,width=480 ] in figure ( [ fig : rho.snv2d.t=2.5 ] ) we show a time slice of the 2d sn simulation . the forward shock is seen at ` a ' , the ` reverse shock ' is at ` b ' and one of the many bubbles in the density profile is located at ` c ' . we note that although our applied neutrino heating is symmetric about the equator ( @xmath42 ) the resulting density profile is not . we attribute this to the inherent instability of multi - dimensional accretion shocks , potentially excited by simulation numerics . in what follows we discuss the behavior and properties of the various features of the profile . the forward shock is a generic feature of supernovae simulations . a stalled forward shock was present in the initial profile inserted into the hydrodynamical code and the heating we introduced was meant to revive it s outward motion . after revival the shock propagates out through the star and is the explosion feature furthest from the proto - neutron star . the forward shock is visible in figure ( [ fig : rho.snv2d.t=2.5 ] ) as the large jump in density at larger radii and we note that the forward shock in the 2d simulation is slightly oblate due to the aspherical heating of the material . we study the behavior of the forward shock with various explosion strengths using the 1d simulations . the density as a function of the radius in a 1d sn model with @xmath43 at @xmath44 ( dotted ) , @xmath45 ( solid ) , @xmath46 ( long dashed ) and @xmath47 ( dash dot).,width=480 ] in fig . ( [ fig : rho.snv1d.1e35.t=0.9,1.8,3.6,7.2 ] ) we show four snapshots of the density profile in a one - dimensional simulation in which the total energy deposition was @xmath43 . actual sn are thought to be more energetic than this so we regard the results from this simulation as being at the lower end of realistic possibilities . this energy is also significantly less than that used in the simulation shown in fig . ( [ fig : rho.snv2d.t=2.5 ] ) and , consequently , the profile is much simpler . the forward shock is clearly visible and we note that the fractional jump in density across it does not vary . for normal shocks the density jump across the shock is given by @xmath48 where @xmath49 the mach number and @xmath50 is determined by the ratio of specific heats . in our simulations we adopted a fixed equation of state and the mach number of the shock is always sufficiently large that we find that the density jump does not vary to any great extent with time and/or with @xmath38 . but the position of the forward shock as a function of time is greatly influenced by @xmath38 with larger values leading to more rapidly moving forward shocks . the density as a function of the radius in a 1d sn model with @xmath51 at @xmath44 ( dotted ) , @xmath45 ( solid ) , @xmath46 ( long dashed ) and @xmath47 ( dash dot).,width=480 ] this can be seen in fig . ( [ fig : rho.snv1d.6e35.t=0.9,1.8,3.6,7.2 ] ) which are snapshots at the same moments as those in fig . ( [ fig : rho.snv1d.1e35.t=0.9,1.8,3.6,7.2 ] ) for a 1d simulation with @xmath51 . this figure also displays the low - density cavity that can form behind the forward shock as @xmath38 increases . this profile is very similar to the profiles used by fuller & schirato @xcite and fogli _ et al . _ @xcite . the heating that led to the regeneration of the forward shock continues to accelerate the material above the proto - neutron star even after the shock has been revived and is moving outwards . a wind is created with a velocity that increases with radius . when the velocity of the material becomes larger than the local sound speed a reverse shock is formed . the reverse shock feature was not present in the ` initial ' profile : it develops only later . if a reverse shock is to form enough energy must be deposited to create a sufficiently strong wind . a reverse shock can be seen in figure ( [ fig : rho.snv2d.t=2.5 ] ) where it is the step - up in density at the back of the turbulent zone behind the forward shock . the reverse shock is oblate due to the aspherical heating of the material in the simulation . the density as a function of the radius in a 1d sn model where @xmath52 at @xmath53 ( dotted ) , @xmath54 ( solid ) , @xmath55 ( long dashed ) , @xmath41 ( dash - dot ) and @xmath56 ( dash double dot).,width=480 ] once again we can use 1d simulations to study this feature more easily . no reverse shock was seen in the profiles shown figure ( [ fig : rho.snv1d.1e35.t=0.9,1.8,3.6,7.2 ] ) or ( [ fig : rho.snv1d.6e35.t=0.9,1.8,3.6,7.2 ] ) but when we increase @xmath38 to @xmath52 we obtain the results presented in figure ( [ fig : rho.snv1d.8e35.t=1,1.5,2,2.5,3 ] ) . we notice that the forward shock is , again , moving more rapidly compared to the results shown in figures ( [ fig : rho.snv1d.1e35.t=0.9,1.8,3.6,7.2 ] ) and ( [ fig : rho.snv1d.6e35.t=0.9,1.8,3.6,7.2 ] ) due to the larger energy deposition @xmath38 . the figure also indicates that the reverse shock is smaller than the forward shock . as @xmath38 increases further the density jump across the reverse shock increases and it moves closer to the forward shock . we find that it is also possible for the reverse shock to penetrate to densities lower than the forward shock . the density as a function of the radius in a 1d sn model with @xmath57 at @xmath58 ( dashed ) , @xmath54 ( solid ) and @xmath59 ( long dashed).,width=480 ] both of these behaviors are shown in figure ( [ fig : rho.snv1d.2e36.t=1,1.5,2 ] ) where we show snapshots of the profile for a 1d simulation with @xmath57 . the density jump across the reverse shock is still smaller than across the forward but they are almost equivalent , and the density immediately behind the reverse shock is lower than the density immediately in front of the forward shock . it was mentioned by toms _ _ @xcite that this feature was also seen at early times in their simulations . but realistically sn may struggle to achieve explosions with comparable energies so we regard the results of this simulation as being at the upper end of possibilities . in both figures ( [ fig : rho.snv1d.8e35.t=1,1.5,2,2.5,3 ] ) and ( [ fig : rho.snv1d.2e36.t=1,1.5,2 ] ) the reverse shock was driven radially outward by a wind generated by the material heated above the proto - neutron star . our energy deposition decreased exponentially with time which led to a gradual decrease in wind strength . in turn , as the wind abates , the outward motion of the reverse shock slows and , eventually , we find that its motion can be completely halted . this stalling of the reverse shock also occurred in the two - dimensional simulation with the shock stalling at different times for different polar angles . we also find that in both 1d and 2d simulations the density jump across the reverse shock decreases as the shock is about to turn around . after the reverse shock stalls both the 1d and 2d simulations indicate that the reverse shock feature then moves back towards the core . in figure ( [ fig : rho.snv1d.8e35.t=1,1.5,2,2.5,3 ] ) we see this stalling of the reverse shock and the backwards motion can be seen after comparing the profiles at @xmath41 and @xmath56 . the reverse shock in this simulation actually reached it s furthest radial position at about @xmath60 . similar backwards motion for the reverse shock may be seen in figure ( 1 ) of toms _ et al . _ @xcite . eventually the reverse shock reaches the core whereupon our simulations indicate that it is reflected and subsequently becomes a weak forward shock . the density as a function of the radius in the same 1d sn model shown in figure ( [ fig : rho.snv1d.8e35.t=1,1.5,2,2.5,3 ] ) i.e. with @xmath52 , at @xmath61 ( dotted ) , @xmath62 ( solid ) and @xmath63 ( long dashed).,width=480 ] further snapshots of the density profile taken from the simulation with @xmath52 are shown in figure ( [ fig : rho.snv1d.8e35.t=2.8,3.1,4.3 ] ) where we see more clearly the backwards motion of the reverse shock and its later reflection . the radial position at which the reverse shock stalls depends upon the energy deposition . additional snapshots from the simulation with slightly larger energy deposition , @xmath57 , are show in figure ( [ fig : rho.snv1d.2e36.t=4,4.5,4.9 ] ) . though the reverse shock in this simulation also attained it s maximum radial position at @xmath60 the figure shows that it was located further out before it was turned around . the density as a function of the radius in the same 1d sn model shown in figure ( [ fig : rho.snv1d.2e36.t=1,1.5,2 ] ) , that is @xmath57 , at @xmath64 ( dotted ) , @xmath65 ( solid ) and @xmath66 ( long dashed).,width=480 ] the reverse shock is an interesting feature of the sn and , in contrast with the forward shock , can move both outwards and inwards . for very energetic explosion it may penetrate to lower densities than that immediately in front of the forward shock . the density jump across the reverse shock can vary with time and does so most noticeably when the shock is about to stall . the difference in the hydrodynamics between the one and two dimensional sn simulations can be traced back to the aspherical heating of the material above the proto - neutron star in the two dimensional case . non - spherical heating can lead to turbulent fluid flow creating eddies and bubbles as shown in figure ( [ fig : rho.snv2d.t=2.5 ] ) . the density at a polar angle of @xmath67 as a function of the radius in a 2d sn model at @xmath68 ( solid ) , @xmath69 ( dotted ) and @xmath70 ( dashed).,width=480 ] the density as a function of the radius in a 2d sn model at @xmath41 . the angular slices are at @xmath71 ( solid ) , @xmath72 ( dotted ) and @xmath73 ( dash dot).,width=480 ] the neutrinos released by the proto - neutron star propagate along radial slices of the profile . in fig . ( [ fig : rho.snv2d.t=3.9,4.8,5.7.slice=25 ] ) we show the density profile from the 2-d model at the polar angle of @xmath74 for snapshots at @xmath68 , @xmath69 and @xmath70 while in figure ( [ fig : rho.snv2d.t=2.5.slice=45,105,165 ] ) we present the density along three different polar angles at @xmath41 , i.e. taken from figure ( [ fig : rho.snv2d.t=2.5 ] ) . in both we see the forward and reverse shocks and , as mentioned earlier , figure ( [ fig : rho.snv2d.t=2.5.slice=45,105,165 ] ) also shows that the radial position of both the forward and reverse shocks varies with the polar angle . figure ( [ fig : rho.snv2d.t=3.9,4.8,5.7.slice=25 ] ) also indicates that between @xmath69 and @xmath70 the reverse shock ( along this radial slice ) stalled and began to move back to the core . the density at a polar angle of @xmath75 as a function of the radius in a 2d sn model at @xmath65 ( dash dot ) , @xmath76 ( solid ) , @xmath77 ( dotted ) , and @xmath47 ( dashed).,width=480 ] for other radial slices the reverse shock turnaround time will be different : in figure ( [ fig : rho.snv2d.t=4.5,5.4,6.3,7.2.slice=125 ] ) we plot radial slices along the @xmath75 line of sight at various times . the reverse shock in the figure , located just beyond @xmath78 , turns around between @xmath79 and @xmath80 . in all our 2d figures the general shape of the density profile between the two shocks is reminiscent of that shown in figure ( [ fig : rho.snv1d.8e35.t=1,1.5,2,2.5,3 ] ) for the 1d simulation with a similar value for @xmath38 - i.e. a lower density region in front of the reverse shock , a wall of higher density material behind the forward shock - but clearly there are large ` fluctuations ' upon this basic trend both between the shocks and behind the reverse shock . the forward shock , the reverse shocks , contact discontinuities and the local bubbles / cavities are the features in the density profile with the greatest potential to alter the state of any neutrinos propagating through the sn since they represent the locations within the sn where the density gradient is largest . large density gradients lead to non - adiabatic evolution of the neutrinos and significant differences compared to the neutrino propagation through the undisturbed profile . the vast majority of the neutrinos emitted by the supernova last interact with matter at the neutrinosphere located at the surface of the proto - neutron star . as they propagate outwards a small percentage will be absorbed by nucleons thereby transferring energy to the supernova and , it is thought , reviving the stalled shock . but even though the remainder of the neutrinos survive the passage through the material overlying the proto - neutron star what emerges is not the same as what was emitted . the change in the neutrinos is due to neutrino oscillations and the presence of matter modulates this mixing . neutrino oscillations arise due to a distinction between the interaction eigenstates ( otherwise known as the flavor eigenstates @xmath81 ) and the eigenstates of the free hamiltonian ( known as the mass eigenstates ) with masses @xmath82 . since there are three basis states a general neutrino wavefunction is described by three complex coefficients and evolves according to the schrodinger equation . in the vacuum the hamiltonian is diagonal in the mass basis but possesses off - diagonal terms in the flavor basis that are the cause of flavor oscillations . in the presence of matter a potential , @xmath83 , that takes into account coherent forward scattering of the neutrinos , is included in the hamiltonian . since we are only concerned with mixing between active neutrino flavors ( i.e. all the flavors that have ordinary weak interactions ) we may subtract off the common neutral current contribution ( which will contribute only an overall phase ) leaving just the charged current contribution to the @xmath84 component of @xmath83 . this contribution is the well - known @xmath85 where @xmath86 is fermi s constant , @xmath87 is the mass density , @xmath88 is the electron fraction and @xmath89 is the nucleon mass . the effect of the matter upon the antineutrinos differs from that of the neutrinos - the potential @xmath90 that appears in the antineutrino hamiltonian has the same magnitude but the opposite sign . due to the inclusion of @xmath91 the hamiltonian is a function of position and is neither diagonal in the mass basis nor the flavor basis . one may try and diagonalize @xmath92 but the unitary transformation that relates the flavor basis to the new ` matter ' basis also varies with the position . consequently the gradient of the unitary transformation is non - zero and one finds that the schrodinger equation in this new matter basis - that was meant to diagonalize @xmath93 - picks up off - diagonal terms . thus it is not possible , in general , to find a basis that diagonalizes the hamiltonian and therefore oscillations of the amplitudes that describe the wavefunction occur in every basis . in general , the three complex components of the wavefunction oscillate simultaneously . exactly what occurs depends upon the energy of the neutrino @xmath94 , the differences between the squares of the masses @xmath82 , the mixing angles that describe the relationship between the flavor and mass basis , and , of course , @xmath91 . since there are three mass eigenstates there are three separate mass splittings @xmath95 ( though only two are independent ) and the relationship between the neutrino flavor and mass bases is described by the matrix @xmath96 which , in turn , is parameterized by three mixing angles @xmath97 and @xmath98 plus a cp - phase @xmath99 . the structure of @xmath96 is @xmath100 where @xmath101 , @xmath102 . mixing has been observed in the neutrinos emitted by the sun and the neutrinos produced by cosmic rays striking the atmosphere . both have been confirmed with terrestrial experiments . each observation of neutrino mixing can be described by a single @xmath103 - @xmath11 pair of parameters and experimentally the ` solar ' and ` atmospheric ' mass splittings differ by around a factor of @xmath104 with the solar mass splitting , @xmath105 , being the smaller of the two . this observation permits us to consider the evolution of the general , three component , neutrino wavefunction as being factored into spatially distinct , two - neutrino mixes . the factorization simplifies matters greatly and from it one can demonstrate that there are two resonances in the supernova density profile : the so - called ` l ' resonance and the ` h ' resonance . the l resonance , at lower density and thus further from the proto - neutron star , involves mixing between matter states @xmath106 and @xmath107 , the relevant mass splitting is @xmath108 - which is approximately the solar mass splitting @xmath105 - and the mixing angle is @xmath109 - which is approximately the mixing angle determined by the solar neutrino experiments @xmath110 . the relevant mass eigenstates and mass splitting for the h resonance , at higher density and closer to the core , depend upon the sign of @xmath111 ( or , equivalently , @xmath112 ) and this is not currently known . if @xmath111 is positive ( a normal hierarchy ) then the h resonance involves mixing between states @xmath107 and @xmath113 and the mass splitting @xmath111 . if @xmath111 is negative ( an inverted hierarchy ) then at the h resonance it is the antineutrinos states @xmath114 and @xmath115 that mix and the relevant mass splitting is @xmath116 . in either case the mixing angle is the unknown , but small , @xmath117 . quite generally the coherent matter basis wavefunction that arrives at a distance , @xmath118 , from the core of the sn is related to the initial matter basis wavefunction at the proto - neutron star via the equation @xmath119 while for the anti - matter states @xmath120 . the matrices @xmath121 and @xmath122 are the s - matrices for the neutrinos and antineutrinos respectively , for a discussion of this approach to neutrino oscillations , see e.g. @xcite . to determine @xmath123 and @xmath124 we need to know both the initial states @xmath125 and @xmath126 and the two matrices @xmath121 and @xmath122 . the matrices @xmath127 and @xmath128 can be factored as @xmath129 and @xmath130 which are sufficiently general to accommodate our lack of knowledge of the hierarchy . the two matrices @xmath131 and @xmath132 represent the neutrino propagation from the surface of the sn through the vacuum to earth and both are diagonal in the mass basis . the matrices @xmath133 , @xmath134 and @xmath135 represent the change to the initial wavefunction due to the neutrino s or antineutrino s passage through the h or l resonance have multiple h resonances therefore the matrices @xmath134 and @xmath135 represent their combined effect of all the h resonances and @xmath133 all the l resonances . we assume that the neutrinos encounter all l - type resonances after all h - type resonances . this is the case for all profiles we have generated ] . there is no matrix @xmath136 because antineutrinos do not experience an l resonance in matter . the two matrices @xmath134 and @xmath135 are evaluated somewhere between the h and l resonances while @xmath137 is evaluated at the surface of the sn . these evaluation positions will not affect our result if they are sufficiently far from the resonances . the structures of @xmath133 , @xmath134 and @xmath135 follow from our knowledge of the matter states that mix at either the h or l resonance and are thus @xmath138 after omitting irrelevant phases . from these matrices we define @xmath139 which are the crossing probabilities for neutrinos or antineutrinos at the two resonances . the resonances are said to be ` adiabatic ' or ` non - adiabatic ' depending upon whether the crossing probability is close to zero or closer to unity . indeed these are the two natural values since the crossing probability is determined by the ratio of the resonance width to the local oscillation lengthscale and typically one is significantly larger than the other . our interest now turns to the initial states . the density at the proto - neutron star is so large that the matter eigenstates and the flavor eigenstates are strongly aligned there . from a full 3-neutrino mixing calculation we find that the initial matter basis spectra for a normal hierarchy ( nh ) are @xmath140 , @xmath141 while all other states , @xmath142 and @xmath143 are equal to the @xmath144 spectrum which we call @xmath145 . with an inverted hierarchy ( ih ) the initial states are @xmath146 , @xmath147 and this time @xmath148 and @xmath149 are all equal to @xmath145 . putting the initial spectra and definitions of the s - matrices together one finds that the flux of matter state @xmath150 a distance d from the supernova is given by @xmath151 where @xmath152 and @xmath153 are the initial spectra of the matter states . but what are detected at earth are , of course , the flavor states . during their flight from the supernova to earth any coherence between the matter eigenstates is lost so the flux of flavor @xmath154 that arrives at earth is the incoherent sum @xmath155 these flavor fluxes may be rewritten in terms of the fluxes emitted by the neutrinosphere by introducing the survival probabilities @xmath156 and @xmath157 for the electron neutrinos and antineutrinos respectively since these are the two flavors with distinct initial spectra . written this way the detectable flavor fluxes are @xmath158 , \\ & & f_{\bar{\nu}_e}(d , e ) = \frac{1}{4\,\pi\,d^{2}}\;\left[\,\bar{p}(e)\,\phi_{\bar{\nu}_e}(0,e ) + ( 1-\bar{p}(e))\,\phi_{\nu_x}(0,e)\right ] , \\ & & 4f_{\nu_x}(d , e ) = \frac{1}{4\,\pi\,d^{2}}\;\left [ ( 1-p(e))\,\phi_{\nu_e}(0,e ) + ( 1-\bar{p}(e))\,\phi_{\bar{\nu}_e}(0,e ) + ( 2+p(e)+\bar{p}(e))\,\phi_{\nu_x}(0,e)\right]\nonumber\\ & & \end{aligned}\ ] ] where we have adopted the notation of dighe & smirnov @xcite by denoting @xmath159 by @xmath160 . from equations ( [ eq : fi ] ) and ( [ eq : fbari ] ) , and the initial spectra , we find that @xmath156 and @xmath157 are related to the elements of @xmath161 and @xmath162 via @xmath163 where @xmath164 , @xmath165 for a normal hierarchy , and @xmath166 , @xmath167 for an inverted hierarchy . after inserting the definition of the crossing probabilities @xmath156 and @xmath157 become @xmath168 these formulae are quite general but from our knowledge of the various neutrino oscillation parameters we can be more specific about what exactly happens to the neutrinos and antineutrinos . the solar mixing parameters are such that the neutrino l resonance is almost completely adiabatic until , perhaps , the very end of the supernova neutrino signal . for this reason @xmath169 . thus at the end we find that determining the neutrino fluxes at earth comes down to computing @xmath170 or @xmath171 . the schrodinger equation forms a starting point by which @xmath170 or @xmath171 can be determined . but if one navely applies , for example , a runge - kutta integrator to this equation one quickly discovers that this is a difficult problem from a numerical standpoint because such algorithms are simply not suited to differential equations where the solution is a highly oscillatory function . a number of alternate methods have been developed for calculating @xmath170 or @xmath171 , such as application of the the landau - zener result or the semi - analytic method by balantekin & beacom @xcite , but for one reason or another these alternate approaches can break down for complex profiles . some authors @xcite have recognized that the evolution of the neutrino wave can be recast as a scattering problem and we adopt in this paper the algorithm outlined in kneller & mclaughlin @xcite which computes the evolution of the neutrino wavefunction by a monte carlo integration . for this paper we selected @xmath172 and for @xmath117 we use @xmath173 as a representative value . with this choice of @xmath117 the unperturbed profile is just sufficiently adiabatic to suppress the transformation effects of all but the most significant density features . the current experimental limit on @xmath117 is set by the chooz experiment @xcite , and for our choice of @xmath103 that limit is @xmath174 . finally , in addition to the effects caused by the propagation through the supernova there are potentially earth matter effects that can arise . these are straightforward to compute and we do not consider them here since their presence ( or absence ) in the neutrino signal is a function of the position of the supernova with respect to the detector when the event occurs . with the sn simulations complete and a method for calculating the crossing probabilities in hand we turn to study the effects of various features in the profiles upon @xmath175 both as a function of time and energy . the forward shock is a generic feature of all our sn simulations . initially the shock is located close to the core and at high density and then , after it is regenerated , begins to move outwards and to lower densities as shown in the figures from section [ section : simulations ] . the density as a function of the radius in a 1d sn model with @xmath43 at @xmath44 ( dotted ) , @xmath45 ( solid ) , @xmath46 ( long dashed ) and @xmath47 ( dash dot ) . the horizontal dashed lines are ( from top to bottom ) the resonance densities for @xmath176 , @xmath1 , @xmath177 , @xmath178 and @xmath179 neutrinos.,width=480 ] in fig . ( [ fig : rho.snv1d.1e35.t=0.9,1.8,3.6,7.2.+rhores ] ) we reproduce the results from fig . ( [ fig : rho.snv1d.1e35.t=0.9,1.8,3.6,7.2 ] ) and also superimpose the resonance densities for @xmath176 , @xmath1 , @xmath177 , @xmath178 and @xmath179 neutrinos . the radii where the horizontal dashed lines intercept the profile are the locations of the resonances . the h resonance crossing probability @xmath175 as a function of neutrino energy for the 1d sn simulation with @xmath43 . in the top panel the snapshot times are @xmath55 ( solid ) and @xmath41 ( dashed ) , in the middle panel @xmath56 ( solid ) and @xmath180 ( dashed ) and in the bottom panel @xmath181 ( solid ) and @xmath65 ( dashed).,width=480 ] and in fig . ( [ fig : pc.snv1d.1e35.t=2,2.5,3,3.5,4,4.5 ] ) we show @xmath175 as a function of neutrino energy at six snapshots of this same simulation . the crossing probability as a function of the energy clearly evolves with time . if @xmath117 is not too small then the evolution of the neutrinos through the undisturbed progenitor profile is almost adiabatic i.e. @xmath182 . as the forward shock arrives at the h - resonance for @xmath183 neutrinos - fig . ( [ fig : rho.snv1d.1e35.t=0.9,1.8,3.6,7.2.+rhores ] ) indicates this occurs at shortly after @xmath184 - the evolution becomes non - adiabatic i.e. @xmath185 because the density jump across the shock straddles the resonance densities of this neutrino energy . as time progresses and the shock moves outwards to lower densities we see from fig . ( [ fig : rho.snv1d.1e35.t=0.9,1.8,3.6,7.2.+rhores ] ) that the shock will begin to affect @xmath186 neutrinos at @xmath46 . the non - adiabaticity sweeps up through the neutrino spectrum from low energy to high . as time progress further eventually the shock ceases to affect the neutrinos of a particular energy and so their propagation returns to being adiabatic . for this simulation fig ( [ fig : rho.snv1d.1e35.t=0.9,1.8,3.6,7.2 ] ) indicates that the evolution of @xmath183 neutrinos returns to being adiabatic at around @xmath46 and that for @xmath186 the return occurs roughly at @xmath47 . note that the @xmath183 neutrinos were affected for a much briefer period ( @xmath187 ) than the @xmath186 neutrinos ( @xmath188 ) this is due to the shape of the density profile . the h resonance crossing probability @xmath175 as a function of time for selected neutrino energies through the 1d sn model with @xmath43 . the curves are : @xmath189 ( solid ) , @xmath190 ( long dashed ) , @xmath191 ( dash - dot ) , @xmath192 ( short dashed ) , and @xmath193 ( dash double - dot).,width=480 ] a complimentary perspective is to look at how particular neutrino energies evolve with time such as those in fig . ( [ fig : pc.snv1d.1e35.e=10,15,20,25,30 ] ) . the temporary transition to non - adiabaticity for each neutrino energy is clearly visible in the figure . this figure makes it most obvious that the lower energies are affected before the higher and also that the duration of the non - adiabatic period increases with the neutrino energy . the extent of the shock feature in neutrino energy seen in fig . ( [ fig : pc.snv1d.1e35.t=2,2.5,3,3.5,4,4.5 ] ) is related to the density jump , @xmath194 , across the shock . at any given time , if @xmath195 is the highest neutrino energy affected by the shock and @xmath196 is the range of neutrino energies then the density jump across the shock is @xmath197 which is independent of the mixing parameters . from equation ( [ eq : delta rho ] ) we saw that , for normal shocks , @xmath194 is a function of the mach number and ratio of specific heats @xmath50 . if @xmath49 is large then @xmath194 , and consequently @xmath196 , are essentially just a function of @xmath50 but if the mach number @xmath49 is not too large - which may be the case - then the width of this feature could be used to infer @xmath49 if @xmath50 is known . the shock position as a function of time , @xmath198 , for the 1d sn simulation where @xmath43 . an artificial time delay of @xmath199 has been added to mimic the stalling of the shock at @xmath200.,width=480 ] as the sn proceeds to explode @xmath195 moves up through the neutrino spectrum . the correspondence between energy and resonance density means that if we know , or assume , a progenitor profile shape then @xmath195 measures the shock position @xmath201 . from the locations of @xmath195 shown in fig . ( [ fig : pc.snv1d.1e35.t=2,2.5,3,3.5,4,4.5 ] ) , and the initial profile , we show in fig . ( [ fig : rs.snv1d.1e35 ] ) @xmath201 as function of time for this particular simulation . the change of @xmath201 with time of course measures the shock speed but we can also use @xmath202 to extrapolate the shock position both forward in time , if necessary , but , more interestingly , also backwards toward the proto - neutron star . if this extrapolation can be accomplished successfully then we can infer the moment , @xmath203 , when @xmath200 . the neutronization burst ( if detected ) supplies a zero of time so @xmath204 is the brief period the shock was stalled . thus the neutrino signal , and in particular the location of the forward shock , contains evidence of whether a key component of the core - collapse supernova paradigm is correct and can provide quantitative data with which to compare with more sophisticated sn simulations than ours . we illustrate this idea in fig . ( [ fig : rs.snv1d.1e35 ] ) where we have artificially included an offset in time . the reverse shock was a feature seen in the more powerful explosions from section [ section : simulations ] and the 2d result . this feature was not present in the initial profile and appeared later on when when the velocity of the wind superseded the local sound speed . after its formation the reverse shock initially moved outwards into the star but , as the energy deposition faded with time and the strength of the wind abated , eventually the reverse shock stalled and then headed back to the core . like the forward shock , the reverse shock affects the adiabaticity of the evolution of the neutrino wavefunction through the profile so that neutrinos with resonance densities straddled by the density jump across the reverse shock evolve non - adiabatically as they pass through it . note that any neutrino with a resonance density straddled by the density jump across the reverse shock will also experience two other resonances : one before the reverse shock and one after . the density as a function of the radius in a 1d sn model with @xmath57 at @xmath58 ( dashed ) , @xmath54 ( solid ) and @xmath59 ( long dashed ) . the horizontal dashed lines are ( from top to bottom ) the resonance densities for @xmath176 , @xmath177 and @xmath179 neutrinos.,width=480 ] this is shown in fig . ( [ fig : rho.snv1d.2e36.t=1,1.5,2.+rhores ] ) where we have superimposed the resonance densities for @xmath176 , @xmath177 , and @xmath179 neutrinos upon the profiles shown in fig . ( [ fig : rho.snv1d.2e36.t=1,1.5,2 ] ) . we see from the figure that it is possible for some neutrinos to be affected by both shocks . if this occurs , and if we neglect the effects from other resonances in the profile , then we might expect that the net effect upon such doubly shock - affected neutrinos to be zero @xcite . the h resonance crossing probability @xmath175 as a function of neutrino energy for three snapshots taken from the 1d sn model where @xmath52 . in the top panel the time is @xmath45 , in the middle @xmath59 , and in the bottom panel @xmath60.,width=480 ] however this expected cancellation is not seen in our results for @xmath175 as a function of the neutrino energy shown in fig . ( [ fig : pc.snv1d.8e35.t=1.8,2,2.4 ] ) for the model where @xmath52 . the crossing probabilities plotted in the figure show some similarities to those plotted in fig . ( [ fig : pc.snv1d.1e35.t=2,2.5,3,3.5,4,4.5 ] ) at the higher neutrino energies affected by the forward shock where there is change from adiabatic to non - adiabatic evolution . but for those energies where cancellation is naively expected we see instead that @xmath175 oscillates wildly . these rapid oscillations are phase effects due to the interference between the two shocks . similar rapid oscillations in the crossing probability have been seen previously in sn neutrino calculations by fogli _ @xcite in a profile with a forward shock and then a bubble cavity behind it , in the sn test case used by kneller & mclaughlin @xcite , and in the results of dasgupta & dighe @xcite where their significance was emphasized . the h resonance crossing probability @xmath175 as a function of neutrino energy for three snapshots taken from the 1d sn model where @xmath57 . in the top panel the time is @xmath205 , in the middle @xmath206 , and in the bottom panel @xmath207.,width=480 ] fig . ( [ fig : pc.snv1d.2e36.t=1.1,1.4,3.0 ] ) we display the crossing probability for the model where @xmath57 at @xmath205 , @xmath206 and @xmath207 . again phase effects are seen . compared to the results shown in fig . ( [ fig : pc.snv1d.8e35.t=1.8,2,2.4 ] ) for the weaker explosion where @xmath52 and in fig . ( [ fig : rs.snv1d.1e35 ] ) for @xmath43 , in this case there is no indication of the characteristic transition from adiabatic to non - adiabatic propagation associated with the forward shock . this occurs because , as noted earlier about this model , the reverse shock penetrates to lower densities than the forward shock at these early times . for this particular model the reverse shock affects the adiabaticity of a particular neutrino energy before the forward shock . the presence of phase effects in @xmath175 are caused by interference between forward and reverse shocks . but it would be nice to find in the neutrino signal a clean signature of the reverse shock that is not contaminated by the forward shock or other features of the profile . a potential signature could arise from the behavior noted earlier which is that the reverse shock stalled as the energy deposition into the material above the proto - neutron star faded and then headed back towards the core . as the reverse shock moves backwards its effects move down through the neutrino spectrum and might eventually cease to overlap with the forward shock . since the density jump across the reverse shock becomes quite small by the time this behavior occurs the reverse shock will produce a narrow spectral feature . for the simulation where @xmath52 , shown in fig . ( [ fig : rho.snv1d.8e35.t=1,1.5,2,2.5,3 ] ) , this expected pattern is difficult to see in the neutrino signal because the forward shock has not swept through the neutrino spectrum by the time the reverse shock turns around . but in the simulation with slightly larger energy deposition , @xmath57 , the forward shock moves much more quickly and has largely swept through the h resonance region by the time the reverse shock starts to make its way back to the core . for this simulation the backwards moving reverse shock is more visible in the neutrino signal . the crossing probability for this simulation at the snapshots shown in fig . ( [ fig : rho.snv1d.2e36.t=4,4.5,4.9 ] ) are presented in fig . ( [ fig : pc.snv1d.2e36.t=4,4.5,4.9 ] ) . the backwards moving reverse shock is responsible for the phase effects at @xmath208 in the middle panel . and then once the reverse shock has reached the core our boundary conditions there led to its reflection and so the reverse shock became a outward moving , weak forward shock and re - appeared in the neutrino spectrum as a spectral feature moving from low to high neutrino energies . this is the cause of the bump in @xmath175 seen at @xmath209 in the bottom panel . the h resonance crossing probability @xmath175 as a function of neutrino energy for the 1d sn model where @xmath57 . from top to bottom the snapshot times are @xmath181 , @xmath65 and @xmath66.,width=480 ] in summary , though it is apparent that the neutrino signal can vary considerably depending upon exactly how the reverse shock behaves the presence of strong phase effects - rapid oscillations with large amplitude in both time and energy of the crossing probability - are a notable signature of the presence of multiple shocks . therefore , there exists tremendous potential in a future detection of the a supernova neutrino signal for probing the inner hydrodynamics of the explosion . the notable difference between the one - dimensional and the two - dimensional models was the local density fluctuations that appeared in the latter . these local density enhancements / cavities affect only neutrinos along particular lines of sight so if one could determine their presence in the signal we would obtain an indication of the asphericity of the explosion . the most obvious method would be to compare different lines of sight to the same sn but obviously this is not practical . instead we must find another signature of asphericity that one might use . the h resonance crossing probability @xmath175 as a function of neutrino energy for a radial slice at @xmath74 through the 2d sn model . from top to bottom the snapshot times are @xmath60 , @xmath76 and @xmath210.,width=480 ] in figure ( [ fig : pc.snv2d.t=2.4,5.4,6.4 ] ) we show the crossing probability @xmath175 as a function of neutrino energy for the snapshot @xmath60 , @xmath76 and @xmath210 and a line of sight @xmath74 . features in the figure resemble those seen in the 1d results : the forward shock leads to a range of energies with @xmath211 as seen in the top panel , phase effects due to presence of the reverse shock lead to the high ` frequency ' oscillations of @xmath175 seen in the middle panel , and the turn around of the reverse shock can be seen in the bottom panel at @xmath212 where it appears , as before , as a narrow range of high frequency phase effects moving down through the spectrum . while a detailed analysis of these results might indicate a difference that is due to asphericity there is no striking feature that one can point to . the h resonance crossing probability @xmath175 as a function of neutrino energy for a radial slice at @xmath74 through the 2d sn model . from top to bottom the snapshot times are @xmath213 , @xmath214 and @xmath215.,width=480 ] however , at late times we found a change in the qualitative appearance of @xmath175 versus neutrino energy . the high frequency phase effects no longer dominate the entire spectrum and now lower frequency portions show up . we did not see anything comparable at late times in any of our 1d simulation results . the changing quality of the phase effects are seen in fig . ( [ fig : pc.snv2d.t=7.4,8,9 ] ) which shows the crossing probability at the h resonance as a function of neutrino energy for the same line of sight at @xmath74 through the 2d model . high frequencies can be seen above @xmath216 in the top panel and above @xmath217 in the middle panel but they have disappeared by @xmath215 . in all three panels we see that at lower neutrino energies the curve is smoother with only narrow patches of high frequency phase effects that moved up through the spectrum . the shocks have largely ceased affecting the lower neutrino energies by this time so it is the ` fluctuations ' upon the profile - both between the shocks and behind the reverse shock - which produce mildly adiabatic resonances , and/or resonances that are located very close to one another ( and possibly even overlap ) that give rise to these features in the figure . the effect upon neutrino flavor transport of supernovae density fluctuations was considered by loreti et al @xcite and later by fogli _ et al _ @xcite and friedland & gruzinov @xcite . these authors suggest that @xmath175 should tend to 1/2 in the presence of many small scale fluctuations . we note , however , that due to the exponential growth of the radial grid spacing the scale of features we can resolve by @xmath218 is of order @xmath219 and this is somewhat larger than the scale considered by fogli _ et al _ and friedland & gruzinov . in our formulation of neutrino mixing using s - matrices we can motivate this expected result of @xmath220 by dividing the density profile into domains such that within each there is a single neutrino resonance . the passage of a neutrino ( or antineutrino ) through each interval is described by an s - matrix which has exactly the same structure as equations ( [ eq : shnu ] ) or ( [ eq : shnubar ] ) as appropriate . the magnitudes of the elements @xmath154 and @xmath221 lie between zero and unity depending upon the adiabaticity of the resonance in that region . the multiplication of all these s matrices together ( in the correct order ) to achieve the net effect of passing through all the resonances will tend to produce a net result where either @xmath222 and @xmath223 or @xmath224 and @xmath225 are of equal magnitude . in our simulations we had cases where we found up to 13 resonances but have not been able to reproduce the anticipated limiting behavior . nevertheless , as shown in figs . ( [ fig : pc.snv2d.t=2.4,5.4,6.4 ] ) and ( [ fig : pc.snv2d.t=7.4,8,9 ] ) , we see a clear distinctive qualitative change in the crossing probabilities at late time in aspherical models . since the detection of neutrinos from sn 1987a in imb @xcite and kamiokande @xcite the number and size of the neutrino detectors have both increased in scale so that many thousands or even tens of thousands of neutrinos will be detected from the next galactic supernova . though many neutrino detectors are water cerenkov detectors that are best suited to detecting the electron antineutrinos from supernovae via the inverse-@xmath221 reaction on protons , other neutrino detection technologies have been demonstrated or explored that allow one to see the electron neutrinos and the @xmath226/@xmath32 flavors via nucleus - neutrino reactions such as @xmath227 , or neutral current reactions such as @xmath228 . the detection of these other neutrino flavors would provide a great deal of valuable information we could then use to learn about the sn explosion . for the purposes of demonstrating the temporal evolution of supernovae neutrino signals we shall consider two detector types : the water cerenkov detector because the majority of neutrino detectors are of this type , and a heavy water detector ( such as was used in the sudbury neutrino observatory ) since this detector was also considered by schirato & fuller @xcite . the largest neutrino detectors are water cerenkov detectors . these detectors are sensitive to both electron neutrinos and electron anti neutrinos via the reactions @xmath229 and @xmath227 @xcite with the inverse-@xmath221 reaction dominating the signal . the cross sections for the neutrino - nucleon reactions are well known but a good deal of uncertainty exists in the important neutrino - nucleus cross sections . a heavy water detector is sensitive to both charged - current and neutral current events so could be used to detect the @xmath226 and @xmath32 flavors . however obtaining energy resolution for the neutral current events is difficult so the observable quantity considered by schirato & fuller @xcite was the ratio of total charged - current to neutral current events . though temporal evolution of this ratio is somewhat smothered due to the fact that the event rates are integrated over the neutrino spectrum schirato & fuller demonstrated that variations of the ratio would occur at the level of @xmath230 . lead detectors are also capable of detecting the @xmath226/@xmath32 flux and have been studied by fuller , haxton & mclaughlin @xcite engel , mclaughlin & volpe @xcite and kolbe et al @xcite . to determine what any given detector will see we must fold in three effects : the initial neutrino spectra emitted by the proto - neutron star , the energy dependence of the cross sections in the detector and , thirdly , the energy resolution of the detector . there is a considerable degree of uncertainty about the emitted neutrino spectra . furthermore , neutrino self interactions - described in samuel @xcite , pastor , raffelt & semikoz @xcite and friedland & lunardini @xcite - in the region immediately above the proto - neutron star can alter the spectra after it s emission from the proto - neutron star but before the flux reaches the h resonance layer . there is a possibility that in some cases the spectra may be ` equilibrated ' by the time they get to the h resonance @xcite . in such circumstances the changes in the density profile and the neutrino mixing described in previous sections would not lead to any observable changes in a detector except for the gradual fading of the signal . we shall consider the case where such equilibration does not occur and experiment with an ` alpha fit ' power law for the spectrum exiting the neutrinosphere motivated both by its analytic simplicity and by recent supernova core monte carlo simulations @xcite . for the time dependence of the luminosity we use an exponential decay with a time constant of @xmath231 . the emitted differential spectra are then @xmath232\,\exp[-t/\tau ] . \label{eq : powlaw}\ ] ] where @xmath233 is the luminosity , @xmath234 the mean energy and @xmath154 is the ` pinch ' parameter . we adopt a hierarchy of mean energies : @xmath235 , @xmath236 and @xmath237 ( where , again , @xmath238 stands for either @xmath239 or @xmath240 ) and the luminosities are @xmath241 , @xmath242 . fluxes of electron antineutrinos ( top panel ) and the sum of @xmath243 neutrino fluxes ( bottom panel ) for completely adiabatic , i.e. @xmath244 ( solid ) and non - adiabatic , i.e. @xmath245 ( dashed ) evolution . the hierarchy is inverted , the pinch parameter was set to @xmath246 and the vacuum mixing angles are @xmath247 , @xmath248 . the distance to the sn was set at @xmath249 and the time @xmath31 to @xmath250.,width=480 ] the fluxes of electron neutrinos ( top panel ) and the sum of @xmath243 neutrino fluxes ( bottom panel ) for completely adiabatic , i.e. @xmath244 ( solid line ) and non - adiabatic , i.e. @xmath245 ( dashed ) evolution . here the hierarchy is normal but otherwise all parameters are the same as those used in fig . ( [ fig : fnu.alpha=3.ih.phbar=0,1]).,width=480 ] in figs . ( [ fig : fnu.alpha=3.ih.phbar=0,1 ] ) and ( [ fig : fnu.alpha=3.nh.ph=0,1 ] ) we present the neutrino spectra that would be observed at earth from a sn at a distance of @xmath251 if the evolution were either completely adiabatic or completely non - adiabatic for example through a single h resonance , i.e. @xmath252 or @xmath245 in either the inverted or normal hierarchy . the two curves in each panel of the figures therefore represent extremes in the sense that the actual flux at a given energy would fall somewhere in between the two curves . we do not plot the electron antineutrino flux for the normal hierarchy or the electron neutrino flux for the inverted hierarchy because these fluxes are always adiabatic and would only fade exponentially with time . in each panel of the two figures the pinch parameter was set to @xmath246 , other than a change in shape the general behavior of the curves at other values of @xmath154 is unchanged . there are many points to notice in each figure . the first is that for any given neutrino flux and hierarchy there is a particular energy where the two curves cross . this will occur as long as the average energies of @xmath253 flavor neutrinos are greater than that of the @xmath0 and @xmath254 . the crossing can be seen best in the @xmath255 spectrum in the normal hierarchy . we define the energies where the curves cross as the critical energies . the critical energies vary with both flavor and the hierachy . at the critical energy for a particular hiearachy and flavor the adiabaticity of the neutrinos as they propagate through the profile is irrelevant and the flux of that flavor and at that energy through a detector is independent of @xmath175 . the other significant feature to notice is the large difference between the two curves for the electron neutrinos and electron antineutrinos for energies below @xmath256 . this wide separation of the extremal fluxes at low to medium neutrino energies , @xmath257 , suggests that we focus our attention upon this portion of the spectrum because it is here that changes in adiabaticity will lead to the most significant changes in the flux . the flux at higher energies will also change with the evolving adiabaticity but to a lesser degree because the fluxes are so small . thus if the evolution of the neutrinos evolves from being adiabatic to non - adiabatic then , in the case of a normal hierarchy , the flux of electron neutrinos will increase for those energies below the critical energy and decrease above . similar behavior will occur for antineutrinos in the inverted hierarchy . the decrement above the critical energy is difficult to see in the figures because the critical energies are all far into the tail of the spectra . for the @xmath258 neutrino fluxes the opposite change occurs : a transition to non - adiabaticity leads to lower fluxes of these flavors at energies below the critical energy and higher fluxes above . having determined the effects of the evolving density profile upon the neutrino mixing we are now in a position to examine the actual neutrino signal within a detector . as discussed in section [ sec : pfcp ] , the actual neutrino flux of a particular flavor through a detector differs from that emitted from the neutrinosphere . in the simplest scenario , if the neutrinos propagate through the outer layers of the sn adiabatically then the survival probabilities @xmath259 and @xmath260 are found by setting @xmath261 in equations ( [ eq : ppbarnh ] ) and ( [ eq : ppbarih ] ) . for the normal hierarchy @xmath262 and @xmath263 : for the inverse hierarchy @xmath264 and @xmath265 . using the value of @xmath266 we used to calculate @xmath175 and @xmath267 based upon the solar mixing parameters , we see that for the normal hierarchy the amount of the emitted @xmath268 flux present in the electron neutrino flux at a detector here on earth is minuscule . similarly we see that for an inverted hierarchy the amount of the emitted electron antineutrino flux present in any detected electron antineutrino flux is equally small . thus if the mass hierarchy is normal , @xmath269 is not too small , and the neutrino propagation adiabatic then the electron neutrino detectors on earth will essentially be detecting the emitted @xmath270 and @xmath271 flux . or if the hierarchy is inverted , again @xmath269 is not too small , and the neutrino propagation adiabatic then the electron antineutrino flux at earth is the emitted @xmath272 and @xmath273 . in order to see the emitted electron neutrino or antineutrino flux we would need to detect the @xmath270 , @xmath272 , @xmath273 and/or @xmath271 flavors or seek the small amount of the emitted @xmath268/@xmath254 present in the detected @xmath0/@xmath255 flux . in either case , it will be very difficult to reconstruct the the @xmath268 or @xmath254 spectrum that was originally emitted from the proto - neutron star . the evolution of the density profile changes this conclusion . depending upon the scenario , either @xmath259 ( for a normal hierarchy ) or @xmath260 ( for an inverted hierarchy ) will change with time and so the proportions of the emitted fluxes present in the fluxes at earth will also change . note that @xmath259 and @xmath260 do not both evolve with time within in the same scenario ; one will stay constant . this means there is a potential to separate the time evolution of the emitted flux from the neutrinosphere from the evolution of the density profile through which the neutrinos propagate . as the star explodes and the forward shock eventually reaches the h resonances the passage of portions of the neutrino spectrum through the density profile becomes non - adiabatic . whether it is the neutrinos or antineutrinos that are affected is determined by the hierarchy . when the passage of the neutrinos is completely non - adiabatic we see that @xmath259 or @xmath260 change to @xmath264 or @xmath263 . again using @xmath274 and @xmath267 we see that @xmath275 , @xmath276 . in effect the forward shock of such explosions acts like a window sweeping across either the neutrino or antineutrino spectra through which we observe an altered mixture of the emitted components . for neutrino or antineutrino energies below the critical energies this leads to an enhanced @xmath0 or @xmath255 flux as shown in figs . ( [ fig : fnu.alpha=3.ih.phbar=0,1 ] ) and ( [ fig : fnu.alpha=3.nh.ph=0,1 ] ) . the ratio of charged current event rates to neutral current event rates in a heavy water detector as a function of time for neutrino propagation through the @xmath67 radial slice of the the 2d simulation . the top panel is for an inverted hierarchy , the lower panel for a normal hierarchy . in each the lines represent different spectral parameters : @xmath277 ( solid ) , @xmath278 ( dotted ) , @xmath246 ( long dashed ) , @xmath279 ( dash dot ) and @xmath280 ( long dash - double dot).,width=480 ] the ratio of charged current event rates to neutral current event rates in a heavy water detector as a function of time for neutrino propagation through the 1d simulation with @xmath43 . the top panel is for an inverted hierarchy , the lower panel for a normal hierarchy . in each the lines are : @xmath277 ( solid ) , @xmath278 ( dotted ) , @xmath246 ( long dashed ) , @xmath279 ( dash dot ) and @xmath280 ( long dash - double dot).,width=480 ] the ratio of charged current event rates to neutral current event rates in a heavy water detector as a function of time for neutrino propagation through the 1d simulation with @xmath52 . the top panel is for an inverted hierarchy , the lower panel for a normal hierarchy . in each the lines are : @xmath277 ( solid ) , @xmath278 ( dotted ) , @xmath246 ( long dashed ) , @xmath279 ( dash dot ) and @xmath280 ( long dash - double dot).,width=480 ] the ratio of charged current event rates to neutral current events rates in a heavy water detector as a function of time for neutrino propagation through the 1d simulation with @xmath57 . the top panel is for an inverted hierarchy , the lower panel for a normal hierarchy . in each the lines are : @xmath277 ( solid ) , @xmath278 ( dotted ) , @xmath246 ( long dashed ) , @xmath279 ( dash dot ) and @xmath280 ( long dash - double dot).,width=480 ] the first detector type we consider is a heavy water , sno like , detector and the signal we focus upon is the ratio of charged current ( cc ) - both from electron neutrinos and electron antineutrinos - to neutral current ( nc ) event rates . this ratio of event rates is the same signal considered by schirato & fuller @xcite even though the @xmath281 counters and then the nacl inserted into sno gave the detector some capability of distinguishing between the @xmath0 and @xmath255 events . to some extent the uncertainty of the initial spectra can be removed by considering this ratio of event rates though by no means is it eliminated . for the results that follow we calculate the event rates using the nakamura _ @xcite neutrino - deuteron cross sections . the basic understanding of what happens to the @xmath0 or @xmath255 flux helps us understand the temporal evolution of the cc / nc ratio shown in fig . ( [ fig : ccnc.snv2d ] ) for the 2d simulation and for the 1d simulations in figs . ( [ fig : ccnc.snv1d.1e35 ] ) , ( [ fig : ccnc.snv1d.8e35 ] ) and ( [ fig : ccnc.snv1d.2e36 ] ) . for all combinations of hierarchy , pinch parameter and explosion energy the ratio begins to vary at around @xmath282 into the simulation . the initial increase in the ratio is due to larger charged current rates ( the neural current rate is flavor blind ) caused by greater electron neutrino flavor content ( for the nh ) or electron antineutrino flavor ( for the ih ) . the change is a clear signature of a transition from adiabaticity to non - adiabaticity that occurs when the forward shock reaches the h resonances for neutrinos with @xmath283 . as expected , the ratio varies more strongly for an inverted hierarchy due to the greater change in the electron antineutrino spectrum when the profile becomes non - adiabatic than for the electron neutrinos as indicated by figs . ( [ fig : fnu.alpha=3.ih.phbar=0,1 ] ) and ( [ fig : fnu.alpha=3.nh.ph=0,1 ] ) . more detailed examination of the figures reveals that for the particular case of the weak explosion - fig . ( [ fig : ccnc.snv1d.1e35 ] ) - there is evidence of the shock effects moving up through the neutrino spectra because at late times we see the ratio dips beneath it s @xmath284 value . for the other simulations the cc / nc ratio exhibits a great deal of variability at the level of a few percent that continues for several seconds . we also notice that the variability of the @xmath285 and @xmath286 1d simulations ceased after @xmath287 when the forward shock had swept through the spectrum and the reverse shock had turned around while the cc / nc ratio for the 2d simulation was sustained for the entire @xmath288 . while observing a period of fluctuations in the cc / nc ratio would indicate that the explosion and the density profile was dynamic it is difficult to be more quantitative about the presence and motion of the various features of the profile from the use of this ratio of event rates alone . the second example of a detector we examine is a water cerenkov type since this is the dominant supernova neutrino detection technology . we focus upon the positron spectrum created by inverse @xmath221 reactions of antineutrinos upon protons . the temporal variations of the neutrino signal we have discussed in the paper would therefore require an inverted hierarchy in order to be visible in this signal . the positron spectrum seen in a detector , @xmath289 , is given by @xmath290 where @xmath291 is the number of protons in the detector , @xmath292 is the electron antineutrino flux and @xmath293 is the differential cross section . for our analysis we used the cross sections of strumia and vissanti @xcite . the positron spectrum allows us to see the effect of the density profile upon the neutrino flux as a function of neutrino energy but , before we show our results , we firstly consider the relationship between the neutrino energy and the positron energy and our ability to resolve the features of the neutrino spectrum . the ` mean ' positron energy @xmath294 as a function of the neutrino energy ( solid line ) given by equation ( 16 ) in strumia and vissanti @xcite . the dashed line is @xmath295.,width=480 ] the positron energy and neutrino energy are closely correlated as shown in fig . ( [ fig : meaneae.eps ] ) . features in the neutrino spectrum at a particular energy will appear in the positron spectrum at more or less the same energy . the super - k energy resolution error , @xmath296 , ( solid line ) and the range of positron energies , @xmath297 , ( dashed ) as a function of the positron kinetic energy.,width=480 ] but the positron spectrum is smoothed relative to the neutrino spectrum by two effects : the distribution of positron energies for a given neutrino energy , and the energy resolution of the detector . for a given neutrino energy the positrons emerge with a range of energies which we may invert so that for a given positron energy , @xmath298 , there is a range of neutrino energies , @xmath297 , from which it could have arisen . this range . ] is given by @xmath299 where @xmath300 is the momentum of the positron and @xmath301 , @xmath302 and @xmath303 are the masses of the neutron , proton and electron respectively . @xmath297 is shown in fig . ( [ fig : skerror , spread ] ) as a function of the positron kinetic energy . and then for the energy resolution of the detector we used a filter so that the spectrum of positrons at energy @xmath298 is given by @xmath304 and we take @xmath305 to be a normalized gaussian with a mean @xmath306 and a variance @xmath307 given by @xmath308}\right)^{0.3467}.\ ] ] this formula is our fit to the energy resolution of the super - k detector @xcite . the energy resolution of the detector , @xmath296 , is also shown in figure ( [ fig : skerror , spread ] ) as a function of the positron kinetic energy . both smearing effects , the range , @xmath297 , of neutrino energies that can produce a positron with energy @xmath298 , and the detector resolution , @xmath296 , increase with @xmath298 and we notice that beyond @xmath309 it is the range @xmath297 that determines that determines the scale of the features we can resolve . below @xmath309 the detector resolution limits out ability to see features of the neutrino spectrum . future detectors will improve upon the energy resolution used here and so - since this is also the energy range where we should focus our attention because it is where the electron antineutrino flux will vary the largest due to the changing adiabaticity - in all our figures we plot both the spectrum as it would be measured with the super - k energy resolution and with a detector that has perfect resolution . of course there is a third effect one needs to consider : the poisson error due to the finite number of detected events . this does not smooth out features of the spectrum but rather make them difficult to detect . in what follows we shall ignore the poisson statistics . . in the top panel the times are @xmath250 ( solid ) and @xmath41 ( long dashed ) , in the middle panel @xmath310 ( solid ) and @xmath64 ( long dashed ) , and in the bottom panel @xmath311 ( solid ) and @xmath312 ( long dashed ) . in all panels the short dashed lines indicate the spectrum convolved with our adopted detector energy resolution . [ fig : faevse.1e35],width=480 ] in fig . ( [ fig : faevse.1e35 ] ) we plot the positron spectrum as a function of positron energy in the case of the inverted hierarchy at various snapshot times using the neutrino flux we calculated for the weakest 1d simulation , @xmath313 . with this spectral information , we can begin to see the features of the explosion . the shock is clearly seen at @xmath314 at @xmath41 in the perfectly resolved spectrum . by @xmath310 the sharp change in spectrum has moved up to @xmath315 and by @xmath64 we begin to see the backside of the shock at @xmath314 and again at @xmath311 at @xmath316 and @xmath312 at @xmath317 . the shock is a less dramatic feature in the spectra convolved with our adopted detector energy resolution but is visible particularly at @xmath41 . given enough events even present detector technology would allow us to see the shock move up through the positron spectrum from which we could begin to glean information about the explosion . . the curves are @xmath189 ( long dashed ) , @xmath191 ( dash dot ) , and @xmath193 ( solid ) . the short dashed lines indicate the spectrum convolved with our adopted detector energy resolution . [ fig : faevst.1e35],width=480 ] the explosion features also appear in the evolution of the positron spectrum at particular energies as a function of time . this is shown in figure ( [ fig : faevst.1e35 ] ) where we plot the positron spectrum for @xmath189 , @xmath191 and @xmath193 . the general trend for each energy is @xmath318 since this was the time dependence we used for the emitted spectra but on top of this exponential decay the shock is obviously visible for all three chosen energies in both filtered and unfiltered curves . the width of the forward shock feature grows with @xmath94 as we noted earlier . . in the top panel the times are @xmath54 ( solid ) and @xmath319 ( long dashed ) , in the middle panel @xmath60 ( solid ) and @xmath310 ( long dashed ) , and in the bottom panel @xmath46 ( solid ) and @xmath66 ( long dashed ) . in all panels the short dashed lines indicate the spectrum convolved with our adopted detector energy resolution.[fig : faevse.8e35],width=480 ] in fig . ( [ fig : faevse.8e35 ] ) we plot the positron spectrum for the more powerful explosion with @xmath285 at various snapshot times . the phase effects we saw in the crossing probability for this model appear also in the spectrum for perfect detector resolution . they are largest in the range of @xmath320 where the difference between adiabatic and non - adiabatic electron antineutrino flux is largest and where the distribution of positron energies for a given neutrino energy is narrow . at higher energies the phase effects largely disappear because of the smoothing effect of the positron distribution . , and an inverted hierarchy . the curves are @xmath191 ( dash dot ) , and @xmath193 ( solid ) . the short dashed lines indicate the spectrum convolved with our adopted detector energy resolution.[fig : faevst.8e35],width=480 ] this smoothing is also seen in the plot of the spectrum at various energies as a function of time shown in fig . ( [ fig : faevst.8e35 ] ) since the @xmath321 positrons has much greater variability than then @xmath322 positrons even in the filtered spectrum . . in the top panel the time is @xmath64 , in the middle panel @xmath65 , and in the bottom panel @xmath66 . in all panels the short dashed lines indicate the spectrum convolved with our adopted detector energy resolution.[fig : faevse.2e36],width=480 ] we can even find the signature of a reverse shock that stalled and headed back to the core . in fig . ( [ fig : faevse.2e36 ] ) we plot the positron spectrum for the more energetic @xmath286 1d explosion . the snapshot times are the same as those shown in fig . ( [ fig : rho.snv1d.2e36.t=4,4.5,4.9 ] ) for the density and fig . ( [ fig : pc.snv1d.2e36.t=4,4.5,4.9 ] ) for the crossing probability for this model . the returning reverse shock appears at @xmath323 at @xmath65 which matches what we expect from the crossing probability seen in the middle panel of fig . ( [ fig : pc.snv1d.2e36.t=4,4.5,4.9 ] ) . ( solid ) and @xmath60 ( long dashed ) , in the middle panel @xmath210 ( solid ) and @xmath213 ( long dashed ) , and in the bottom panel @xmath214 ( solid ) and @xmath215 ( long dashed ) . in all panels the short dashed lines indicate the spectrum convolved with our adopted detector energy resolution.[fig : faevse.snv2d],width=480 ] finally in fig . ( [ fig : faevse.snv2d ] ) we plot the spectrum for the 2d simulation at various snapshot times selected so that they match as closely as possible the snapshots we have shown of the profile and/or the crossing probability . the forward shock and phase effects can be seen here too in the unfiltered spectrum and the inward moving reverse shock is responsible for the phase effects seen at @xmath316 at @xmath210 and in fig . ( [ fig : pc.snv2d.t=2.4,5.4,6.4 ] ) . other features upon the positron spectrum can also be related to various features shown in the crossing probability for this model . the next galactic supernova has the potential to provide a great deal of information about both neutrino mixing and about the supernova . the temporal and spectral evolution of the signal is altered as the star explodes due to the changes in the density profile of the star . the forward shock should be easily detected even with present detector technology as either a feature sweeping up through the positron spectrum in a water cerenkov detector or as a sudden increase in event rates for a detector with no spectral information such as a heavy water detector . once identified , and if the progenitor were known , the information can be used to measure the shock speed , the strength of the shock and potentially also the stalled shock time delay . the reverse shock can be spotted qualitatively as the cause of fluctuations in either event rates or upon the positron spectrum . there is also the possibility for the reverse shock to be seen if its motion is reversed and it heads back to the core . the neutrino signal may also contain hints that the supernova was aspherical as a qualitative change in the phase effects . the authors are very grateful to john blondin for his helpful advice with the hydrodynamic simulations . this work was supported by the u.s . department of energy at umn under grant de - fg02 - 87er40328 and at ncsu under grant de - fg02 - 02er41216 .. j. f. beacom , r. n. boyd and a. mezzacappa , , * 63 * , 073011 ( 2001 ) [ arxiv : astro - ph/0010398 ] . g. c. mclaughlin and r. surman , , * 75 * , 023005 ( 2007 ) [ arxiv : astro - ph/0605281 ] . d. h. hartmann , r. d. hoffman , s. e. woosley and w. c. haxton , nucl . a * 527 * , 663c ( 1991 ) . a. heger , e. kolbe , w. c. haxton , k. langanke , g. martinez - pinedo and s. e. woosley , phys . b * 606 * , 258 ( 2005 ) [ arxiv : astro - ph/0307546 ] . s. e. woosley , j. r. wilson , g. j. mathews , r. d. hoffman and b. s. meyer , , * 433 * , 229 ( 1994 ) . r. surman , g. c. mclaughlin and w. r. hix , , * 643 * , 1057 ( 2006 ) [ arxiv : astro - ph/0509365 ] . j. pruet , r. surman and g. c. mclaughlin , , * 602 * , l101 ( 2004 ) [ arxiv : astro - ph/0309673 ] . g. c. mclaughlin , g. m. fuller and j. r. wilson , , * 472 * , 440 ( 1996 ) [ arxiv : astro - ph/9701114 ] . g. c. mclaughlin and g. m. fuller , , * 464 * , l143 ( 1996 ) [ arxiv : astro - ph/9603082 ] . b. s. meyer , g. c. mclaughlin and g. m. fuller , , * 58 * , 3696 ( 1998 ) [ arxiv : astro - ph/9809242 ] . t. yoshida , t. kajino , h. yokomakura , k. kimura , a. takamura and d. h. hartmann , arxiv : astro - ph/0606042 . g. m. fuller , w. c. haxton and g. c. mclaughlin , , * 59 * , 085005 ( 1999 ) [ arxiv : astro - ph/9809164 ] . a. s. dighe and a. y. smirnov , , * 62 * , 033007 ( 2000 ) [ arxiv : hep - ph/9907423 ] . j. engel , g. c. mclaughlin and c. volpe , , * 67 * , 013005 ( 2003 ) [ arxiv : hep - ph/0209267 ] . c. lunardini and a. y. smirnov , jcap * 0306 * , 009 ( 2003 ) [ arxiv : hep - ph/0302033 ] . a. friedland and c. lunardini , , * 68 * , 013007 ( 2003 ) [ arxiv : hep - ph/0304055 ] . r. f. sawyer , , * 72 * , 045003 ( 2005 ) [ arxiv : hep - ph/0503013 ] . h. duan , g. m. fuller , j. carlson and y. z. qian , , * 97 * , 241101 ( 2006 ) [ arxiv : astro - ph/0608050 ] . s. hannestad , g. g. raffelt , g. sigl and y. y. y. wong , , * 74 * , 105010 ( 2006 ) [ arxiv : astro - ph/0608695 ] . j. p. kneller and g. c. mclaughlin , , * 73 * , 056003 ( 2006 ) [ arxiv : hep - ph/0509356 ] . b. dasgupta and a. dighe , [ arxiv : hep - ph/0510219 ] . blondin , j. m. , mezzacappa , a. , & demarino , c. 2003 , , * 584 * , 971 blondin , j.m . and mezzacappa , a. , , * 642 * , 401 ( 2006 ) ohnishi , n. , kotake , k. and yamada , s. , , * 641 * , 1018 ( 2006 ) scheck , l. , kifonidis , k. , janka , h .- th . and mueller , e. , astro - ph/0601302 a. burrows , e. livne , l. dessart , c. d. ott and j. murphy , [ arxiv : astro - ph/0610175 ] . schirato , r.c . and fuller , g.,m . , astro - ph/0205390 takahashi , k. , sato , k. , dalhed , h.e . and wilson , j. r. , astropart . phys , 20 , 1989 ( 2003 ) fogli , g. l. , lisi , e. , and mirizzi , a. and montanino , d. , , * d68 * , 033005 ( 2003 ) a.b . balantekin , and j.f . beacom , , * 54 * , 6323 ( 1996 ) a.n . ioannisian and a.yu . smirnov , , * 93 * , 241801 ( 2004 ) a.n . _ et al . _ _ , , * 71 * , 033006 ( 2005 ) e.kh . akhmedov , m. maltoni and a.yu . smirnov , , * 95 * , 211801 ( 2005 ) f. n. loreti , y. z. qian , g. m. fuller and a. b. balantekin , , * 52 * , 6664 ( 1995 ) [ arxiv : astro - ph/9508106 ] . fogli , g. , lisi , e. , mirizzi , a. and montanino , d. , , jcap , * 6 * , 12 ( 2006 ) friedland , a. and gruzinov , a. , arxiv : astro - ph/0607244 ( 2006 )
the dynamical behavior of large populations of nonlinearly coupled oscillators may describe many phenomena in physics , biology and medicine , @xcite . in particular synchronization of mean - field coupled phase oscillators with different natural frequencies is nicely illustrated by kuramoto s well - known and extesively analyzed model @xcite . to describe certain biological phenomena , inertial effects should be added to this model . in @xcite , ermentrout revisited the special problem of self - synchronization in populations of fireflies of a certain kind ( _ the pteroptyx malaccae _ ) . compared to observed behavior , the approach to oscillator synchronization as described by the kuramoto model seems to be too fast . thus a more appropriate adaptive frequency model has been proposed in @xcite , where the natural frequency of an oscillator is a new independent variable , which is allowed to vary with time . thus an oscillator is described by its phase and frequency . from the mathematical standpoint , the new model is governed by a system of coupled second - order differential equations containing inertial terms , in contrast to the system of first - order differential equations governing the kuramoto model . indeed inertia slows down synchronization and this may result in better agreement between theory and experimental measurements . other possible biological applications of ermentrout - type models include after - effects in alterations of circadian cycles in mammalians , cf . @xcite a different set of applications for oscillators with inertia include power systems described by the swing equations @xcite , or by hamilton equations @xcite . an important technologically relevant application is the study of superconducting josephson junctions arrays @xcite . here inertial terms describe the effect of nonzero electrical capacitance . such effect is often far from being negligible , and it is absent in the kuramoto model , @xcite . in this paper , we consider the model equations of ref . @xcite , @xmath2 where @xmath3 , @xmath4 and @xmath5 denote phase , frequency and natural frequency of the @xmath6th oscillator , respectively . the natural frequencies are distributed with probability density @xmath0 , which may have a single maximum ( _ unimodal _ distribution ) , or several peaks ( _ multimodal _ distribution ) . the positive parameters @xmath7 and @xmath8 are the `` inertia '' and the coupling strength , respectively . the complex order parameter defined by @xmath9 measures phase synchronization : @xmath10 if the oscillators are synchronized and @xmath11 if not . finally , @xmath12 s are independent identically distributed gaussian white noises , with @xmath13 . white noise terms were not included in @xcite . when the inertial terms vanish , @xmath14 , eqs . ( [ 1 ] ) and ( [ 2 ] ) are exactly the kuramoto model . typically , _ n _ is very large , and oscillator synchronization is conveniently analyzed in the limiting case of infinitely many oscillators . in this limit , models with mean - field coupling are described by an evolution equation for the one - oscillator probability density , @xmath15 , @xcite . for the present model this equation is @xcite @xmath16 -\omega\frac{\partial \rho}{\partial\theta}\ , , \label{fpe}\end{aligned}\ ] ] where the order parameter is now given by @xmath17 equations ( [ fpe ] ) and ( [ order ] ) should be supplemented with appropriate initial and boundary data ( @xmath18 is @xmath19-periodic in @xmath20 and has suitable decay behavior as @xmath21 ) plus the normalization condition , @xmath22 differentiating @xmath23 with respect to time , and then using eq . ( [ fpe ] ) itself , together with periodicity in @xmath20 and decay in @xmath24 , we find that the left side of ( [ norma ] ) is time independent . normalization to unity of the initial probability density then implies ( [ norma ] ) for the solution of ( [ fpe ] ) . in this paper , we study oscillator synchronization and transition from incoherence to synchronization in the model ( [ fpe ] ) - ( [ norma ] ) . the incoherent solution of ( [ fpe ] ) - ( [ norma ] ) ( or simply _ incoherence _ ) is a stationary solution which is independent of @xmath20 . this solution asigns equal probability to all angles and has @xmath25 ( no order ) , so it corresponds to lack of oscillator synchronization . there are synchronized solutions which branch off from incoherence as the coupling among oscillators is increased . these bifurcations describe the synchronization transitions , which we have analyzed and compared to the corresponding ones in the kuramoto model . our main results are that inertia : ( i ) may stabilize incoherence , making it harder to synchronize oscillators , and ( ii ) it may harden the synchronization transition . in the kuramoto model ( @xmath14 ) or with oscillators with identical natural frequencies , the synchronization transition is soft ( supercritical bifurcation ) , whereas it may become hard ( subcritical bifurcation ) if the distribution of natural frequencies has a nonzero spread ( unimodal lorentzian distribution ) or several peaks ( e.g. , a discrete bimodal distribution ) . the methods we have used in our analysis are similar to those previously employed in the kuramoto model @xcite : linear stability of incoherence , bifurcation analysis , high - frequency singular perturbations and numerical solutions . an important difference is that now we do not have an explicit functional form for stationary solutions ( as it was the case for the kuramoto model ) . this has led us to use mode - coupling expansions of the solution and solving the corresponding mode - coupling equations . solutions of these equations in close form are not always accesible , so that we have introduced some closure assumptions . the results of these uncontrolled assumptions have been compared to direct simulations or to approximate amplitude equations and found reasonable in the limit of small inertia . the rest of the paper is as follows . in section ii , we find the incoherent solution and study its linear stability for several natural frequency distributions . results are compared with those obtained in the massless case @xcite . it is found that the critical coupling needed to destabilize incoherence increases with @xmath7 for `` unimodal '' frequency distributions of the lorentzian type . the critical coupling is independent of @xmath7 when @xmath26 . in this case the time needed to reach synchronization increases as @xmath7 increases . if @xmath27/2 $ ] ( discrete bimodal distribution ) , the critical coupling may grow or decrease with @xmath7 depending on the values of @xmath28 . in section iii , we construct other stationary solutions by two procedures : an amplitude expansion for solutions branching off from incoherence and a general expansion in hermite polynomials which is appropriately truncated . an exact analytical solution is obtained if @xmath29 ( cf . @xcite ) , while analytical approximations for small @xmath7 and @xmath30 are available in the general case . we have observed that inertia tends to _ harden _ the synchronization transition : in the kuramoto model ( @xmath14 ) or with oscillators with identical natural frequencies , the synchronization transition is soft ( supercritical bifurcation ) , whereas it becomes hard ( subcritical bifurcation ) in the cases of unimodal lorentzian or discrete bimodal frequency distributions . in section iv , we obtain approximations to stable time - dependent solutions of eq . ( [ fpe ] ) in the `` high frequency limit '' , @xmath31 @xcite . there are partially synchronized nonlinearly stable solutions of standing wave type , as in the kuramoto model @xcite ) . finally , numerical results are presented in section v , and compared to the approximate or exact solutions of previous sections . two appendices at the end are devoted to technical details . the incoherent solution is a @xmath20-independent stationary solution of ( [ fpe ] ) . its order parameter is @xmath25 according to ( [ order ] ) . then eqs.([fpe ] ) , decay as @xmath21 and the normalization condition ( [ norma ] ) yield the incoherent solution : @xmath32 to analyze its linear stability , let us consider a small disturbance about incoherence , @xmath33 where @xmath34 . normalization of @xmath15 then implies @xmath35 we now introduce ( [ perturb ] ) into ( [ fpe ] ) and ( [ order ] ) and equate like terms in @xmath36 . to order @xmath36 , the result is @xmath37 - \frac{d}{m^2}\frac{\partial^2 \eta}{\partial \omega^2 } = \nonumber\\ -\frac{k\ , \frac{\partial\rho_{0}}{\partial\omega}}{m}\ , \int_0^{2\pi}\int_{-\infty}^{+\infty}\int_{-\infty}^{+\infty } \eta(\phi,\omega,\omega , t)\ , \nonumber\\ \times \sin(\phi-\theta)\ , g(\omega ) d\omega d\omega d\phi . \label{eta}\end{aligned}\ ] ] we now insert a trial solution @xmath38 ( which is @xmath19-periodic in @xmath20 ) into ( [ eta ] ) , thereby obtaining @xmath39 where we have defined the scalar product @xmath40 notice that @xmath41 and that @xmath42 because of the normalization condition ( [ zero ] ) . equation ( [ eqb ] ) can be transformed into a nonhomogeneous parabolic cylinder equation by the following change of variable : @xmath43\ , \beta_n ( w;\omega,\lambda ) , \label{beta1}\\ w = \sqrt{{m\over d}}\ , ( \omega - \omega + 2ndi ) . \label{beta2}\end{aligned}\ ] ] inserting ( [ beta1 ] ) and ( [ beta2 ] ) into ( [ eqb ] ) , we obtain @xmath44\ , \beta_n = \quad \nonumber\\ i\pi k { \partial\rho_{0}\over \partial \omega } e^{{1\over 4}(w-2i\sqrt{md})^{2}}\ , \langle 1 , e^{-{1\over 4}(w-2i\sqrt{md})^{2 } } \beta_1 \rangle\ , \delta_{n,1}. \label{beta3}\end{aligned}\ ] ] ( recall that @xmath45 when using the definition of scalar product ) . let us assume now that @xmath46 and that @xmath30 is a fixed real number . then the right hand side of ( [ beta3 ] ) is zero and the resulting equation has the following eigenvalues @xmath47 associated to the eigenfunctions @xmath48 which are independent of @xmath49 and @xmath30 . in this formula , @xmath50 and @xmath51 are the parabolic cylinder function and the hermite polynomial of index @xmath52 , respectively @xcite . the eigenvalues @xmath53 of ( [ beta4 ] ) , with @xmath54 , @xmath55 and @xmath30 belonging to the support of @xmath0 , constitute the continuous spectrum of the linear stability problem . in fact , a nonhomogeneous linear problem with a homogeneous part given by ( [ beta3 ] ) can not be solved for an arbitrary source term if @xmath56 . notice that the continuous spectrum lies to the left side of the imaginary axis if @xmath57 and @xmath58 . then the `` eigenvalues '' ( [ beta4 ] ) have negative real parts ( and therefore correspond to stable modes ) . as we have already said , the neutrally stable modes with @xmath59 have zero amplitude due to the normalization condition ( [ zero ] ) . if @xmath60 , we can solve ( [ eqb ] ) by means of an expansion in eigenfunctions @xmath50 , @xmath61 . to obtain the generalized fourier coefficients of @xmath62 , we multiply both sides of ( [ eqb ] ) by @xmath50 and integrate over @xmath63 . as @xmath64 ( orthogonality condition , cf . 7.711.1 of @xcite ) , the result is @xmath65 where @xmath66 once we have found @xmath62 , we can calculate the scalar product @xmath67 . since this scalar product appears as a factor in both sides of the resulting expression , we can divide by it , thereby obtaining an eigenvalue equation for @xmath68 : @xmath69 in appendix a , we show that this equation may be rewritten as @xmath70 where @xmath71 is defined as @xmath72 the result of evaluating this integral is ( cf . appendix a ) : @xmath73 inserting ( [ beta11 ] ) in ( [ beta9 ] ) , we obtain @xmath74 as @xmath75 , this equation coincides with that obtained for the kuramoto model , @xcite . equation ( [ dispersion3 ] ) can be rewritten in terms of incomplete gamma functions as follows @xcite : @xmath76 { g(\omega ) d\omega\over ( md)^{m(\lambda+d+i\omega ) } } \nonumber\\ = \int_{-\infty}^{\infty } \left[1 - { ( \lambda + i\omega)m e^{m d } \gamma ( m(\lambda+d+i\omega),md)\over ( md)^{(\lambda+d+i\omega)m } } \right ] \nonumber\\ \times g(\omega)\ , d\omega\ , , \label{gammaf}\end{aligned}\ ] ] where @xcite @xmath77 > from now on , we analyze eq . ( [ dispersion3 ] ) for special frequency distributions : + \(a ) _ unimodal frequency distribution _ , @xmath1 . in this case we show that if @xmath78 , then @xmath79 . thus , the eigenvalues that may acquire a positive real part are real . then the critical coupling is obtained by setting @xmath80 . by subtracting from eq.([dispersion3 ] ) its complex conjugate , we obtain @xmath81 where @xmath82}.\ ] ] notice than the even function @xmath83 decreases monotonically with im@xmath84 . on the other hand , @xmath85 tends to zero , as @xmath86 , and @xmath87 thus @xmath85 does not vanish at finite values of im@xmath68 , and therefore the only solution of ( [ zeros ] ) is im@xmath80 . setting now @xmath80 , equation ( [ dispersion3 ] ) yields the critical coupling @xmath88 , @xmath89 = 1 , \quad \mbox{and therefore } \nonumber\\ k_c=2d.\label{unimodalkc}\end{aligned}\ ] ] figures [ lsu1 ] and [ lsu2 ] show the largest eigenvalue @xmath68 as a function of @xmath7 and @xmath8 . to compute @xmath68 numerically , we used from eq . ( [ perturb ] ) , and ( [ trial ] ) , that the amplitude of the order parameter is @xmath90 close to incoherence . then , the goal is to simulate the evolution of the system , choosing the initial condition sufficiently close to the incoherent solution , and obtain numerically the amplitude order parameter @xmath91 . [ lsu2 ] shows that different eigenvalue curves ( for different @xmath7 ) intersect the horizontal axis , @xmath80 , _ at the same value of _ @xmath8 , as expected from eq . ( [ unimodalkc ] ) . \(b ) _ unimodal lorentzian frequency distribution _ , @xmath92 . eq . ( [ dispersion3 ] ) becomes @xmath93 = \frac{k}{2}m\,e^{md } \nonumber\\ \times ( m\,d)^{-m(\lambda+d+\varepsilon ) } \left[(m\,d)^{-m(\lambda+d+ \varepsilon)-1}\ , e^{-m\,d } \right . \nonumber\\ \left . -\gamma ( m(\lambda+d+\varepsilon),m\,d)\left(\frac{\lambda+d+ \varepsilon}{d}-1\right ) \right ] \label{lorentz}\end{aligned}\ ] ] an explicit solution for @xmath68 can not in general be found . thus , we consider several limiting cases corresponding to physically interesting parameter choices . in the small noise limit , @xmath94 , we consider the cases @xmath95 @xmath96 fixed , and @xmath97 @xmath98 . it is remarkable that the expansion @xmath99 where @xmath100 , @xmath101 , holds in both cases . if @xmath102 , @xmath103 , ( [ tt ] ) holds as a convergent expansion , @xcite . if @xmath104 , and @xmath105 ( with fixed @xmath68 and @xmath36 of order 1 ) , ( [ tt ] ) holds as an asymptotic expansion , @xcite . inserting ( [ tt ] ) in eq . ( [ lorentz ] ) , we obtain @xmath106 \label{lorentz1}\end{aligned}\ ] ] similarly to the unimodal case , it is possible to prove that @xmath68 is always real . to this purpose , notice that replacing @xmath107 in eq . ( [ lorentz ] ) with @xmath68 in eq . ( [ dispersion3 ] ) ( setting @xmath108 ) we obtain the same equation . the critical coupling @xmath88 is then found by setting @xmath80 in ( [ lorentz1 ] ) . in case @xmath95 , we have @xmath109 in the limit of vanishing mass , we recover the result @xmath110 valid for the kuramoto model @xcite . another important limit is @xmath111 , which reproduces the unimodal distribution . we find @xmath112 , _ independently _ of mass . thus the spread in frequency distribution plays an important role in synchronizing populations of oscillators affected by inertia . in case @xmath97 , eq . ( [ lorentz1 ] ) yields @xmath113 from which @xmath114.\ ] ] this quantity is always real and vanishes for @xmath115 . note that @xmath116 grows roughly linearly with @xmath7 . thus oscillator synchronization is made harder by increasing inertia in the limit of vanishing noise . this behavior is slightly different from that described in @xcite . there numerical simulations seemed to show that incoherence remains stable up to a critical coupling , which was independent of @xmath7 . the singular nature of the limit @xmath117 makes the cause of this discrepancy unclear , although we should mention that no stability analysis was conducted in @xcite . in the opposite limit @xmath118 , @xmath119 , and incoherence is always stable . the stability diagram in the parameter space ( @xmath36,@xmath8 ) is shown in figure [ lsl1 ] for @xmath120 , and compared to that of the kuramoto model ( @xmath14 ) . this diagram is obtained from eq . ( [ lorentz ] ) with @xmath80 , for fixed @xmath121 and @xmath7 . in this figure , we have also plotted the evolution of the order parameter amplitude for the parameter values marked in the stability diagram by ( 1 ) to ( 4 ) . in all cases , the initial condition is taken sufficiently close to the incoherent solution , @xmath25 . \(c ) _ bimodal frequency distribution _ , @xmath122 $ ] . eq . ( [ dispersion3 ] ) becomes @xmath123 } { [ m(\lambda+d)+n]^2+m^2\omega_0 ^ 2 } = 1.\label{beta12}\end{aligned}\ ] ] in the high frequency limit , @xmath124 , we can find an analytical formula for @xmath68 by inserting the following asymptotic expansion for the incomplete gamma function in ( [ gammaf ] ) @xcite , @xmath125 where @xmath126 , and @xmath127 . the result is @xmath128 which yields @xmath129 re@xmath130 gives the same critical coupling as the kuramoto model , @xmath131 , for the same bimodal frequency distribution @xcite . figure [ lsb1 ] shows the stability diagram , which is obtained from eq . ( [ beta12 ] ) with re@xmath80 , for @xmath132 and three different mass values @xmath133 . the stability diagram of the kuramoto model ( @xmath14 ) is also depicted . notice how the curves corresponding to differerent masses tend to @xmath134 as @xmath124 , as expected . fig.[lsb2 ] displays the evolution of the order parameter amplitude in different regions of the stability diagram corresponding to @xmath135 and @xmath132 . in conclusion , increasing @xmath7 , @xmath121 , @xmath36 ( inertia , noise , and frequency spread ) makes it more difficult to synchronize the oscillator population via stationary bifurcations from incoherence . inspired by the previous linear stability analysis , we shall expand the distribution function using a basis of parabolic cylinder functions ( or , equivalently , hermite polynomials ) of unit mean square norm @xcite , @xmath136 in ( [ expansion ] ) , we have defined @xmath137 so that @xmath138 . the functions @xmath139 are @xmath19-periodic in @xmath20 , and we have @xmath140 as it follows from the normalization of @xmath141 . we shall find a system of mode - coupling equations for the coefficient functions @xmath142 . then we shall try to find stationary solutions for different frequency distributions @xmath0 . this is not so easy in the general case , so that we shall follow a standard approach : we shall recognize in the system of equations for the coefficient functions a particular stationary solution corresponding to incoherence . then we shall try to find a bifurcation equation for other stationary solutions branching off from incoherence . we shall see that even this requires different approximation schemes in order to succeed . let us insert eq . ( [ expansion ] ) into the fokker - planck equation ( [ fpe ] ) . we then obtain the following hierarchy of coupled partial differential equations for @xmath139 , @xmath143 here we have defined the operator @xmath144\ , f. \label{oper}\end{aligned}\ ] ] in terms of the functions @xmath142 , the equation for the order parameter becomes @xmath145 the _ incoherent _ @xmath20-independent solution , @xmath146 , depends only on @xmath30 . it can readily be obtained from the above hierarchy of equations , by ignoring all derivatives . the result is @xmath147 inserting this into the expansion in eq.([expansion ] ) , we indeed recover eq.([incoherent ] ) . in particular setting @xmath108 , this yields the simplest incoherent solution , @xmath148 , corresponding to the unimodal frequency distribution . other interesting stationary solutions are partially synchronized distributions , which depend on @xmath20 . notice that eq . ( [ 0systemc ] ) implies that @xmath149 does not depend on @xmath20 . in the following we analyze stationary solutions bifurcating from incoherence for different frequency distributions . \(a ) _ discrete unimodal frequency distribution _ , @xmath1 . let us look for _ stationary _ solutions with finitely many nonzero coefficients @xmath150 for @xmath151 and @xmath152 . eq.([systemc ] ) yields @xmath153 , and therefore , @xmath154 ( @xmath155constant ) . inserting this result in ( [ systemc ] ) for @xmath156 , we find @xmath157 . the corresponding solution is not periodic in @xmath20 unless @xmath158 . repeating this argument , we obtain the stationary solution @xmath159 where @xmath160 for @xmath161 and the normalization condition has been used . observe that the resulting distribution , @xmath162 , is factorized with respect to its two arguments , @xmath20 and @xmath24 , cf . it is remarkable that @xmath163 is independent of @xmath7 , and coincides with that obtained for the kuramoto model . therefore , from ( [ rc0jer ] ) the order parameter does not depend on inertia , and the bifurcation diagram for @xmath164 is exactly the same as that for the kuramoto model . \(b ) _ general frequency distributions . _ for general frequency distributions , we could try to find stationary solutions of the mode - coupling equations ( [ systemc ] ) which bifurcate from incoherence . it is however more direct to work with the stationary fokker - planck equation ( [ fpe ] ) as follows . we consider stationary solutions as functions of a fixed value of the synchronization parameter @xmath164 . then we expand these solutions in power series of @xmath164 and write a hierarchy of equations for the coefficient functions . the first coefficient function should be the incoherent solution of synchronization parameter @xmath25 . inserting the power series probability density function into eq.([order ] ) , we find the amplitude equation for stationary solutions bifurcating from incoherence . this procedure is explained in appendix b. we quote here the result : @xmath165 where @xmath166 , \label{alphateo}\end{aligned}\ ] ] and @xmath167 can be calculated numerically by solving a system of differential equations . as the inertia vanishes , @xmath168 , @xmath169 becomes @xmath170 notice that @xmath169 in equation ( [ alphateo ] ) coincides with @xmath171 in eq . ( [ dispersion3 ] ) provided @xmath80 and @xmath172 . this means that the critical coupling for bifurcation towards stationary synchronized states is obtained at @xmath173 , no matter what the symmetric frequency distribution may be . before interpreting the amplitude equation ( [ ralpha ] ) , we shall outline a procedure to obtain its coefficients based upon an uncontrolled closure assumption which is accurate for small values of @xmath174 . consider the expansion ( [ expansion ] ) . we expect that the coefficients @xmath142 of stationary solutions close to incoherence do not differ much from the coefficients of the latter . in view of the functional form ( [ cnbar ] ) , we anticipate that the coefficients @xmath142 approach zero as @xmath175 _ faster for smaller values of the mass_. thus we shall now consider stationary solutions for general frequency distributions , such that @xmath160 in ( [ systemc ] ) , for all @xmath176 . by solving equation ( [ systemc ] ) for such a stationary solution with @xmath177 , we obtain @xmath178\ , c_1(\omega ) . \label{c2}\ ] ] we now insert this expression into ( [ 1systemc ] ) . the resulting equation is solved for a @xmath179 , which is @xmath19-periodic in @xmath20 and obeys the normalization condition ( [ norma ] ) . we find @xmath180\nonumber\\ \times e^{-\frac{1}{d}[\omega \eta+ k\,r\,\cos ( \psi-\theta-\eta ) ] } d\eta , \label{varphi}\\ z(\omega ) = \int_0^{2\pi } e^{\frac{k\,r}{d}\cos ( \psi-\theta ) } \varphi(\theta,\omega)\ , d\theta , \label{zeta}\\\end{aligned}\ ] ] in ( [ c0stat ] ) , @xmath164 should be determined so that @xmath181 holds . the function @xmath182 can be obtained by integrating ( [ 1systemc ] ) with respect to @xmath20 and using the normalization condition for @xmath179 together with the @xmath19-periodicity in @xmath20 of @xmath179 and @xmath183 : @xmath184.\label{c1a}\ ] ] then , the stationary distribution can be approximated by @xmath185\ , e^{-m\omega^2/4d}. \label{sdis}\end{aligned}\ ] ] notice that a nonvanishing @xmath182 in ( [ expansion ] ) implies that the probability density is no longer even in @xmath24 ( @xmath186 is an odd function of @xmath24 ) . by ( [ c1a ] ) , this occurs in the synchronized phase for the case of nonidentical oscillators . for the bimodal frequency distribution , fig . [ ss0 ] shows @xmath149 as function of @xmath28 for two different values of @xmath7 . both the approximate expression ( [ c1a ] ) and results of direct numerical simulations are depicted . notice that the agreement between our approximation and the numerical result improves as @xmath7 decreases . thus we observe that for each fixed nonzero @xmath30 and each fixed @xmath20 , the distribution function is no longer peaked at @xmath187 . however , the _ instantaneous frequency distribution _ , defined by @xmath188 may turn out to be even in @xmath24 . this is certainly true for the approximate stationary distribution ( [ sdis ] ) , for @xmath189 as it follows from the definition of the order parameter and the expansion ( [ expansion ] ) . another indication that the exact instantaneous frequency distribution may be even in @xmath24 is that the average frequency tends to zero as @xmath190 . this would occur if the stable stationary distribution is even in @xmath24 ( although , admittedly , distributions which are not even in @xmath24 may have zero mean ) . the result can be shown directly from the fokker - planck equation ( [ fpe ] ) . we multiply that equation by @xmath191 and integrate with respect to all variables . then we obtain the following equation for the mean value @xmath192 : @xmath193 the right hand side of this expression is zero if @xmath194 . then @xmath195 tends to zero exponentially fast as @xmath196 . let us now go back to the problem of obtaining the coefficients in the amplitude equation ( [ ralpha ] ) . let us expand ( [ c0stat ] ) in powers of @xmath164 and insert the result in ( [ rc0 ] ) . we obtain an approximation to the coefficients of the amplitude equation ( [ rc0 ] ) . @xmath169 is again given by ( [ ralpha ] ) , and the expression for @xmath167 is @xmath197\ , g(\omega ) d\omega . \label{alphabeta2}\end{aligned}\ ] ] we now interpret the amplitude equation ( [ ralpha ] ) in the usual way . notice that the nonzero solution is approximately given by @xmath198 then the critical value of @xmath8 is @xmath199 , and the sign of @xmath167 determines the direction of the bifurcating branch of stationary solutions . assume @xmath200 . then the bifurcating stationary solution exists for @xmath201 if @xmath202 ( _ supercritical bifurcation _ ) . if @xmath203 , we have a _ subcritical bifurcation _ and the partially synchronized stationary solution exists for @xmath204 . the critical coupling tends to infinity as @xmath205 , and the bifurcation does not occur at positive couplings if @xmath206 . given the formulas ( [ alphateo ] ) or ( [ alphabeta1 ] ) , @xmath169 could become negative , depending on the value of @xmath7 . let us now analyze the bifurcation diagram corresponding to two different frequency distributions . * _ unimodal lorentzian frequency distribution_. + for an unimodal lorentzian frequency distribution , the coefficients @xmath169 and @xmath167 of ( [ alphabeta1 ] ) and ( [ alphabeta2 ] ) are approximately given by @xmath207.\label{lorss}\end{aligned}\ ] ] apparently , the sign of @xmath169 could again be negative provided @xmath208 is sufficiently large . however , evaluation of the exact expression ( [ alphateo ] ) shows that @xmath209 . figure [ sslor1 ] shows @xmath169 as a function of @xmath36 for two different masses . notice that the approximate expression for @xmath169 fits better the numerical result as @xmath7 and @xmath36 decrease . + in contrast with @xmath169 , @xmath167 can really change sign for a lorentzian frequency distribution . fig . [ sslor2 ] shows the good qualitative agreement between the approximate expression for @xmath167 and its numerical evaluation from the exact equations ( [ z3sys ] ) . + there is a critical mass @xmath210 ( @xmath36 and @xmath121 are kept fixed ) for which @xmath211 . this mass separates the supercritical and subcritical bifurcation regions . setting @xmath211 in eq . ( [ lorss ] ) yields @xmath212 figure [ sslor3 ] shows the qualitative agreement between the approximate expression ( [ criticallor ] ) , and the numerical result obtained from the solution of ( [ z3sys ] ) . + note that in the massless case , the stationary solution always branches off supercritically , independently of @xmath36 . in the presence of inertia and for appropriate values of @xmath36 , we have found a subcritical bifurcation . this prediction is illustrated by figures [ sslor4 ] ( parameters corresponding to a supercritical bifurcation ) and [ sslor5 ] ( parameters corresponding to a subcritical bifurcation ) . figure [ sslor4 ] is a typical supercritical bifurcation diagram ( order parameter amplitude versus @xmath8 ) . in the subcritical case , figure [ sslor5 ] shows the different evolution of the order parameter amplitude for two different initial conditions . notice the bistability between incoherence and the stable synchronized solution typical of a subcritical bifurcation . * _ bimodal frequency distribution _ + for the bimodal frequency distribution , @xmath169 may be zero . this means that there is a critical frequency @xmath213 above which @xmath214 . in fact , we have shown in fig . [ lsb1 ] , looking to the branch corresponding to @xmath80 , that in case of a bimodal frequency distribution , it is possible to find some values of @xmath28 without its corresponding @xmath8 in the stability diagram of incoherence , due to the existence of an horizontal asymptote in the space parameter @xmath215 . this means that there is not a finite coupling @xmath216 where the stationary solution branches off . on the other hand , this never occurs in the bimodal kuramoto model [ @xmath14 , @xmath217 , because in this case there is not such an horizontal asymptote . ( [ alphabeta1 ] ) shows that the critical frequency at which @xmath218 is @xmath219 figure [ ssb1 ] compares the previous approximate expression to the exact critical frequency obtained from ( [ alphateo ] ) . note that the approximation improves as @xmath7 decreases . [ ssb2 ] shows that @xmath216 increases as @xmath28 does . @xmath216 becomes infinity for @xmath220 . + for @xmath200 , there is another important critical frequency . in this case , the sign of @xmath167 decides whether the bifurcation is subcritical ( @xmath203 ) or supercritical ( @xmath202 ) . the sign of @xmath167 depends on @xmath7 and @xmath28 . for small mass , we can use the approximations ( [ alphabeta1 ] ) and ( [ alphabeta2 ] ) ( ignoring terms which are quadratic in the mass ) . then we find that the critical frequency at which @xmath211 is @xmath221 we have solved numerically the system of equations ( [ z3sys ] ) , and compared with the approximation ( [ alphabeta2 ] ) for @xmath167 . in figure [ ssb3 ] the coefficient @xmath167 is plotted as a function of @xmath28 for different masses . similarly , figure [ ssb4 ] shows how the critical frequency @xmath222 varies as a function of @xmath7 . note that the analytical approximation improves as @xmath7 goes to zero ( as expected ) . + notice that @xmath223 . then the branch of synchronized stationary states bifurcates _ subcritically _ from incoherence at @xmath224 , provided @xmath225 . summarizing , inertia favors the subcritical character of bifurcations describing the transition from incoherence to the partial synchronized state . in fact in the bimodal case , the transition of incoherence to synchronization will most likely occur as a subcritical bifurcation as inertia increases . for a continuous unimodal lorentzian frequency distribution , inertia may turn subcritical the supercritical bifurcation to stationary synchronized states , which is always found in the masslesss kuramoto model . the high frequency limit of eq . ( [ fpe ] ) can be analyzed by means of a multiscale method . for the kuramoto model @xcite , this method lead to the result that the probability density is ( to leading order ) a superposition of different components corresponding to the different peaks of the oscillator frequency distribution . to apply this method to the present model , we shall assume that the frequency distribution @xmath0 has @xmath7 maxima located at @xmath226 , so that @xmath227 tends to the distribution @xmath228 as @xmath124 . in order to simplify the calculations , we change variables to a comoving frame : @xmath229 and let @xmath230 , where @xmath231 then we obtain the following equations : @xmath232 where @xmath233 , and @xmath234 , @xcite . we now define fast and slow time scales , @xmath235 and @xmath236 , and make the ansatz : @xmath237 inserting this into the governing equations , we obtain the following hierarchy : @xmath238 \right\ } \ , , \label{hierar}\end{aligned}\ ] ] where @xmath239 elimination of secular terms yields the following condition : @xmath240\right\ } = 0.\end{aligned}\ ] ] notice that this equation corresponds to the equation of a unimodal frequency distribution , already studied in @xcite . its solution evolves towards the following stationary state as time evolves : @xmath241 where @xmath242 incoherence of a component corresponds to @xmath243 . we know that as @xmath244 surpasses @xmath245 , a stable synchronized solution bifurcates from incoherence . in the particular case of a symmetric bimodal frequency distribution , the bifurcation value is @xmath134 , independently of the inertia @xmath7 . this agrees with the results of the linear stability analysis for @xmath124 . all the results previously obtained for the kuramoto model can be applied to the present model without any modification @xcite . thus , both a stable standing wave solution ( sw ) and an unstable travelling wave solution ( tw ) bifurcate supercritically from incoherence at @xmath134 @xcite . in figure [ sw1 ] illustrates the comparison between the asymptotic solution ( [ swsol ] ) and the result of direct numerical simulation for a large enough value of the frequency @xmath28 . finally , fig.[bdiag ] shows the global bifurcation diagram of the bimodal case for the case of positive @xmath169 and @xmath167 . as explained in the previous section , the branch of synchronized stationary states bifurcates _ subcritically _ from incoherence at @xmath224 , provided @xmath225 . thus @xmath246 and the end of the sw and tw branches in fig.[bdiag ] should extend to infinity as @xmath247 . four different numerical methods have been used . numerical simulations of the system of langevin equations ( [ 1 ] ) were carried out for a large number of oscillators ( n=20 000 ) , using an euler method . a standard finite difference method was used to solve numerically the fokker - planck equation ( [ fpe ] ) , or the system of partial differential equations ( [ systemc ] ) . in addition , we have used a simple spectral method , which generalizes the one proposed in @xcite . the idea is to solve a set of ordinary differential equations for moments of @xmath18 : @xmath248\,d\theta , \label{momentx}\\ ( y_i^j)_{k}:= \int_0^{2\pi } c_i(\theta,\omega_k , t ) \sin [ j(\psi-\theta)]\,d\theta , \label{momenty}\end{aligned}\ ] ] @xmath249 the coefficient functions @xmath250 are the same as in the parabolic cylinder expansion ( [ expansion ] ) . the integral in ( [ rnumerical ] ) will be approximated by a suitable quadrature formula , as @xmath251 for instance , the gauss - laguerre quadrature has been chosen case for the case of a lorentzian frequency distribution . the system of ordinary differential equations is given by : @xmath252 -\frac{i}{m}(x_i^j)_{k}\nonumber\\ + \sqrt{i+1}\sqrt{d}{m}j ( y_{i+1}^j)_{k}-j\dot{\psi}(y_i^j)_{k},\\ ( \dot{y}_i^j)_{k}=-j\sqrt{i}\sqrt{\frac{d}{m } } ( x_{i-1}^j)_{k}+\sqrt{i}\frac{1}{\sqrt{m\,d } } \omega_k(y_{i-1}^j)_{k}\nonumber\\ + \sqrt{\frac{i}{m\,d}}\frac{k\,r}{2 } \left[(x_{i-1}^{j-1})_{k}- ( x_{i-1}^{j+1})_{k}\right ] -\frac{i}{m}(y_i^j)_{k}\nonumber\\ -\sqrt{i+1}\sqrt{d}{m}j ( x_{i+1}^j)_{k}+j\dot{\psi}(x_i^j)_{k},\\ i=1,\ldots , n,\,\ , j=1,\ldots , m , \nonumber\end{aligned}\ ] ] @xmath253 in order to numerically simulate this system , it is necessary to truncate the hierarchy after a reasonable number of modes @xmath254 , and @xmath255 . these numbers will depend on the inertia @xmath7 , the spread , and the coupling strength . they should be chosen large enough , so that the numerical results do not depend on @xmath254 and @xmath255 . [ nr1 ] shows the ratio @xmath256 as a function of @xmath49 , for the stationary solution and for various mass values . @xmath257 is the maximum value of the stationary coefficient @xmath258 . it has been obtained from a finite - difference solution of eq.([fpe ] ) . note that this ratio increases as inertia does . thus the truncation approximation ( [ sdis ] ) ceases to make sense for larger values of inertia . whether truncating ( at higher order ) the moment equations ( [ momentx ] ) and ( [ momenty ] ) yields good numerical results , is tested in fig . this figure shows how closely the previous system ( with moments of order 4 or 10 ) resembles the direct solution of the langevin equations . we notice that the system containing moments of order 10 is rather close to the solution of the langevin equations , but it does not contain the fluctuations unavoidable in stochastic methods . if we are interested in solving the nonlinear fokker - planck equation , the system of moment equations seems a good alternative to solving langevin equations for a large number of oscillators . we have investigated synchronization properties of a model of globally and nonlinearly coupled phase oscillators , where the effects of white noise , inertia and spread in the natural frequency distribution are all considered . the linear stability of the incoherent solution is rigorously analyzed . stationary and time - dependent solutions of the standing and travelling wave type are obtained by a variety of perturbative methods . these include finding and approximately solving mode - coupling equations for expansions of the probability distribution in parabolic cylinder functions , finding amplitude equations near bifurcation points , and hierarchy - closure assumptions . numerical simulations of the different equations and the original model favorably agree with the different perturbative results . inertia changes the stability boundaries of the incoherent solution in a non trivial way . in the case of an unimodal lorentzian frequency distribution , incoherence is stabilized , but the effect of mass completely drops out if the frequency spread vanishes . for a discrete bimodal frequency distribution , both stability boundaries and the character of the transition from incoherence to synchronized states depend on the values of the natural frequency @xmath28 and on inertia . the effect of inertia on the stationary solution is dramatic in some cases . in general , inertia hardens the synchronization transition : it may render subcritical ( hard ) originally supercritical ( soft ) transitions ( in unimodal lorentzian frequency distributions ) , or it increases the region in parameter space where the transition is subcritical ( discrete bimodal frequency distributions ) . analytical as well as numerical calculations confirm these findings . this work was supported , in part , by the european union tmr contract erb fmbx - ct97 - 0157 , by unesco under contract uvo roste 875.629.9 , by the spanish dges grant pb98 - 0142-c04-c01 , and by the gnfm of the italian c.n.r . all numerical simulations were conducted at caspur rome . in ( [ beta8 ] ) there appear two integrals over @xmath63 : @xmath259 the second integral is directly @xmath260 . the first integral equals @xmath261 because of ( [ beta7 ] ) . an equivalent expression is @xmath262 this can be written as @xmath263 provided we set @xmath264 after performing the calculations . thus we see that @xmath265 this expression and the previous one for @xmath266 yield ( [ beta9 ] ) . let us now obtain @xmath71 for integer @xmath52 and @xmath267 . by using that @xmath268 is even for even @xmath52 and odd for odd @xmath52 , we can write @xmath269 where @xmath270 for @xmath52 even and @xmath271 for @xmath52 odd . we can find ( cf . 7.741 of @xcite ) : @xmath272 provided @xmath267 . from these formulas , we obtain @xmath273 these two expressions are equivalent to ( [ beta11 ] ) . we will now calculate coefficients @xmath169 and @xmath167 of eq . ( [ ralpha ] ) . the idea is simply to expand the right hand side in the definition of the order parameter , @xmath274 as a power series in @xmath164 . to this end , we fix @xmath164 and @xmath275 in eq . ( [ fpe ] ) , expand its stationary solution in powers of @xmath164 and insert the result in ( [ order1 ] ) . as we do not have a closed formula for the stationary solution of ( [ fpe ] ) , we shall first derive a hierarchy of equations for its coefficients in the series in powers of @xmath164 . thus we have @xmath276 then @xmath277 where we have absorbed @xmath275 in the definition of @xmath278 . comparison with eq . ( [ ralpha ] ) , establishes that @xmath279 , @xmath280 , and @xmath281 . from ( [ g ] ) and ( [ fpe ] ) , we obtain @xmath282-\omega \frac{\partial g_n}{\partial \theta } = -\frac{n}{m}k\sin \theta \frac{\partial g_{n-1}}{\partial \omega},\ ] ] with @xmath283 . since we are trying to find solutions bifurcating from incoherence , we should have @xmath284 , i.e.the @xmath20-independent incoherent solution ( [ incoherent ] ) . this directly confirms that @xmath285 . the unknowns @xmath286 are @xmath19-periodic in @xmath20 , so that we may fourier expand them as @xmath287 where @xmath288 here @xmath289 for @xmath290 because @xmath291 is the @xmath20-independent incoherent solution . the condition that the probability density function be real yields @xmath292 , which in turn implies @xmath293 . inserting these expressions into ( [ r_nformula ] ) , we find @xmath294 the unknowns @xmath295 satisfy the following hierarchy of equations : @xmath296-i\,l\omega z_n^l=-\frac{n}{m}\frac{k}{2i } \frac{d}{d\omega}[z_{n-1}^{l-1}-z_{n-1}^{l+1 } ] . \label{odeznl}\ ] ] the normalization condition for @xmath18 and the incoherent solution together with ( [ znl ] ) imply @xmath297 in order to obtain @xmath169 , and @xmath167 , we should show that @xmath298 and calculate @xmath299 and @xmath300 . we do this by means of ( [ odeznl ] ) , with @xmath301 . + ( a ) case n=1 . ( [ odeznl ] ) becomes @xmath302-i\,\omega z_1 ^ 1\nonumber\\ = -\frac{1}{m}\frac{k}{2i } \frac{dz_0 ^ 0}{d\omega } , \label{z11}\end{aligned}\ ] ] where @xmath303 is the incoherent solution ( [ incoherent ] ) , and we have used @xmath304 . we can solve eq . ( [ z11 ] ) by means of the solution of eq . ( [ eqb ] ) , in which we set @xmath80 and replace @xmath305 with @xmath306 , and the right - hand side with @xmath307 . then we obtain @xmath308 the previous analogy allows us to calculate @xmath299 from ( [ rn ] ) and equations ( [ beta8 ] ) , ( [ beta10 ] ) and ( [ beta11 ] ) ( with @xmath80 ) . the result is @xmath309.\end{aligned}\ ] ] \(b ) case n=2 . we now show that @xmath298 . + in order to analyze @xmath310 , it is necessary to consider @xmath311 , which is the solution to the following equation : @xmath312 -i\,\omega z_2 ^ 1=-\frac{2}{m}\frac{k}{2i } \frac{d}{d\omega}[z_{1}^{0}-z_{1}^{2}].\ ] ] we shall show that @xmath313 . if this is so , the resulting homogeneous equation can be transformed to the parabolic cylinder equation with a quadratic potential having complex coefficients . its solution can not decay to zero both as @xmath21 unless it is identically zero . the reason @xmath314 and @xmath315 are zero is similar . the equation for @xmath314 is homogeneous ( @xmath316 for @xmath290 ) , and it has a solution @xmath317 . the normalization condition ( [ condzn0 ] ) then implies @xmath318 . @xmath319 again obeys a homogeneous equation which can be transformed into the parabolic cylinder equation with a complex potential . the only solution which decays to zero as @xmath21 is again @xmath320 . \(c ) case n=3 . + the equations for @xmath321 are @xmath322-i\,\omega z_3 ^ 1=-\frac{3}{m}\frac{k}{2i } \frac{d}{d\omega}[z_2 ^ 0-z_2 ^ 2],\nonumber\\ \frac{d}{m^2}\frac{d^2 z_2 ^ 2}{d \omega^2}+ \frac{1}{m } \frac{d}{d\omega}[(\omega-\omega)z_2 ^ 2]-2 i\,\omega z_2 ^ 2=-\frac{2}{m}\frac{k}{2i } \frac{d}{d\omega}z_1 ^ 1,\nonumber\\ \frac{d}{m^2}\frac{d^2 z_2 ^ 0}{d \omega^2}+ \frac{1}{m } \frac{d}{d\omega}[(\omega-\omega)z_2 ^ 0]= -\frac{2}{m}\frac{k}{2i } \frac{d}{d\omega}[z_1^{-1}-z_1 ^ 1]=\frac{2k}{m}\ , im\frac{d}{d\omega}z_1 ^ 1,\nonumber\\ \frac{d}{m^2}\frac{d^2 z_1 ^ 1}{d \omega^2}+ \frac{1}{m } \frac{d}{d\omega}[(\omega-\omega)z_1 ^ 1]-i\,\omega z_1 ^ 1=-\frac{1}{m}\frac{k}{2i } \frac{d}{d\omega}z_0 ^ 0.\label{z3sys}\end{aligned}\ ] ] this system of equations has to be solved numerically in order to obtain the coefficient @xmath167 . y. kuramoto , in _ international symposium on mathematical problems in theoretical physics _ , h. araki ed . , lecture notes in physics , vol.39 ( springer , berlin ) ( 1975 ) ; _ chemical oscillations , waves , and turbulence _ , ( springer , berlin ) ( 1984 ) .
the deconfined interacting quark gluon plasma phase ( qgp ) is the _ equilibrium _ state of matter at high temperature and/or density . it is believed that this state has been present in the early universe , 10 - 20@xmath2s into its evolution . the question is if , in the short time , @xmath3@xmath4 s , available in a laboratory heavy ion collision experiment , the color frozen nuclear phase can melt and turn into the qgp state of matter . there is no valid first principles answer to this question available today , nor as it seems , will a first principles simulation of the dynamic heavy ion environment become available in the foreseeable future . to address this issue we study qgp experimentally , which requires development of laboratory experiments and suitable observables . to form qgp in the laboratory we perform relativistic heavy ion collisions in which a domain of ( space , time ) much larger than normal hadron size is formed , in which color - charged quarks and gluons are propagating constrained by external ` frozen vacuum ' , which abhors color @xcite . we expect a pronounced boundary in temperature and baryon density between confined and deconfined phases of matter , irrespective of the question if there is , or not , a true phase transition . we search for a boundary between phases considering the size of the interacting region and the magnitude of the reaction energy . detailed study of the properties of the deconfined state shows that qgp is rich in entropy and strangeness . the enhancement of entropy @xmath1 arises because the color bonds are broken and gluons can be created . enhancement of strangeness @xmath0 arises because the mass threshold for strangeness excitation is considerably lower in qgp than in hadron matter . moreover there are new mechanisms of strangeness formation in qgp involving reactions between ( thermal ) gluons . thus @xmath1 and @xmath0 are the two elementary observables which are explored with soft hadronic probes , for further theoretical details and historical developments see our book @xcite . the numerical work presented here was carried out with the public package of programs share @xcite . this report is a self contained summary of our recent results , see @xcite . entropy enhancement , observed in terms of enhanced hadron multiplicity per net charge , has been among the first indications of new physics reach of cern - sps experimental heavy ion program @xcite . the enhancement of strange hadron production both as function of the number of participating baryons , and reaction energy has been explored in several experiments at at bnl - ags , cern - sps and and bnl - rhic . we refrain from extensive historical survey of these results and present perhaps the latest , star - rhic result in @xcite . in this presentation one sees the yield per participant @xmath5 divided by a reference yield obtained in @xmath6 reactions . we observe that the enhancement rises both with the strangeness content in the hadron and with the size of the reaction region , indicating that the cause of this enhancement is a increased yield of strange quarks in the source , a qualitative expectation we will address in our quantitative analysis below . the gradual increase of the enhancement over the range of @xmath5 is an important indicator of the physics mechanisms at work . this behavior agrees with our studies of kinetic strangeness production and strangeness yield increasing with the size of the reaction region . this enhancement of strange antibaryons which demonstrates that a novel strangeness production mechanism is present has been extensively studied at sps energy range , where it was originally discovered @xcite . 0.2 cm further evidence for parton dynamics prior to final state hadronization is obtained from the study of strange hadron transverse energy spectrum . the identity of hyperon and antihyperon spectra in particular @xmath7 and @xmath8 implies that both strange matter and antimatter must have been produced from a common source by the same fundamental mechanism . furthermore they were not subject to interactions in their passage through the baryon rich hadron gas present at the sps energy range . the @xmath9 annihilation in baryon - rich hadron gas is strongly momentum dependent . this should deform the shape of the antihyperon spectra as compared to the spectra of hyperons . thus symmetry of the hyperon - antihyperon spectra also implies that there was no appreciable annihilation of the @xmath10 after their formation . the working hypothesis is therefore that hadronization of the qgp deconfined phase formed in high energy nuclear collision is direct , fast ( sudden ) and occurs without significant sequel interactions . that can be further tested in a study of yields and spectra of unstable resonances @xcite . at rhic the parton level dynamics is convincingly demonstrated by quark content scaling of azimuthal asymmetry of the collective flow @xmath11 . further evidence is derived from the consideration of quark recombination formation of hadrons from qgp . we refer to the recent comprehensive survey of the rhic result @xcite , and presentations at the meeting addressing these very interesting and recent developments . we will assume in this report that the case of quark - parton dynamics prior to hadronization is convincing , and will use our analysis of soft hadron production to find the thresholds of the onset of deconfinement and to determine the properties of the deconfined fireball at the time of its breakup into hadrons . in next section [ stathad ] we introduce the statistical hadronization method of analysis of hadron production . we discuss data analysis as function of impact parameter and energy dependence of soft hadron production in section [ depen ] . we will present both the systematics of statistical model parameters , and the associated physical properties . in section [ compare ] , we address the physical qgp signatures indicating presence of a phase boundary , giving particular attention to the explanation of the ` horn ' in the @xmath12 ratio , and the strangeness to entropy relative yield . we discuss , in section [ final ] , the role these results play in understanding of the phase boundary separating qgp from normal confined matter . we also discuss briefly possible new soft hadron physics at lhc . to describe the yields of particle produced we employ the statistical hadronization model ( shm ) . shm is by definition a model of particle production in which the birth process of each particle fully saturates ( maximizes ) the quantum mechanical probability amplitude , and thus , the relative yields are determined solely by the appropriate integrals of the accessible phase space . for a system subject to global dynamical evolution such as collective flow , this is understood to apply within each local co - moving frame . when particles are produced in hadronization , we speak of chemical freeze - out . hadron formation from qgp phase has to absorb the high entropy content of qgp which originates in broken color bonds . the lightest hadron is pion and most entropy per energy is consumed in hadronization by producing these particles abundantly . it is thus important to free the yield of these particles from the chemical equilibrium constraint . the normalization of the particle yields is , aside of the freeze - out temperature @xmath13 , directly controlled by the particle fugacity @xmath14 , where @xmath15 is particle ` @xmath16 ' chemical potential . since for each related particle and antiparticle pair , we need two chemical potentials , it has become convenient to choose parameters such that we can control the difference , and sum of these separately . for example , for nucleons @xmath17 , and , respectively , antinucleons @xmath18 the two chemical factors are : @xmath19 @xmath20 the ( baryo)chemical potential @xmath21 , controls the baryon number , arising from the particle difference . @xmath22 , the phase space occupancy , regulates the number of nucleon antinucleon pairs present . there are many different hadrons , and in principle , we could assign to each a chemical potential and then look for chemical reactions which relate these chemical potentials . however , a more direct way to accomplish the same objective consists in characterizing each particle by the valance quark content . the relation between quark based fugacity and chemical potentials ( @xmath23 ) and the two principal hadron based chemical potentials of baryon number and hadron strangeness @xmath24 is : _ b=3_q_s=1 3 _ b-_s , _ s=_q_s . an important ( historical ) anomaly is the negative s - strangeness in @xmath0-carrying - baryons . we will in general follow quark flavor and use quark chemical factors to minimize the confusion arising . in the local rest frame , the particle yields are proportional to the momentum integrals of the quantum distribution . as example , for the yield of pions @xmath25 , nucleons @xmath17 and antinucleons @xmath26 we have : @xmath27 n&=&{v}g_n\!\!\int\!\!\frac{d^3p}{(2\pi)^3 } \frac{1}{\gamma_q^ { -3}\lambda_q^ { -3 } e^{e_n / t}+1},\quad \overline { n}= { v}g_n\!\!\int\!\!\frac{d^3p}{(2\pi)^3 } \frac{1}{\gamma_q^ { -3 } \lambda_q^ { + 3 } e^{e_{\bar n}/t}+1}.\end{aligned}\ ] ] there are two types of chemical factors @xmath28 and @xmath29 , and thus two types of chemical equilibriums . these are shown in table [ parameters ] . the absolute equilibrium is reached when the phase space occupancy approaches unity , @xmath30 . the distribution of flavor ( strangeness ) among many hadrons is governed by the relative chemical equilibrium . 0.3 cm .[parameters]four quarks @xmath31 and @xmath32 require four chemical parameters ; right : name of the associated chemical equilibrium [ cols= " < , < , < " , ] in order to arrive at the full particle yield , one has to be sure to include all the hadronic resonances which decay feeding into the yield considered , _ e.g. _ , the decay @xmath33 feeds into @xmath34 and @xmath25 yields . the contribution is sensitive to temperature at which these particles are formed . inclusion of the numerous resonances constitutes a book keeping challenge in study of particle multiplicities , since decays are contributing at the 50% level to practically all particle yields . a public statistical hadronization program , share ( statistical hadronization with resonances ) has simplified this task considerably @xcite . the resonance decay contribution is dominant for the case for the pion yield . this happens even though each resonance contributes relatively little in the final count . however , the large number of resonances which contribute compensates and the sum of small contributions competes with the direct pion yield . for the more heavy hadrons , generally there is a dominant contribution from just a few , or even from a single resonance . the exception are the @xmath35 which have no known low mass resonances , and also @xmath36 the known resonances are very heavy and very few . a straightforward test of sudden hadronization and the shm is that within a particle ` family ' , particle yields with same valance quark content are in relation to each other well described by integrals of relativistic phase space . the relative yield of , _ e.g. _ , @xmath37 and @xmath38 or @xmath39 and @xmath17 are controlled by the particle masses @xmath40 , statistical weights ( degeneracy ) @xmath41 and the hadronization temperature @xmath13 . in the boltzmann limit one has ( star denotes the resonance ) : @xmath42 validity of this relation implies insensitivity of the quantum matrix element governing the coalescence - fragmentation production of particles to intrinsic structure ( parity , spin , isospin ) , and particle mass . the measurement of the relative yield of hadron resonances is a sensitive test of the statistical hadronization hypothesis and lays the foundation to the application of the method in data analysis . the method available to measure resonance yields depends in its accuracy significantly on the the precise nature of the hadronization process : the resonance yield is derived by reconstruction of the invariant mass of the resonance from decay products energies @xmath43 and momenta @xmath44 . should the decay products of resonances after formation rescatter on other particles , then often their energies and momenta will change enough for the invariant mass to fail the acceptance conditions set in the experimental analysis . generally , the rescattering effect depletes more strongly the yields of shorter lived resonances , as a greater fraction of these will decay shortly after formation , when elastic scattering of decay products on other produced particles is possible . we further hear often the argument that the general scattering process of hadrons in matter can form additional resonance states . in our opinion , the loss of observability ( caused by _ any _ scattering of _ any _ of the decay products is considerably greater than a possible production gain . the loss of resonance yield provides additional valuable information about the freeze - out conditions ( temperature and time duration)@xcite . in order to explore the properties of the fireball at hadronization as function of the volume at the top rhic energy @xmath45gev au au , we study the 11 centrality bins in which the @xmath46 and @xmath47 rapidity yields @xmath48 for @xmath49 have been recently presented , see table i and table viii in ref . these 6 particle yields and their ratios change rapidly . on the other hand , the additional two experimental results , the star @xmath50 @xcite , and @xmath51 @xcite show little centrality dependence . in addition , three supplemental constraints help to determine the best fit : + a ) strangeness conservation , _ i.e. _ , the ( grand canonical ) count of @xmath0 quarks in all hadrons equals such @xmath52 count for each rapidity unit ; + b ) the electrical charge to net baryon ratio in the final state is the same as in the initial state ; + c ) the ratio @xmath53 , which helps constrain the isospin asymmetry . + this last ratio appears redundant , as we already independently use the yields of @xmath54 and @xmath55 . these yields have a large systematic error and do not constrain their ratio well , and thus the supplemental constraint is introduced , since share allows for the isospin asymmetry effect . the 7 shm parameters ( volume per unit of rapidity @xmath56 , temperature @xmath13 , four chemical parameters @xmath57 and the isospin factor @xmath58 are in this case studied in a systematic fashion as function of impact parameter using 11 yields and/or ratios and/or constraints , containing one ( pion ratio ) redundancy . although the number of degrees of freedom in such analysis is small , the @xmath59 minimization yielding good significance is easily accomplished , showing good consistency of the data sample . the resulting statistical parameters are shown in , as function of participant number . we show on left results for the full non - equilibrium model allowing @xmath60 ( full circles , blue ) and semi - equilibrium setting @xmath61 ( open circles , red ) . from top to bottom the ( chemical ) freeze - out temperature @xmath13 , the occupancy factors @xmath62 , @xmath63 and together in the bottom panel the baryon @xmath64 and hyperon @xmath65 chemical potentials . on left , in , we present the results for the centrality dependence , as function of participant number , on right as function of energy . to study the energy dependence , we must assemble several different experimental results from different facilities and experiments @xcite . the results of this extensive analysis are shown on right hand side in . we note that when we are able to consider the total particle abundances , the number of participating nucleons is a tacit fit parameter . the lowest energy result is from our ags study @xcite , the sps data we used are from na49 energy dependence exploration at the cern - sps @xcite . these results are for the total particle yields . the highest energy two results are based on studies of rhic data at 130 and 200 gev at central rapidity and address the @xmath48 particle yields , with the highest point corresponding to the results presented at greatest centrality on the left hand side in . there are several relevant features in fig.[gammu ] . we see on left , that , for @xmath66 , there is no centrality dependence in freeze - out temperature @xmath13 and chemical potentials @xmath67 ( up to the variation which can be associated with fluctuation in the data sample ) . however , there is a change in values of the chemical potentials with reaction energy . this is result of rapidly decreasing baryon density , which due to reduced stopping is distributed over a wider range of rapidity as the reaction energy increases . the middle sections , in , address the phase space occupancies which were obtained in terms of hadron particle yields . the quark - side occupancy parameters could be considerably different , indeed as model studies show , a factor 2 lower at the discussed conditions @xcite : the hadron side phase space size is in general different from the quark - side phase space , since the particle degeneracies , particle spectra are quite different . we see similar behavior of @xmath62 as of @xmath13 both for volume ( that is , ` wounded ' participant number @xmath68 ) and energy dependence seen in fig.[gammu ] : the two lowest energy bins ( top ags and lowest sps energy ) deviate from the behavior seen at all other energies , as do the bins with @xmath69 . @xmath63 as function of centrality rises steadily indicating longer lifespan of the fireball with increasing size . as function of energy , @xmath63 reaches a plateau at 30 @xmath68gev , with further rise only seen in the central rapidity results at rhic . we note that , in our analysis , there is no saturation of the @xmath70 as we approach the most central reactions . this is inherent in the data we consider which includes the yields of @xmath36 , and @xmath71 . this result is consistent with the implication that strangeness is not fully saturated in the qgp source , though it appears over - saturated when measured in the hadron phase space . the deviation at the most peripheral centrality bins and at lowest reaction energy from trends set by other results could be an indication of the change in the reaction mechanism . as a threshold of centrality and/or energy are crossed a relatively small value of @xmath72 grows rapidly to the maximum allowed by pion condensation condition , @xmath73 . this behavior signals a transformation of a chemically under - saturated phase of matter into something novel , where chemical equilibration is easy , and results in hadronization in a over - saturated phase of matter . independent of the chemical ( non-)equilibrium assumption , the baryochemical potential @xmath74 mev is seen across 10 centrality bins . similarly , we find strangeness chemical potential @xmath75 mev ( related to strange quark chemical potential @xmath76 ) . the most notable variation , in fig.[gammu ] , is the gradual increase in strangeness phase space occupancy @xmath63 and thus strangeness yield with collision centrality . this effect was predicted and originates in an increasing lifespan of the fireball @xcite . the over - saturation of the phase space has been also expected due to both , the dynamics of expansion @xcite , and/or reduction in phase space size as a parton based matter turns into hg @xcite . this latter effect is also held responsible for the saturation of light quark phase space @xmath77 . a systematic increase of @xmath70 with collision centrality has been reported for several reaction energies @xcite . let us now discuss more in depth the magnitude of the hadronization temperature which at high energy / central collisions we find at @xmath78 mev . some prefer the statistical hadronization to occur at higher temperature , perhaps as high as@xmath79 mev , a point argued at this meeting in great detail by dr . peter braun - mnzinger . in his presentation we heard that he believes that the lattice results will reach to such high temperature near to @xmath80 and this is where one should expect to see hadronization we disagree with both claims . for one , we note that _ chemical equilibrium qcd lattice _ @xcite results are mature and yield @xmath81 mev when extrapolated to physical quark mass scale . this result is in a very good agreement with the prior work on 2 and 3 flavor qcd which brackets this result by @xmath82 mev @xcite . moreover , the heavy ion collisions present a highly dynamical environment and one has to pay tribute to this especially regarding the value of hadronization temperature . for this reason we do not expect to find that the observed hadronization condition @xmath83 will line up with the phase boundary curve obtained in study of a statistical system in thermodynamic limit on the lattice : \a ) + a widely discussed effect which displaces the hadronization condition from the phase boundary is the expansion dynamics of the fireball . when the collective flow occurs at parton level , the color charge flow , like a wind , pushes out the vacuum @xcite , adding to thermal pressure a dynamical component . this can in general lead to supercooling and a sudden breakup of the fireball . we find that this can reduce the effective hadronization temperature by up to 20 mev @xcite . \b ) + lattice result have been discussed for 2-flavor lattice qcd at corresponding to @xmath84 ( called ) and for 2 + 1 flavor , corresponding to @xmath85 . while the precise nature of the phase limit is still under study it appears that for 2-flavor case the phase boundary temperature rises by about 7 - 10 mev compared to the 2 + 1 case . we refer to the recent review of lattice qcd for further details @xcite . similarly the phase limit in pure gauge case corresponding in loose sense to @xmath86 was seen near or even above @xmath87 mev . these results do suggest that presence and the number of quarks matters regarding the precise location of the phase boundary and its nature . its importance could be greatly enhanced , should the over - saturation of quark phase space have the same effect as would additional quark degrees of freedom . these are known to cause even for @xmath88 the conversion of the phase crossover into a 1st - order phase transition which would , with these additional degrees of freedom , be expected at just the temperature we find in the shm analysis . we can be nearly sure that the chemical conditions matter and can displace the transition temperature . because the degree of equilibration in the qgp depends on the collision energy , as does the collective expansion velocity , we can not at all expect a simple hadronization scheme appropriate for the hadronization of nearly adiabatically expanding universe . leaving this issue we note that , in the data analysis assuming chemical equilibrium , we find @xmath89 mev for the chemical equilibrium and strangeness non - equilibrium freeze - out , see . the error is our estimate of the propagation of the systematic data error , combined with the fit uncertainty ; the reader should note that the error comparing centrality to centrality is negligible . the freeze - out temperature is for the semi - equilibrium and equilibrium model about 10% greater than the full chemical non - equilibrium freeze - out . this result for @xmath13 , in the equilibrium case , is in mild disagreement ( 1.5 s.d . ) with earlier equilibrium fits @xcite . this , we believe , is due to some differences in data sample used , specifically , the hadron resonance production results used provide a very strong constraint for the fitted temperature , and more complete treatment by share of hadron mass spectrum . interestingly , it seems that the general consensus about the chemical equilibrium best analysis result is in gross disagreement with the results advanced at this meeting by dr . peter braun - mnzinger . we now turn our attention to the physical properties of the hadronizing fireball obtained summing individual contributions made by made of each of the hadronic particles produced . often particles observed experimentally dominate ( _ e.g. _ pions dominate pressure , kaon strangeness yield etc ) . however , it is important to include in this yields of particles predicted in terms of the shm fit to the available results . again , on left in fig.[phys ] . we show the behavior as function of impact parameter and on right as function of energy . from top to bottom we show the pressure @xmath90 , energy density @xmath91 , entropy density @xmath92 , and the dimensionless ratio @xmath93 . all contributions are evaluated using relativistic expression , see @xcite . when we fitted particle rapidity yields , the global fitted yield normalization factor is @xmath94 , for total particle yields it is @xmath95 which is also a function of centrality trigger . when we consider ratios of two bulk properties e.g. @xmath96 , the results are in general smoother indicating cancellation in the error propagating from the fit . the overall error is of the magnitude of the particle yield , for example pressure is dominated by pions and hence its precision is limited by this error . however , on left , the point - to - point error is minimal as the systematic error is common . on right , the absolute error matters as the fluctuations in the results presented show . we note , in , that as the reaction energy passes the volume threshold @xmath97 and even more so , the energy threshold @xmath98 the hadronizing fireball becomes much denser . the entropy density jumps by factor 34 , and the energy and baryon number density by a factor 23 . the hadron pressure jumps up from @xmath99 initially by factor 2 and ultimately more than factor 3 . there is a gradual increase of @xmath100 at low reaction energy to 0.165 at the top available energy . it is important to note that exactly the same behavior of the fireball physical properties arises both as function of reaction volume and reaction energy . this is the case for both the physical properties and the statistical parameters . we believe that this shows a common change in the physical state created as function of energy available and reaction volume . one of the most interesting questions is if there is an energy threshold for the formation of a new state of matter . an important observable of the deconfined phase of matter is aside of strangeness , the high entropy content , which is arising from broken color bonds . the observable for both is the k@xmath101 yield ratio @xcite . this ratio has been studied experimentally @xcite and a pronounced horn structure arises . we can describe this structure in our study of the particle yields only within the chemical non - equilibrium model . although this change is associated a rather sudden modification of chemical conditions in the dense matter fireball , this effect is caused by two distinct phenomena : the rapid rise in strangeness @xmath52 production below , and a rise in the antiquark @xmath102 yield above a energy reaction threshold . -.4 cm -.25 cm -14.75 cm 4.0 cm 2.6 cm the measured @xmath103 ratio by na49 is shown at top left of , where for comparison we also show the @xmath6 results . on right top , we present our results reduced to the correct relative scale , both for the total yield ratio for the ags rhic energy range , and for the central rapidity results from rhic . the solid line guides the eye to the fit results we obtained . to show that the @xmath104 ratio drop is due to a decrease in baryon density which leads to rise in the @xmath102 yield , we show in bottom section of , the nearly baryon density independent @xmath105 double ratio ratio , on left as function of @xmath106 , and on right as function of centrality of the reaction for @xmath107 gev : = . both upper and lower portion of are also drawn on same relative scale . seen how the horn specifically arises in the one @xmath104 , one can wonder if this is really a physical effect and how , in qualitative terms , a parameter @xmath62 , which controls the light quark yield , can help explain the horn structure seen in top of . we observe that this horn structure in the @xmath12 ratio traces out the final state valance quark ratio @xmath108 , and in language of quark phase space occupancies @xmath28 and fugacities @xmath109 , we have : @xmath110 in chemical equilibrium models @xmath111 , and the @xmath12 ratio and its horn must arise solely from the variation in the ratio @xmath112 and the change in temperature @xmath13 which both are usually smooth function of reaction energy . the isospin factor @xmath58 is insignificant in this consideration . as collision energy is increased , increased hadron yield leads to a decreasing @xmath113 . we recall the smooth decrease of @xmath114 with reaction energy seen in bottom panel in . the two chemical fugacities @xmath115 and @xmath116 are coupled by the condition that the strangeness is conserved . this leads to a smooth @xmath112 . the chemical potential effect is suggesting a smooth increase in the k@xmath117 ratio . for the interesting range of freeze - out temperature , @xmath118 is a smooth function of @xmath13 . normally , one expects that @xmath13 increases with collision energy , hence on this ground as well we expect an monotonic increase in the @xmath12 ratio as function of reaction energy . with considerable effort , one can arrange the chemical equilibrium fits to bend over to a flat behavior at @xmath119 . it is quasi impossible to generate the horn with chemical equilibrium model . to accomplish this , an additional parameter appears necessary , capable to change rapidly when hadronization conditions change . this is @xmath62 . it presence also allows @xmath13 to vary in non - monotonic fashion , as is seen in . the full shm is capable to describe the data , and we now show that it can pinpoint the properties of the phase of matter that was created early on in the reaction . to see this we consider what we learn from the final state data about strangeness and entropy production . for this purpose we consider both the specific per baryon and per entropy yield of strangeness . in addition , we look at the cost in thermal energy to make strangeness . all these quantities are nearly independent of the dynamics of hadronization , since they are related to processes occurring early on , ` deep ' inside the collision region , and long before hadronization . in the qgp , the dominant entropy production occurs during the initial glue thermalization , and the thermal strangeness production occurs in parallel and/or just a short time later @xcite . the entropy production occurs predominantly early on in the collision during the thermalization phase . strangeness production by gluon fusion is most effective in the early , high temperature environment , however it continues to act during the evolution of the hot deconfined phase until hadronization @xcite . both strangeness and entropy are nearly conserved in the evolution towards hadronization and thus the final state hadronic yield analysis value for @xmath120 is closely related to the thermal processes in the fireball at @xmath1212 fm / c . we believe that for reactions in which the system approaches strangeness equilibrium in the qgp phase , one can expect a prescribed ratio of strangeness per entropy , the value is basically the ratio of the qgp degrees of freedom . we estimate the magnitude of @xmath120 deep in the qgp phase , considering the hot stage of the reaction . for an equilibrated non - interacting qgp phase with perturbative properties : = = 0.027 . here , we used for the number of flavors @xmath122 and @xmath123 . we see that the result is a slowly changing function of @xmath116 ; for large @xmath124 , we find at modest sps energies , the value of @xmath120 is reduced by 10% . considering the slow dependence on @xmath125 of @xmath126 there is minor dependence on the much less variable temperature @xmath13 . the dependence on the degree of chemical equilibration which dominates is easily obtained separating the different degrees of freedom : = 0.027 _ s^qgp . we assume that the interaction effects are at this level of the discussion canceling . seen we expect to see a gradual increase in @xmath120 as the qgp source of particles approaches chemical equilibrium with increasing collision energy and/or increasing volume . we repeat that it is important to keep in mind that this ratio @xmath120 is established early on in the reaction , the above relations , and the associated chemical conditions we considered , apply to the early hot phase of the fireball . at hadron freeze - out the qgp picture used above does not apply . gluons are likely to freeze faster than quarks and both are subject to much more complex non - perturbative behavior . however , the value of @xmath120 is nearly preserved from the hot qgp to the final state of the reaction . how does this simple prediction compare to experiment ? given the statistical parameters , we can evaluate the yields of particles not yet measured and obtain the rapidity yields of entropy , net baryon number , net strangeness , and thermal energy , both for the total reaction system and also for the central rapidity condition , also as function of centrality . in passing , we note that , in the most central reaction bin at rhic-200 , @xmath127 baryons per unit rapidity interval , implying a rather large baryon stopping in the central rapidity domain . the rise of strangeness yield with centrality is faster than the rise of baryon number yield : @xmath128 is seen in the top left panel in fig.[pest ] . the solid ( blue ) lines are for the chemical nonequilibrium central rapidity yields of particles at rhic-200 . solid ( green ) lines , on right , are for total hadron yields and thus total yields of the considered quantities , _ e.g. _ strangeness , entropy . for the most central head - on reactions , we reach at rhic-200 @xmath129 . in the middle panel in we compare strangeness with entropy production @xmath120 , which we just evaluated theoretically . again , on left , as function of participant number @xmath68 at rhic-200 and , on right , for the total reaction system for the most central 5 - 7% reactions . on left , we see a smooth transition from a flat peripheral behavior where @xmath130 to smoothly increasing @xmath120 reaching @xmath131 in most central reactions . this indicates that even at rhic-200 for the more central reactions some new mechanisms of strangeness production becomes activated . on right , we see that the change on @xmath120 is much more drastic as function of reaction energy . after an initial rapid rise the further increase occurs beyond the threshold energy at slower pace . in the bottom panel , on left , we see the thermal energy cost @xmath132 of producing a pair of strange quarks as function of the size of the participating volume ( _ i.e. _ @xmath68 ) this quantity shows a smooth drop which can be associated with transfer of thermal energy into collective transverse expansion after strangeness is produced . thus , it seems that the cost of strangeness production is independent of reaction centrality . the result is different when we consider @xmath106 dependence of this quantity , see bottom panel on right . there is a very clear change in the energy efficiency of making strangeness at the threshold energy . we will return to discuss possible reaction mechanisms below . the chemical freeze - out conditions we have determined presents , in the @xmath13@xmath114 plane , a more complex picture than naively expected , see . considering results shown in , we are able to assign to each point in the @xmath13@xmath114 plane the associate value of @xmath106 . the rhic @xmath48 results are to outer left . they are followed by rhic and sps @xmath133 results . the dip corresponds to the 30 and 40 @xmath68gev sps results . the top right is the lowest 20 @xmath68gev sps and top 11.6 @xmath68gev ags energy range . to guide the eye , we have added two lines connecting the fit results . we see that the chemical freeze - out temperature @xmath13 rises for the two lowest reaction energies 11.6 and 20 @xmath68 gev to near the hagedorn temperature , @xmath134 mev , of boiling hadron matter . once the chemical non - equilibrium is allowed for , the data fit turns to be much more precise , and the picture of the phase boundary with smaller ` measurement ' error reveal a much more complex structure , and contains physics details prior analysis based on a rudimentary model could not uncover . the shape of the hadronization boundary , shown in in the @xmath13@xmath114 plane , is the result of a complex interplay between the dynamics of heavy ion reaction , and the properties of both phases of matter , the inside of the fireball , and the hadron phase we observe . the dynamical effect , capable to shift the location in temperature of the expected phase boundary is due to the expansion dynamics of the fireball which occurs at parton level , and effects of chemical nonequilibrium , see subsection [ tphase ] for full discussion . possibly , not only the location , but also the _ nature _ of the phase boundary can be modified by variation of @xmath28 . we recall that for the 2 + 1 flavor case , there is a critical point at finite baryochemical potential with @xmath135 mev @xcite . for the case of 3 massless flavors , there can be a 1st order transition at all @xmath114 @xcite . considering a classical particle system , one easily sees that an over - saturated phase space , _ e.g. _ , with @xmath136 for the purpose of the study of the phase transition acts as being equivalent to a system with 3.2 light quarks and 1.6 massive ( strange ) quarks present in the confined hadron phase . considering the sudden nature of the fireball breakup seen in several observables @xcite , we conjecture that the hadronizing fireball leading to @xmath137 super - cools and experiences a true 1st order phase boundary corresponding also at small @xmath114 . the system we observe in the final state prior to hadronization is mainly a quark antiquark system with gluons frozen in prior expansion cooling of the qcd deconfined parton fluid . the quark dominance is necessary to understand _ e.g. _ how the azimuthal asymmetry @xmath11 varies for different particles @xcite . these quarks and antiquarks have , in principle , at that stage a significant thermal mass . the evidence for this is seen in in its two bottom panels , showing the dimensionless variable @xmath96 . the energy end entropy per particle of non - relativistic and semi - relativistic classical particle gas comprising both quarks and antiquarks is ( see section 10 , @xcite ) : @xmath138 it is thus possible to interpret the fitted value @xmath139 in terms of a quark matter made of particles with @xmath140 , @xmath141 which is close to what is expected based on thermal qcd @xcite . we expect considerably more violent transverse expansion of the fireball of matter created at lhc . the kinetic energy of this transverse motion must be taken from the thermal energy of the expanding matter , and ultimately this leads to local cooling and thus a reduction in the number of quarks and gluons . the local entropy density decreases , but the expansion assures that the total entropy still increases . primarily , gluons are feeding the expansion dynamics , while strange quark pair yield , being weaker coupled remain least influenced by this dynamics . model calculations show that this expansion yields an increase in the final qgp phase strangeness occupancy @xmath70 @xcite . this mechanism , along with the required depletion of the non - strange degrees of freedom , in the feeding of the expansion , assures an increase in the @xmath142 ratio given the nearly 30-fold increase of collision energy . depending on what we believe to be a valid hadronization temperature for a fast transversely expanding fireball , the possible maximal enhancement in the @xmath142 ratio may be in the range of a factor 23 . perhaps even more interesting than the @xmath142 ratio enhancement would be the enhancement anomaly in strange ( antibaryon ) yields . with @xmath143 , we find that the more strange baryons and antibaryons are more abundant than the more ` normal ' species . specifically of interest would be @xmath144 , @xmath145 , and @xmath146 which should show a major , up to an order of magnitude shift in relative production strength . detailed predictions for the yields of these particles require considerable extrapolation of physics conditions from the rhic to lhc domain and this work is in progress @xcite . ultimately , strange @xmath0 , and @xmath52 quarks can exceed in abundance the light quark component , in which case we would need to rethink in much more detail the distribution of global particle yield . the ratios of neutral and charged hadrons would undergo serious change . regarding charm , we note that situation will not become as severe . given the large charm quark mass , we expect that most of charm quark yield is due to first hard interactions of primary partons . for this reason , the yield of strange and light quarks , at time of hadronization , exceeds by about a factor 100 or more that of charm at central rapidity . thus , even though charm phase space occupancy at hadronization may largely exceed the chemical equilibrium value , seen the low hadronization temperature , _ e.g. _ , @xmath147 , it takes a factor @xmath148 to compensate hadron yield suppression due to the high charm mass . said differently , while strange quarks can compete in abundance with light quarks for @xmath149 , charm ( and heavier ) flavor(s ) will remain suppressed , in absolute yield , at the temperatures we can make presently in laboratory experiments . we have shown that strangeness , and entropy , at sps and rhic are well developed tools allowing the detailed study of hot qgp phase . our detailed discussion of hadronization analysis results has further shown that a systematic study of strange hadrons fingerprints the properties of a new state of matter at point of its breakup into final state hadrons . we have shown that it is possible to describe the ` horn ' in the @xmath104 hadron ratio within the chemical non - equilibrium statistical hadronization model . we have shown that appearance of this structure is related to a rapid change in the properties of the hadronizing matter . of most theoretical relevance and interest are the implications of non - equilibrium hadronization on the possible change in the location and _ nature _ of the phase boundary . in summary , we have presented interpretation of the experimental soft hadron data production and discussed the production of strangeness and entropy that this analysis infers . we have seen , in quantitative way , how the relative strangeness and entropy production in most central high energy heavy ion collisions agrees with quark - gluon degree of freedom counting in hot primordial matter where the values of these quantities have been established . work supported in part by a grant from : the u.s . department of energy de - fg02 - 04er4131 . lpthe , univ.paris 6 et 7 is : unit mixte de recherche du cnrs , umr7589 . jr thanks bikash sinha and jan - e alam and the organizers of the 5th international conference on physics and astrophysics of quark gluon plasma , february 8 12 , 2005 salt lake city , kolkata , india for their kind hospitality . dedicated to professor bikash sinha on occasion of his 60th anniversary . 19 assessments by the rhic experimental collaborations : _ hunting the quark gluon plasma : results from the first 3 years at rhic _ bnl-73847 - 2005 formal report , april 18 , 2005 , to appear in nucl . phys . a ( 2005 ) . j. letessier and j. rafelski , _ hadrons and quark - gluon plasma _ cambridge monogr . . cosmol . * 18 * 1 - 397 ( 2002 ) ( cambridge , uk , 2002 ) . available on line to read freely with win and mac platforms at http://site.ebrary.com/pub/cambridgepress/doc?isbn=0521385369 g. torrieri , w. broniowski , w. florkowski , j. letessier and j. rafelski , [ arxiv : nucl - th/0404083 ] , computer physics communications in press , see : www.physics.arizona.edu/@xmath150torrieri/share/share.html j. rafelski , j. letessier and g. torrieri , `` centrality dependence of bulk fireball properties at rhic , '' arxiv : nucl - th/0412072 . j. letessier and j. rafelski , `` hadron production and phase changes in relativistic heavy ion collisions , '' arxiv : nucl - th/0504028 . j. rafelski and j. letessier , acta phys . b * 34 * , 5791 ( 2003 ) [ arxiv : hep - ph/0309030 ] ; and j. phys . g * 30 * ( 2004 ) s1 [ arxiv : hep - ph/0305284 ] . j. letessier , a. tounsi , u. w. heinz , j. sollfrank and j. rafelski , phys . * 70 * , 3530 ( 1993 ) . h. caines [ star collaboration ] , j. phys . g * 31 * , s1057 ( 2005 ) . g. e. bruno [ na57 collaboration ] , j. phys . g * 30 * , s717 ( 2004 ) [ arxiv : nucl - ex/0403036 ] . m. gazdzicki _ et al . _ [ na49 collaboration ] , j. phys . g * 30 * , s701 ( 2004 ) [ arxiv : nucl - ex/0403023 ] . s. s. adler _ et al . _ [ phenix collaboration ] , phys . c * 69 * , 034909 ( 2004 ) [ arxiv : nucl - ex/0307022 ] . h. b. zhang [ star collaboration ] , `` delta , k * and rho resonance production and their probing of freeze - out dynamics at rhic , '' , poster presentation at qm2004 , oakland , january 2004 [ arxiv : nucl - ex/0403010 ] ; + j. adams [ star collaboration ] , [ arxiv : nucl - ex/0412019 ] , phys . rev . c. ( 2005 ) in press . j. adams _ et al . _ [ star collaboration ] , phys . b * 612 * , 181 ( 2005 ) [ arxiv : nucl - ex/0406003 ] . j. letessier , a. tounsi and j. rafelski , phys . b * 389 * ( 1996 ) 586 . j. rafelski and j. letessier , phys . b * 469 * , 12 ( 1999 ) [ arxiv : nucl - th/9908024 ] . j. rafelski and j. letessier , nucl . a * 702 * , 304 ( 2002 ) [ arxiv : hep - ph/0112027 ] . b. kampfer , j. cleymans , p. steinberg and s. wheaton , heavy ion phys . * 21 * , 207 ( 2004 ) . a. peikert , f. karsch , e. laermann and b. sturm , nucl . * 73 * , 468 ( 1999 ) . c. bernard _ et al . _ [ milc collaboration ] , phys . rev . d * 71 * , 034504 ( 2005 ) . h. z. huang and j. rafelski , aip conf . proc . * 756 * , 210 ( 2005 ) [ arxiv : hep - ph/0501187 ] .
the two - degree field galaxy redshift survey ( 2dfgrs : @xcite ) and the main galaxy sample @xcite of the sloan digital sky survey ( sdss : @xcite ) , have provided detailed maps of the local structure from hundreds of thousands of galaxies of all types at a mean redshift of @xmath4 , while the sdss is also generating a large sample of more distant luminous red galaxies ( lrgs ) with @xmath5 @xcite . the present paper describes a new survey which combines the precision of the sdss photometric survey with the spectroscopic capabilities of the two - degree field instrument ( 2df ) on the 3.9 m anglo - australian telescope ( aat ) , to extend the lrg survey from @xmath6 to @xmath7 . lrgs constitute ideal tracers of large - scale structure at intermediate redshifts ( @xmath8 ) because they are intrinsically luminous and spectroscopically homogeneous , and can be reliably identified photometrically . lrgs are strongly clustered , with a bias towards high density regions in ways that are believed to be well understood . the shift in the focus of observational cosmology from determining cosmological parameters to attempting to constrain galaxy evolution and formation has accelerated over the past decade , especially since the release of the first wmap results @xcite . the spectroscopic information extracted from the 2dfgrs and sdss has provided important clues as to the nature of galaxies in the local universe and their recent history @xcite . although galaxies are generally believed to form hierarchically through a succession of mergers , it appears that the many of the most massive elliptical galaxies are very old as judged from their stellar content . thus it seems that the rate of the merger process depends strongly on the local density . a large survey of lrgs at @xmath9 therefore has two objectives : revealing the large scale structure and clustering of matter when the universe was about two thirds of its present age , and understanding the evolution of the most massive galaxies themselves . the sdss lrg survey @xcite is limited to @xmath10 by the fixed exposure time of 45 m , set for sdss spectroscopic observations in the lower redshift main galaxy survey . this is well above the limit to which lrgs can be selected reliably from the sdss imaging data . the original 2dfgrs included all galaxies with blue magnitude @xmath11 , which reached @xmath12 and required exposures of less than an hour on the aat . by increasing the exposure time to 4h and targetting only the relatively rare lrgs for spectroscopy in the red spectral region , 2df can reach @xmath13 and @xmath14 . the surface density of lrg targets requires only 200 of the 400 2df fibres within each @xmath15 diameter field , so all the lrgs were observed in one of the two 2df spectrographs . the other was used for a lower resolution survey of faint qsos @xcite which extends the earlier 2df qso redshift survey ( 2qz : @xcite ) . the two new surveys together comprise the 2df - sdss lrg and qso survey , 2slaq . a bonus of 2slaq is that there is some overlap in redshift range for the lrgs and qsos , which will enable a direct comparison between their spatial distributions . there is also scope for investigating environmental and gravitational lensing effects , through the comparison of foreground lrgs and the absorption spectra of nearby background qsos ( cf . @xcite ) . the 2slaq targets were selected within two narrow equatorial strips , each @xmath15 wide and about @xmath16 long running through the northern and southern galactic caps , chosen for having good photometry at the time of the first sdss data release ( dr1 : @xcite ) . the aim of the lrg survey was to obtain spectra for 10,000 galaxies in the redshift range @xmath17 . this paper describes the photometric target selection , spectroscopic observations and data analysis for the lrgs , and summarises the properties of the sample . basic cosmological parameters are being derived from the full sample , such as luminosity functions @xcite and spatial correlation functions and clustering @xcite . @xcite investigate the incidence of star formation as a function of redshift and have identified rare ` k+a ' galaxies which had a substantial burst of star formation within the last billion years , while @xcite identify and determine the cosmic evolution of almost 400 2slaq lrgs which are catalogued radio sources . the 2slaq lrg redshifts have been used by @xcite and by @xcite and @xcite to validate the determination of sdss photometric redshifts . distant lrgs were selected on the basis of sdss photometric data ( see @xcite for the definition of the @xmath18 @xmath19 system ) , using a two - colour plot of ( @xmath20 ) against ( @xmath21 ) and the @xmath22band magnitude . effectively , the selection uses a crude determination of photometric redshift as the 4000 break moves through the @xmath23 bands . the colour criteria are similar to those used for the cut ii sample of sdss lrgs @xcite , although the selection here is easier since beyond @xmath24 there is less confusion with lower - redshift galaxies . targets were also required to have non - stellar images . the selection of targets into several priority classes was done originally using the best sdss photometry available in 2003 . the later dr4 photometry @xcite has been used in the final redshift lists and for the diagrams in this paper . for most objects the changes amount to at most a few hundredths of a magnitude , with r.m.s . scatters of @xmath25 mag and negligible mean shifts of only @xmath26 mag in each colour . the photometry for some galaxies changed by larger amounts , due mainly to changes in how composite images were de - blended : as a result , a few tens of galaxies ( @xmath27% ) now appear to have colours inconsistent with their original selection classification . after some experimentation two main samples of lrgs were defined . the primary sample ( sample 8 in the data lists ) has a surface density of about 70 @xmath28 , chosen to maximise the completeness and spatial uniformity of 2df coverage for lrgs with @xmath29 ( see section 3.2 ) . the secondary sample 9 consists of galaxies with @xmath30 , to utilise fibres which could not be assigned to primary targets . a few targets falling outside the selection boundaries were included as ` fillers ' ( sample 0 ) . [ 2colsel ] illustrates the colour selection boundaries superimposed on the sdss photometric data . most galaxies of all types lie along a common locus in the lower left hand corner of this plot , becoming redder in @xmath21 with increasing redshift until the 4000 break moves into the @xmath31band at @xmath32 . thereafter , the @xmath20 colour becomes rapidly redder until the break moves into the @xmath22band at @xmath33 . thus the most massive and luminous intermediate redshift galaxies , i.e. lrgs with a dominant passively - evolving population , are expected to lie along a vertical track with @xmath34 in fig . [ 2colsel ] . 2-colour diagram . the primary ( sample 8) lrg targets lie in the top zone in the plot , above the diagonal lines @xmath35 and the upper @xmath36 cut at 0.65 ( see text for details ) . the secondary ( sample 9 ) objects lie between the two parallel @xmath36 cuts at 0.65 and 0.55 . most of the points below the @xmath37 cut ( in the bottom right hand quadrilateral ) represent lower redshift lrgs from the initial observing runs . passively evolving lrgs with @xmath38 are expected to lie along an approximately vertical locus in this plot , with @xmath34 and with @xmath20 increasing from @xmath39 to @xmath40 as redshift increases from 0.4 to 0.7 . nearly all lower redshift galaxies of all types lie below the @xmath41 boundary in the bottom left hand corner . the points with @xmath42 are mainly due to errors in the sdss @xmath43-band photometry for such faint objects.,width=359 ] cuts above lines of constant @xmath44 ( cf . @xcite ) select early - type galaxies at increasingly high redshift . the top priority primary sample has @xmath45 while the secondary sample has @xmath46 . a second cut with @xmath47 eliminates later - type galaxies . the colours used here are the extinction - corrected @xmath48 colours : the definition of these and other sdss parameters can be found in @xcite or at ` http://www.sdss.org/dr4/algorithms/photometry.html ` . the third main selection parameter is the de - reddened magnitude in the @xmath22band : the 2slaq lrgs have @xmath49 where @xmath50 is the total magnitude based on a fit to a de vaucouleurs profile and @xmath51 is the extinction in the @xmath22-band . this choice of a fixed limiting magnitude enables the selection of bright lrgs out to @xmath52 , although it also means that the sample at @xmath53 includes a substantial number of fainter early - type galaxies with luminosity @xmath54 . further cuts @xmath55 eliminate objects too far from the main lrg locus , probably mostly composite objects or the result of photometric errors . the photometric selection criteria are summarised in table [ selcrit ] . the first two rows define the primary sample 8 and secondary sample 9 , which comprise 67% and 27% respectively of the final lrg survey . 4% of the objects are in samples 36 , observed in march and april 2003 using somewhat different criteria ( section 2.2 ) . the final 2% fell outside the selection boundaries and have been assigned to sample 0 . @ccccccl sample & @xmath36 & @xmath41 & @xmath50 & @xmath21 & @xmath20 + s8 & @xmath56 & @xmath57 & @xmath58 & @xmath59 & @xmath60 + s9 & @xmath61 & @xmath57 & @xmath58 & @xmath59 & @xmath60 + + s3 & @xmath62 & @xmath57 & @xmath63 & @xmath64 & @xmath60 + s4 & @xmath65 & @xmath57 & @xmath63 & @xmath64 & @xmath60 + s5 & @xmath66 & @xmath57 & @xmath63 & @xmath64 & @xmath60 + + s6 & @xmath62 & @xmath57 & @xmath67 & @xmath64 & @xmath60 + star / galaxy separation based on the sdss images eliminates most stellar contamination from the sample . two criteria were used , @xmath68 and @xmath69 but some cool m - dwarf stars remain and these comprise about 5% of all targets . objects too diffuse to yield useful spectra using the 2 arcsec diameter 2df fibres were eliminated by requiring @xmath70 ( such objects are liable to be spurious in any case ) . the sdss 3 arcsec fibre diameter is used for convenience , although the 2df fibres are only 2 arcsec in diameter . the scatter of points on the red side of the main clump in fig . [ 2colsel ] , beyond @xmath71 , is mainly due to photometric errors in the sdss @xmath43-band . most points in this region have @xmath72 and correspond to galaxies with @xmath73 and hence @xmath74 . at this very faint level the mean error in @xmath43 is more than 0.3 mag . different cuts and priorities were used for the initial observing runs in march and april 2003 . the original magnitude limit was @xmath75 but it became apparent that reliable redshifts could be determined to somewhat fainter limits . tests showed that @xmath76 was a realistic limit in exposure times of 4 hours . the number counts are sufficiently steep that this relatively small change had two very significant consequences : the proportion of high redshift galaxies with @xmath77 doubled to @xmath78% and half of the primary targets are in the faint extension . effectively the original limit became the new median redshift , at @xmath79 . three colour - selected samples were defined originally : a top priority class consisting of all objects with @xmath80 and two sparse - sampled classes with @xmath81 and @xmath82 ( samples 3 , 4 and 5 respectively ) . these targets all had @xmath83 and @xmath84 . a special faint sample 6 with @xmath85 was observed in one test field , d10 , and was used to determine the optimum magnitude limit for the remainder of the survey . in addition to the different sample specifications , there were subsequent small revisions to the sdss dr1 photometry . most of the early observations could be reassigned to samples 8 or 9 and nearly all of the fields have been re - observed to maintain the uniformity of the survey , but about 900 galaxies remain outside the primary and secondary samples . the aat 2df fibre system @xcite is ideally suited to this project . the lrg surface density is @xmath8670 per sq deg down to a magnitude limit of @xmath13 , corresponding to an sdss fibre magnitude of @xmath87 . 2df can provide reliable redshifts for over 90% of such targets in four hours , in average conditions . atmospheric refraction limits the maximum time for which a given 2df configuration can be efficiently observed to about 3h for fields on the equator , so most fields were observed on two consecutive nights . the most practical survey design was two narrow strips along the celestial equator , one in each galactic hemisphere , given the early sdss photometric coverage and the requirement to work through complete nights . there was also a desire to overlap with previous surveys such as 2qz @xcite and the millenium galaxy survey @xcite and to enable follow - up observations from large telescopes in both hemispheres . the northern strip runs from 8.2h to 15.3h in ra , broken into five sub - strips to utilise the best photometric data ; the southern strip runs from 20.6h to 4.0h . each strip can be covered by a single row of overlapping 2df fields . the ends of the strips come close to galactic latitude @xmath88 and thus suffer from significant extinction of up to 0.4 mag in @xmath43 ( based on the maps of @xcite ) , plus substantial foreground star contamination . priority was therefore given to fields nearer the centre of each strip with @xmath89 and extinction @xmath90 mag . [ fields ] shows the layout of the target strips and the 2df fields actually observed . there is considerable imbalance between the two strips , with the northern strip containing more than two thirds of the data and having much higher completeness . this was an accidental consequence of the nights scheduled and variable observing conditions . open circles represent fields with completeness @xmath91% or with non - standard selection criteria . the letters ` a'-`e ' and ` s ' signify the sub - strips and are used as the first character of the field names.,width=377 ] the principal objective was to maximise the total number of objects in a photometrically well - defined sample , with good coverage on different spatial scales and minimal incompleteness . for simplicity it was decided to use a fixed field separation . for the initial observations in march and april 2003 the centres were set @xmath92 apart but this was increased to @xmath93 for all subsequent observations , following modelling tests of the yield . two strategies helped to maximise the fraction of targets which were observed . about 28% of the targets lie in the overlap regions of adjacent 2df fields . targets which had been observed in the configuration for one field were given lower priority in the adjacent fields . a sequence of alternate fields was observed first , followed by the partially overlapping fields in a second pass along the strip . this led to independent repeat observations for a few percent of the total sample , giving a valuable check on the internal accuracy of the data . the second strategy arose because each field was observed on two or more nights . a significant fraction of the targets , mainly the brighter galaxies and most of the m - stars , had good spectra and unambiguous redshifts after only one night . typically some 1020% of the fibres could then be re - allocated to new targets . this helped in crowded fields in particular , since due to physical constraints 2df fibres have a minimum separation of at least 30 arcsec . tests of the 2df configure software showed that the algorithm which assigns fibres to targets could introduce significant patterns into the distribution of observed objects across the 2df field . the main effect was a discontinuity in density at @xmath94 radius , if the mean number of targets per field in the input catalogue was much higher than the number of fibres . there could also be systematic effects depending on the ordering of the input catalogue . both effects were avoided in the 2slaq surveys by keeping the surface density of the top priority targets below 70 @xmath28 and by randomising the order of the targets in the input catalogue . there are instrumental constraints on the targets accessible to 2df . the nominal @xmath15 field can be up to @xmath95 in diameter , but this varied slightly during the survey due to both hardware and software limits . there are also 20 small triangular regions around the edge of the field which are inaccessible to the fibres feeding one spectrograph . the net result is that the effective area of the survey can be approximated by a row of overlapping @xmath15-diameter circles . [ xy ] shows the distribution of targets across the 2df field , for all objects in the final redshift catalogue . there is a slightly lower density of points towards the right hand edge of the field due to the asymmetrical distribution of sky fibres ( see section 4.2 ) . both this density gradient and the empty triangles on the eastern and western edges are largely eliminated when the data from overlapping fields are combined , but the northern and southern triangles remain . field diameter ; the solid circle at @xmath92 radius indicates the effective area covered . the empty triangles around the edge are inaccessible to the fibres feeding the single spectrograph used for the lrgs . the lower density towards the right - hand side of the field is due to the allocation of 2030 of the western fibres to sky : those fibres feed the central rows in the spectrograph camera.,width=359 ] in addition to the inaccessible regions shown in fig . [ xy ] , each fibre button has a large ` footprint ' with a long tail due to the fibre , which runs across the front of the 2df field plate . it is never possible to access targets within 30 arcsec of another which has already had a fibre allocated , and in some directions the exclusion zone is much larger . there is therefore a strong intrinsic bias against observing close pairs of galaxies , and very few fibres can be allocated to members of one cluster . this bias is partially overcome within the overlap regions of adjacent 2df fields , and a few fields were independently observed in different runs . the bias is also alleviated by the policy of re - allocating some fibres after the first night on a given field ( section 3.1 ) . however , there must remain a substantial bias against close pairs of targets and the level of incompleteness will have to be assessed by comparing the objects actually observed with the full input catalogue . there is a similar conflict between the lrgs and qsos , since the two classes of objects were observed simultaneously . thus , although one of the strengths of 2slaq is having samples of lrgs and qsos in the same fields and within the same redshift range , there is a bias against qsos which lie near lrgs since the lrgs were given higher priority than the qsos in the allocation process . the lrgs were observed with 2df spectrograph number 2 , using a 600 lines mm@xmath96 @xmath97 grating . the detector was a tek1024 ccd with @xmath98 pixels , giving a dispersion of 2.2 pixel@xmath96 and an effective resolution of about 5 or @xmath99 . almost all the spectra were taken with the central wavelength set to 6150 , covering the range 5050 to 7250 . this enables secure determination of redshifts for @xmath100 using the ca ii h&k lines and the detection of [ o ii ] 3727 down to @xmath101 . for the first observing run in march 2003 the central wavelength was set to 6350 but almost no useful data were obtained beyond 7250 , due to telluric features which did not cancel well in the low dispersion 2df spectra . more details of the spectrographs and their performance are given in @xcite . a subset of the 2df fibres was assigned to clear sky positions , after checking the sdss images . these are used to determine a median sky spectrum which is subtracted from all target spectra . for each frame 30 fibres were assigned to sky ( later reduced to 20 , after tests showed that this resulted in negligible degradation ) . these sky fibres were selected to lie in the central half of the ccd ( i.e. between spectra 50 and 150 ) to keep the sky spectra uniform , although this results in some mismatch to spectra at the edges of the ccd due to optical distortions in the camera . in principle it is possible to correct for these but this has not yet been implemented in the 2df software . the fibres which feed the centre of the ccd happen to lie on the western side of the 2df field , which causes the slight density gradient seen in fig . [ xy ] . the standard pattern was to take sets of seven exposures per field : a quartz lamp flat - field , an arc lamp exposure using 4 cu ar lamps and 2 he ar lamps , @xmath102s exposures on the targets and a final arc exposure . thus in good conditions four or occasionally five sets of observations were obtained per night and four 2df fields could be fully observed in two nights . in poor conditions the observations were extended over extra nights , until the required mean signal - to - noise per pixel ( s / n ) ( and hence redshift completeness fraction ) was reached . the 2slaq surveys were supported approximately equally by the australian and british time assignment panels , plus a few director s discretionary nights , to give a total allocation of 87 nights between march 2003 and august 2005 . on average about 60% of the time was useable . a total of 102 datasets was obtained for 80 2df fields , 54 in the northern strip and 26 in the south . observations were deemed to be complete ( i.e. reliable redshifts for at least 85% of the primary targets ) for 72 fields while 8 are either incomplete or used non - standard selection criteria . the final yield was very close to one completed field per allocated night , about half of the best possible rate . this is lower than the overall fraction of clear time because 2df can not be re - configured quickly . the 80 distinct fields which were observed are illustrated in fig . [ fields ] and all the datasets are listed in appendix a , along with various quality parameters which are discussed in section 5.5 . some fields have more than one dataset because data were obtained in separate observing runs , sometimes with different parameters or target selections , and were reduced independently . there is some scope for combining such datasets before analysis which would lead to better quality spectra for repeated targets , but the gain in redshift completeness will be very small . given that it was impossible to cover entire strips in the time available , the aim was to observe several continuous sub - strips within each galactic hemisphere . in the event , the time allocated and the observing conditions led to more than twice as much data being obtained in the northern strip as in the south . nine separate sub - strips have been observed with varying levels of completeness . the proportion of primary ( sample 8) input targets for which reliable redshifts were obtained reaches 90% in the best regions such as northern sub - strips ` b ' and ` d ' , each of which is @xmath103 long . each of these sub - strips covers a volume of space comparable to that observed in the northern 2dfgrs strip , since the angular extent of the latter was about five times larger but the mean redshift was only a fifth as high . all of the raw data have been analysed using the aao 2dfdr software @xcite . for each field , the location of the fibres on the ccd was determined using a quartz lamp exposure which was also used as a flat field to remove pixel - to - pixel sensitivity variations . two arc exposures provided wavelength calibration . all spectra were scaled according to the relative throughput of the fibres , as determined from the strongest night sky lines , and a median sky spectrum was subtracted from each object spectrum . the different frames for each field were combined using mean flux weighting , which takes account of the variable signal levels arising from changes in ` seeing ' , transparency or exposure time ; cosmic ray events were removed during this final step . the 2dfdr software was developed for the analysis of the 2dfgrs/2qz data . for those surveys , the data for each field consisted of several similar frames with precisely the same 200 targets , all taken on the same night . the 2dfdr software was modified during the course of the 2slaq surveys to cope with data taken on different nights , sometimes with significant changes to the central wavelength and often with altered allocations of fibres to targets . the lrg redshifts were derived using a modified version of the zcode fortran program , developed by @xcite and others for the 2dfgrs spectra of low redshift galaxies @xcite . that program determined two independent redshifts whenever possible , one based on discrete emission lines and the other using cross - correlation with a set of template spectra . the higher resolution 2slaq lrg spectra cover only half the wavelength range and the targets have been photometrically selected to be predominantly passively evolving early - type galaxies at much higher redshifts ( @xmath104 ) . thus it is rarely possible to derive a secure redshift using emission lines alone : the only line which is common is [ o ii ] at 3727 , detected by the zcode in about 20% of the spectra . the 2slaq version of the zcode derives redshifts by cross - correlating each spectrum against a set of templates and uses any emission lines only as a check on the cross - correlation results . the cross - correlations are done using the method of @xcite , which involves fourier transformation of all the spectra and templates . the program selects the most plausible redshift for each object and assigns a quality parameter ` q ' , which is a measure of the consistency and reliability of the initial set of redshift estimates . another indicator of the reliability of the redshift is given by the @xcite parameter ` @xmath31 ' ( called @xmath105 here to avoid confusion with the photometric @xmath31 ) , which measures the height of the cross - correlation peak relative to the general noise level . the spectrum , cross - correlation function and redshift for each target are shown on an interactive graphic display . the operator can either accept the automatic redshift or choose a different value ( e.g. by making alternative identifications of strong features or by selecting a different template ) . the operator assigns an independent quality parameter to the reliability of each redshift . the lrg version of the zcode can be run in several modes : the basic fully interactive mode in which every spectrum is visually checked ; a quick mode in which only spectra with low quality redshifts are inspected ; and a fully automatic mode . the advantage of inspecting every spectrum is that unusual or extreme objects can be identified , including composite spectra ; the automatic mode has the advantages of being more consistent and objective . the semi - automatic ` quick ' mode provides a good compromise between the two methods and is normally used for initial data reductions at the telescope . given the rather homogeneous 2slaq spectra , it has proved possible to tune the automatic zcode to give very reliable results for most objects . fewer than 1% of the galaxies give significantly discrepant redshifts from two independent analyses of the same dataset or from repeat observations of the same targets ( section 6.1 ) . a majority ( 72% ) of the redshifts for the initial data release have been derived automatically , without visual inspection . some datasets from each observing run have been re - analysed in fully interactive mode to verify that the error rate remains low . all redshifts have been corrected to be heliocentric . information on the templates and other details specific to the redshifting of the 2slaq lrgs are given in appendix b. the principal indicator of lrg redshift reliability is the parameter @xmath106 . note that @xmath106 does not necessarily indicate the quality of the actual spectrum . although the two are strongly correlated in general , there are cases where a very poor spectrum yields an unambiguous redshift , and conversely . two quality values are output by the zcode : the automatic value ( @xmath107 ) set by the code itself , and the operator value ( @xmath108 ) set during the visual inspection of the data . the preferred value is generally the operator value , where both exist . the definitions of the operator values are as follows : @xmath109 : a good spectrum with an unambiguous redshift ( usually @xmath110 ) , based on several strong features . @xmath111 : a single strong cross - correlation peak ( with @xmath112 ) and some obvious features ( most often h&k , g - band and/or balmer lines ) . @xmath113 : a plausible redshift but not completely convincing ( @xmath114 ) , or where alternative redshifts are possible . @xmath115 : a spectrum where no believable redshift can be found . in each case , the presence of [ o ii ] 3727 emission ( or , rarely , [ o iii ] at 5007 ) at the absorption redshift can be used to increase the @xmath106 value by 1 unit ( but with @xmath109 remaining the maximum for @xmath108 ) . checks on multiply - observed objects from overlapping fields , and from repeat observations of the same sets of targets , show that the reliability of the different classes is @xmath116% for @xmath109 , @xmath117% for @xmath111 and @xmath118% for @xmath113 ( see also section 6.1 ) . the automatic quality parameter @xmath107 is derived somewhat differently but produces very consistent results . the evaluation is done in two stages : first the cross - correlation results are compared for the full set of templates to determine the parameter @xmath119 for the most probable redshift ( i.e. the one which gives the highest @xmath105 value ) , and then the list of possible emission lines is checked for any identifications at the same redshift . the @xmath107 value is also based on the consistency of the redshift across the set of templates . the m - stars are dealt with slightly differently , since they only ever fit one template ( and that template , t8 , never gives a correct redshift to any galaxy ) . @xmath107 takes values in the range 0 to 5 . comparisons between automatic and manual ( i.e. fully interactive ) analyses of the same datasets show good agreement between the two sets of @xmath106 values . in particular , it is very rare to get any discrepancies between redshifts for which both methods give @xmath120 , which comprise 90% of the targets for most datasets . a few percent of targets lie on the @xmath121 borderline and these sometimes show discrepant redshifts , as is to be expected . the overall quality of each dataset ( i.e. each set of combined frames for a given 2df field ) can be evaluated in several ways , using four global parameters listed in table [ fldlist ] in appendix a. these parameters are the number of raw data frames , @xmath122 ; the mean s / n per pixel ; the mean difference between the logarithm of the counts obtained and the expected magnitudes of the targets , @xmath123 , using here the sdss @xmath124 parameter ; and the percentage of targets with reliable redshifts . the redshift success rate , defined as the fraction of targets which have redshift reliability @xmath106 of 3 or more , is the most important one for the lrg survey , while the mean s / n should be the most reliable measure of overall quality . the observing strategy attempts to minimise the variations in mean s / n by taking more frames in poor conditions , but this is only partially successful since non - statistical errors begin to become significant when many frames are combined . [ yields ] shows the correlation between @xmath125 success rate and mean s / n ( per pixel , per object ) for all the datasets listed in table [ fldlist ] . the northern galactic strip has systematically better data than the southern strip , mainly due to better weather conditions . datasets for which @xmath126 always have @xmath125 completeness @xmath127% while @xmath128 normally results in completeness of more than 85% , the level chosen as the practical limit for deciding when to terminate the observations of a field . most of the datasets with low completeness were re - observed to give higher final redshift completeness . data from fields with less than 85% final completeness are retained in the lrg catalogue but must be treated with caution in any statistical studies . nearly all the datasets with s / n@xmath129 are for northern fields observed early in 2003 , when a brighter magnitude cut - off was used . the 2slaq lrg survey is spatially incomplete ( see fig . [ fields ] ) and has a variable redshift success rate ( table [ fldlist ] ) . however , a particular effort has been made to maximise the completeness for the primary targets within several continous sub - strips . the top panel of fig . [ completeness8 ] shows the number of potential targets in each field of the northern strip for samples 8 and 9 , with considerable cosmic variance between fields . the middle panel shows the percentage of primary targets with reliable redshifts , which is close to 90% on average and reasonably uniform . this is the overall completeness , i.e. the fraction of targets observed multiplied by the redshift success rate . the bottom panel of fig . [ completeness8 ] gives the equivalent plot for the secondary sample 9 objects , for which the completeness is much lower and much less uniform , ranging from 20% up to 90% . sample 9 utilised only those 2df fibres which could not be placed on primary targets : it tends to be least complete in the highest density fields . the completeness is particularly poor for the low galactic latitude fields a01 and a02 , which were observed only with the original 2003 selection criteria . similarly , the southern field s01 was abandoned after its high levels of stellar contamination and foreground reddening were recognised . it is difficult to define the precise boundaries of the 2slaq survey , given the overlapping tiling pattern , the inaccessible triangles near the edge of the field ( fig . [ xy ] ) and the fibre re - allocation observing strategy . one simple way to derive a very complete sample is to work in rectangles which lie completely within the patterns of overlapping 2df circles . two such inscribed rectangles can be drawn in the well - observed northern sub - strips ` b ' and ` d ' , containing over 30% of the entire lrg survey . the targetting completeness is 94.5% for the primary sample 8 and the redshift completeness is 96.7% , giving an overall completeness of 91.4% . the best way to evaluate the completeness of the 2slaq survey , and to calculate its effective total area , is to create a spatial mask including all objects in the input lists which could in principle have been accessed by 2df from at least one of the observed field centres . this has been done by @xcite and the resulting mask will be made available as part of the survey data . two ` ascii ` listings are generated by the zcode for each 2df field . one , ` filenamez.rz ` , gives the target names , positions , fibre numbers , redshift parameters and some photometric and quality data , with one line for each spectrum analysed . the second , ` filenamez.zlog ` , gives much more data on the redshifting process , including the results of cross - correlating every target against each template and lists of strong emission lines found . some overall statistics , such as the numbers of reliable redshifts and the mean s / n of all the spectra , are also recorded . the ` * z.rz ` files are concatenated and sorted to generate the final output redshift lists for the lrg survey . the target names follow the standard iau convention , i.e. using the truncated j2000 position in right ascension and declination . these are quoted to a precision of 0.1 arcsec , which ensures unambiguous identification of the objects . however , one drawback is that sometimes the names can change slightly when the input sdss photometric data are revised , or when positions are converted between arcsec ( and seconds of time ) and decimal degrees or radians . the only safe way to make cross - identifications with other catalogues is to look for positional coincidences , rather than identical names . multiple observations were made for about 20% of the targets , either because they lay in the overlap regions of adjacent 2df fields or because entire fields were re - observed . the 2dfdr and zcode software originally assumed that all observations for a given dataset were taken using the same configuration of 200 fibres with the same spectrograph set - up . thus two or more independent spectra were obtained for many objects and two catalogues have been generated , one listing every spectrum obtained and the other with a single entry for each discrete object observed . the latter , which is the basic redshift catalogue , simply uses the best available redshift for each object . the selection is based on the redshift quality parameter @xmath106 , or on mean s / n when two spectra give the same @xmath106 . better spectra could be obtained for some repeated objects by combining the independent spectra from different observing runs . every spectrum should also be checked visually , to pick up any very peculiar objects or ones where the automatic redshifting has missed some obvious feature such as a composite spectrum . however , this will make very little difference to the actual redshifts : there should simply be a small increase in the number which are reliable . the radio sources which comprise about 3.5% of the 2slaq lrgs @xcite have been used as a quality control sample and confirm this expectation . the redshift lists are held in the 2slaq team archive at the university of queensland , ` http://lrg.physics.uq.edu.au/ ` and can be accessed via links at the university of portsmouth ( ` http://www.2slaq.info ` ) , at the aao ( ` http://www.aao.gov.au ` ) and elsewhere . the archives also contain the full sets of 2df reduced data , the individual lrg spectra and sets of sdss postage stamp images . a catalogue of the best available redshift for every object observed is being released in mid-2006 , together with the basic sdss photometric data and other parameters . the final full set of 2df spectra and sdss photometry will be made publicly available once the data re - reductions are complete . a number of representative galaxy spectra , plus one of the contaminating m stars , are illustrated in fig . [ spectra ] . [ spectra](a ) is a good spectrum of j110954.40 + 004843.1 at @xmath130 , a typical 2slaq lrg close to the median redshift . the h&k lines of ca ii and strong 4000 break are unambiguous determinants of the redshift and several other stellar absorption features are very obvious , but no emission lines are detectable . \(b ) is the galaxy j022459.77 - 000156.4 at @xmath131 , with both [ o ii ] and the [ o iii ] doublet in emission . this galaxy is close to the lower redshift cut - off for 2slaq lrgs and shows why the spectra very rarely yield reliable emission - line redshifts , since both [ o ii ] and [ o iii ] fall within the 2slaq spectral window for only a very narrow redshift range . \(c ) is j225439.77 - 001501.2 at @xmath132 . the best cross - correlation used the k - type stellar template but the redshift reliability was low . the redshift is secure only because high excitation emission lines of [ ne v ] 3426 and [ ne iii ] 3870 are clearly present . \(d ) is j212907.12 + 002405.4 at @xmath133 , one of the relatively rare ` e+a ' or ` k+a ' galaxies in the 2slaq sample ( @xcite ) . these galaxies show very strong balmer absorption lines , with h-@xmath134 making the ca ii h line apparently much stronger than k , and a relatively weak g - band and 4000 break . [ o ii ] 3727emission is often present . these galaxies are understood to have had a substantial episode of star formation up to @xmath135 years ago . \(e ) is j122554.82 - 000752.4 at @xmath136 , one of the highest redshift galaxies in the sample . the continuum is close to zero over much of the spectrum , but the strong h&k lines plus [ o ii ] yield an unambiguous redshift . like many high redshift lrgs , this galaxy shows ` k+a ' features similar to fig . [ spectra](d ) . this spectrum was obtained in march 2003 with the wavelength range shifted 200redwards compared with all later spectra : it is apparent that the data beyond 7250 were not useful . \(f ) is j124853.73 - 010332.7 at @xmath137 , a typical m - star with very characteristic molecular absorption bands . about 5% of the targets are foreground galactic m - dwarfs ; contamination by other stellar types is negligible . considerable effort has gone into making the catalogue of 2slaq lrg redshifts complete and reliable . this involves not only determining good redshifts for as many lrgs as possible but also identifying those objects that are not lrgs within the redshift range of primary interest . usually @xmath138% of the targets give unambiguous redshifts in a first pass through the data . by far the most common cause of failure is low s / n , due to either target faintness or fibre misplacement . a few targets ( @xmath139% ) fail to give a reliable redshift due to other causes such as instrumental artefacts or proximity to bright stars , but these are not liable to introduce any systematic bias into cosmological analyses of the lrgs . about twenty composite spectra have been found and more would no doubt be revealed by systematic searches : the zcode software finds only the most probable redshift for each target . most of the composites consist of an m - star plus a galaxy while a few show a second emission line redshift on top of the primary spectrum . the photometric data are unlikely to be meaningful when a strong m - star signature is present so the redshift is generally given as zero in such cases . any secondary redshift is noted since it may still be useful , e.g. if a target is identified as a radio source . several internal checks have been carried out by comparing different analyses of the same data set , different sets for the same field , and finally targets which have been observed independently in different fibres from different configurations . the best datasets yield very high redshift completeness , i.e. reliable ( @xmath120 ) redshifts for @xmath140% of the targets . comparisons between analyses of such sets by different zcode operators show excellent agreement in the preferred redshift , with only a handful of objects differing , mostly among those with @xmath113 . there can be more spread in the @xmath106 ( @xmath108 ) values since this is a somewhat subjective measure , but nearly all agree within 1 unit . several pairs of datasets for the same field have been compared . the number of significant redshift discrepancies , i.e. where two different redshifts both with @xmath141 are claimed , averages less than 1% . most of these are cases where one or both cross - correlation redshifts are on the borderline between quality 2 and 3 , a few are due to the automatic code missing an obvious emission line fit , a few are due to manual operator errors , and a couple have been composite spectra with two valid redshifts . the largest set of independent spectra , mostly taken in different configurations using different fibres , consists of repeat observations in 2005 of 1253 objects from the first two years . just 9 pairs were significantly discrepant , in that two apparently good redshifts with @xmath142 differed by more than 0.01 . three of these were composite star+galaxy spectra where both redshifts were valid and one was a very unusual emission line galaxy . the remaining five each had one unambiguous redshift while the other was a marginal redshift close to the boundary between @xmath111 and @xmath113 . thus the error rate is only about 0.5% . an external check of redshift reliability is provided by the sdss lrg survey which has some overlap in the redshift range @xmath143 ( see section 6.3 ) . the redshifts agree to within 0.002 for 143 of 145 galaxies in common , the only discrepancies being for 2 of the 3 galaxies for which the 2slaq redshift were unreliable , i.e. with @xmath113 . the @xmath142 redshifts are evidently 100% reliable in this comparison set . the redshift reliability depends strongly on s / n . histograms of s / n for the objects with and without reliable redshifts are shown in fig . [ snhist ] . the overall success rate ( i.e. @xmath142 ) for all targets in samples 8 and 9 was 92% . this rises to 99% for spectra with s / n @xmath144 but falls to just over 50% for the @xmath145% of spectra with s / n@xmath146 . are adequate for determining redshifts.,width=355 ] fig . [ snz ] shows the run of s / n per pixel with redshift , for spectra which yielded reliable redshifts . the average s / n decreases as @xmath125 increases , from a median value of about 5.0 at @xmath147 to 3.0 at @xmath148 . this is partly because the galaxies become systematically fainter with redshift and partly because only a small section of spectrum redwards of the 4000 break remains within the 2slaq window . night sky emission lines and atmospheric absorption features beyond 6800 exacerbate the latter effect . the overall shape of the distribution is a reflection of the plot of redshift against magnitude ( see fig . 10 , section 8) . a few galaxies are listed with negative s / n , which is clearly non - physical . these are usually spectra where the continuum drops below zero due to over - subtraction of the sky or scattered light . the redshift code can sometimes still give a reliable @xmath125 value in such cases . , for all primary ( sample 8) and secondary galaxies with reliable redshifts . a few objects fall below the s / n = 0 boundary ( dotted line ) due to errors in sky subtraction , but retain sufficient features for a reliable redshift determination.,width=355 ] the comparison with the sdss lrgs provides a quantitative check on the accuracy of both sets of redshifts . there is a mean shift of 0.00031 between the two sets , arising from the use of air wavelengths for the 2slaq templates and vacuum wavelengths by the sdss . the rms difference for the sample is 0.00040 , after removing two discrepant 2slaq redshifts and a further three outliers with @xmath149 . this implies that the internal accuracy of the @xmath142 2slaq redshifts is no worse than @xmath150 . a similar check can be derived from repeated 2slaq spectra . the rms scatter between sets of data taken in different years was typically about 0.0006 . some of this scatter arose from differences between the templates , partly intrinsic and partly due to calibration errors . when the velocity zero - points were corrected ( appendix b1 ) the rms scatter fell to 0.00045 . this again implies an internal error in a single redshift of @xmath150 . the overall statistics for the 2slaq lrg survey are summarised in table [ zstats ] . nearly 18500 spectra were obtained over three years , for 14978 discrete objects . 13784 of these ( 92% ) have good ( @xmath151 ) redshifts . about 5% of the targets turn out to be foreground m stars , leaving a total of 13121 galaxies of which 11451 have @xmath29 . these statistics are for all 80 discrete 2df fields which were observed . a few fields have low quality data or were observed with only the original photometric selection criteria . in particular , fields a01 , a02 , s01 and s12 have less than 50% completeness for the primary sample 8 targets and should be ignored in statistical analyses ( see table [ fldlist ] ) . three columns of figures are given in table [ zstats ] . column 1 is for the primary sample 8 which is the most complete and homogeneous sample , column 2 for the secondary sample 9 and column 3 for all objects observed . sample 9 is photometrically homogeneous with high redshift completeness and only 1% contamination by m - stars , but it has very variable spatial completeness . the first two rows of table [ zstats ] give the number of discrete objects observed and the number which have reliable redshifts . row 3 gives the number of contaminating stars ( virtually all m - stars ) . row 4 is the number of _ bona fide _ lrgs and row 5 gives their median redshift . the final three rows give the numbers of galaxies in the primary redshift target range of @xmath17 and in the low and high redshift tails . most of the @xmath152 galaxies are either in sample 9 or were observed in the first semester , before the selection criteria were refined . virtually all of the highest redshift galaxies , with @xmath153 , are from sample 8 . @lrrr selection & sample 8 & sample 9 & all + observed & 10072 & 3977 & 14978 + @xmath125 ( @xmath142 ) & 9307 & 3624 & 13784 + + stars & 551 & 37 & 663 + lrgs & 8756 & 3587 & 13121 + median @xmath125 & 0.55 & 0.47 & 0.522 + + @xmath17 & 8289 & 2647 & 11196 + @xmath154 & 214 & 935 & 1664 + @xmath155 & 253 & 5 & 261 + the total area covered by the 2slaq survey is approximately 180 @xmath1 , calculated as the number of fields observed multiplied by the effective area of each field , corrected for edge effects and the overlap between adjacent fields . however , this is not a particularly useful number since the completeness of the survey varies significantly from field to field and between the different samples in each hemisphere . further complications arise from the constraints on placing 2df fibres in close proximity , mentioned in section 3.3 . a more useful statistic for comparison with other surveys is the effective area of the survey , defined as the total number of targets with reliable redshifts divided by their mean density in the input catalogues . this gives an effective area of approximately 135 @xmath1 for the sample 8 targets and 90 @xmath1 for sample 9 . a more careful calculation by @xcite , using the detailed survey mask of @xcite , revises these areas to 141.7 @xmath1 and 93.5 @xmath1 for samples 8 and 9 respectively . these figures imply that the overall completeness of the primary sample is only about 75% , falling to 50% for the secondary sample . however , as noted in section 5.6 , the completeness of the primary sample rises to @xmath138% in the best - observed regions . a histogram of the redshifts of all the 2slaq galaxies is shown in fig . [ zhist ] . it is obvious that the sdss colour and magnitude selection criteria produce a very clean sample of high redshift galaxies . almost all the targets have @xmath104 , apart from @xmath145% contamination by foreground m - stars . the dotted line is for the entire set of 2slaq lrgs , including the early samples with somewhat different selection criteria which contribute most of the low - redshift tail . the solid line is for the primary sample which has median redshift 0.55 and 80% of the redshifts within @xmath17 , as desired . the sharp lower cut - off is set by the colour selection while the declining tail for @xmath156 is due to the @xmath157 magnitude limit . the secondary ( sample 9 ) galaxies , shown by a dashed line , have median redshift 0.47 . represents the 5% contamination by foreground m - stars . the effectiveness of the 2slaq photometric selection for @xmath158 galaxies is evident.,width=317 ] fig . [ zhistcomp ] compares the redshift distribution of the 2slaq lrgs with that of the sdss lrgs @xcite and the 2dfgrs galaxies of all types @xcite , in each case counting only the galaxies lying within the same @xmath15 wide northern equatorial strip . the three surveys are complementary and together give good redshift coverage from @xmath159 to @xmath33 . there is significant overlap between the 2slaq and sdss lrg samples at @xmath160 but 2slaq provides a much higher space density of galaxies with @xmath161 than any other current survey , as is needed for mapping the 3d structure . -wide northern equatorial strip . the 2slaq lrg redshift distribution ( shaded histogram ) is compared with that of the sdss lrg survey ( dotted line ) and galaxies of all types in the 2dfgrs ( dashed line).,width=317 ] a plot of redshift against sdss @xmath50 magnitude ( fig . [ zi89 ] ) illustrates clearly the importance of a faint @xmath22band limiting magnitude in selecting high @xmath125 lrgs . the number of galaxies increases with increasing @xmath22 magnitude at all redshifts . at the highest redshifts only a small number of intrinsically very luminous galaxies are observed , while at lower redshifts the sample includes early - type galaxies with luminosity @xmath54 , somewhat fainter than the ` classical ' definition of an lrg as having @xmath162 @xcite . the sharp lower boundary near @xmath147 demonstrates again the high efficiency of the sdss photometric selection criteria . -band magnitude for 11,000 2slaq galaxies ( samples 8 and 9 ) . the upper diagonal boundary represents the most luminous galaxies at each redshift ; the lower boundary at @xmath163 is set by the colour selection criteria.,width=359 ] the colour distribution of the targets varies strongly as a function of redshift . this is illustrated in fig . [ 2col8multi ] for the primary ( sample 8) data . each panel is for a different redshift bin in steps of 0.05 in @xmath125 , apart from the low and high-@xmath125 tails . _ versus _ @xmath164 diagram as a function of @xmath125 , for the primary ( sample 8) objects . each panel is for a successively larger redshift range , starting with the m - type stars at top left and ending with the highest redshift objects at bottom right . the dotted line in each panel is the @xmath165 lower boundary to sample 8 , for reference ( a few points appear below this line due to revisions of the photometry ) . the main clump of lrgs moves to redder @xmath20 colours at approximately constant @xmath21 as @xmath125 increases , as the 4000 break moves through the @xmath31band.,width=317 ] the top left panel shows the m - stars , which mostly occupy a vertical strip with @xmath166 and extending to very red @xmath20 colours . for the galaxies in the remaining seven panels , the centroid of the main concentration of points moves to higher values of @xmath167 at approximately constant @xmath164 as redshift increases , due to the 4000 break moving through the @xmath31band . for @xmath168 the main concentration of lrgs moves clear of the colour selection boundaries , indicating that the primary lrg sample has high completeness above the @xmath22band apparent magnitude limit . careful analysis of the numbers of sdss and 2slaq lrgs @xcite shows they have almost identical luminosity functions at @xmath169 and @xmath170 , indicating that most lrgs were formed at higher redshift and are evolving passively . this conclusion is confirmed by the composite lrg spectra for different redshift ranges presented by @xcite : these are all very similar to the spectrum of a nearby giant elliptical galaxy such as ngc 3379 ( template 2 , fig . [ tempspec ] ) and show only slight evolution for @xmath171 . the 2-colour distribution in fig . [ 2col8multi ] becomes more diffuse as @xmath125 increases , which may be an indication of increasing levels of current or recent star formation affecting the colours at higher redshifts . however , the errors in the colours also become significant for the faintest and reddest objects , especially in the @xmath43-band . objects with @xmath172 have not been plotted , following the discussion of photometric errors in section 2.1 . such a cut eliminates 3% of the primary sample , including virtually all objects with @xmath42 . for @xmath168 the colour distribution appears to be bi - modal , with a clump centred near @xmath173 and a second population towards the lower left corner of each plot , truncated on the blue side by the @xmath174 diagonal limit ( cf . [ 2colsel ] ) . this bimodality is shown more clearly in fig . [ cparhist ] , a histogram of the distribution of @xmath41 for galaxies in the two highest redshift bins from fig . [ 2col8multi ] . further evidence for the reality of this division comes from the distribution of galaxies with strong [ o ii ] emission , as detected during the redshifting process and indicated by the dotted histogram in fig . [ cparhist ] . the em - line galaxies are strongly concentrated towards the blue boundary . the equivalent widths of the emission lines are not large enough to cause the colour shift directly in true lrgs : it appears most likely that the ` bluer ' objects represent later - type galaxies with recent star formation , spilling into the 2slaq sample from below the colour selection boundaries . @xcite and @xcite discuss this effect in some detail , since it has to be corrected for in determining the luminosity functions of the lrgs and in measuring the evolution in the rate of star formation with redshift . the fraction of emission line galaxies becomes particularly high for the highest redshift group , with @xmath155 . this is evidence for higher rates of activity ( either star - forming or agn ) at earlier epochs , although the statistics have to be corrected for observational selection since the useful spectral range is short and affected by telluric features ( see fig . [ spectra](e ) ) . sometimes such galaxies can only be assigned reliable redshifts when [ o ii ] 3727is present , so there is a bias towards recognising the more active galaxies at high redshift . photometric parameter for galaxies in the primary lrg sample with @xmath175 and @xmath176 . the full line is for all lrgs , the dotted line is for those with [ o ii ] emission . the sharp boundary at @xmath177 marks the lrg colour selection cut - off . the bins are independent , supporting the reality of the dip in the full distribution at @xmath178 . the highly skewed distribution of emission line galaxies indicates the existence of two separate populations of high redshift galaxies , with only the redder peak corresponding to true lrgs.,width=317 ] the template statistics given in appendix b support the identification of two types of galaxy among the highest redshift 2slaq lrgs . the main ( redder ) concentration of galaxies nearly all fit the composite lrg spectrum ( t1 ) best and must be true lrgs , while many of the bluer galaxies lying close to the @xmath41 boundary are best fit by the later - type emission - line galaxy ngc 5248 ( t3 ) ( see fig . [ tzhist ] ) . the appearance of emission - line or star forming galaxies along the @xmath41 boundary is evident in the lower redshift bins as well , at least down to @xmath161 , which probably indicates that the two basic populations occur throughout the present sample as in other galaxy redshift surveys . this population is a contaminant in the lrg sample , and has to be distinguished from the subset of true lrgs which have emission lines due to either transient star formation or agn activity , as discussed by @xcite . the two - colour distribution of the primary targets which were observed but failed to yield a reliable redshift is shown in fig . this distribution is not the same as for the full input sample ( fig . [ 2colsel ] ) or any of the redshift subsamples in fig . [ 2col8multi ] . evidently the redshift failure rate is a complex function of redshift itself , and of other parameters such as magnitude and colour . [ noz ] suggests that a considerable fraction of the ` failed ' objects are high redshift galaxies with @xmath179 , while there is also a significant population near the @xmath180 cut - off with colours similar to @xmath181 galaxies . there are few failed redshift targets within the main clump of @xmath38 lrgs seen in the central panels of fig . [ 2col8multi ] , despite these being the dominant population in the sample , indicating that the identification of the true passive lrgs is very nearly complete . ( again omitting data with @xmath172 ) . this distribution is significantly different from that of the full sample or any of the redshift sub - samples plotted in fig . [ 2col8multi ] . a relatively high proportion of objects with extreme colours fail to give good redshifts , while _ bona fide _ lrgs in the clump with @xmath38 are under - represented . the ` failed redshift ' objects include a mixture of high redshift galaxies , galaxies with peculiar spectra , composite objects , lower redshift galaxies which are not lrgs , and lrgs which simply had weak spectra because of bad fibre placement or low throughput.,width=317 ] the overall distribution of the failed redshifts shows no significant dependence on apparent @xmath22-band magnitude or on the general quality of the datasets , as measured by redshift completeness or mean s / n ( section 5.4 ) . quantitatively , the fraction of failed redshift objects with colours within the main concentration of lrgs ( taken here as the objects with @xmath182 ) is only 5% compared with 8% for all objects in sample 8 . by contrast , 12% of the reddest objects with @xmath183 fail to yield reliable redshifts and this fraction rises to 17% for the bluest objects with @xmath184 . these variations demonstrate that there are real correlated differences in colour and spectral type within the primary sample : they can not simply be due to either poor spectra or photometric errors . many of the reddest and bluest objects must have spectra which are not so well matched to any of the template spectra as are the bulk of the lrgs . some may be active galaxies of various types , some may be composite spectra due to chance alignments . the ` failed redshift ' list may also include some very high redshift objects with @xmath185 , given that such objects are difficult to recognise in the 2slaq data . a few spectra with high s / n also fail to yield redshifts ; in at least one case this was due to contamination of the spectrum by light from a nearby bright star with an almost featureless spectrum in the 2slaq wavelength range . the discussions of the colour distributions above , of sample completeness in section 5.5 and of redshift reliability in section 6 , together indicate that the 2slaq sample of true passively evolving lrgs is over 95% spectroscopically complete and reliable for redshifts between 0.5 and 0.65 . few lrgs can have been missed , and few non - lrgs included , down to the @xmath186 magnitude limit . the primary sample 8 is also almost spatially complete in the most fully observed regions , assuming that the sdss input catalogue is itself almost complete , so that it should be possible to define a sample of lrgs with an absolute completeness of better than 90% . the primary sample 8 2slaq lrgs therefore provide a good sample of co - moving test particles for cosmological purposes and for constraining lrg evolution . in particular , most of the lrgs seen at redshift 0.65 must be the progenitors of lrgs at @xmath187 , and of the passively evolving lrgs seen at lower redshift in other surveys . the secondary sample 9 2slaq lrgs are much less useful for studying large scale structure , since their spatial completeness is very variable from field to field . however , this sample has a much tighter colour distribution than sample 8 ( fig . [ 2colsel ] ) and so comprises a more homogeneous set of galaxies , with high redshift completeness and little contamination , which will be very useful for studying the evolution of lrgs . a map of the spatial distribution of the lrgs within the more complete northern slice is shown in fig . [ pw_wedge ] , a plot of redshift ( or co - moving radial distance , assuming @xmath188 , @xmath189 and @xmath190kms@xmath96mpc@xmath96 ) against ra along the equatorial strip . the 2slaq lrgs are plotted as large black dots and lie in discrete sectors corresponding to the sub - strips with the best sdss photometry . the distribution appears non - random with higher density clumps and filaments and lower density voids . the thin pencil - beam extensions down to @xmath12 come from the 2df fields observed early in 2003 , using lower - redshift selection criteria . the grey points with @xmath5 are the sdss lrgs which lie within the same @xmath15-wide equatorial strip @xcite and the fine structure with @xmath191 comes from a combination of the sdss main @xcite and 2dfgrs @xcite galaxies of all types , also within the same narrow strip . the apparently finer structure seen at low redshifts is mainly an artefact of the much higher surface density of points available . this wedge plot gives an excellent impression of how the four surveys complement each other in terms of redshift and spatial coverage . the redshift histogram shown previously as fig . [ zhistcomp ] is effectively a projection of fig . [ pw_wedge ] along the redshift axis . the large - scale structure in the 2slaq lrgs can be seen more clearly at higher magnification . [ wedge_d89 ] is an enlargement of part of fig . [ pw_wedge ] corresponding to sub - strip ` d ' , one of the largest areas with contiguous coverage . a pattern of voids and super - clusters linked by filaments or walls becomes apparent , on the same @xmath192 mpc scales as seen at @xmath193 in surveys such as the 2dfgrs @xcite and sdss main survey @xcite . for sub - strip ` d ' at ra@xmath194h . this shows the spatial distribution of 1737 lrgs in the primary and secondary 2slaq samples ( samples 8 and 9 ) with @xmath195 . a pattern of voids and super - clusters linked by filaments or walls is apparent , on scales of hundreds of mpc.,width=317 ] the primary objective of the 2slaq lrg survey , to obtain redshifts for more than 10,000 galaxies with @xmath17 , has been achieved . the galaxies lie in two thin sectors or wedges , one in each galactic hemisphere , corresponding to narrow strips running along the celestial equator . the input target selection was based on sdss @xmath23 photometry and has proved to be very efficient with more than 90% of the galaxies falling in the desired high redshift range . more than two thirds of the galaxies are in the northern slice . reliable redshifts have been obtained for 92% of the objects observed . a large - scale structure of clumps , filaments and voids is apparent in the most complete regions , similar to what was seen in earlier large surveys at lower redshift . two thirds of the galaxies are in the photometrically well - defined primary sample and most of the rest ( 27% ) in the secondary sample . the primary sample has high spectroscopic completeness in the sense that good redshifts have been obtained for @xmath138% of the targets within the best - observed fields . the survey also has high absolute completeness for lrgs with @xmath196 and de - reddened total @xmath22-band magnitude brighter than 19.6 , since such galaxies lie well clear of the photometric selection boundaries . the sdss-2df lrg survey thus offers a significant advance in our understanding of the most massive galaxies over the past 8 gyr : the census of such objects must be almost complete . it should be possible to distinguish unambiguously between continuing hierarchical ( i.e. merger ) and early ( high redshift ) models for the formation of lrgs . for the immediate future , the success of 2slaq shows that the new aaomega spectrograph for 2df , which gives a gain of a factor of up to four in throughput together with better quality and higher resolution , can be used for cosmological surveys targetting either higher redshifts or much larger samples , or both . we particularly thank all members of the aao staff who helped to run and maintain 2df , and its complex software , during the course of the 2slaq survey . the 2slaq surveys were conducted by a substantial team of people , coordinated by rcn with ace and mjd as principal investigators for the lrgs in the uk and australia respectively . rdc wishes to acknowledge their crucial roles , and can claim credit only for contributing to the observing and data analysis . he also thanks will sutherland and will saunders for helpful discussions about the zcode redshifting software . ace acknowledges support through the royal society . kap acknowledges support through an epsa university of queensland research fellowship and a uqrsf grant . rdc is grateful for the hospitality of the department of astrophysics at oxford , where the final version of this paper was completed . funding for the creation and distribution of the sdss archive has been provided by the alfred p. sloan foundation , the participating institutions , nasa , nsf , the us department of energy , the japanese monbukagakusho , and the max planck society . the sdss web site is ` http://www.sdss.org ` . the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the university of chicago , fermilab , the institute for advanced study , the japan participation group , the johns hopkins university , los alamos national laboratory , the max - planck - institute for astronomy , the max - planck - institute for astrophysics , new mexico state university , the university of pittsburgh , princeton university , the united states naval observatory , and the university of washington . abazajian , k. , et al . , 2003 , aj , 126 , 2081 adelman - mccarthy , j. k. , et al . , 2006 , apjsupp , 162 , 38 bailey , j. a. , heald , r. , croom , s. m. , 2004 , the 2dfdr data reduction system users manual , aao bell r.a . , gustafsson b. , 1978 , a&as , 34 , 229 blake , c. , collister , a. , bridle , s. , lahav , o. , 2006 , mnras , in press (= _ astro - ph_/0605303 ) bolton , a. s. , et al . , 2004 , aj , 127 , 1860 colless m. m. , dalton g. b. , maddox s. j. , et al . , 2001 , mnras , 328 , 1039 collister a. , lahav o. , blake c. , et al . , 2006 , submitted (= astro - ph_/06nnnnn ) croom , s. m. , smith , r. j. , boyle , b. j. , shanks , t. , miller , l. , outram , p. j. , loaring , n. s. , 2004 , mnras , 349 , 1397 croom , s. m. , et al . , 2006 , in preparation driver , s. p. , liske , j. , cross , n. j. g. , de propris , r. , allen , p. d. , 2005 , mnras , 360 , 81 eisenstein d. j. , annis j. , gunn j. e. , szalay a. s. , connolly a. j. , nichol r. c. , et al . , 2001 , aj , 122 , 2267 eisenstein d. j. , et al . , 2003 , apj , 585 , 694 fukugita , m. , et al . , 1996 , aj , 111 , 1748 heavens , a. , panter , b. , jimenez , r. , dunlop , j. , 2004 , nature , 428 , 2474 kauffmann , g. , et al . , 2003 , mnras , 341 , 54 lewis i. j. , cannon r. d. , taylor k. , et al . , 2002 , mnras , 333 , 279 liske , j. , lemon , d. j. , driver , s. p. , cross , n. j. g. , couch , w. j. , 2003 , mnras , 344 , 307 padmanabhan , n. , et al . , 2005 , mnras , 359 , 237 pandey b. , bharadwaj s. , 2005 , mnras , 357 , 1068 pimbblet k. a. , drinkwater m. j. , hawkrigg m. c. , 2004 , mnras , 354 , l61 richards g. t. , et al . , 2005 , mnras , 360 , 839 roseboom i. et al . , 2006 , mnras submitted ross , n. , et al . , 2006 , mnras in preparation sadler , e. m. , et al . , 2006 , mnras in preparation schlegel , d. j. , finkbeiner , d.p . , & davis , m. , 1998 , apj , 500 , 525 spergel , d. n. , et al . , 2003 , ap j suppl , 148 , 175 stoughton , c. , et al . , 2002 , aj , 123 , 485 strauss m. a. , weinberg d. h. , lupton r. h. , narayanan v. k. , et al . , 2002 , aj , 124 , 1810 sutherland w. j. , 1999 , the 2df redshift code - user guide ( available via the aao ) tonry j. , davis m. , 1979 , aj , 84 , 1511 wake , d. , et al . , 2006 , mnras submitted (= _ astro - ph_/06nnnnn ) york , d. , et al . , 2000 , aj , 120 , 1579 data were obtained for 80 discrete 2df fields during the 2slaq survey , some with two or more sets of observations . in general , data obtained in different observing runs have been treated as independent datasets , since the target lists were usually modified between runs . a total of 104 datasets are listed in table [ fldlist ] . the first seven columns give the name and centre of each field ( equinox j2000 ) . column 8 identifies each discrete dataset by field name and date , usually in the format yymmxx : these are the datasets identified in the final redshift catalogues for individual objects ( the ` g ' distinguishes galaxies from qsos , while the final ` 2 ' denotes the spectrograph , in practice always no . 2 for the lrgs ) . most fields were observed on two or more nights during an observing run and all the data combined , in which case the final two characters are either ` 00 ' or ` fi ' . occasionally data were obtained on only one night , in which case the last two digits give the date . a few specially combined datasets are identified by other character combinations . where the same field was observed in different runs the datasets are listed separately since the target lists will be different , although often with many objects in common . column 9 gives the number of exposures combined in each dataset , @xmath122 , normally each of 30 min duration . column 10 gives the mean s / n achieved , averaged over all pixels for all targets , and @xmath123 in column 11 is the mean offset in magnitudes in a plot of log(counts ) versus @xmath31-band fibre magnitude for each dataset ( total @xmath50 magnitude was used instead of @xmath197 for some early datasets : these account for most of the negative values of dmag ) . both of these are indicators of quality . the final three columns give the number of objects in each dataset , the number for which reliable ( @xmath141 ) redshifts were obtained and the proportion these are of the total . the observations of a field were normally deemed to be complete when the proportion of reliable redshifts was at least 85% ( strictly , this criterion was applied to the original set of targets if some fibres were re - allocated after the first night , but the numbers quoted here apply to the full datasets ) . the footnotes give further details on some datasets . the third column of table [ templist ] lists small zero - point velocity offsets which have been derived in two ways : from the mean differences for a set of lrgs , all cross - correlated with each template , and by cross - correlating each template with a synthetic k - type stellar spectrum generated using the _ ssg _ code of @xcite . the two techniques gave consistent results to within @xmath198kms@xmath96 , a small error compared with the accuracy of @xmath192kms@xmath96 for individual galaxy redshifts . these relative velocities are all tied to the composite lrg template t1 , whose 85kms@xmath96 velocity is the difference between the vacuum wavelengths used by the sdss and the air wavelengths used by 2dfdr . the effects on the final redshifts are of course small , amounting to no more than 0.0003 . the fourth column of table [ templist ] is a redshift offset applied to some templates , to optimise the interactive display for most of the targets . the four galaxies were shifted to @xmath199 , the mean redshift of the lrgs , while the stars were left at @xmath200 . the derived redshifts were not affected significantly by these offsets . in practice the only stars which contaminate the lrg sample are m - type dwarfs ( 5% of the targets ) , but the g and k - type stellar spectra often provide useful confirmation of the galaxy redshifts , especially for @xmath201 . the final column in table [ templist ] gives the percentage of galaxies for which each template was used . the template which gives the best redshift estimate for each galaxy is a function of galaxy type . [ tzhist ] shows the redshift distribution for most of the templates . three of the templates , t4 , t7 and t8 for a- , k- and m - type stars respectively , have been omitted because they rarely if ever gave the best fit to a galaxy . not surprisingly , the mean sdss lrg spectrum ( t1 ) provides the best fit to by far the largest fraction of the targets , about 60% of the total . the ngc 3379 ( t2 ) spectrum , which is very similar but without the uv tail , was used for a further 16% : for most objects these two templates give virtually identical results . thus at least three quarters of the 2slaq sample matches very well the spectrum of a lower redshift quiescent elliptical galaxy or lrg . note that the ordinate scales in fig . [ tzhist ] are different for these two templates . the other two galaxy templates , the emission - line s0 galaxy ngc 5248 ( t3 ) and a ` k+a+em ' composite ( t9 ) , are each used for 8% of the sample . these include a higher proportion of active galaxies , either with active galactic nuclei ( agns ) or star - forming ( sf ) ( these two classes can not be easily distinguished in the 2slaq spectra , since often the only clue is the presence of [ o ii ] at 3727 ) , and they tend to be at higher redshifts . note that the emission lines in t3 and t9 are blanked out for the cross - correlations and do not contribute to the redshift determinations . for the great majority of galaxies with reliable redshifts , all four galaxy templates agree . the f- , g- and k - type stellar templates together give the best fit to about 6% of the lrgs , although again the galaxy templates usually give the same redshifts . there is a tendency for the stars to be used for the lowest redshift galaxies with @xmath202 , perhaps because such galaxies are not true lrgs . the upper cut - off at @xmath148 for t5 and t6 is an artefact of the zcode . the a - type stellar spectrum gives an apparent best fit in only @xmath203% of cases and is often clearly wrong when the fit is checked visually . the m - type spectrum is used only for foreground m - dwarfs .